Final Peer Review Report

Evaluation of OSHA’s Disposition of 2009 Reviewer Comments

and 

Reviewer Responses to Oral and Written Testimony Presented at OSHA’s
Informal Hearings on the Proposed Rule on Respirable Crystalline Silica,
March 2014

Contract No.  DOLJ129F32859

Task Order 16

Submitted to:

David O’Connor

Directorate of Standards and Guidance

Occupational Safety and Health Administration

200 Constitution Avenue, NW

Washington, DC  20210

Submitted by:

Eastern Research Group, Inc.

110 Hartwell Avenue

Lexington, MA  02421-3136

April 7, 2014TABLE OF CONTENTS

Introduction	1

  HYPERLINK \l "_Toc251578075"  Bruce Allen, Bruce Allen Consulting,
Chapel Hill, NC 	3

  HYPERLINK \l "_Toc251578076"  Kenneth Crump, Louisiana Tech University
Foundation, Inc., Ruston, LA	 15

  HYPERLINK \l "_Toc251578078"  Gary Ginsberg, Connecticut Department of
Public Health, Hartford, CT 	19

  HYPERLINK \l "_Toc251578079"  Brian Miller, IOM Consulting Ltd.,
Scotland, UK 	27

  HYPERLINK \l "_Toc251578080"  Andrew Salmon, Private Consultant,
Lafayette, CA 	33

Introduction

Introduction 

In 2009, under contract to OSHA, Eastern Research Group, Inc. (ERG)
coordinated an independent, scientific peer review of two draft
documents prepared by OSHA: 1) Preliminary Health Effects Section for
Silica; and 2) Preliminary Quantitative Risk Assessment for Silica.
These documents were developed by OSHA to provide the Agency with
quantitative information on these risks to support development of a
proposed rule governing exposure to crystalline silica in general
industry, construction, and maritime.  

2009 Peer Review Process

ERG conducted a search for nationally recognized experts in occupational
epidemiology, biostatistics and risk assessment, animal and cellular
toxicology, and occupational medicine who had no conflict of interest
(COI) or apparent bias in performing the review. Interested candidates
submitted evidence of their qualifications and responded to detailed COI
questions. ERG also searched the Internet to determine whether qualified
candidates had made public statements or declared a particular bias
regarding silica regulation. From the pool of qualified candidates, ERG
selected seven to conduct the review.

Reviewer	Affiliation	Reviewed Health Effects Section	Reviewed
Quantitative Risk Assessment

Bruce Allen	Bruce Allen Consulting	(	(

Kenneth Crump	Louisiana Tech University Foundation

(

Murray Finkelstein	McMaster University, Ontario	(	(

Gary Ginsberg	Connecticut Department of Public Health	(	(

Brian Miller	IOM Consulting Ltd., Scotland	(	(

Andrew Salmon	Private Consultant	(	(

Noah Seixas	University of Washington, Seattle	(	(



Reviewers were provided with a letter of instruction, the charge, review
documents, and hundreds of relevant references to conduct the review.
The experts submitted their individual written reviews, which were
submitted as received.  ERG also compiled the comments by charge
question in preparation for a follow-up conference call.  ERG
facilitated the call with the reviewers and OSHA, during which reviewers
provided clarifications and  answered any questions OSHA had regarding
their comments.  After the conference call, reviewers submitted their
final peer review comments which were compiled as submitted into ERG’s
final peer review report.

At the close of the peer review, ERG informed the experts that OSHA
would be addressing and incorporating comments received from the peer
review and the public, with the intent of proposing the rule in the fall
of 2009.  At that time, OSHA would hold informal public hearings on the
proposed rule on respirable crystalline silica.  Through its OSHA
contract with ERG, the peer reviewers would be invited to attend the
hearings in Washington, DC, to listen to oral testimony as it applied to
the health effects and quantitative risk assessment sections, read
additional written comments, and provide any additional comments on
scientific issues presented in written or oral testimony at during the
public hearings.  In addition, the reviewers would be asked to read the
revised health effects and quantitative risk assessment sections and the
summary of these sections in the proposed rule, and provide comments to
OSHA on whether they felt that OSHA had or had not adequately addressed
and/or considered their peer review comments when revising the rule, and
to reiterate any concerns or issues that they felt were still
outstanding and should be addressed as OSHA moves forward to finalize
the rule.   

2014 Completion of Peer Review

In 2014, under contract to OSHA, ERG completed the last phase of the
peer review of OSHA’s Preliminary Health Effects and Preliminary
Quantitative Risk Assessment sections for Silica, which entailed the
tasks described above. 

Of the seven original peer reviewers, five (5) were available to conduct
the pre-hearing work and attend the two days of public hearings in
Washington, DC, on March 19-20, 2014.  The five reviewers included Bruce
Allen, Kenneth Crump, Gary Ginsberg, Brian Miller, and Andrew Salmon. 
ERG secured their services and sent them a letter of instruction with
guidance on accessing the written comments in the docket, the revised
sections of the health effects and quantitative risk assessment sections
in the rule, and the summary of same, copies of the oral testifiers’
written comments, and their original peer review comments in preparation
for attending the public hearings and completing the last step of the
peer review.  

This report provides the peer reviewers’ comments on the disposition
of their original peer review comments in the proposed rule, as well as
comments peer reviewers provided in response to written and oral
testimony presented at OSHA’s Informal Public Hearings on the Proposed
Rule on Occupation Exposure to Respirable Crystalline Silica on March 19
and 20, 2014, held at the Department of Labor, Washington, DC.  The
appendix consists of the original peer review report from 2009.



Bruce Allen

Bruce Allen Consulting, Chapel Hill, NCFrom: Bruce Allen 

Sent: Wednesday, April 02, 2014 7:43 PM

To: Kate Schalk

Subject: RE: OSHA Silica - Final comments

Attachments: Comments on OSHA Docket-2010-0034_Allen.docx

Kate:

My comments are attached.

I have reviewed the original review materials, my 2009 peer review
comments and their disposition by OSHA in the current proposed
rulemaking. I feel that OSHA has adequately addressed my comments in
their updated Health Effects and QRA sections. I have included my
comments on these areas in my attached submission. I have also read the
written testimony of the public commenters and heard their oral
testimony. I have included comments on some of the commenters’
testimony as it applies to the QRA.

Bruce Allen

Comments on Docket OSHA-2010-0034

Submitted by Bruce Allen

April 2, 2014

Response to Peer-Review Comments (2009)

I have read OSHA’s responses to the peer-review comments that were
submitted in 2009.  I am satisfied with the manner in which OSHA
responded to those comments, and particularly to the ones that I
submitted.  I continue to believe that the low-exposure studies by Brown
and Rushton (2005a, 2005b) and Ulm et al. (1999) can be useful, but
probably mostly if a formal meta-analysis were to be conducted.  I am
satisfied that, for the analyses that have been done and reported in the
QRA section, the exclusion of those studies is acceptable.

Observations and Responses Related to Written and Oral Testimony

The following comments are provided in response to the public comments
that were submitted to the docket and which were espoused in oral
testimony on March 19-20, 2014.  The following comments are organized by
discussion point rather than specific written or oral testimony.  Many
of the discussion points were raised by various commenters; I have
attempted to cite as many of them as possible. 

Declining Silicosis Mortality

The submission, 01_OSHA-2010-0034-DRAFT-3708_Borak.pdf, has a graph of
declining silicosis mortality (Figure 1, and see below).  The written
testimony makes it clear that there are two possible contributors to
that decline, reduced exposure and declining number of exposed
individuals.  However, during oral testimony, no mention was made of the
decline in the exposed population; the graphic was proposed to show the
effect of reduced exposure levels associated with “historic
implementation” of OSHA and MSHA PELs.

The fact that this graph in no way adjusts for the declining employment
in jobs with silica exposures makes its interpretation problematic.  To
emphasize the contribution of historical declines in exposure as the
underlying cause is spurious; no information is given to allow one to
account for declining employment.  Moreover, the scale for death rate
(right-hand vertical axis) is misleading insofar as it falsely gives the
impression that the rate of silicosis deaths among exposed workers has
declined to about 1 per million.  That rate is for the entire US
population, unexposed predominantly, not for the workers who are the
population of concern with respect to OSHA regulations.



Figure 1: Silicosis: Number of deaths, crude and age-adjusted death
rates, US residents age 15 and over, 1968-2004 (Source: (2))

Thresholds

The issue of thresholds in dose-response modeling and risk assessment is
a contentious one.  On the one hand, biologists and toxicologists often
assume that thresholds for toxic responses exist, based on arguments
that “a single molecule” of a substance would not be enough to
trigger such responses.  On the other hand, it is essentially impossible
to distinguish between dose-response patterns that represent a threshold
and those that do not.  Any observations that are consistent with no
dose-related change in response are also consistent with a slight,
nonzero dose-related change.

Several written and oral testimonies discussed the potential existence
and effect of thresholds for either silicosis or lung cancer (see
03_OSHA-2010-0034-2230_Long-Valberg.pdf,
08_OSHA-2010-0034-2307_attachment6_Cox.pdf, and associated oral
testimony from Drs. Valberg and Cox, among others).  The common theme of
these testimonies was that there is a threshold for silica-related
health effects and that the modeling by OSHA, which did not include a
threshold parameter, was therefore flawed.  

During the oral testimony, when asked about the assumption that a
proposed threshold was constant across individuals, several of the
presenters acknowledged that such a threshold would vary in a
population; even if the concept of a threshold is accepted, then not all
individuals would have the same threshold for response.  When
considering a diverse population with variations in exposures to various
substances, physiological states, and genetics, it makes little sense to
suppose that every individual would be identical with respect to a
threshold exposure level.  

Yet, in the written testimony and throughout the oral testimony, the
presenters continued to refer to “a threshold” as if such a
non-varying entity existed (see, for example,
03_OSHA-2010-0034-2230_Long-Valberg.pdf, Figure 2).  The claim was made
that models without a threshold could not “reveal threshold
concentrations” (03_OSHA-2010-0034-2230_Long-Valberg.pdf, p. 20).  

My following response suggests that the critiques in those testimonies
are misguided at best.  Let us take silicosis morbidity as an example. 
And let us assume that the dose-response relationship, for individual j
in a population, is given by

			P(d) 	= 0				if d < th(j)

				= 1-exp(-beta*(d-th(j)))		if d > th(j)			Eq. 1

where P(d) is the lifetime probability of silicosis morbidity, d is the
exposure level, th(j) is the exposure threshold for that individual, and
beta is a parameter to be estimated.  Once the idea of a varying
threshold value (individual th(j) values, for all individuals in the
population), then one must consider a distribution of such thresholds. 
For this example we assume the distribution is lognormal with some
specified mean and variance.  One could even assume that there is some
absolute value above which all the individual thresholds must fall
(analogous to frequently heard arguments that “a single molecule” of
a substance would not cause a toxic response in any individual).  Thus,
the distribution of thresholds could be characterized as

			[th(j) – α] ~ LN(μ, σ2)  

where α is that absolute minimum for any individual threshold, and
LN(μ, σ2) represents a lognormal distribution with mean μ and
variance σ2.  From these assumptions, we can derive what the population
level (frequency of response) pattern would be.

To complete this example, let us set some specific values for the
distribution of thresholds.  Let us make this correspond to the claims
made by various presenters that there was no evidence of increased
response (and therefore a “threshold”) at 100 μg/m3.  Since we must
admit varying thresholds, let us make the mean of the distribution equal
to 100.  If we let α= 5 (arbitrary small value) then the mean of th(j)
is actually 105.  Let the GSD be 1.5 (only 50% variation in the
threshold distribution, which corresponds to a log-scale variance of
0.16).  The resulting distribution of thresholds is shown here:

Note that the individual thresholds can be quite large (above 100
μg/m3) but all are greater than or equal to 5 μg/m3.  

To complete the example, let the estimate of beta be such that, at 500
μg/m3, the risk of silicosis morbidity is the smallest of the values
presented in Table VI-2 of the Proposed Rules published in the Federal
Register (Vol. 58, No. 177, p. 56321), which is 626 per 1000 workers. 
That value is chosen because the presenters expressed no concerns about
estimates of risk for such high exposure levels, and because it is the
lowest (leading to least protective) value for silicosis morbidity
presented by OSHA for that exposure level.  Given the values of the
other parameters, a beta value of 0.0025 (per μg/m3) achieves that
match.  The following graph shows the resulting population-level dose
response curve:

If we focus on the lower end of the dose and response curve, we observe
the following:

Even with a mean threshold at 100 μg/m3, the predicted response rate at
50 μg/m3 is 0.6 per 1000.  The 1 per thousand risk level is exceeded at
54 μg/m3.  The expected frequency of response at 100 μg/m3 is 32 per
1000 workers.

The major points illustrated by this example are as follows:

It makes no sense to discuss a single threshold value.

Given, then, that thresholds must be envisioned as a distribution in the
population, then there is substantial population-level risk even at the
mean threshold value, and unacceptably high risk levels at exposures far
below the mean threshold.

It is NOT, therefore, inappropriate to model the population-level
observations using a non-threshold model.  The population-level
dose-response curves shown in the previous two graphs do NOT have a
threshold (except at a meaninglessly small exposure level; here, for
example, at 5 μg/m3).  Since such small exposure levels are not
included in the epidemiological data used for modeling, such vanishingly
small thresholds can be ignored when fitting models to aggregated
dose-response data.  In fact, I would claim that it is inappropriate to
include ANY threshold models (i.e., those that assume a single threshold
value) when modeling epidemiological data.  A non-threshold model for
characterizing population dose-response behavior is theoretically and
practically the optimal approach.

One additional set of comments about thresholds and their detection is
warranted.  Several presenters claimed there was a clear “threshold”
evidenced in the dose-response pattern estimated by Steenland et al.
(2001) (their Figure 1, and Figure 3 in
03_OSHA-2010-0034-2230_Long-Valberg.pdf, p. 21).  It is well known that
presenting the x-axis (dose) on a log-scale enhances the visual
appearance of threshold-like behavior.  The presenters made no attempt
to go beyond visual examination, and therefore misconstrue the results.



Fallacious Analogies by Dr. Cox

The written and oral testimonies by Dr. Cox are replete with
misrepresentations of various aspects of OSHA’s risk assessment.  I
will make note of several of them here.  They reflect poorly on his
reliability as an expert suited to comment on, let alone conduct, a
meaningful risk assessment.

First, consider his use of the phone vs. CHD mortality plot (Figure 1 of
his written testimony).  While no one would suggest that per capita
phone ownership is causally linked to CHD mortality, the association is
still not without value.  The plot shows results for 15 countries.  If a
per capita phone data point was found for a 16th county, I contend that
it would still be appropriate to use the graph in his Figure 1 to
predict CHD mortality in that country (in the absence of better, causal
data).

His rejoinder is that if you changed the phone ownership in one of those
countries, you would not expect the CHD mortality to change in that
country.  True.  But that is not analogous to the situation faced by
workers exposed to silica.  The observational units in the phone example
are countries; the observational units in a silica exposed population
are the individual workers.  Unlike with the countries, where phone
ownership could be changed and the effect on the CHD mortality in that
country could be monitored, the workers exposed to silica cannot be
tracked, even with a change in exposure levels,  for changes from dead
to alive, or silicosis-diseased to disease-free.  Unfortunately for
them, the response variable is not “changeable” from the state of
having the adverse event (death or silicosis) to a state of being
without such an event.

In fact, that is the problem with Dr. Cox’s entire argument in favor
of causal analysis over standard dose-response modeling.  There is no
identifiable “change” that can be tracked in the observation units
following a change in the exposure conditions.  The effects of previous
exposures are not separable from those of the changed (presumably lower)
exposure conditions.  Therefore his entire criticism of the standard
dose-response approach is erroneous.  We are trying to predict the
effects of various exposure scenarios on hypothetical, future workers
who would spend their working lives under new exposure conditions.  The
best basis for doing that is to examine what has occurred in workers
exposed to historical levels, and use model-based extrapolations as
needed to predict what would happen under revised circumstances.  I.e.,
one is not predicting what would happen to the observational units if
exposure changed, but rather predict what would happen to new
observational units under different (lower) exposure conditions.

In the case of silica and its associated diseases, the extrapolation
mentioned above is not that far below what has occurred historically. 
Dr. Borak, in his oral testimony, acknowledged that so-called
“over-exposures” in the range of 120 to 480 μg/m3 – i.e., not
that much greater than the proposed PEL – do pose health hazards.  And
those acknowledged hazardous levels were more likely than not determined
to be hazardous based on durations of exposure less than, often much
less than, the assumed OSHA scenario of 45 years of exposure.  I saw no
recognition on the part of the Chamber-of-Commerce- or ACC-sponsored
presenters that the demonstrated hazards occurred usually with exposure
durations less than the rule-based 45 years.

Another fallacious analogy presented by Dr. Cox relates to Figure 2 of
his written testimony.  He states that linear modeling of absolute risk
gives spurious associations even when the relationship between the x and
y variables is random.  In his oral testimony, he further clarified that
that result was achieved by fixing the intercept to be zero
(constraining the model to pass through the origin).  The problem is
that this misrepresents the process of the risk assessments that OSHA
has done.

No modeling, linear or otherwise, for silica-related health effects has
been done on risk per se.  The modeling was performed with probability
of response or relative risk as the effect.  Neither of those effect
measures should be (or were, to my knowledge) constrained to pass
through the origin; it would be ridiculous to do so, especially for
relative risk, which should be at or around 1 for zero exposure.  It is
only by computation from more fundamental relationships that one
estimates absolute risk: from relative risk one computes lifetime
probability of response using life tables; from lifetime probability of
response one computes risk by subtracting the prediction for 0 exposure
from the probability for the dose of interest.  The fact that the
absolute risk curve passes through the origin is a (necessary)
by-product of those computations, regardless of assumptions about or
estimates of the background relative risk or probability of response.

Therefore, his claims about model misspecification (top of p. 21 of his
written testimony) simply do not apply.  The “linear absolute risk
model” he proposes as misspecified is a fiction.  It is therefore
inappropriate on that basis to claim, as Dr. Cox does, that this has
resulted in a false positive association between silica exposure and
silicosis or lung cancer.

Related to that mistake are claims that Dr. Cox makes about apparent
positive associations being more generally erroneous.  He cites an
article by Dr. Ioannidis and claims that it shows that most published
research findings are wrong.  An examination of that manuscript reveals
that under certain circumstances, the predictions of Dr. Ioannidis,
which are model-based rather than empirical, of the likelihood of a
correct finding can reach or exceed 80%.  Dr. Cox has done no analysis
to suggest where the silica studies fall with respect to the variables
used to estimate the proportion of correct results.  Apparently Dr. Cox
prefers to just use the spector of a false result as a scare tactic
rather than doing an investigation that would put the silica data base
in the correct context with respect to its predicted likelihood of a
true or a false result. 

One last issue relates to the claim made by that Dr. Cox that a Monte
Carlo analysis of exposure uncertainty is inappropriate and is just
repeating the same mistake many times over.  His claim is that only by
using an errors-in-variables approach can one correctly characterize the
effect of uncertainty in the exposure metrics.  In that approach, the
independent variable(s) are assume to have an error structure and this
is factored directly into the analyses and parameter estimates.  In
contrast, I contend that the Monte Carlo approach that OSHA has used is
analogous to a parametric bootstrap analysis.  That is, rather than
specifying the error structure for the independent variables and
analytically solving for model parameter estimates and corresponding
confidence intervals (a classical statistical approach),  the bootstrap
approach relies on sampling observations (in this case independent
variable values via Monte Carlo sampling) and characterizing the
uncertainty via the resulting distribution of estimates.  In both
approaches, a good representation of the structure of the variance
(uncertainty) is required.  But given that, I see no reason why the
Monte Carlo approach would be less satisfactory than the analytical
approach.  It may in fact be advantageous especially if analytical
solutions are difficult to obtain or require assumptions that are
inappropriate for the data under consideration.



  HYPERLINK \l "_Toc251578066"  Kenny Crump

Private Consultant, Ruston, LA  Final Peer Review Comments on OSHA’s
Health Effects Analysis and Quantitative Risk Assessment for Crystalline
Silica

Kenny Crump

March 2014

Comments on OSHA’s response to my comments made in 2009 on the earlier
draft of OSHA’s Quantitative Risk Assessment

I believe that my comments have been fairly taken into account in the
current draft and I have no further comments to make.

Comments based on listening to testimony presented during March 19-20,
2014 at the OSHA Hearings

OSHA, in its Preliminary Quantitative Risk Assessment (OSHA 2014),
stated that "available information cannot firmly establish a threshold
exposure for silica-related effects." At the OSHA Hearings “For the
Proposed Rule on Occupational Exposure to Respirable Crystalline Silica
Hearings” March 19 – April 4, 2014, a number of commenters
recommended that OSHA reevaluate this conclusion.  E.g., Dr. Jonathan
Borak stated “OSHA has failed to adequately consider the evidence of a
threshold response function for silicosis and silica-related lung
cancer” and Dr. Peter Valberg stated that “there is a need for OSHA
to carefully test whether the available epidemiologic data allow for the
identification of a threshold.”  

However, OSHA is on very solid ground in the statement that "available
information cannot firmly establish a threshold exposure for
silica-related effects."  In fact, in response to Dr. Valberg’s
suggestion, data (and particularly epidemiologic data) can never allow
one to conclude that a threshold has been identified, not just for
silica effects on lung cancer or silicosis but for any effect of any
contaminant.  Scientific data never permit the conclusion that a
threshold has been identified.  (Any data set is consistent with a
different response at a particular dose than the background response.) 
Thus, the hypothesis that a particular dose response does not have a
threshold is not falsifiable.   Similarly, the hypothesis that a
particular dose response does have a threshold is not falsifiable.  (If
a response is confirmed at a particular dose, there could always be a
threshold at a lower dose.).  The scientific method has been described
as posing falsifiable hypotheses and using data to test those
hypotheses.  Thus, accordingly, an attempt to distinguish between
threshold and non-threshold dose responses is not even a scientific
exercise (Crump 2011).  The best that can be done is to attempt to place
bounds on the amount of risk at particular exposures consistent with the
available data, which is what OSHA had done in their risk assessment.

Nevertheless, it has been claimed that some dose response analyses
appearing in the OSHA docket have statistically identified a threshold. 
To illustrate the fallacy of these claims I will consider an analysis
that is in the OSHA docket and has been published in the peer reviewed
literature.  From fitting a threshold dose response model to data from
an epidemiological study of quartz dust exposure and silicosis
incidence, Morfeld et al. (2013) concluded that their analysis
“indicates a respirable quartz dust exposure … concentration
threshold … possibly as high as 0.25 mg/m3,” which was the threshold
estimate obtained from the model that provided the smallest AIC out of
several alternative threshold models applied.  The Morfeld et al. (2013)
modelling approach was quite elegant and involved defining a threshold
in terms of concentration in air but assuming that silicosis incidence
was a function of the resulting (possibly lagged) cumulative exposure. 
At OSHA’s public hearing to gather information to assist in revising
the exposure standard, a number of presenters recommended that OSHA use
the Morfeld et al. results, or a similar threshold modelling exercise,
to set an occupational standard for quartz dust exposure (OSHA, 2014). 

 However, the Morfeld et al. threshold modeling effort suffers from the
same limitation that all such attempts statistically to estimate a
threshold must – the results are not reliable because the threshold
estimates so obtained are highly unstable.  In fact, it is easy to see
that there is always a dose response model that has no threshold which
will fit the underlying data essentially as well as a particular
threshold model.  For example, one can always simply add a linear
exposure term (or any one of many other non-threshold exposure terms) to
the threshold model, and by making the size of the linear term
sufficiently small (i.e., making the exposure coefficient sufficiently
small but greater than zero), make the modified model virtually
indistinguishable from the original model.  However the modified model
will have no threshold and consequently will predict a positive risk at
every exposure.  

To make this argument specific to the Morfeld et al. analysis, the
Morfeld threshold model for relative risk can be written as (using their
terminology)

  otherwise.  However, it is clear that the slightly modified model,

 , could even be larger than OSHA would deem allowable.  Consequently
the threshold identified by Morfeld et al. is model dependent in the
extreme and consequently should not be used by OSHA to set an exposure
standard.   Of course there is nothing necessary about the specific
modification I have made to the Morfeld et al. model as many other
modifications could be made that would lead to the same conclusion. 
Similarly there is nothing specific to the Morfeld et al. model in this
argument, as a similar conclusion could be reached for any other
threshold estimate obtained by fitting a threshold model to
dose-response data. 

References

Crump KS. 2011. Use of Threshold and Mode of Action in Risk Assessment.
Critical Reviews in Toxicology 41(8): 637–50.

Morfeld P, Mundt KA, Taeger D, Guldner K, Steinig O, Miller BG.  2013. 
Exposure and Silicosis Incidence among Workers in the German Porcelain
Industry, JOEM 55(9): 1027-1034.

OSHA (Occupational Safety and Health Administration) 2014.  Health
Effects Analysis and Quantitative Risk Assessment for Crystalline
Silica.  Docket OSHA-2012-0030, Room N-2625, U.S. Department of Labor,
200 Constitution Avenue, NW., Washington, DC 20210.



  HYPERLINK \l "_Toc251578067"  Gary Ginsberg 

Private Consultant, Hartford, CT 

Comments of Gary Ginsberg, Ph.D.

On The

Revised (2013) OSHA Silica Health Effects Review and Quantitative Risk
Assessment

Docket: OSHA 2010-0034

March 27, 2014

I have reviewed my comments from 2009 in relation to the original draft
Silica Health Effects and QRA documents prepared by OSHA and have
reviewed the OSHA response.  Further I have reviewed the updated OSHA
health effects and QRA synthesis available in docket OSHA 2010-0034 and
have read submissions to the document and heard testimony during the
March 19 & 20th hearings.  Outstanding issues with respect to my initial
comments remain in the areas of addressing the potential for a threshold
and on the potential for underestimation of silicosis rate based upon
radiographic evidence.  Regarding the underestimation of silicosis from
radiographic evidence, OSHA’s response was to view that as a
qualitative “significance of risk” issue.  I understand that OSHA
must make its determinations based upon the available data and that
there is no way to estimate across the breadth of epidemiological
literature on this subject the amount of silicosis that goes
undiagnosed.  However, OSHA should consider this potential
underestimation of cases as a significant uncertainty in its
quantitative risk assessment.  Thus in addition to the potential
influence of measurement error, as discussed extensively at the
hearings, diagnostic error may have also influenced non-cancer dose
response and estimation of risk.    

Analysis of Potential for a Threshold in Silica-Related Cancer,
Morbidity and Mortality

Regarding my second concern, there is a natural question as to whether a
threshold exists in the silica dose response, especially  for noncancer
endpoints.  Non-cancer endpoints are often conceived of as having a
threshold below which exposure is not expected to create a tangible
risk.   A number of commenters during the March 19-20th hearing
questioned the linear approach for both cancer and non-cancer endpoints
and urged OSHA to consider modeling forms that include the possibility
of threshold.  Some of their comments relative to measurement error
tending to blur a threshold and thus lead to statistically significant
coefficients in linear or log-linear regressions (Drs. Cox and Long)
have some merit.  OSHA may want to acknowledge this point but also
recognize that measurement error can also weaken associations and
decrease the significance of regression coefficients such that those
studies in which the regression coefficients are significant may be
because they had less measurement error (or other confounders) or in
which the dose response is strong enough to overcome the hurdle of
measurement error.  OSHA may also want to acknowledge that a
statistically significant linear regression coefficient  may come about
for three other reasons which have nothing to do with measurement error:
 1) there is no threshold for the toxic mechanism as it involves direct
attack on cellular protein, lipid or DNA such that there is a finite
probability of adverse consequence on the biomolecular level that has
the potential to propagate through amplification phenomena (e.g.,
secondary messengers and altered gene expression at the level of hormone
receptors and oxidant signaling to clonal growth of initiated cells at
the level of DNA mutation).  This type of mechanism is usually
considered most plausible for direct-acting carcinogens and although it
may also be possible for other types of agents, the general belief is
that cellular defenses, homeostatic mechanisms, biological redundancy
(e.g., numerous receptors or enzymes that do the same thing) and tissue
repair generally act to create a threshold for the adverse effect.  2)
there is a threshold but it is below the dose levels tested.  In this
case, the linear-appearing regression reflects the fact that the
exposure is capable of producing the effect through the range of
observation.  3) there is a threshold at the individual level but each
individual has their own threshold such that the population can be
characterized as having a distribution of vulnerability.  This
distribution may be due to differences in levels of host defenses that
come with differences in age, co-exposure to other chemicals, the
presence of interacting background disease processes, non-chemical
stressors and a variety of other host factors.  For example, of
relevance to silica is that obesity is a systemic inflammatory state
which has been demonstrated to cause exaggerated responses to inhaled
particulate matter and ozone in rats (Moon et al. 2014) and humans
(Mancuso 2010).  Additionally, pre-existing respiratory disease  such as
emphysema, chronic bronchitis, COPD or asthma may theoretically enhance
the crystalline silica effect as these are lung inflammatory conditions
to which silica-induced inflammation would be additive.  Even if these
diseases are not clinically present but the pre-clinical condition is
(lung diseases are understood to be a gradual process that have various
stages), this may create increased vulnerability and caused a greater
number of individuals to be higher risk on the population level.  The
National Academy of Sciences report Science and Decisions (NAS 2009)
describes these concepts and their relevance to dose response assessment
of non-cancer endpoints, particularly with respect to the lack of
threshold at exposure levels in population studies.  Given the multiple
sources of vulnerability that may exist in a population of workers
(including smoking which was often poorly controlled for in silica
studies) it is logical for OSHA to strongly consider inter-subject
variability in vulnerability (the third alternative to the explanation
of measurement error) as the reason for linearly-appearing regression
slopes in silica-related non-cancer and cancer studies.  This
explanation does not imply an artifact but that the linear (or log
linear) regression coefficient extending down to low dose reflects the
inherent variability in susceptibility such that the effect of concern
(e.g., silica-induced cancer, silicosis or other non-malignant diseases)
may occur in some individuals at doses well below what might be a
threshold in others.  This biological variability is important to
consider in criteria development and use of a population regression
coefficient avoids the use of NOAELs and somewhat arbitrary uncertainty
factors.  

There are other reasons to decrease emphasis on the measurement error
explanation for linear-appearing dose response for silica.  A
cornerstone of the measurement error hypothesis is the Brauer et al.
2002 demonstration that error in particulate matter (PM) measurements
are capable of washing out a built in threshold in PM dose response for
mortality.  Brauer et al. took personal monitoring measurements for 16
individuals for both particulate matter and sulphate and compared these
measurements to pollution measurements at regional monitors.  The PM
measurement error was large when relying on central monitoring locations
indicating that local and personal factors play a large role in the
value obtained.  Brauer et al (2002) then go on to demonstrate that the
degree of measurement error in their PM data was sufficient to smooth
out the dose response and obscure a threshold they had built into their
dose response model for PM-induced mortality.    Rhomberg et al. (2011)
reviewed the cases of PM, nitrogen oxide and ozone measurement
variability with respect to obscuring thresholds in air pollution
epidemiology studies.  Dr. Long who testified at the silica hearings
about measurement error was co-author on this review and mentioned this
paper in his testimony.  The Rhomberg review points out the importance
of indoor/outdoor difference in pollutant measures that could create
considerable measurement error when basing personal exposure on central
monitors, given the amount of time people spend indoors.  While there
may be numerous sources of measurement error in silica workplaces, the
source of error focused upon by Rhomberg et al (2011) – indoor/outdoor
differences, would not apply.  But beyond this, Rhomberg et al (2011) do
not consider the example of sulphate pollution even though this is a key
part of the Brauer et al. (2002) study.  In contrast to the particulate
matter results, sulphate measures from personal monitors correlated
quite well with central outdoor air monitors across the 16 subjects. 
This led Brauer et al. to demonstrate that given the small measurement
error for sulphate it is easy to show a threshold in population studies
should one exist (they simulated 3 different thresholds, each reliably
reproduced for sulphate when the sulphate measurement error was built
into the model).  This suggests that unlike PM, measurement error should
not substantially mask population thresholds for sulphate or sulphur
dioxide, which total sulphate is a surrogate measure for.  Looking back
at the population dose response for sulphate -related mortality  relied
upon by Brauer et al. (2002) (WHO 1999), and an update of that in WHO
2006, it is clear that many sulphate population studies are associated
with linear-appearing dose response.  The author of the sulphur dioxide
chapter of the WHO 2006 report, Dr. Morton Lippman synthesized the
sulphur dioxide evidence by saying that: “As with ozone and PM, no
obvious threshold levels have so far been identified in these
population-based studies. “  According to the Brauer et al. 2002
analysis, measurement error is unlikely to explain the lack of threshold
for sulphur dioxide increased mortality in the population studies
reviewed by Lippman.  The sulphate example makes the case that other
factors besides measurement error can cause a linear dose response down
to low dose in population studies for inhaled irritants.  These other
factors (e.g., inter-human variability) are summarized above for the
general case.  

With respect to the body of silica epidemiology literature, perhaps the
case with the least amount of measurement error is of US industrial sand
workers wherein many measurements were made with filter samples and XRD
determination of crystalline silica and in which there was very careful
estimation of historical exposure for both silica and smoking (Macdonald
et al. 2005; Steenland and Sanderson 2001; Hughes et al. 2001).  In
categorical analysis these studies demonstrated what appears to be a
monotonic dose response (increasing odds ratio with increasing category
of exposure) for silicosis and what might reflect a threshold for lung
cancer as the two lowest categories of exposure had similar cancer odds
ratio which then became significantly elevated along a dose response
pattern at higher doses (MacDonald et al. 2005).  If the lung cancer
data does evidence a threshold, it was thru the >300-<1,100 ug/m3-yrs
group, which is well below the current OSHA PEL (100 ug/m3 for 45 years
is 4,500 ug/m3-yrs).  In fact, the dose category in which the cancer
risk became elevated (OR= 2.24) was still below the current PEL
equivalent (>1,100 – 3,300 ug/m3-yrs).   

Beyond air pollutants, there are numerous examples of population studies
in which linear slopes occur down to low levels of exposure without
evidence of threshold.  In many of these cases measurement error is
unlikely to explain the lack of threshold since the index of exposure is
biomonitoring data.  Examples include linear appearing slopes for
arsenic-induced cancer and non-cancer effects in relation to urinary
inorganic arsenic (NAS, 2013),  for mercury-induced neurodevelopmental
deficits in relation to maternal hair mercury (USEPA, 2001, NAS, 2000)
which has been modelled as a linear relationship by Axelrad et al.
(2007), cadmium-induced decline in GFR which has been found to have
significant linear regression to low levels of urinary cadmium (Akesson
et al 2005) and lead-induced neurodevelopmental deficits in relation to
childhood blood lead (Bellinger 2008).  In the last case, the lack of a
demonstrable threshold has caused USEPA to not develop a threshold type
toxicity target (reference dose or RfD) for lead.  

It may be argued that there is still some error in the biomonitoring or
XRD silica results and that epidemiology studies will always have issues
of exposure misclassification or other types of error that may create
uncertainty when it comes to model specification.  However, these types
of error will also bias correlations to the null such that if they were
sufficiently influential to obscure a threshold they may also
substantially weaken regression results and underestimate the true risk.
  Further, the arguments for measurement error as the explanation have
not considered the influence that inter-subject variability in response
can have on the lack of a threshold at the population level as discussed
above and highlighted in NAS’s Science and Decisions (NAS 2009).  

Animal evidence for thresholds in silica response is also important to
consider to the extent that such data exist.  Perhaps the most relevant
study is from Kuempel et al. (2001) in which a constant dose of silica
(15 mg/m3) was delivered to rats by inhalation 6 hrs/day, 5 days per
week for varying lengths of time.  Even though only one dose level was
applied, this represents a dose-response study in terms of silica lung
loading relative to the emergence of initial (inflammation) and
secondary (fibrotic) lung damage.  While OSHA already describes this
study, its importance to the threshold concept may merit additional
discussion in a revised OSHA document.  

The critical loading of lung tissue with quartz required for PMN
infiltration and impairment of macrophage clearance in this study was
0.39 mg silica/g lung.  Kuempel et al. (2001) considered this to be a
threshold for silica-induced inflammation in their animal model system
although the actual data did not exhibit a threshold for PMN
infiltration (lowest dose used was a LOAEL) and the critical loading
dose was an extrapolation from a toxicodynamic model.   Taking this
reported threshold loading dose at face value leads to a threshold
inhaled dose of 0.036 mg/m3 for a 45 year occupational span of exposure.
 Kuempel et al. (2001) recognized that this critical loading value is
variable such that some workers may be more sensitive and estimated the
lower 95% confidence limit on the critical threshold dose as 0.049 mg
quartz/g lung which corresponds to an inhaled concentration for 45 years
of 0.005 mg/m3 (Kuempel et al. 2001).   In a followup analysis this same
research group recognized that the toxicokinetics of inhaled quartz will
vary across humans particularly with regard to lung clearance rates and
estimated a 2.5 fold variability in silica loading (95th% to median) for
a given inhaled concentration (Kuempel et al. 2001b).  Thus the Kuempel
et al. (2001, 2001b) rat analysis of lung threshold loading and
extrapolation to human dosimetry leads to the conclusion that in the
median case this threshold is approximately 3 times below the current
OSHA PEL and that in those who are more sensitive either because of
toxicokinetic or toxicodynamic factors (or both), the threshold may be
an order of magnitude or more below the PEL.  In terms of what this
threshold represents (inflammatory response), Kuempel et al. (2001)
projected this threshold loading out to cancer risk based upon a
combination of rat and human silica cancer data.  While more data
describing thresholds in rats and humans for various silica-related
outcomes would be helpful, the little direct evidence that does exist
suggests that the threshold for the initial stages of lung pathology
(inflammation, disruption of macrophage function) may occur below the
current PEL.   

In summary, OSHA may consider incorporating a section on Model
Specification for Low Dose Extrapolation that explicitly describes the
choices relative to the threshold concept for the silica dose response
dataset.  This section could include discussion of specific dose
response datasets that are consistent with a linear or a threshold-type
model, if a threshold seems likely, where was it seen relative to the
current and proposed PEL, and a general discussion of mechanism of
action, measurement error and population variability as concepts that
can help us understand silica dose response for cancer and non-cancer
endpoints.  

High Dose Saturation of Response

Several studies (e.g., Park et al. 2002, diatomaceous earth workers, Lui
et al 2013, Chinese pottery workers and miners; Vacek et al. 2013,
Vermont granite workers) found evidence of silica effect at lower doses
but response at high dose was attenuated in some manner.  OSHA can
consider pulling such studies into a section that describes this type of
non-linearity, the reasons why it appears repeatedly in the silica
literature, and the modeling approaches to address it.  Measurement
error may be considered as a key reason for this apparent saturation of
effect as the highest doses in these studies are associated with the
oldest datasets in which there was the greatest opportunity for
measurement error (inexact job descriptions and work histories, older
methods of silica measurement).  Another possibility is that at high
dose various intercurrent causes of mortality related to silica exposure
are more likely to eliminate subjects before completing the latency
period for cancer than at lower dose.   Other explanations put forward
by various authors (e.g., healthy worker/survivor effect) may also be
described.  The value of this section is that it would more clearly
describe the statistical and biological rationale for either excluding
high exposure groups or using log conversions of the data.  The current
OSHA draft mentions the high dose problem in the data and exclusion of
some data without complete explanation.  

IARC 2012 Monograph 

The current OSHA document describes the IARC 1997 review of crystalline
silica.  The updated IARC review should be highlighted particularly with
respect to its reliance on the numerous meta-analyses and their
consistent showing of a cancer risk.  Further, IARC 2012 can be cited
for mechanism of action information and for data regarding the
requirement (or lack thereof) of silicosis for the induction of
silica-related lung cancer.  Along these lines, the Lui et al. 2013
study should also be highlighted.  

References not cited in the OSHA 2013 Draft Document

Axelrad DA, Bellinger DC, Ryan LM, Woodruff TJ. 2007. Dose-response
relationship of prenatal mercury exposure and IQ: an integrative
analysis of epidemiologic data. Environ Health Perspect. 115(4):609-15.

Bellinger DC. 2008. Very low lead exposures and children's
neurodevelopment. Curr Opin Pediatr. 20(2):172-177.

Brauer M, Brumm J, Vedal S, Petkau AJ. 2002. Exposure misclassification
and threshold concentrations in time series analyses of air pollution
health effects. Risk Anal. 22(6):1183-93.

Kuempel ED, Tran CL, Smith RJ, Bailer AJ. 2001b. A biomathematical model
of particle clearance and retention in the lungs of coal miners. II.
Evaluation of variability and uncertainty. Regul Toxicol Pharmacol.
34(1):88-101.

Mancuso P. 2010. Obesity and lung inflammation. J Appl Physiol 
108(3):722-8.

Moon KY(1), Park MK, Leikauf GD, Park CS, Jang AS. 2014. Diesel Exhaust
Particle-Induced Airway Responses are Augmented in Obese Rats.  Int J
Toxicol. 33(1):21-28.

NAS 2000.  The Toxicological Effects of Methyl Mercury.  National
Academy Press, Washington, DC. 

NAS (National Acad Science) 2009.  Science and Decisions.  Advancing
Risk Assessment.  National Academy Press, Washington DC.  

NAS 2013. Critical Aspects of EPA’s IRIS Assessment of Inorganic
Arsenic.  Interim Report.  National Academy Press, Washington DC.  

Rhomberg LR, Chandalia JK, Long CM, Goodman JE.  2011. Measurement error
in environmental epidemiology and the shape of exposure-response curves.
 Crit Rev Toxicol. 41(8):651-71.

USEPA 2001.  IRIS Profile for Methyl Mercury.  Available at:   HYPERLINK
"http://www.epa.gov/iris/subst/0073.htm" 
http://www.epa.gov/iris/subst/0073.htm  

WHO (World Health Org) 1999. Air Quality Guidelines.  

WHO 2006. Air quality guidelines: global update 2005: particulate
matter, ozone, nitrogen dioxide, and sulfur dioxide. Copenhagen,
Denmark; World Health Organization, 2006, pg 395.  



  HYPERLINK \l "_Toc251578068"  Brian Miller

IOM Consulting Ltd., Scotland, UK From: Brian G Miller 

Sent: Tuesday, April 01, 2014 10:11 AM

To: Kate Schalk; 

Subject: RE: OSHA Silica - Reminder - Final comments due no later than
Wed, April 2nd

Dear Kate

I have reviewed the original review materials, my 2009 peer review
comments and their disposition by OSHA in the current proposed
rulemaking.  I feel that OSHA has adequately addressed my comments in
their updated Health Effects and QRA sections.  I have included one
correction in my attached submission.  I have also read the written
testimony of the public commenters and heard their oral testimony.  I
have included comments on some of the commenters’ testimony as it
applies to the strength of the evidence for causality and/or effects at
exposures to low concentrations.

Best wishes

Brian Miller

Principal Epidemiologist

Institute of Occupational Medicine

Research Avenue North

Riccarton

Edinburgh

EH14 4AP

Web:   HYPERLINK "http://www.iom-world.org"  http://www.iom-world.org 

 



	

 

The Institute of Occupational Medicine (IOM) is a company limited by
guarantee, registered in Scotland (No.SC123972) and a  Registered
Scottish Charity (No.SC000365). IOM Consulting Ltd is a wholly owned
subsidiary of IOM and a private limited company registered in Scotland
(No. SC205670). Registered Office: Research Avenue North, Riccarton,
Edinburgh, EH14 4AP, Tel +44 (0)131 449 8000.

This email and any files transmitted with it are confidential and
intended solely for the use of the individual or entity to whom they are
addressed. If you have received this email in error please notify the
sender - immediately and delete the message from your system. All IOM
email is virus scanned but the recipient should also check this email
and any attachments for the presence of viruses.

  To protect the environment please do not print this e-mail unless you
really need to.



Brian Miller - Comments on Docket OSHA-2010-0034

Comments on Docket OSHA-2010-0034

I am satisfied that my comments on the 2009 drafts were taken on board
in the redraft.

On P321, last paragraph, the sentence ‘Normal radiographs are
categorized as 0/-.’ is both incorrect and superfluous, and should be
deleted.  It may be that other parts of this paragraph are made
redundant by the more detailed descriptions above.

General comments

During the testimony, it was good to see that both industry and others
expressed the need to avoid exposures that would lead to health risks.

Industry made strong representations on a number of points that, in
their view, implied that a reduction in the PEL is unnecessary,
unwarranted on the available evidence and/or unachievable. I have no
opinion on achievability, but have comments on other aspects.

Much was made of the paucity of evidence on lung cancer risks at
concentrations at or below the proposed PEL, and it is hard to disagree
with that assessment.  In addition, there was discussion of whether
silicosis is a prerequisite for increased lung cancer risk. I consider
this issue unanswerable, given that we cannot investigate for early
fibrotic lesions in the living, but must rely on radiographs.  However,
there is no doubt that silica is causal for silicosis, and regulating to
eliminate that disease is the sensible starting point.  It is to be
expected that reducing exposure to that end will have side-benefits in
reducing cancer risks.

That expectation is predicated on a belief that silica exposure is
causal for cancer.  The testimony of Dr LA Cox emphasised strongly that
association does not necessarily imply causality, and illustrated this
with some trite (and irrelevant) pathological examples.  He also listed
a series of potential biases that could affect epidemiological studies
if not adjusted for.  However, in epidemiology we are often limited to
observational studies from which associations of health outcomes with
prior exposures are observed and quantified.  Some of the studies on
which OSHA relies are not simply cross-sectional, but show responses to
changing patterns of exposure.  This brings much greater assurance that
causality is at work.  When challenged on the issue of causality in
silicosis, Dr Cox accepted that silica was causal, but questioned
whether there was evidence for increased risks at low concentrations;
i.e. whether there was a threshold.  The Morfeld paper was quoted, which
points at a threshold around 0.25 mg/m3 of respirable quartz.  While
that analysis is consistent with the possibility of a true threshold in
the porcelain industry, it is based on very low prevalence, which may be
a reflection of the mediating effect of clay minerals; and its results
rest on some of the observed cases having developed at zero relevant
(above-threshold) exposure concentrations.  It remains to be seen
whether the same effect can be demonstrated in other industries can be
replicated in other industries, for example in the absence of clay
minerals.

Dr Cox also had many criticisms of studies of the health effects of
quartz. On P97 of his written submission, he deals with IOM papers on
effects in Scottish coal workers. He states ‘excluding age increases
the effects of age-related confounder, … undermining causal
interpretation…’. This ignores the fact that there are no
confounders, age-related or otherwise, of silicotic changes at the 2/1
level, which are caused by silica alone. He then criticises the detailed
model-building and testing that the papers describe, characterising them
as biased from ‘uncorrected multiple testing’, a criticism that I
would refute.  The various model steps were taken in the data-based
knowledge of the history of exposures at that colliery, in an attempt to
build a reality-driven and ‘best-fitting’ model, which Cox describes
on P4 as ‘standard practice’. Finally, he characterises the final
model of the Buchanan reanalysis as biased by ‘dichotomizing a
continuous variable’.  This misrepresents the final model entirely,
which does not dichotomise but fits a change-point model with dual
slopes on exposures experienced at concentrations above and below the
change-point of 2mg/m3. It is not clear how these misrepresentations
have arisen, but they do not inspire confidence in the acuity of Cox’s
criticisms of other studies.

In the end of the day, it’s true that there is little current
scientific evidence of what risks will be experienced by workers exposed
under the current or proposed PEL.  On all existing evidence, it seems
likely that those risks will vary by industry, e.g. by whether exposure
can be mediated by clay minerals. However, it would seem that OSHA has
little choice but to regulate for the highest risks, e.g. in industries
with exposure to pure respirable quartz.  In my own view, whether to
adopt a 0.1/0.05 regulatory regime or a 0.05/0.025 one is a matter for
OSHA’s judgment, but whichever is adopted it must include regular
monitoring and health surveillance.  In addition, I agree with several
of those giving testimony that a ban on sandblasting in the US is long
overdue.       

Dr Brian G Miller

Institute of Occupational Medicine

Edinburgh, UK

1 April 2014 



  HYPERLINK \l "_Toc251578069"  Andrew Salmon 

Private Consultant, Lafayette, CA 

March 31, 2014

Dear Sir,

Re: Proposed Rule: Occupational Exposure to Respirable Crystalline
Silica

“I have reviewed the original review materials, my 2009 peer review
comments and their disposition by OSHA in the current proposed
rulemaking.  I feel that OSHA has, with one minor exception, adequately
addressed my comments in their updated Health Effects and QRA sections. 
I have included my comments on that area in my attached submission.  I
have also read the written testimony of the public commenters and heard
their oral testimony.  I have included comments on some of the
commenters’ testimony, specifically as it relates to thresholds,
latency, causality, new epidemiological analysis published by Liu et al.
(2013) and measurement error vs. variability.”

Yours sincerely,

Andrew G. Salmon M.A., D.Phil., C.Chem. M.R.S.C.

Proposed Rule: Occupational Exposure to Respirable Crystalline Silica

Analysis of responses to 2009 peer review and 2014 public comments

A.G. Salmon M.A., D.Phil.

Responses to 2009 Peer review

For the most part OSHA has addressed my comments on the earlier draft of
the Health Assessment and QRA sections thoroughly and adequately.  The
improved description and consideration of the South African studies of
silicosis (Hnizdo and Sluis-Cremer, 1993; Churchyard et al., 2004) is
particularly welcome.

The only area which remains somewhat less than completely resolved is
the question of the shape of the dose-response curve (for silicosis
particularly) in the lower-dose range.  I understand that OSHA’s
interest for the purpose of the rulemaking is the shape and steepness of
the dose response over the range of exposures corresponding to the
current and proposed PELs and action level.  In this respect the
analysis resembles that used for instance by US EPA’s criteria air
pollutant standards, where health effects are clearly demonstrable at
the concentrations currently observed and under consideration as
standards.  This is in contrast to the usual approach for toxic
substances where it is usual to attempt to define a “safe”
concentration below which health effects are either absent or
negligible.  Under the circumstances of the proposed silica regulation
and the difficulties (noted by OSHA) of control and measurement at
levels much below the proposed action level, this is entirely
reasonable.  However it does not replace the need to consider the
dose-response curve in the lower ranges, as illustrated by several of
the public comments presented on the issues of thresholds and the
relationship between clinical silicosis and lung cancer.  This aspect of
the debate will be addressed in the section on my consideration of some
of the public comments.

Analysis of 2014 public comments

A number of commenters drew attention to the question of whether there
is a threshold dose for the induction of silicosis.  These comments
varied from essentially anecdotal claims that silicosis had not been
observed in occupational groups with exposures generally within the
current PEL, to more analytical analyses of the epidemiological data
(e.g. Morfeld).  However it is important to understand what is implied
by this supposed “threshold” concept.  Many of the so-called
thresholds seen in epidemiological studies represent thresholds of
observability rather than thresholds of disease incidence.  The
appearance of silicosis is progressive with both dose and time since
first exposure, so studies (and anecdotal observations) with less
statistical power and shorter post-exposure followup (or none) will
necessarily fail to see the less frequent and later-appearing responses
at lower doses.  This creates an apparent threshold which is higher in
these studies than the apparent threshold implied by studies with
greater statistical power and longer follow-up.  This effect is also
clearly seen for studies which use more severe radiographic criteria to
define a positive response, since it has been established in the OSHA
analysis that radiographic diagnosis of silicosis is in fact a
relatively insensitive measure, and again the severity of response is a
function of both dose and timing of observation.  The best illustration
of the impact of these factors in seen in the risk assessment for
chronic inhalation of crystalline silica by California’s Office of
Environmental Health Hazard Assessment (OEHHA, 2005 in the references
listed by OSHA).  In that report a benchmark dose analysis was used to
identify a point of departure based on the data on South African gold
miners (referenced by OSHA as Hnizdo and Sluis-Cremer, 1993).  The
averaged and time-weighted concentration at this point was 11.4 μg/m3,
significantly below the proposed PEL, let alone the existing one.  This
analysis also illustrates a further important feature of these so-called
“thresholds”.  The point of departure identified by the benchmark
dose analysis can be regarded as a threshold of observability, and is
usually chosen to have roughly similar properties to the NOAEL
identified by more traditional risk assessment procedures.  However, it
is clearly established as a statistically based measure with a
distribution of plausible values reflecting both measurement uncertainty
and actual variability in the population being analyzed.  It is
definitely not a single specific value below which it can be confidently
asserted that no cases will be observed, as appears to be implied in
some of the public comments.  Another peer reviewer (Crump) has
eloquently argued that these observational thresholds are essentially
artifacts of certain types of analysis rather than genuine features of
the data.  The observational threshold or point of departure is also
quite explicitly not a statement about a possible mechanism involving a
threshold response.  If this is to be established it needs substantial
additional mechanistic information such as identification of the nature
and dose-response characteristics of precursor effects.

In a somewhat similar vein, several commenters referred to the
“latency” for appearance of silicosis or lung cancer some
considerable time after the causative exposure to crystalline silica. 
This term is often used to describe a delay term included in the
equations used to fit time of incidence data in epidemiological
analysis.  However, similarly to the point of departure used in
dose-response analysis, it is an error to describe this as a cutoff
point before which no cases will be observed.  It is in fact a best-fit
estimate of the median value for a parameter in a very simplistic
equation used to describe time dependence.  As such it represents a
distribution of possible values reflecting both uncertainty and
variability, as well a random variation in the time of appearance of the
disease, rather than a single value.  The type of time dependence
implied by use of this value is not necessarily a particularly good
representation of the actual dependence in the data, although it is
often used as a mathematically convenient approximation.  Where the data
are of sufficient quality and quantity to permit, epidemiologists have
often used more sophisticated models of the time dependence for
incidence of observable effects, which make clear the distributional
nature of parameters such as “latency”.

Some commenters (notably Cox) appeared to question the causality of the
association between silicosis or lung cancer and exposure to respirable
crystalline silica.  Cox’s argument in both written and oral testimony
appears to be that, from a statistical standpoint he is unable to
establish the causality of a relationship, which he illustrates with a
trivial example involving cell phone use and coronary heart disease
mortality.  This point is both true and obvious: it could have been made
more succinctly without any loss of effectiveness.  The identification
of causality as opposed to statistical association is, as described by
Bradford Hill in his well-known criteria, based mainly on
non-statistical considerations such as consistency, temporality and
mechanistic plausibility:  the role of statistics is mostly limited to
establishing that there is in fact a quantitatively credible association
to which causality may (or may not) be ascribed.  OSHA correctly cites
the substantial body of evidence supporting the association and
causality for silicosis and lung cancer following silica exposure, and
also quotes previous expert reviews (such as IARC).  The causal nature
of these associations has already been established beyond any reasonable
doubt, and OSHA’s analysis sufficiently reflects this.  Cox also
appears to suggest that even if the causal association is proven at high
dose levels it may not be present at lower dose levels.  The assumption
of continuity in scientific hypotheses (such as that effects at one dose
level would be consistent with those at other doses, and connected by
continuous quantitative relationships without singularities) is usually
treated as a “hard” assumption which would not be questioned unless
there were actual data providing evidence to the contrary.  No such
evidence was offered in this case.  Cox also asserts that there is a
bias towards finding a positive association even if the actual data are
randomly distributed, showing a field of points with all positive, but
randomly distributed x and y values, and a least squares fitted line
(his figure 2 and slide).  This argument is irrelevant on two counts: a)
the least squares fit he uses forces the assumption y = 0 at x=0.  OSHA
does not make any such assumption in its model fits, instead fitting to
actual measured data points within the range of observation.  Indeed
models such as the spline and logarithmic fits explicitly exclude that
zero-point assumption.  b)  If indeed the data shown in this figure were
independent observations of effect at specific real doses, and the
incidence at zero dose were known to be zero (or some positive but much
lower effect incidence), a statistical analysis would correctly conclude
that the data showed a positive association between exposure (x) and
effect (y), although the data clearly do a poor job of estimating the
value of the actual slope factor.

The paper (Liu et al., 2013) cited in Steenland’s written testimony,
and commented on by several others, is an important extension of the
earlier work on the Chinese cohorts exposed in various industries to
crystalline silica.  The major difference between this and previous
analyses is the exclusion of workers considered to have substantial
confounding exposures to polycyclic aromatic hydrocarbons or arsenic,
both of which are known causes of lung cancer.  The details of this
censoring are stated as being based on measurements in the previously
published studies of thee workers.  Exact details would only be
accessible by a line-by-line analysis of the source data which is
unlikely to appear in a published paper.  However the overall conclusion
is important, in that with the exclusion of these confounders (likely
substantial if not complete) the association of silica exposure with
lung cancer is maintained with similar magnitude to that reported in the
earlier studies.  This implies that although these earlier studies were
potentially susceptible to these confounders their effect is probably
relatively small both in the earlier studies and in Liu et al (2013). 
An important new conclusion in the new analysis the near multiplicative
interaction of silica exposure and smoking in the induction of lung
cancer.  This is consistent with observations on the interaction of
smoking with other dusts and fibers, such as asbestos, and has
significant implications for public health policy and education.  Also
significant is the apparently clear finding that although both lung
cancer and silicosis are correlated with silica exposure, they appear
independently, i.e. the clinical observation of silicosis is not a
prerequisite for the appearance of lung cancer.  This point has been
subject to extensive debate, with some results supporting the hypothesis
that silicosis is a necessary precursor to lung cancer and others
suggesting the opposite.  The study by Liu et al. (2013), by virtue of
its size and finding on this point is a strong argument for the
independent appearance of lung cancer.  This does not of course argue
that the initiating events for lung cancer and silicosis are not part of
a common adverse outcome pathway, but it reinforces the perception that
radiographically observable silicosis is a separate and relatively late
event.  In any event, this new publication is an important addition to
the literature which in some respects clarifies and reinforces earlier
conclusions, and OSHA should include consideration of this in its final
determination.  It is unlikely that this would result in any change to
the numerical recommendations, but it would help to resolve some of the
qualitative and mechanistic uncertainties.

Several comments addressed actual or potential uncertainties in
measurement, both in epidemiological studies used as the basis of
evidence and in workplace measurements to establish compliance with new
or existing PELs and action levels.  While this discussion is
interesting and important, it appears that in general OSHA has sound and
data-based analyses of these questions.  However an additional source of
variation in outcome which was given less attention in public comments
and discussions is interindividual variations between members of the
workforce.  This applies both to variations which can affect exposure,
such as behavior, exact location etc., and to variations in individual
susceptibility to the effects, which may be genetic or related to age
and health status inter alia.  In general these variations, which are
real features of the population and its response rather than
“errors”, are likely to be at least as large as uncertainties in
measurement.

APPENDIX A

External Peer Review of OSHA’s Draft

“OSHA Preliminary Health Effects Section for Silica”

and

 “Preliminary Quantitative Risk Assessment for Silica”

Peer Review Comments

Contract No.  GS-10F-0125P

OSHA BPA No. DOLQ059622303

Submitted to:

David O’Connor

Directorate of Standards and Guidance

Occupational Safety and Health Administration

200 Constitution Avenue, NW

Washington, DC  20210

Submitted by:

Eastern Research Group, Inc.

110 Hartwell Avenue

Lexington, MA  02421-3136

January 13, 2010

TABLE OF CONTENTS

  TOC \o "1-3" \h \z \u    HYPERLINK \l "_Toc251578060"  Introduction	 
PAGEREF _Toc251578060 \h  1  

Health Effects

  HYPERLINK \l "_Toc251578061"  Technical Charge to External Peer
Reviewers For Review of 

OSHA’s Preliminary Health Effects Section for Silica	  PAGEREF
_Toc251578061 \h  3  

  HYPERLINK \l "_Toc251578062"  External Peer Review of OSHA’s Draft
“OSHA Preliminary Health Effects Section for Silica”   

Peer Review Comments	  PAGEREF _Toc251578062 \h  7  

  HYPERLINK \l "_Toc251578063"  Additional Comments/References	  PAGEREF
_Toc251578063 \h  39  

  HYPERLINK \l "_Toc251578064"  Individual Review Comments	  PAGEREF
_Toc251578064 \h  41  

  HYPERLINK \l "_Toc251578065"  Bruce Allen Bruce Allen Consulting,
Chapel Hill, NC	  PAGEREF _Toc251578065 \h  43  

  HYPERLINK \l "_Toc251578066"  Murray Finkelstein McMaster University,
Thornhill, ON Canada	  PAGEREF _Toc251578066 \h  53  

  HYPERLINK \l "_Toc251578067"  Gary Ginsberg Connecticut Department of
Public Health, Hartford, CT	  PAGEREF _Toc251578067 \h  63  

  HYPERLINK \l "_Toc251578068"  Brian Miller IOM Consulting Ltd.,
Scotland, UK	  PAGEREF _Toc251578068 \h  71  

  HYPERLINK \l "_Toc251578069"  Andrew Salmon Private Consultant,
Lafayette, CA	  PAGEREF _Toc251578069 \h  83  

  HYPERLINK \l "_Toc251578070"  Noah Seixas University of Washington,
Seattle, WA	  PAGEREF _Toc251578070 \h  95  

Quantitative Risk Assessment for Silica

  HYPERLINK \l "_Toc251578071"  Technical Charge to External Peer
Reviewers Peer Review of 

OSHA’s Preliminary Quantitative Risk Assessment for Silica	  PAGEREF
_Toc251578071 \h  107  

  HYPERLINK \l "_Toc251578072"  External Peer Review of OSHA’s
“Preliminary Quantitative Risk Assessment for Silica”   

Peer Review Comments	  PAGEREF _Toc251578072 \h  109  

  HYPERLINK \l "_Toc251578073"  Additional Comments/References	  PAGEREF
_Toc251578073 \h  139  

  HYPERLINK \l "_Toc251578074"  Individual Peer Review Comments	 
PAGEREF _Toc251578074 \h  147  

  HYPERLINK \l "_Toc251578075"  Bruce Allen Bruce Allen Consulting,
Chapel Hill, NC	  PAGEREF _Toc251578075 \h  149  

  HYPERLINK \l "_Toc251578076"  Kenneth Crump Louisiana Tech University
Foundation, Inc., Ruston, LA	  PAGEREF _Toc251578076 \h  159  

  HYPERLINK \l "_Toc251578077"  Murray Finkelstein McMaster University,
Thornhill, ON Canada	  PAGEREF _Toc251578077 \h  177  

  HYPERLINK \l "_Toc251578078"  Gary Ginsberg Connecticut Department of
Public Health, Hartford, CT	  PAGEREF _Toc251578078 \h  183  

  HYPERLINK \l "_Toc251578079"  Brian Miller IOM Consulting Ltd.,
Scotland, UK	  PAGEREF _Toc251578079 \h  191  

  HYPERLINK \l "_Toc251578080"  Andrew Salmon Private Consultant,
Lafayette, CA	  PAGEREF _Toc251578080 \h  201  

 

Introduction

In 2009, Eastern Research Group, Inc. (ERG), under contract to the
Occupational Safety and Health Administration (OSHA), conducted an
independent, scientific peer review of two draft documents prepared by
OSHA: 

Preliminary Health Effects Section for Silica.

Preliminary Quantitative Risk Assessment for Silica. 

This report documents the review process and provides the reviewers’
final comments and recommendations.  

Background 

OSHA developed two documents to support development of a proposed rule
governing exposure to crystalline silica in the general, construction,
and maritime industries: 1) a scientific assessment of the literature
and other information on the health effects of silica, and 2) a
quantitative risk assessment for silicosis, lung cancer, and other
diseases resulting from occupational exposure to crystalline silica. To
ensure a sound scientific basis for decision making, the Agency
contracted with Eastern Research Group, Inc. (ERG) to conduct a letter
peer review of these two documents.  

Peer Review Process

Using selection criteria provided by OSHA, ERG conducted a search for
nationally recognized experts in occupational epidemiology,
biostatistics and risk assessment, animal and cellular toxicology, and
occupational medicine who had no conflict of interest (COI) or apparent
bias in performing the review. Interested candidates submitted evidence
of their qualifications and responded to detailed COI questions. ERG
also searched the Internet to determine whether qualified candidates had
made public statements or declared a particular bias regarding silica
regulation.   

From the pool of qualified candidates, ERG selected seven to conduct the
review, based on:

Their qualifications, including their degrees, years of relevant
experience, number of related peer-reviewed publications, experience
serving as a peer reviewer for OSHA or other government organizations,
and committee and association memberships related to the review topic. 

Lack of bias and lack of any actual, potential, or perceived conflict of
interest. 

The need to ensure that the panel collectively was sufficiently broad
and diverse to fairly represent the relevant scientific and technical
perspectives and fields of knowledge appropriate to the review.  

OSHA reviewed the qualifications of the candidates proposed by ERG and
verified that the range of the candidates’ qualifications met the
technical selection criteria. ERG then contracted with the following
reviewers to perform the review. Six experts reviewed both documents;
Dr. Crump reviewed only the quantitative risk assessment:

Reviewer	Affiliation	Reviewed Health Effects Section	Reviewed
Quantitative Risk Assessment

Bruce Allen	Bruce Allen Consulting	   (	(

Kenneth Crump	Louisiana Tech University Foundation

(

Murray Finkelstein	McMaster University, Ontario	(	(

Gary Ginsberg	Connecticut Department of Public Health	(	(

Brian Miller	IOM Consulting Ltd., Scotland	(	(

Andrew Salmon	Private Consultant	(	(

Noah Seixas	University of Washington, Seattle	(	(



Reviewers were provided with the charge, review documents, and hundreds
of relevant references and given six weeks to conduct the review. 

Early in the review schedule, ERG organized and facilitated a briefing
call to ensure that reviewers understood the peer review process. OSHA
representatives were available on the call to respond to technical
questions of clarification. Reviewers were invited to submit any
subsequent questions of clarification to ERG via email. 

ERG checked reviewers’ written comments to ensure that each reviewer
had clearly responded to all charge questions and then submitted the
individual reviewer comments, unedited, to OSHA. ERG also organized the
comments by charge questions for both documents and submitted this
version to OSHA and the reviewers in preparation for a follow-up
conference call.

The conference call, organized and facilitated by ERG, provided an
opportunity for OSHA to seek clarification of individual reviewer’s
comments. After the call, ERG directed reviewers to revise their written
comments to include their clarifications. ERG submitted the revised
comments to OSHA, organized by charge question, and prepared this final
peer review report. Section 2 of this report provides peer review
comments for the health effects section and Section 3 provides reviewer
comments for the quantitative risk assessment. Both sections first
present the charge to reviewers and then provide reviewer comments
organized by charge question and, subsequently, by individual reviewer. 



Technical Charge to External Peer Reviewers

For Review of OSHA’s Preliminary Health Effects Section for Silica

ERG Contract No. GS-10F-0125P

BPA No. DOLQ059622303

ERG Task No. 0193.15.064.001

Instructions to Peer Reviewers

You have been selected to participate in an external peer review of the
OSHA preliminary health effects section for silica that will be part of
a proposed rule that would amend the existing regulation for
occupational exposure to silica.  The health effects section includes a
discussion of the adverse effects that result from silica exposure and
the evidence that supports those associations.  The health effects
section is not a quantitative risk assessment, but rather the primary
analysis upon which the Agency must rely when making its determination
that employees exposed to silica at the current permissible exposure
limit (PEL) face a significant risk of material impairment to their
health and that the proposed standard will substantially reduce that
risk.  The quantitative risk assessment is based on a more limited set
of health endpoints than is discussed in the health effects section. 
The quantitative risk assessment is discussed in Section VI, which you
are also reviewing under this task, addressing a different set of charge
questions.

In the discussion of the health effects and their association with
silica exposure, the interpretations of studies and conclusions drawn
should be reasonable, sound, and consistent with the underlying science.
 The overriding goal of the peer review is to ensure that the
presentation of the studies, interpretations of the study results, and
preliminary conclusions drawn are clear and scientifically credible. 
OSHA has provided the following questions to guide and focus your review
divided into two types: 1) general questions that apply to all parts of
the health effects section; and 2) specific questions that apply to
individual sections/topics in the health effects section of the proposed
rule.  Your review should, at a minimum, address each question,
providing a discussion and rationale for any “yes” or “no”
responses, and providing other comprehensive comments to clarify your
comments and/or recommendations.

General Questions

The draft health effects section reviews and summarizes a number of
studies.  Are the studies described in sufficient detail for the reader
to understand how the studies were done?  Are the descriptions of the
studies and results accurate?  Are the strengths and limitations of the
studies adequately discussed and appropriately evaluated?  Are OSHA’s
interpretations of the study results reasonable and explained
adequately?  Are discussions of studies clear and easy to follow?  If
not, let OSHA know what needs to be added to improve the information and
the manner in which it is presented.

The draft health effects section covers a selection of studies in each
topic area.  The discussion does not cover every study ever done on the
topics.  Do you think that OSHA should have discussed additional or
different studies on a specific health endpoint?  Are you aware of
significant studies that were omitted?  If yes, inform OSHA of those
studies.  



The draft health effects section covers a number of health endpoints. 
Are you aware of additional significant health endpoints affected by
exposure to crystalline silica that should be discussed?  If so, inform
OSHA of the additional health endpoints and studies that address them.

The draft heath effects section contains conclusions for each topic area
as well as an overall conclusion.  Are these conclusions reasonable in
light of the available data?  Is it clear in the discussion how OSHA
arrived at those conclusions based on the studies discussed? 

Questions on Section A. Introduction

A-1.	Is OSHA’s review of the history of dust exposure monitoring
helpful to the reader in understanding the needs and limitations of
exposure conversions to a modern mass per volume metric in reference to
the exposure-response relationship?

Questions on Section B. Silicosis

B-1.	OSHA presents background information on the use of radiography,
imaging techniques, pulmonary function studies, and other diagnostic
tools used in clinical settings and for in epidemiological studies. 
Does OSHA correctly present the benefits and shortcomings of these
techniques?  Are you aware of any additional information that would
either strengthen this discussion or contradict what is written?  

B-2.	OSHA attempted to describe the available silicosis surveillance
data.  In your opinion, were any important topics left uncovered?  Any
data sources not discussed?  Is the subject of under-reporting
sufficiently covered?  Are the conclusions supported by the discussion
and stated succinctly?  Are you aware of any important studies in this
area that OSHA did not cover?

B-3.	For the discussion of progression of silicosis, OSHA did not review
every study that addressed this issue.  Are you aware of additional
studies that OSHA should consider, especially any studies that
illustrate progression in a cohort exposed to silica in the range of
OSHA’s current PEL (0.1 mg/m3)?   

B-4.	Results of studies covered in the PFT section are not always
consistent.  Did OSHA do a fair job of presenting these results and
drawing conclusions?  Are there any important studies that you are aware
of that would strengthen this section or the conclusions of this
section?

Questions on Section C. Carcinogenic Effects of Silica (Cancer of the
Lung and Other Sites)

C-1.	With such a large amount of epidemiological literature to review,
OSHA may be subjected to criticism of selection bias.  Is the selection
criteria used by OSHA clear?  Have we missed any important studies that
have good or at least reasonable exposure estimates?  Do you recommend
any alternate or additional selection criteria?



C-2.	OSHA concludes that the studies involving four cohorts among the
20+ cohorts and case controls reviewed were the strongest in terms of
several criteria in record availability and data quality (exposure,
smoking, worker and facility history) as well as data treatment and
analyses?  Do you agree with this conclusion?  Would you add or delete
any of these studies?  

C-3.	Do the tables add to the text presentation?  Do you have any
recommendations to improve their content or presentation?  Would you add
or delete another table?  Provide an explanation for any proposed
deletion or a mock-up of any proposed addition.

C-4.	OSHA has reviewed a number of studies that examined the
relationships between exposure to silica and cancer at sites other than
the lung, including larynx and nasopharynx, stomach, and esophagus. 
Have we missed any important studies of cancer at other sites? If yes,
inform OSHA of those studies.

C-5.	OSHA has preliminarily concluded that an association has not been
established between silica exposure and excess mortality from cancer at
sites other than the lung.  Do you agree with this interpretation of the
data?  If not, please provide a detailed rationale for your different
interpretation.

Questions on Section D. Other Nonmalignant Respiratory Disease

D-1.	This draft section discusses COPD, NMRD, bronchitis, emphysema,
airways disease, etc.  Is it clear in the discussion how these health
endpoints are related or distinct entities?  If not, tell OSHA how to
improve the presentation.

Questions on Section E. Renal and Autoimmune Effects

E-1.	OSHA has preliminarily concluded that studies on renal and
autoimmune effects have not provided sufficient data (in quality or
amount) to support quantitative estimates of risk.  From your knowledge
and understanding of these studies, do you agree with OSHA’s
conclusion?  If you believe that there is sufficient data to support
quantitative estimates of risk, please tell OSHA specifically what data
you think could be used.

Questions on Section F. Physical Factors that May Influence Toxicity of
Crystalline Silica

F-1.	The draft “Physical Factors…” section contains a discussion
of proposed mechanisms of action by which silica causes silicosis and
lung cancer.  Does the discussion of the mechanism of action reflect the
most recent thinking on this subject?  If not, tell OSHA what the most
recent thinking is, as you understand it.

F-2.	The draft Mechanism of Action section contains a discussion on the
possible role of silicosis as a precursor or prerequisite for lung
cancer.  Does this discussion accurately reflect the current scientific
opinion on this issue?  If not, tell OSHA what the current scientific
opinion is on this issue, as you understand it.



F-3.	In the draft “Physical Factors…” section OSHA preliminarily
concludes that the available information on the physical factors that
may influence crystalline silica toxicity cannot be used at this time to
refine quantitative estimates of the lung cancer and silicosis
mortality.  Does this seem to be an appropriate conclusion based on the
information presented?  Is it clear in the discussion how OSHA arrived
at this conclusion?  Would you arrive at a different conclusion based on
the information presented?  If so, tell OSHA what your conclusion would
be and explain how you arrived at it?  



External Peer Review of OSHA’s Draft

“OSHA Preliminary Health Effects Section for Silica”

Peer Review Comments

Specific Questions

The draft health effects section reviews and summarizes a number of
studies.  Are the studies described in sufficient detail for the reader
to understand how the studies were done?  Are the descriptions of the
studies and results accurate?  Are the strengths and limitations of the
studies adequately discussed and appropriately evaluated?  Are OSHA’s
interpretations of the study results reasonable and explained
adequately?  Are discussions of studies clear and easy to follow?  If
not, let OSHA know what needs to be added to improve the information and
the manner in which it is presented.

Bruce Allen	In general, the descriptions of the studies appear to be
sufficient for understanding what has been done and I have no reason to
suspect that they have been misrepresented.  By way of more succinct
summary of strengths and weaknesses, I would suggest a summary table
cross-classifying the studies with major items that affect strengths and
weaknesses (e.g., study size, length of follow-up, issues related to
measuring effects, etc.).  This would allow readers to appreciate the
major differences among the studies and their relative strengths and
weaknesses for those issues that OSHA considers of primary importance. 
Just having such a succinct list of what OSHA considers important would
provide a more transparent means to evaluate those items and to
determine if that list corresponds to what informed readers would
consider important.

From a formatting perspective, it was sometimes difficult to follow when
discussion of one study or one topic ended and another began.  I would
suggest more extensive use of headers and sub-headers (bolded,
italicized or something, with perhaps section numbering and subsection
numbering) to delineate the sections and make it easier to find the
summary conclusions for each issue.

Murray Finkelstein	I think that the studies are, in general, well
described and that OSHA’s interpretations are reasonable and
adequately explained. The mass of text makes it difficult, however, to
locate specific studies and results. I think that the presentation could
be improved by including summary tables of results. As an example of one
possible layout, see the tables in the Text and Appendices of the
Institute of Medicine review “Asbestos: Selected Cancers”	ISBN-10:
0-309-10169-7 and  ISBN-13: 978-0-309-10169-1

Gary Ginsberg

Gary Ginsberg

	The descriptions of studies are well developed and provide good
background information for the risk assessment.  The conclusions about
silica-induced diseases and the variables that can lead to inconsistency
or modify dose response are generally well founded.  However, this
document is disconnected from the risk assessment in that the study
descriptions are not tied to the overall purpose of developing a
quantitative analysis.  Most study descriptions lack critical evaluation
of whether that particular study has sufficient exposure assessment,
numbers of subjects, outcome measures and quality control for inclusion
in quantitative dose-response models.  OSHA’s identification of
industries which are least confounded and thus most useful is helpful
and seemingly accurate.  However, for the most part, the critical
sorting stops at that level.   For example, it is unclear why the 1989
Danish stone cutter study is not incorporated into quantitative models
while other studies are.  The stone cutting and sand industries are
identified as having little confounding and studies from these cohorts
are not included in Section VI.  There may be good reason for that, but
without those reasons explained, one does not know whether the
quantitative assessment could be more robust or is in some manner biased
by the exclusion of useful cohorts.  The description of the Italian
brick workers (pg 138) is an example where OSHA discusses study
limitations that help the reader understand why it might not have been a
good candidate for quantitative analysis.  However, this is the
exception rather than the rule and it appears that OSHA is relying upon
study selection made by others (e.g., NIOSH, 2002; IARC 1997; the
various meta analyses; pre-existing quantitative analyses).  A
study-by-study critique (not just review, as is now presented) is needed
to sort out the value of individual studies for quantitative analysis
and to document that the array of studies relied upon in Section VI is
reasonably complete.     

Better use of tables would greatly facilitate use of this document. 
V-C-1 is a good start in terms of laying out key studies, but a 2nd
(perhaps 3rd as well) table is/are needed which log key information for
each study in terms of study design (cohort size, measurements made,
exposure range, etc.), results and most importantly utility for
quantitative analysis.  Also, tables would be more useful if organized
in a discrete section in the back of the report and if they were listed
in the TOC.

Brian Miller	In the main, the descriptions seem good.  I have annotated
some corrections, suggestions and comments by page number within each
section of this document.  There were some studies whose size made the
lack of statistically significant results almost a foregone conclusion: 
the concept of power of a study was not prominent, and could perhaps
have been emphasized more.

Andrew Salmon	The format used for descriptions of studies in the health
effects section is appropriate, and effectively lays out the content,
methodology, strengths and weaknesses of the studies described.  In
general the section is well written and clearly explains OSHA’s
interpretation of the available data.  There are a few cases where
studies have been either ignored completely, or given less consideration
than they deserve, (or, alternatively, over-emphasized), but these
difficulties are isolated and will be identified below where
appropriate.  The level of detail given in this section is sufficient
for its purpose, which is mainly a qualitative description of the health
effects of silica exposure: more details, including report of the
detailed quantitative evaluations of the studies by Toxichemica Inc.
(2004), are appropriately deferred to the section on quantitative risk
assessment.  The section represents a commendable attempt to reduce an
enormous literature to manageable proportions.  This is most successful
in the consideration of lung cancer, where the previously published
review by IARC (1997) and the meta-analysis by Steenland et al. (2001a)
provide an overall perspective.

Noah Seixas	In sum, yes.  OSHA has clearly and thoroughly evaluated the
extensive literature relating silica exposure to the key health effects
known or suspected to be related to this exposure.  The review is
extensive, covers all of the major studies to my knowledge, and provides
an appropriate level of discussion of each study.  In particular, the
methods, results, limitations and strength of the evidence are
considered for each study reviewed. Most of the document is clearly
written.

However, I think the document is lacking an adequate “Introduction.”
 The section labeled introduction is a detailed discussion of sampling
methods.  While this is relevant, I don’t believe it should be labeled
as introductory material.  In particular, I think the document as a
whole is lacking clear definitions of many of the concepts and
terminology used throughout the document.  There is even no clear
definition of the terms silica, crystalline silica, quartz, etc.  These
terms should be clearly defined at the beginning of the document, and
used consistently throughout.



The draft health effects section covers a selection of studies in each
topic area.  The discussion does not cover every study ever done on the
topics.  Do you think that OSHA should have discussed additional or
different studies on a specific health endpoint?  Are you aware of
significant studies that were omitted?  If yes, inform OSHA of those
studies.

Bruce Allen	I believe the selection of studies was adequate for the
purpose of conveying the important issues and conclusions about exposure
to silica.  I am not aware of any significant studies that were omitted.

Murray Finkelstein	The literature on silica health effects is
voluminous. I think that OSHA has identified and discussed the major
reports. I am aware of no significant studies that were omitted.

Gary Ginsberg	The selection of studies reviewed appears to be broad and
extensive.  Not starting from a background in silica
toxicology/epidemiology, I can’t really say whether important studies
have been omitted from this review.  As mentioned above, the selection
of studies for quantitative assessment needs better justification.

Brian Miller	No, I am not aware of other studies.  Those I was familiar
with were included, along with many others that were new to me.

Andrew Salmon

Andrew Salmon

	The discussion of silicosis morbidity and mortality starts from the
premise that silicosis is the inevitable and unique result of exposure
to respirable crystalline silica particles, and provides only a very
brief mention of some of the historical reports underlying this
conclusion.  It then proceeds to review further only those studies which
the authors consider to provide information on specific topics
qualifying that association, such as the precision, repeatability and
sensitivity of the ILO radiographic method for diagnosis, the
association with other diseases such as tuberculosis, the probability
and rate of progression after initial diagnosis, etc.  While this is
somewhat reasonable in view of the large literature, it results in an
unbalanced view of the literature and produces some odd results.  Thus
this section does not describe some of the key studies used for
quantitative analysis in the following section, or only refers to them
in the context of some subsidiary point, without laying out the main
findings of a dose relationship between exposure to respirable dust
containing crystalline silica and observation of silicosis as defined by
radiography, lung function and/or pathological criteria.  It would
improve the overall perspective of this part of the narrative if these
important studies were described here, in addition to the analysis of
surveillance data.  I understand the logic of keeping a discussion of
the dose-response characteristics to the following section where they
are treated in detail.  However this section would be improved by a
demonstration that such a relationship does exist.  It would also help
the discussion of important qualitative differences such as the
distinction between acute silicosis (a high-dose effect) and the
longer-term responses.

The surveillance data which are reviewed certainly support the
assumption of a relationship between silicosis and exposure to silica,
but provide a somewhat strange perspective due to their individual and
clinical, rather than statistical, perspective.  Additionally they may
seriously understate the prevalence of the disease.  Not only is
mortality a severe endpoint which may not affect more than a minority of
those affected in less extreme ways, but as succinctly demonstrated in
the analysis presented here is prone to substantial under-reporting. 
Much of the surveillance data cited here provides a limited perspective.
 For a start it is largely confined to United States data which of
itself ignores the international dimension of the problem, and may
produce other weaknesses resulting from the peculiarities of the United
States health care system.  In general, apart from special projects
following pre-defined cohorts, United States data are significantly
lacking compared to those in other countries (especially Scandinavia and
the EU) where population-wide health care systems exist.  With regard to
morbidity statistics, in the case of silicosis these are mainly related
to workers’ compensation systems, which include an element of bias
towards a negative diagnosis (thus avoiding payment) unless the evidence
is overwhelming: thus these reports are likely to show a high degree of
accuracy but low sensitivity compared to an analysis without this bias,
even if the same nominal criteria are used.

Noah Seixas	I am not aware of any significant studies on silica and
health effects that were not reviewed in this document. There are
studies which consider alternative dose metrics for silica, but these
are more appropriately addressed in the risk assessment document.  There
is a substantial literature on exposures, and exposure control
techniques for silica in various industries.  These studies are not
reviewed here, I think appropriately given the goals of the Health
Effects document.



The draft health effects section covers a number of health endpoints. 
Are you aware of additional significant health endpoints affected by
exposure to crystalline silica that should be discussed?  If so, inform
OSHA of the additional health endpoints and studies that address them.

Bruce Allen	I am not aware of other health endpoints that were omitted
from the discussion.

Murray Finkelstein	I am not aware of any additional significant health
endpoints.

Gary Ginsberg	The selection of health endpoints appears to be inclusive
and it is hard to imagine that other endpoints exist that would
materially add to the quantitative assessment.  However, my knowledge
base in this area is very limited.

Brian Miller	This section covered all the effects that I was aware of as
being potentially related to silica.

Andrew Salmon	I am not aware of any additional endpoints which should be
considered.

Noah Seixas	The document addresses silicosis, lung function limitation
including obstructive lung disease, lung cancer, cancers other than
lung, renal and immunological endpoints.  These are known health effects
associated with silica exposure.  I am not aware of other significant
health effects of silica exposure.



The draft heath effects section contains conclusions for each topic area
as well as an overall conclusion.  Are these conclusions reasonable in
light of the available data?  Is it clear in the discussion how OSHA
arrived at those conclusions based on the studies discussed?

Bruce Allen	It appears that OSHA’s conclusions are reasonable in light
of the data that they presented.  The basis for those conclusions does
appear to be clear.

Murray Finkelstein	The discussion is clear and the conclusions are
reasonable.

Gary Ginsberg	The conclusions for the various topic areas appear to be
supported by the data.  However, in some cases the conclusions are not
complete thoughts or do not tie to the quantitative assessment.  For
example, Page 396 states that many cases of silicosis may be missed with
standard radiographic and CT techniques.  This connects to the evidence
I alluded to above regarding false negatives being discovered when
autopsies are performed.  OSHA does not provide any perspective on how
false negative outcomes might influence their analysis.  On the next
page, associations between pulmonary function decrements and x-ray
evidence of silicosis are discussed with doubt raised about whether
functional declines can be seen at the earlier stages of pathology. 
While the conclusion drawn may be reasonable, how this makes a
difference to the quantitative assessment is unclear.  Does this affect
OSHA’s case definition of silicosis (e.g., only if there is associated
functional decline)?  Does it affect the way OSHA interprets the
epidemiology? Does OSHA have a degree of silicosis it is trying to
prevent – i.e., does the minimal ILO grade constitute an adverse
effect that is in need of prevention through regulation.  This decision
affects how the epidemiology is interpreted and risk assessment
conducted.  For example, if the dose response is from a study whose
sensitivity was for mild to moderate silicosis, OSHA may not want to
weigh that study as heavily as one with a more sensitive case
definition.  It would appear that the effort OSHA spends in Section V
(page 18 and further) to define silicosis would be a good place to
discuss how it approaches the issue of case definition when interpreting
the epidemiology and in building the quantitative risk assessment.

Brian Miller	In the main, I believe the discussions balanced the
strengths of the studies convincingly, although as mentioned above some
discussion of statistical power might have helped here.

Andrew Salmon	The health effects summaries are clear and reasonable.

Noah Seixas	Yes.  I think OSHA has made very reasonable summaries and
interpretations of the various aspects of health effects associated with
silica.  The literature reviews are comprehensive, clear and adequate
for making the determinations provided in this section.



A-1 	Is OSHA’s review of the history of dust exposure monitoring
helpful to the reader in understanding the needs and limitations of
exposure conversions to a modern mass per volume metric in reference to
the exposure-response relationship?

Bruce Allen	Yes, this section is important for understanding the
uncertainties that are associated with exposure reconstruction and the
conversion factors (and their potential biases) that might be present in
that reconstruction.

Murray Finkelstein	OSHA’s review is helpful in providing a historic
perspective and in providing context for the retrospective-prospective
epidemiologic studies.

Gary Ginsberg	The introductory material regarding dust exposure
monitoring is useful for understanding the epidemiology, how the quality
of studies has changed over time, and how certain measurement variables
can be a source of uncertainty.

Brian Miller	Given the disparity in measurement methods over the years,
this is necessary.  However, I’d suggest not describing these as
exposures, but as concentrations or intensities, and reserving the word
‘exposure’ for subjects’ interaction with those intensities, i.e.
including duration.  This is a subtle distinction, but a useful one to
maintain.  It is exposure, not concentration, that engenders risk.    

P5  para 2 Replace ‘measure of exposure’ with  ‘airborne
concentration’.  

P8 para 3 spans -> span.  Give ref(s) for methods?

P12 para 1 last sentence contains extra text.

P14 para 1 explain more fully what ‘respirable’ means?

Andrew Salmon

Andrew Salmon

	This section provides a summary of the major types of measurement
technology which have been used for estimating exposure to respirable
silica.  While accurate as far as it goes, this provides a rather
limited perspective.  OSHA decided some time ago that the ideal method
for measuring exposures to silica dust involves personal sampling using
a cyclone to remove non-respirable particles and a filter to collect the
respirable particles for gravimetric and X-ray crystallographic
analysis.  There are good reasons for choosing this methodology, but
this section does not do a very good job of explaining what these are. 
The section is somewhat dismissive of the various methods based on
particle counting equipment, characterizing these as “historical”
and mainly addressing how their results could be converted to the
OSHA-preferred gravimetric data.  It is not that the gravimetric measure
is intrinsically technically superior, but rather that the availability
of particle counts and size- or surface-area measurements as well as
gravimetric analyses made an important contribution to the eventual
conclusion that 1) long-term cumulative exposure was the chief
determinant of chronic silicosis and 2) the gravimetric measure is the
appropriate dose metric to be related to health effects.  It is curious
that the authors of this section chose to report conversion factors used
in the analysis of various US studies (Table V-A-1), but did not include
any quantitative analysis of the various reports on dust exposures in
the South African gold mines, which represent an important cohort for
study of health effects and for which there are a number of publications
presenting such comparative analysis (Beadle and Bradley, 1970;
Page-Shipp and Harris, 1972; DuToit 1991 and others).  I also do not see
any description of modern electronic particle-counting systems, some of
which are capable of providing a gravimetric description of the
distribution of particle sizes in an atmosphere, on a continuous basis. 
While these may be unsuitable for personal monitoring of workers, they
may provide important data especially for outdoor environments where
dust exposures are a concern.  More detailed consideration of particles
size and surface measures might further inform the later discussion of
factors potentially affecting the potency of specific types of silica
dust.

Noah Seixas

Noah Seixas

	This section of the document is important, given the historical changes
in dust monitoring techniques that have occurred, and the relevance of
these changes to estimating dose for chronic conditions associated with
silica.  I think that this discussion is important to the integrity and
interpretability of the document as a whole.  However, I find this
particular section poorly organized and presented. I think it could be
presented in both simpler and clearer terms.  

In particular, there are several issues within the context of
measurement technique and definitions, that are frequently confused. 
First, the duration of exposure monitoring should be discussed.  Grab
samples, or instantaneous samples should be distinguished from
short-term samples, from full-shift samples (called erroneously, TWA
samples).  Further, the duration of measurement is frequently confused
with the exposure metric – typically cumulative exposure.  These are
not the same issue and should not be confused.  Although some direct
reading instruments (DRI) are now available, they are not discussed. 
Although DRI are generally not relevant to historical exposure datasets,
they are increasingly in use and may be important in future studies –
their use and limitations should at least be addressed briefly.

Second, particle counts vs. mass concentrations should be discussed –
as it is.  The discussion of the conversion of mppcf to mg/m3 is useful
and thorough.  However, the rationale for the superiority of mass
concentration is not addressed adequately.  My understanding is that the
choice of mass fraction is primarily dependent on the relative
simplicity (and lower expense), rather than a biological
appropriateness.  While mass fraction may be a superior metric
biologically, the evidence for this is not presented. This issue should
be considered separately from how to convert between the two measures of
exposure.

Third, the rationale and definitions of particle size fractions are not
well described.  Because different studies, and different periods in
time have used different particulate size fraction measurement systems,
and because they may have different implications for different health
endpoints, this should be addressed more clearly.

Fourth, the differences in measurement of dust, vs. silica, vs.
crystalline silica should be more clearly explained.

Fifth, in the context of describing each of these issues, the use of
different sampling and analytic devices is also relevant and should be
described.  Sampling methods are substantially different over time and
between different countries.  Each of these sampling devices or
techniques have implications in terms of the issues discussed above, and
are therefore, relevant to the comparability of the data derived from
them. 

Finally, the choice of a standard based on crystalline silica
concentration in air, rather than the currently used formula combining
mass concentration and silica content of the dust should be explained. 
I believe this is a defensible approach, but given its shift from
historical practice, it would be useful to point out the limitations of
the earlier standard, and OSHA’s position on the superiority of the
new method of quantification.



B-1	OSHA presents background information on the use of radiography,
imaging techniques, pulmonary function studies, and other diagnostic
tools used in clinical settings and for in epidemiological studies. 
Does OSHA correctly present the benefits and shortcomings of these
techniques?  Are you aware of any additional information that would
either strengthen this discussion or contradict what is written.

Bruce Allen	This is not my area of expertise; I am not able to assess
whether or not the presentation correctly assesses benefits and
shortcomings of the techniques.

Murray Finkelstein	I think that the section on Diagnostic Tools requires
revision. In particular:

CT and HCRT: The study by Begin needs to be revisited by OSHA. There
were only 58 subjects and only 6 did not have radiographic silicosis.
The relevant question is: does CT find silicosis when X-ray does not?
There were 6 in Category 0 by X-ray and 12 in Category 0 by CT.

The Talini paper studies only 27 subjects. There was substantial
disagreement between CT and CXR at low profusions.

I think that the literature search needs to be updated for the
comparison of CT and CXR. The references in the document are old and my
own search on PubMed found additional references.

Gary Ginsberg	I do not have the technical expertise to comment on
whether OSHA descriptions of radiological and pulmonary function tests
are accurate.

Brian Miller

Brian Miller

	I’d like to see the description of the ILO conventions for describing
the abnormalities on chest x-rays spelt out early, around PP14-16,
because that informs a large part of what we mean by ‘silicosis’. 
It appears too late at present.

I’d also suggest that this section might give more of an impression
that ‘chronic’silicosis, like all pneumoconiosis, is an evolving
damage process and that decisions about when a subject has the
‘disease’ involve arbitrary distinctions of severity along a
continuum.  This would also be a good place to have the (currently
later) material on reader variation. 

P18 para 4 The ILO system standardizes the description of chest x-rays,
not their ‘interpretation’.

P19 para 4 however -> but

P23 para 2.  The authors used multiple regression, not multivariate
analysis.

P32 para 1.  kappa  0.49 may be ‘better’ than 0.29, but it’s still
not very good.  Note kappa does not inform about the direction of
disagreement.  Might the relationship between HRCT and ILO be shown in a
table?

P34 para 1.  Note other lung function measurements can be very variable
too.

P35 para 2  Note that values below the 5th percentile may in fact be
perfectly normal?

P37 para 1 (and/or elsewhere in this section).  ?Make more explicit that
the high inter-individual variation in lung function values implies high
variation in %predicted, and that is why (1) %predicted should be used
only as an initial screening measurement and (2) rate of loss over time
in an individual is a much more sensitive marker.?

Andrew Salmon	This chapter provides a good summary of the available
methods for recognition and evaluation of silicosis and related lung
diseases.  It makes an important point about the relatively low
sensitivity of radiography, as shown by Hnizdo et al. (1993), Craighead
and Vallyathan (1980) and others.  This point is important in evaluating
the prevalence of silicosis as reported in the various epidemiological
studies considered.  The discussion of the lung function test results in
patients affected by silicosis (however diagnosed or suspected) is also
useful.

Noah Seixas	The discussion of diagnostic tools, X-ray, CT, HRCT and
pulmonary function tests is appropriate, clear and adequately complete
for assessing their relations to silica-induced disease.  The discussion
is useful in terms of understanding the literature on silica risks,
however, there are no summary or conclusions with based on this
discussion that could be used for designing screening or surveillance
systems for silica-exposed workers.  Presumably, this discussion would
appear elsewhere before OSHA enters into recommendations for medical
surveillance requirements.



B-2	OSHA attempted to describe the available silicosis surveillance
data.  In your opinion, were any important topics left uncovered?  Any
data sources not discussed?  Is the subject of under-reporting
sufficiently covered?  Are the conclusions supported by the discussion
and stated succinctly?  Are you aware of any important studies in this
area that OSHA did not cover?

Bruce Allen	I am not familiar with surveillance data collection and
reporting so I cannot evaluate whether any important topics/data sources
have been left uncovered.  It does appear that the subject of
under-reporting has been sufficiently discussed.  It appears the
conclusions are supported by the data presented.

Murray Finkelstein	I think that the section on surveillance was
adequately done. I am not aware of overlooked studies.



Gary Ginsberg	The surveillance data review and underreporting issue
appear to be adequately described and of relevance to the remainder of
the document as general background information.

Brian Miller	I’m not close to these schemes, so I don’t have any
others to suggest.  The topic of ‘under-reporting’ is linked to the
definition of the minimum severity required to classify a subject as
having silicosis, and that might be brought out a little more clearly.

P48 para 1.  Presumably deaths in the young are from ‘acute’
silicosis?

P48 para 2.  I think YPLL is a misnomer, and capable of
misinterpretation.  Why not Years of Working Life Lost, YWLL, which is
what is meant.

P49 para 2.  Reword last sentence?  Workers in industries without silica
exposure are not likely to die of silicosis at all.

P52 para 2  Define ‘latency’?

P56 para 3.  If the three states are getting their silicosis cases from
hospital discharge records, presumably they are at the more serious end
of profusion or ‘acute’, or…, certainly not matching a definition
of ILO categories 1/0+.  There could be thousands of case that meet the
ILO definition that never go to hospitals. This highlights the necessity
of keeping it clear that not all uses of the word ‘silicosis’ mean
the same thing.     

P61 para 1 ‘Silicosis’ at ILO 1/0 is unlikely to be symptomatic. 
Such cases are only found if actively sought, e.g. with a radiographic
workplace surveillance program.  The conclusion in P62 para 3 is
therefore not at all surprising.

P68 para 3 Says there is no system that collects silicosis deaths, then
describes NORMS, which (I thought) does just that. But it’s true that
no system collects new live cases.

Andrew Salmon

Andrew Salmon	This description of surveillance data is sufficient as far
as it goes in describing the U.S. data.  There may be reasons why other
sources of this type of data, which certainly exist in various parts of
the world (U.K. in particular, many others), were not considered, but
these reasons are neither presented nor defended.  Specific problems
relating to under-reporting are identified and discussed.  However, as
noted previously in the general comments, the limitations in the
surveillance approach generally and in particular the concentration on
only the most severe effects (mortality, substantial disability and
high-grade radiographic diagnoses) limit the usefulness of this data
source for anything beyond the qualitative identification of disease
end-points.

There is an additional, recently published study that OSHA should
consider, which is related to those using surveillance data by its
objective of examining the prevalence of silicosis in certain population
groups.  This is the study of pneumoconiosis in Californian agricultural
workers exposed to silica and silicate-containing dust, by Schenker et
al. (2009).  Instead of using reported illness or death certificates,
these authors used histological analysis and measurements of accumulated
particles, using lung specimens obtained at coroner’s autopsy. 
Significant degrees of pneumoconiosis and other pulmonary disease
indications were apparent in agricultural workers compared to others not
exposed to the mineral dusts prevalent in dry farming activities.  This
result is interesting both as an indicator of silicosis-type disease in
an occupational group not considered in the OSHA report and as a
possible contribution to the debate about aged vs. fresh silica
particles, surface occlusion by clays and so on.

Consideration of studies from outside the U.S. describing prevalence of
silicosis is entirely lacking: this is a substantial omission if the
objective was to provide a balanced view of the subject.  Some of the
key studies from the U.K., Asia, South Africa etc. are described in the
subsequent discussion of silicosis progression in this section, or in
the dose-response section, but their omission here is peculiar.  And
some important studies are omitted in both places.  It is a downright
eccentricity that the authors refer briefly to the study by Churchyard
et al. (2004) in the lung cancer chapter as a source of gravimetric
measurements of silica in South African mines, but this study is nowhere
mentioned either in this section or the dose-response section for its
presentation of data on the prevalence and dose-response of silicosis in
black South African miners, which is its primary subject.

Noah Seixas	The surveillance information presented is an excellent
summary, covering silicosis mortality and morbidity, silica exposure
surveillance, and estimates of under-reporting of these data.  Lessons
learned from these data systems, along with their limitations, are well
discussed.  However, one comes away with an important question.  Given
the rapid and consistent reduction in mortality, and the continuing
exposure levels over the current PEL, one must ask if the remaining
disease observed is not completely due to illegal (i.e., > PEL)
exposures, rather than the inadequacy of the current standard.  I think
some additional discussion of this, in particular the limitations in
assessing non-silicosis outcomes (e.g. COPD and Lung CA) in relation to
silica exposures, is warranted.  That is, just because there is a large
reduction in silicosis identified through the various surveillance
systems, does not suggest that there is little remaining risk, other
than that where exposures exceed the standard.

Given the limitations of the various surveillance data sources (both
exposure and outcomes), it should be pointed out that the best
information about the true risk of disease must come from well
constructed epidemiologic studies where the cohort can be adequately
defined, exposure can be sufficiently quantified, and outcomes can be
thoroughly ascertained.  

On p 63, I assume the quote from Windau should read…”pneumoconiosis
due to other inorganic dusts.”



B-3	For the discussion of progression of silicosis, OSHA did not review
every study that addressed this issue.  Are you aware of additional
studies that OSHA should consider, especially any studies that
illustrate progression in a cohort exposed to silica in the range of
OSHA’s current PEL (0.1 mg/m3)?

Bruce Allen	I am not aware of any such studies.

Murray Finkelstein	I am not aware of additional studies that OSHA should
consider.

Gary Ginsberg	Not having background in this area, I cannot offer any
additional studies regarding progression of silicosis.  This section is
of general relevance but its unclear how it assists with the
quantitative assessment in Section VI.

Brian Miller	I’m not aware of any other studies that should be
considered.

P79 para 5  Is it worth pointing out that the problems in the Scottish
colliery were spotted as a rapid and atypical burst of radiological
progression?  If so, reference Seaton et al (Lancet, 1981)?  

P80 para 2 ‘occupational history since leaving the colliery’ was
particular to the 1990/91 follow-up survey.  (regular PFR surveys did
not include retirees.)

P81 para 1.  delete ‘, as well as the mean and maximum of the non-zero
exposure estimates’?  (That looks out of place.)

P83 para 4.  Should that say ‘it progresses quickly once
established’?

P84 para 4 silicosis defined how?

P102 para 3 presumably the incidence rate what was compared, not the
ratio.

P105 para 2 odd -> odds

P108 para 3 where part of the study cohort was exposed…for a period of
some years.

P109 What does ‘sufficient data were not provided’ mean?  Should
this be ‘there have been no follow-up surveys of this cohort since
1991’?  

P109 para 2 determinate -> determinant

Andrew Salmon	This part of the report does a good job of distilling what
is certainly a very large body of published data.  The key point, that
progression occurs both in the presence and absence of continuing
exposure to crystalline silica dust, is well made.  I do not have any
suggestions for additional studies that should be considered.

Noah Seixas	No.  the discussion is well organized and to the extent of
my knowledge, accurate and complete.  The conclusions concerning
progression, and determinants of progression seem well founded.



B-4	Results of studies covered in the PFT section are not always
consistent.  Did OSHA do a fair job of presenting these results and
drawing conclusions?  Are there any important studies that you are aware
of that would strengthen this section or the conclusions of this
section?

Bruce Allen	I am not familiar with the literature on silicosis and PFT. 
However, based on the results of the studies as presented in the
document, it appears that the conclusions reasonably reflect the
information that is available, including the uncertainties associated
with sometimes inconsistent findings.

Murray Finkelstein	OSHA did a fair job of presenting the studies.
However, most of these studies were small, and of low statistical power.
I suggest that OSHA consider pooling the results (meta-analysis) of
comparable studies in order to increase the ability to draw conclusions
from the data.

I am not aware of important omitted studies.

Comments with respect to: PRELIMINARY CONCLUSIONS: SILICOSIS AND DISEASE
PROGRESSION (Pg 107)

Pg 111: “Both of these studies report progression in terms of
percentage of workers whose chest x-ray categories progress, and their
results are close (2 percent to 6 percent). Based on these data, one can
estimate an annual progression rate of 2-6 percent for this exposure
range. This means that every year approximately 2 to 6 percent of
workers exposed to respirable silica dust in the range of 0.12 to 0.48
mg/m3 are expected to progress at least 1 sub-category on the ILO
scale.”

This statement omits to mention one crucial factor: latency. Progression
will depend upon the time from onset of exposure. It is almost certainly
not true, for example, that in the first 10 years of exposure every year
approximately 2 to 6 percent of workers exposed to respirable silica
dust in the range of 0.12 to 0.48 mg/m3 are expected to progress at
least 1 sub-category on the ILO scale.” The OSHA authors need to bring
the time frame (latency) into the discussion.

Gary Ginsberg	The pulmonary function test results section appears to be
a reasonable discussion on the evidence surrounding PFT changes and
radiographic evidence of pathology.  I have no studies to add.

Brian Miller	This section might go a little further to emphasize how the
inter-individual variation on lung function measurements between healthy
subjects of the same height and age makes it difficult to detect
abnormality cross-sectionally.  The idea of the lower 5% being
automatically accepted as ‘abnormal’ is reported rather
uncritically.

Andrew Salmon	I think this section is reasonable and well-considered. 
The conclusions drawn are cautious and defensible.  Given the complex
and conflicting nature of the data I think this chapter does an
excellent job of the analysis.  I am not aware of any additional
relevant studies (although I have to say that since I am not an expert
in clinical lung function testing I defer to others for a definitive
judgment on that point).

Noah Seixas	My reading of this presentation would suggest that OSHA has
been overly conservative in it’s conclusion concerning the association
of pulmonary function with the progression of silicosis.  Independent of
the role of emphysema, the literature supports a finding of decreased
lung function with progression of radiological silicosis.  The evidence
is most convincingly presented by the strongest study designs – those
with longitudinal follow-up of the cohorts.  The statement on page 101,
that, “it is plausible that exposure to silica impairs lung function
in at least some individuals at an earlier point in time than can be
detected on chest radiograph,” is unduly conservative.



C-1	With such a large amount of epidemiological literature to review,
OSHA may be subjected to criticism of selection bias.  Is the selection
criteria used by OSHA clear?  Have we missed any important studies that
have good or at least reasonable exposure estimates?  Do you recommend
any alternate or additional selection criteria?

Bruce Allen	The selection criteria are fairly well laid out on pp.
237-238.  It might have been better to establish the criteria “up
front” in Section C so that as the reader is considering what is
written about each study, s/he can have those criteria as a frame of
reference.  Here, as elsewhere in my comments, I suggest that a summary
table showing how the studies compare to one another with respect to
these criteria would be a useful addition.

Murray Finkelstein	I believe that the selection criteria used by OSHA
are clear. I am not aware of any important studies that have been
missed.

Gary Ginsberg	The selection of studies for this review is described on
Page 115, bottom paragraph.  This description is vague and doesn’t
really describe a selection process with defined criteria.  Rather, it
just states what OSHA reviewed and how this builds on previous reviews. 
If there were selection criteria for inclusion in the review or in the
quantitative analysis, this should be made explicit.

Brian Miller	This section of the literature has been reviewed fairly
extensively, and the criteria for inclusion seem reasonable.

Andrew Salmon	This section does a good job of presenting an overview of
the large and complex topic, which is helped by the earlier reviews by
IARC and NIOSH. Unlike the previous section which dealt with silicosis,
this section does present an overview of the various studies not only in
the U.S. but worldwide.

I do not have any specific suggestions as to additional reports to
include.  OSHAs independent review is certainly comprehensive as regards
both the inclusion of relevant studies, and the breadth of different
industries with silica exposure which were considered.  In response to
the question about selection criteria it would appear that OSHA has
taken the inclusive approach of at least describing any possibly
relevant studies, although obviously giving weight in the final analysis
to those offering reasonable determinations of exposure to crystalline
silica, without complicating concurrent exposures to other potentially
carcinogenic materials.

Noah Seixas	While I believe that the review conducted is comprehensive,
and to the degree that I am aware, has not missed any significant
studies, I don’t see a clear presentation of study selection criteria
for the studies reviewed.  I think it would be useful to state in the
beginning of this section how individual studies were selected. 
Further, I would be useful to identify any studies that were not
included and provide the rationale for why they were not.



C-2	OSHA concludes that the studies involving four cohorts among the 20+
cohorts and case controls reviewed were the strongest in terms of
several criteria in record availability and data quality (exposure,
smoking, worker and facility history) as well as data treatment and
analyses?  Do you agree with this conclusion?  Would you add or delete
any of these studies?

Bruce Allen	This question basically follows on from the question above. 
And my answer relates to two studies that were reviewed but excluded,
for reasons that are not entirely clear.  The Brown and Rushton study of
British industrial sand workers was apparently not selected because of
“the low cumulative exposure profile” (p. 179, line 8).  A similar
comment is made about the Ulm et al. study of German stone and ceramic
workers (p. 191, line 4).  There is some discussion of other limiting
factors for these two studies, but it is not clear how those limitations
relate to the selection criteria on pp. 237-238.  In particular, I would
argue that inclusion of studies with low exposures as part of a
dose-response analysis is very useful precisely because such studies
help constrain and define the low-dose (low-exposure) shapes that can be
considered to be consistent with the data and which are particularly
relevant for determining how low exposure must be to be consistent with
non-significant risk levels.  

It may be that these two studies have some fatal flaws that the selected
studies do not have, but that is not made clear enough.  In particular,
it needs to be shown that, based on the selection criteria listed, these
two studies are not adequate (or the selection criteria need to be
expanded or refined to truly reflect the selection process that OSHA
used).  Again, a summary tabulation of the studies cross-classified
against the criteria would facilitate such an assessment.

Murray Finkelstein	I believe that, given the abundance of studies, it is
appropriate to weight most highly those studies with the least
confounding (due to the presence of other known carcinogens or the
absence of smoking histories). There has been discussion of
heterogeneity. Little mention was made of exposure misclassification as
a cause of heterogeneity. Of all the silica studies, I have seen the raw
exposure data only for the Vermont granite sheds. In this data set there
are many years in which there was no sampling at all, and a number of
jobs for which very few dust measurements were obtained. It is my
opinion that these data are sparse and much interpolation of uncertain
data is required. These data can certainly be used for quantitative risk
assessment, but the resulting risk estimates will have a large measure
of uncertainty attached. I would not be surprised if the same were true
of most of the other cohorts. Nevertheless, I believe that it is
important to make the best use of whatever data are available. I thus
support the pooling of data from individual studies. I recognize that
the pooling of poor quality data will tend to obscure any relations
between exposure and response. Should an exposure-response relationship
emerge from these analyses, despite the weaknesses, then I believe that
this supports the existence of a real association between exposure and
disease.

Gary Ginsberg	I agree that the 4 industries identified by IARC 1997
(Page 114) and highlighted by OSHA appear to be the most useful for
studying the effects of crystalline silica on workers.

Brian Miller	I agree that these appear the strongest at the time of
review.  There is a new paper on mortality in British coalminers that
has results on respirable quartz and lung cancer, just available online.
 This has among the best-characterised exposures of any studies.

Miller BG, MacCalman L. (2009). Cause-specific mortality in British coal
workers and exposure to respirable dust and quartz. Occup. Environ. Med.
(Published Online First: 9 October 2009. doi:10.1136/oem.2009.046151.  

The above paper is summarised from:

Miller BG, MacCalman L, Hutchison PA. (2007). Mortality over an extended
follow-up period in coal workers exposed to respirable dust and quartz.
Edinburgh: Institute of Occupational Medicine.  (IOM Report TM/07/06).  
 HYPERLINK "http://www.iom-world.org/pubs/IOM_TM0706rev.pdf" 
http://www.iom-world.org/pubs/IOM_TM0706rev.pdf 

Andrew Salmon

Andrew Salmon

	I’m not sure exactly which part of the report this charge question is
referring to – the chapter on lung cancer reviews about one hundred
studies in total, with 23 occupational cohort studies, 5 case-control
studies and 5 silicosis cohorts analyzed in detail.  The preliminary
conclusions (on page 237) cite five cohort studies as being “of the
best quality to evaluate exposure to crystalline silica as a risk factor
for lung cancer”.  These were 

Diatomaceous Earth Workers (Checkoway et al. 1993, 1996, 1997 and 1999;
Seixas et al. 1997)

British Pottery Workers (Burgess et al., 1997 and 1998; Vherry et al.,
1998; McDonald et al., 1995)

Vermont Granite Workers (Applebaum et al., 2007; Attfield and Costello,
2004; Costello and Graham, 1998; Davis et al., 1983)

North American Industrial Sand Workers (Hughes et al., 2001; McDonald et
al., 2001, 2005; Rando et al., 2001)

North American Industrial Sand Workers (Sanderson et al., 2000;
Steenland and Sanderson, 2001)

This list certainly identifies a selection of key studies with adequate
statistical power and reliability, and reasonable exposure and health
endpoint determination.  It seems to me that the omission of the South
African gold-mining studies is a mistake, particularly with regard to
the more recent reports (Hnizdo and Sluis-Cremer, 1991; Hnizdo et al.,
1997), about which OSHA comments somewhat favorably (page 236).  While
there are some questions about the exposure measurements used and
possible confounding by radon exposure it appear that most of these have
been addressed by related studies.  This group of miners has been an
important source of data on silica effects over the years and
historically these studies have contributed greatly to the worldwide
concern about, and understanding of, silica exposures and health
effects.  The fact that a modern epidemiological study finds a positive
association with lung cancer in this group is an important and
persuasive part of the whole jigsaw-puzzle.  In the last analysis
however the most persuasive part of the argument is not any one of the
single study reports, but the meta-analysis by Steenland et al. (2001a).

Noah Seixas	I think the selected studies are among the most useful and
complete available. Again, I don’t think that OSHA has clearly stated
the criteria for inclusion, and demonstrated that the selected studies
did conform to these criteria, or that they were necessarily superior
with respect to the criteria than other studies which may not have been
included.  

Further, OSHA has selected 5 studies in four industries as the most
useful studies available.  While the attributes which are used to make
this judgment are provided, it is not until all the other studies are
reviewed that these criteria a stated (on page 237-8).  Stating these
criteria, and reviewing the literature with them as an explicit goal
could make a more convincing argument that the best studies were in fact
selected.



C-3	Do the tables add to the text presentation?  Do you have any
recommendations to improve their content or presentation?  Would you add
or delete another table?  Provide an explanation for any proposed
deletion or a mock-up of any proposed addition.

Bruce Allen	The inclusion of tables appears to be rather hit-or-miss. 
For some study summaries a table is given, while for others the odds
ratio results are just presented in the text.  I would rather see a more
consistent presentation and would tend to prefer one that includes more
tables.  This should enhance the comparability of the assessments if the
tables are as consistent as possible across studies.  I have already
mentioned that summary tables of studies cross-categorized against
selection criteria would also be a useful addition.

Murray Finkelstein

Murray Finkelstein	I like Table V-C-1.

Tables V-C-2, V-C-3, and V-C-4 are orphans. It is not clear why results
from only these studies should be presented in tabular format. I would
delete these tables.

Additional Tables:

1) I think that a summary Table describing the characteristics of each
study, such as:

Type of industry

Cohort or case-control

Other confounders?

Adjustment for smoking?

Quantitative exposure estimates?      would be very helpful as an
overview.

2) I think that much of the material in the text could be usefully
summarized in results tables. These would replace the tables mentioned
above and summarize all the data in one place.

As an example of one possible layout, see the tables in the Text and
Appendices of the Institute of Medicine review “Asbestos: Selected
Cancers” ISBN-10: 0-309-10169-7 and  ISBN-13: 978-0-309-10169-1

(  HYPERLINK
"file:///\\\\ERG-LEX-B\\DATA3\\SHARED\\conf\\Projects\\OSHA%20Silica%202
014\\2014%20Peer%20Review%20Report\\(http:\\www.nap.edu\\catalog.php%3fr
ecord_id=11665)"  http://www.nap.edu/catalog.php?record_id=11665) 

Gary Ginsberg	See my previous comment (under general comments) regarding
potential improvements to the tables.

Brian Miller

Brian Miller	P113 para 1 delete ‘of’

P129 para 5 i.e. -> e.g.

P130 para 2 participants -> participants’

P141 para 2 base -> based

P141 para 2 Is it correct to describe an ‘elevated risk’ for gastric
cancer with a lower CL of 78?

P142 para 1 The cohort comprised white…

P144 para 1 …from regression models including …

P144 para 1 …exposure to dust could cause death from COPD only …

P156 para 5 …mineworkers who were employed…

P164 para 4  I don’t accept as valid a test of significance for a
pneumoconiosis SMR.  The general population does not contract any
pneumoconiosis, so arguably an SMR for this cause is meaningless;  the
implied null hypothesis is untenable.

P169 para 2 …quartile analysis of average respirable silica
concentration…

Table V-C-2 Average exposure -> Average concentration (throughout)

P183 para 4.  Confusing.  If ‘observed number of cases differed
significantly from expectation’, why is it that the ‘SMR based on
local rates barely missed being statistically significant’?  Are these
not the same, or is one using a different set of comparison rates?

P199 para 2 ‘decreased’ does not seem a good summary of the pattern
here.

P200 para 1 Should ‘a modest non-monotonic trend’ not read ‘no
trend’?  Presumably the ORs are in comparison with the lowest exposure
group, assumed at 1.0.  Did this really yield a p-value of 0.007?

P201 para 2 Follow-up was to 1996, not 1966.

P213 Table V-C-4  The column headed ‘Low/no’ puzzled me, because in
a case-control study it’s common to compare other exposure groups with
the lowest, and set this to an arbitrary level of 1.  However, the
column is simply mislabelled, following a typo in the original paper. In
Table 4 of Calvert et al (2003) the column is headed ‘Ever* low/no’,
and the asterisk leads to a footnote ‘*Ever =medium, high, and super
high categories combined’.  This makes it clear that this column is an
average OR for the exposed categories, and the heading in Table 4 should
have been rendered as ‘Ever* v low/no’.  Perhaps here the heading
should be ‘Ever v low/no’.

P213 Table V-C-4.  The ‘trend’ for lung cancer is not very
convincing, and the increases claimed are not large; a reminder that any
difference can be declared statistically significant with large enough
numbers.

P215 para 3  Variables are not ‘correlated in models’;  their
effects may be, i.e. demonstrate confounding.

P217 para 2.  ‘Gender’ (i.e. sex) was matched for. The analysis
method was unconditional; was gender not adjusted for?

P220 para 1 …range of relative risk estimates was…

P220 para 2 SIR is not an ‘incident rate’ but a ‘Standardized
Incidence Ratio’.

P223 para 1 …supplemented by stepwise multiple regression analysis
using proportional hazards modelling.

P223 para 2 SMRs are ‘Standardized Mortality Ratios’, not
‘standard mortality rates’.  The notions that SMRs were
‘elevated’, ‘there were trends’ that ‘became more
pronounced’, but then none of this was statistically significant,
seems like serious over-claiming.  Another way to put it is that there
was no convincing evidence of a relationship.  

P224 para 2 explain what the ‘computer simulations’ were doing here?
 Trying to compensate for an absence of data?

Andrew Salmon	The tables are fine as far as I am concerned

Noah Seixas	Yes, however, there is no table summarizing the strength of
the evidence.  Although it might be more appropriate in the Risk
Assessment section, it might be possible to construct a table
identifying the strengths or weaknesses of each reviewed study (e.g.,
size of cohort, strength of exposure characterization, number of cases,
control of confounding, etc.) and the summary of results in terms of RR
or OR, etc. Such a table could provide an easier way of summarizing the
strength of the evidence, unless the permutations become too unwieldy
for a simple table.



C-4	OSHA has reviewed a number of studies that examined the
relationships between exposure to silica and cancer at sites other than
the lung, including larynx and nasopharynx, stomach, and esophagus. 
Have we missed any important studies of cancer at other sites? If yes,
inform OSHA of those studies.

Bruce Allen	I am not aware of any other studies of cancer at other
sites.

Murray Finkelstein	I am not aware of any important studies that have
been missed.

Gary Ginsberg	I am not aware of any additional cancer sites or studies
suitable for this review.

Brian Miller	I’m not aware of any additional studies regarding these
or other sites.

P252 Something wrong with the reference dates?

P258 para 3 …greater than one hundred times…

Andrew Salmon	I am not aware of any additional studies that should be
considered.

Noah Seixas	I am not aware of any other cancer sites of importance with
respect to silica exposure.



C-5	OSHA has preliminarily concluded that an association has not been
established between silica exposure and excess mortality from cancer at
sites other than the lung.  Do you agree with this interpretation of the
data?  If not, please provide a detailed rationale for your different
interpretation.

Bruce Allen	From the data that have been presented, I would agree that
the risk of cancer at sites other than the lung appears to be much less
well-established and probably less in magnitude (if they exist at all),
compared to the risk of lung cancer.  For that reason I can support
OSHA’s implied decision not to pursue a risk assessment based on
cancers at other sites.

Murray Finkelstein	I agree that definitive associations between silica
exposure and excess mortality from cancer at other sites have not been
established. I think that a potential association between silica
exposure and stomach cancer is provocative, and additional work should
be conducted exploring this connection in a quantitative fashion.

Gary Ginsberg	I agree that the data describing cancer at other target
sites are too limited and in some cases too contradictory to allow OSHA
to make these sites a focus of the risk assessment.

Brian Miller	I agree that the evidence is not convincing for increased
risks at any of these other sites.

Andrew Salmon	I think that is a reasonable conclusion.  There may be
grounds for suspicion of such effects, but they do not reach the level
of certainty to establish such a connection.

Noah Seixas	Given the studies reviewed by OSHA, I would agree with the
assessment that these other cancer sites are not clearly associated with
silica exposure, and if there is an underlying association, there are
insufficient data to provide any strong conclusions at this time.



D-1	This draft section discusses COPD, NMRD, bronchitis, emphysema,
airways disease, etc.  Is it clear in the discussion how these health
endpoints are related or distinct entities?  If not, tell OSHA how to
improve the presentation.

Bruce Allen	It is not clear from the discussion exactly how these health
endpoints are related to one another.  I am not sure it really matters
that much, since OSHA’s conclusion is that silica exposure (probably)
affects all of them, but a simple flow-chart or other visual, graphical
display of the relationships should suffice.

Murray Finkelstein	Pg 264.  Chronic bronchitis and emphysema are NOT
either/or.  Both conditions may be present together.

Gary Ginsberg

Gary Ginsberg	The non-malignant respiratory disease (NMRD) section reads
well and makes a number of important points regarding the ability of
silica to induce these diseases without the need for silicosis to occur.
 Once again, this information is not tied to the risk assessment, in
particular, how the additional disease entities may add to the number of
silicosis cases or add to the cumulative respiratory disease burden.  In
this regard, the OSHA assessment has the potential to underestimate
silica-related non-cancer risks.

D-1 – OSHA clarification request to Ginsberg: You state that OSHA has
not tied the information on NMRD to the risk assessment, in particular,
“how the additional disease entities may add to the number of
silicosis cases or add to the cumulative respiratory disease burden.” 
Could you clarify?

Ginsberg Clarification: My intention with this comment is to point out
the disconnect in the two documents whereby respiratory disease not
classified as silicosis is understood to be related to cumulative silica
exposure in these worker cohorts in a dose response fashion, and yet
there is little analysis of how this additional respiratory disease
burden could be included in a quantitative risk assessment.  The Park et
al. 2002 diatomaceous worker study is included in this section and is
highlighted in the quantitative risk assessment.  Its definition of lung
diseases other than cancer (LDOC) includes the silicosis cases as well
as other diseases logically related to silica exposure to derive a total
lung burden of disease.  The dose response for this endpoint is roughly
10 times more potent than for the pooled silicosis mortality estimate. 
This study (Park) attempts to correct for smoking and has other positive
features (high quality exposure assessment; reasonable sample size).  
In Table VI-12 and associated text, OSHA considers this as an upper
bound on silicosis-related mortality.  Perhaps the better way to view
the issue is that the Park et al.study represents a reasonable estimate
of silica-induced total respiratory mortality and keep that as a
separate outcome relative to the pooled analysis of silicosis mortality.
 This suggestion is also responsive to the followup question regarding
D-1 posed by OSHA. My relevant comments in response to the quantitative
risk assessment charge are provided below.

“OSHA may want to evaluate whether the LDOC mortality (lung disease
other than cancer, as defined by Park et al., 2002) is the more
appropriate endpoint for silica-induced non-cancer mortality risk. 
Right now the focus is on silicosis, which as stated in the document,
can be an underreported condition, sometimes called emphysema or other
lung conditions as a cause of death.  Further, as documented in Section
V, silica can induce a wide variety of lung conditions in addition to,
and without the necessary occurrence of, silicosis.  Therefore, an
additive or more inclusive approach to silica-induced non-cancer risk
assessment (as suggested by the Park et al. 2002 approach) may be more
appropriate.  It is useful that Table VI-12 pulls that data point out
separately but its unclear how much weight will be given to this potency
value.”

Brian Miller	It would be useful to define these terms clearly when they
are first used, including how they are usually defined in both clinical
and epidemiological contexts. It would then be clearer which are subsets
of others. Note that there are several different health endpoints here,
and it may not be appropriate to add together the numbers of cases of
different sorts.

P274 para 1.  Insert a definition of chronic bronchitis at beginning.

P275 para 2 ’reflected’ is vague. Chronic sputum production is a
defining symptom of bronchitis.

P275 para 3 explain “survivor” effect?

P286 para 2  Both ‘dust’ and ‘silica’ exposures are mentioned. 
Are these the same?   If so, maybe use ‘silica’ throughout?  If this
section is addressing both silica and mixed dust, don’t jump between
them so much, deal with them separately.

P289 para 1 multivariate analyses -> multiple regression 

P292 para 2 delete ‘In a bivariant analysis,’

P293 para 1 multivariate -> multiple regression

P298 para 2 spell out COPD

P308 para 2  Do the ‘overall conclusions’ need to distinguish those
findings related to silica exposure from those associated with
‘dust’?



Andrew Salmon	This section provides a clear and reasonable discussion of
the presence and relationship between these disease endpoints in those
exposed to silica dust.  It is important to emphasize that these
different pulmonary disease manifestations, plus the radiographic
diagnosis of silicosis, are all reflections of an underlying disease
process related to silica exposure, and no single diagnosis includes all
possible individual responses.  It is therefore important to look at all
these diseases entities in assessing the overall disease burden from
exposure to silica dust.

Noah Seixas	This section has a high degree of overlap with section B3,
in which respiratory impairment associated with the appearance of
silicosis is discussed.  It is not clear to me why this section is
presented separately.  Nevertheless, the endpoints discussed a
reasonable, well defined and appropriately reviewed.  Integration of
these studies with those on impairment in silicosis would make more
sense to me.



E-1	OSHA has preliminarily concluded that studies on renal and
autoimmune effects have not provided sufficient data (in quality or
amount) to support quantitative estimates of risk.  From your knowledge
and understanding of these studies, do you agree with OSHA’s
conclusion?  If you believe that there is sufficient data to support
quantitative estimates of risk, please tell OSHA specifically what data
you think could be used.

Bruce Allen	It is clear from the OSHA document itself that there are
data that would support a quantitative risk assessment for at least some
renal or autoimmune diseases (see pp, 320-321, for example).  The
document even states (p 323, lines 20-21) that some of the
exposure-response analyses that have been completed are “credible.” 
Coupled with the general tenor of the review of all the studies,
including those for which no quantitative analysis can be completed, I
would conclude that the “sufficiency” of the data (certainly in
terms of amount) is adequate.  I think it is not appropriate to base a
determination of whether or not there are sufficient data for a
quantitative risk assessment of renal or autoimmune disease on a
comparison to what is known about silicosis or lung cancer (p. 344,
lines 14-16).  Certainly, other compounds have been subjects of
quantitative risk assessments with much less data than what is available
for silicosis and lung cancer, and they have been considered adequate
for some level of quantitative evaluation.  One might ask oneself if, in
the absence of the silicosis or lung cancer data, one would use the
renal and/or autoimmune data for silica as the basis of a quantitative
risk assessment and determination of regulatory levels.  I suspect that
one would.

Murray Finkelstein	I agree that there is reasonable evidence to support
an association between silica exposure and renal and autoimmune effects.
I also agree that the data are insufficient in quality to support
quantitative estimates of risk.

Gary Ginsberg	I agree with the OSHA analysis that the data are
insufficient for quantitative risk assessment on these non-cancer
endpoints.

Brian Miller	I agree that a full quantitative assessment is not yet
possible.

Andrew Salmon	OSHA comments in the description of the available studies
that the evidence is convincing for an association between renal disease
and exposure to silica.  This certainly seems to be the case, in that
there are a substantial number of positive epidemiological studies
finding such an effect, as well as a large number of case reports.  The
plausibility of a causal association is enhanced by various laboratory
studies, finding of silica particles in kidney tissues, etc., which lay
the foundation for a reasonable mechanistic explanation of how such an
effect might occur.  The existence of a few “negative” (i.e.
non-positive) reports is not a convincing counter-argument, since these
observations could easily be explained by low power, misclassification
of exposures, competing effects and various other features of those
studies.  In general, one or more good positive epidemiological studies
are much more convincing than any number of studies which failed to find
any clear result.

OSHA’s chapter on this topic also specifically describes at least two
recent and major studies (McDonald et al., 2005 and the pooled analysis
by Steenland et al., 2002a) which explicitly determine a quantitative
relationship between silica exposure and renal disease.  Although
McDonald et al. do not argue in favor of a causative explanation for
their observation, this may be simply a case of those authors confining
their comments to a conservative interpretation of their own data,
rather than attempting to consider the implications of the whole body of
published evidence on the topic.  OSHA’s reluctance to consider a
quantitative analysis of renal disease and mortality therefore seems
unjustified.  It would be entirely reasonable to endorse the existing
quantitative analysis by Steenland et al. (2002a), particularly as many
aspects of this cohort and the accompanying exposure data have been
extensively validated in the consideration of silicosis and lung cancer.
 OSHA’s conclusion on page 345 states that there are “only two
studies on renal disease provide quantitative exposure-response data”
– this in itself is something of an understatement, but even as it
stands it contradicts the following sentence which asserts that the data
are “insufficient”.  OSHA does not define what it means by a
“robust” estimate, but seems to be using this nebulous desideratum
to hide behind in this case.

The data on rheumatoid arthritis and other autoimmune conditions also
include some published quantitative analyses, so it is somewhat
disingenuous of OSHA to contend that the data will not “support
quantitative estimates of risk” when that is precisely what some
authors have successfully provided.  However, the issue is not whether
such estimates can be made, but what level of confidence should be
placed in them and whether they have any implications for setting
health-protective policy.  In the case of the arthritis and autoimmune
responses obviously the data are somewhat less convincing and
comprehensive, due at least in part to the complexity and variability of
this type of disease.  It could certainly be argued that the
quantitative estimates available for these health endpoints are somewhat
uncertain, and the links required to confidently assume causality are
not as well established as for the other endpoints discussed in this
section.  This is not in itself a good argument for dismissing the
effort to apply a quantitative analysis even where substantial
uncertainties are present, although obviously the associated
uncertainties would need to be addressed when considering the results.  
Indeed, it is particularly important to pursue an appropriate
quantitative analysis where possible in such a case, because it is only
by use of quantitative analytical techniques that the situation can be
properly defined, including identification of the sources and extent of
uncertainty, and estimation of confidence limits as well as measures of
central tendency for key dose-response parameters.



Noah Seixas	While the data appear to be sufficient to associate silica
exposure with these endpoints, I would agree that there are insufficient
data for a quantitative risk assessment at this time.  If one required
quantitative estimates of the risk of renal disease and silica exposure,
it would be feasible, given the available studies.  However, I don’t
see a compelling reason to do so, given the much more robust data
resources on silicosis and lung cancer.  Nevertheless, the clear
association of renal and autoimmune disease endpoints with silica
exposure is an important piece of the overall health effects assessment.



F-1	The draft “Physical Factors…” section contains a discussion of
proposed mechanisms of action by which silica causes silicosis and lung
cancer.  Does the discussion of the mechanism of action reflect the most
recent thinking on this subject?  If not, tell OSHA what the most recent
thinking is, as you understand it.

Bruce Allen	I have no independent knowledge of what recent thinking has
been on the topic of proposed mechanisms, so I have no way of addressing
this question.

Murray Finkelstein	I am not competent to discuss mechanisms.

Gary Ginsberg	The proposed mechanisms of action for silicosis and lung
cancer are not spelled out in any detail and I would more refer to these
as “Mode of Action” descriptions (overall process rather than
detailed biochemical explanation).  As a MOA description, the current
document appears to adequately cover the potential pathways involved.  I
would have liked to have seen more about silica accumulation in the lung
and what governs deposition and retention.

Brian Miller	My knowledge and experience do not qualify me to comment on
the biological mechanisms proposed.

P360 para 1 Other work has also shown agreement with surface area as a
metric.  Include references?

P360 para 2 then -> than

P366 para 2 then -> than

P378 para 2 quart -> quartz

P396 para 1 This refers to more severe silicosis, and not to the milder
definitions sometimes used elsewhere (e.g. ILO 1/0+)



Andrew Salmon	As far as I am aware this discussion reflects current
thinking on this topic.  The various identified markers of inflammatory
response, and the likely involvement of reactive oxygen toxicity are
widely thought to play a role in both silicosis and lung cancer,
although obviously at this point there is no absolute and complete
description of the mechanism of action for either endpoint.

Noah Seixas	I am not qualified to discuss physical factors in any depth.
 The data presented by OSHA concerning these aspects of silica exposure
and risk appear to be well documented, thorough, and support the
conclusions made in the document.



F-2	The draft Mechanism of Action section contains a discussion on the
possible role of silicosis as a precursor or prerequisite for lung
cancer.  Does this discussion accurately reflect the current scientific
opinion on this issue?  If not, tell OSHA what the current scientific
opinion is on this issue, as you understand it.

Bruce Allen	I have no independent knowledge of what the current
scientific opinion is on the topic of the role of silicosis as a
precursor for lung cancer, so I have no way of addressing this question.

Murray Finkelstein	I believe that this discussion does accurately
reflect the current scientific opinion on this issue.

Gary Ginsberg	This section makes a major point that silicosis is not a
prerequisite for lung cancer induction as there are mechanisms
independent of tissue damage that can instigate the oncogenic process. 
However, what is unclear from this description and the underlying cited
papers is the extent to which silicosis can enhance or synergize the DNA
damaging effect of silica.   Thus, while there is substantial evidence
that silicosis is not required for oncogenic potential, the degree of
enhancement and how this affects the dose response and risk assessment
is unclear.

Brian Miller	P401 para 2 Silicosis cannot be a ‘prerequisite for lung
cancer’ - many people contract lung cancer without either silicosis or
silica exposure - and so this phrase should never be used unqualified.
Whether any excess silica-related lung cancer risk operates through
fibrosis or other aspects of the silicotic response is a difficult
question, and the text reflects this.

Andrew Salmon

Andrew Salmon

	This discussion presents several key points on this question, which I
believe reflect current thinking.  It seems likely that both silicosis
and lung cancer share some common mechanistic steps at the beginning of
the processes which lead to these responses, but it is simply unknown
how far up the chain of events this commonality extends.  One key
difficulty with the underlying question is what initial stages can be
properly called “silicosis” in this context.  

As has been eloquently described elsewhere in this section, the standard
indicators used to diagnose silicosis are X-radiography and post-mortem
histology, both of which are “late” indicators and will certainly
not serve to identify initial stages of the disease which might, or
might not, be required steps in the eventual  development of lung
cancer.

Noah Seixas	I am not qualified to discuss the biological basis of silica
toxicity in any depth.  The data presented appear well documented,
relevant to our understanding of silica health effects, and
appropriately interpreted.



F-3	In the draft “Physical Factors…” section OSHA preliminarily
concludes that the available information on the physical factors that
may influence crystalline silica toxicity cannot be used at this time to
refine quantitative estimates of the lung cancer and silicosis
mortality.  Does this seem to be an appropriate conclusion based on the
information presented?  Is it clear in the discussion how OSHA arrived
at this conclusion?  Would you arrive at a different conclusion based on
the information presented?  If so, tell OSHA what your conclusion would
be and explain how you arrived at it?

Bruce Allen	It appears from the discussion that there are several and
various factors that could modify the toxicity of silica, both for lung
cancer and silicosis.  At this time, the manner in which they act
(individually or in combination) does not appear to be well understood
(although there are some ideas) and certainly does not appear to be
quantifiable.  Thus, I conclude that an overall (“average”) measure
of risk is all that can be derived and applied at this point in time. 
Basing any regulatory decisions or limits on a range of risk estimates
(as OSHA appears to be doing in the quantitative risk section) should
reflect at least some of the variation that could be associated with the
physical characteristics.  That appears to be a satisfactory approach at
the current time, pending accumulation of additional relevant
information that might allow a more context-specific set of risk
estimates.

Murray Finkelstein	I agree that it is not possible, at the present time,
to include information on physical factors in quantitative risk
assessment. I think that the OSHA discussion is clear.

Editorial Comments on the Text

1. Page 12 Paragraph 1: There is a typo in the last sentence... a phrase
is duplicated

2. The word “data” is plural. Thus, for example, Page 13, last
sentence paragraph 2, “The modern data was” should be “The modern
data were”

3. Pg 16 first sentence and Pg 21.  “Ends of the lung tissue” is a
poorly chosen phrase.

4. Pg 16 Para 4 line 2: “infrequently” should be replaced by “in
low density”.  “Frequency” is not an appropriate word in this
paragraph.

5. Pg 71: “In considering the exposure-response relationship with a
progressive condition like silica, the effect variable (in this case
silicosis) takes on a more continuous and less discrete nature.”.... I
don’t understand what this sentence means.

Gary Ginsberg	The document spends considerable effort to explain various
physical factors and how they may influence crystalline silica potency
in the lung.  This would appear to create a substantial uncertainty in
the risk assessment as the most relevant study for a given occupation
may not be known.  It might be tempting to have a separate potency
factor for each industrial sector based upon prior studies of that
sector.  In theory, the surface properties at play in the published
study may still be present in today’s workers and may differ
substantially from other industrial sectors.  Therefore, to address this
issue, OSHA may want to either select the highest potency factor across
the board to cover the possibility that the workers might be exposed to
the most aggressive form in any given occupation.  Alternatively, OSHA
may apply a potency value specific to the industrial sector in the hopes
of capturing the silica properties specific to that industry.  It is
unclear how the OSHA risk range will in practice be applied, so I
don’t necessarily endorse this approach.

F-3 – OSHA clarification request to Ginsberg:  You suggest OSHA could
assign industry-specific potency values to account for differences in
quartz particle toxicities, but that you don’t necessarily endorse
“this” approach.  Do you mean the estimation of industry-specific
potency factors?

Ginsberg Response: I do not necessarily endorse industry-specific
potency factors.  Given how the physical state of crystalline silica may
vary across industries in a manner that could affect risk, ideally it
would be possible to make such distinctions.  However, this may not be
feasible for a number of reasons.  Rather, I raise the suggestion that
OSHA evaluate whether individual industries show a greater dose
response.  If so, they should decide whether it makes sense to identify
a plausible upper end of the potency range based upon a robust data set
from a specific industry which has defined silica physical
characteristics.  The discussion of this issue in the current draft
represents an uncertainty that is not really dealt with.   

Brian Miller	Experiments in vitro and in vivo suggest that the toxicity
of silica varies between samples from different sources, and that some
external factors can modify the response to either reduce or intensify
the effects.  It’s difficult to see how those results could be
generalized from present knowledge.  The prudent approach is to regulate
for those situations where at is most toxic; I don’t see how you could
adjust the regulation across different industries or situations where
the toxicity differs.

Andrew Salmon	I consider that the description presented here is a clear,
fair and reasonable analysis of an admittedly complicated situation.  It
seems to be the case that different samples of crystalline silica dust
do present quantitatively (although probably not qualitatively)
different hazards  Differences between samples appear to be of basically
unknown origin.  Factors such as recent cleavage, occlusion by other
minerals, surface contaminants and so on may well influence potency, but
it has not been possible to consistently or quantitatively establish
these connections.  The only physical characteristic which does seem to
have been established as definitive is particle size – the particles
need to be respirable to cause the effect.



Noah Seixas	Given the historical suggestion that polymorphs other than
quartz had higher toxicity, this is an important discussion.  I agree
that there is insufficient evidence available to suggest that this is
true, and in fact, most of the evidence presented suggests that the
polymorphs are of comparable toxicity.  Further, the evidence concerning
freshly fractured silica and silica in mixed moiety (e.g. clay) have
lower toxicity is important and compelling.  However, I think OSHA has
presented the evidence as completely as possible, and made reasonable
conclusions based on the evidence.  I agree with OSHA that there is
insufficient evidence or consistency in the data to provide guidance on
differential control of exposure based on physical and surface
characteristics.



Additional Comments/References

Bruce Allen	None.

Murray Finkelstein	None.

Gary Ginsberg	None.

Brian Miller	None.

Andrew Salmon	References

(other than those cited in the Health Effects document)

Beadle DG, Bradley AA. 1970. The composition of airborne dust in South
African gold mines. In: Shapiro HA (ed). Pneumoconiosis. Proceedings of
the International Conference. Johannesburg 1969. Cape Town: Oxford
University Press. pp. 462-6.

Page-Shipp RJ, Harris E. 1972. A study of dust exposure of South African
white gold miners. South Afr Inst Mining Metall. 73:10-24.

Schenker MB, Pinkerton KE, Mitchell D, Vallyathan V, Elvine-Kreis B,
Green FHY (2009).  Pneumoconiosis from Agricultural Dust Exposure among
Young California Farmworkers.  Environmental Health Perspectives 117(6):
988–994.

Noah Seixas	None.



Individual Review Comments

Bruce Allen

Bruce Allen Consulting, Chapel Hill, NC



Review of OSHA’s Preliminary Health Effects Section for Silica

Comments from Bruce Allen

General Questions

1.	The draft health effects section reviews and summarizes a number of
studies.  Are the studies described in sufficient detail for the reader
to understand how the studies were done?  Are the descriptions of the
studies and results accurate?  Are the strengths and limitations of the
studies adequately discussed and appropriately evaluated?  Are OSHA’s
interpretations of the study results reasonable and explained
adequately?  Are discussions of studies clear and easy to follow?  If
not, let OSHA know what needs to be added to improve the information and
the manner in which it is presented.

In general, the descriptions of the studies appear to be sufficient for
understanding what has been done and I have no reason to suspect that
they have been misrepresented.  By way of more succinct summary of
strengths and weaknesses, I would suggest a summary table
cross-classifying the studies with major items that affect strengths and
weaknesses (e.g., study size, length of follow-up, issues related to
measuring effects, etc.).  This would allow readers to appreciate the
major differences among the studies and their relative strengths and
weaknesses for those issues that OSHA considers of primary importance. 
Just having such a succinct list of what OSHA considers important would
provide a more transparent means to evaluate those items and to
determine if that list corresponds to what informed readers would
consider important.

From a formatting perspective, it was sometimes difficult to follow when
discussion of one study or one topic ended and another began.  I would
suggest more extensive use of headers and subheaders (bolded, italicized
or something, with perhaps section numbering and subsection numbering)
to delineate the sections and make it easier to find the summary
conclusions for each issue.

The draft health effects section covers a selection of studies in each
topic area.  The discussion does not cover every study ever done on the
topics.  Do you think that OSHA should have discussed additional or
different studies on a specific health endpoint?  Are you aware of
significant studies that were omitted?  If yes, inform OSHA of those
studies.  

I believe the selection of studies was adequate for the purpose of
conveying the important issues and conclusions about exposure to silica.
 I am not aware of any significant studies that were omitted.

The draft health effects section covers a number of health endpoints. 
Are you aware of additional significant health endpoints affected by
exposure to crystalline silica that should be discussed?  If so, inform
OSHA of the additional health endpoints and studies that address them.

I am not aware of other health endpoints that were omitted from the
discussion.

The draft heath effects section contains conclusions for each topic area
as well as an overall conclusion.  Are these conclusions reasonable in
light of the available data?  Is it clear in the discussion how OSHA
arrived at those conclusions based on the studies discussed? 

It appears that OSHA’s conclusions are reasonable in light of the data
that they presented.  The basis for those conclusions does appear to be
clear.

Questions on Section A. Introduction

A-1.	Is OSHA’s review of the history of dust exposure monitoring
helpful to the reader in understanding the needs and limitations of
exposure conversions to a modern mass per volume metric in reference to
the exposure-response relationship?

Yes, this section is important for understanding the uncertainties that
are associated with exposure reconstruction and the conversion factors
(and their potential biases) that might be present in that
reconstruction.

Questions on Section B. Silicosis

B-1.	OSHA presents background information on the use of radiography,
imaging techniques, pulmonary function studies, and other diagnostic
tools used in clinical settings and for in epidemiological studies. 
Does OSHA correctly present the benefits and shortcomings of these
techniques?  Are you aware of any additional information that would
either strengthen this discussion or contradict what is written?  

This is not my area of expertise; I am not able to assess whether or not
the presentation correctly assesses benefits and shortcomings of the
techniques.

B-2.	OSHA attempted to describe the available silicosis surveillance
data.  In your opinion, were any important topics left uncovered?  Any
data sources not discussed?  Is the subject of under-reporting
sufficiently covered?  Are the conclusions supported by the discussion
and stated succinctly?  Are you aware of any important studies in this
area that OSHA did not cover?

I am not familiar with surveillance data collection and reporting so I
cannot evaluate whether any important topics/data sources have been left
uncovered.  It does appear that the subject of under-reporting has been
sufficiently discussed.  It appears the conclusions are supported by the
data presented.

B-3.	For the discussion of progression of silicosis, OSHA did not review
every study that addressed this issue.  Are you aware of additional
studies that OSHA should consider, especially any studies that
illustrate progression in a cohort exposed to silica in the range of
OSHA’s current PEL (0.1 mg/m3)?   

I am not aware of any such studies.

B-4.	Results of studies covered in the PFT section are not always
consistent.  Did OSHA do a fair job of presenting these results and
drawing conclusions?  Are there any important studies that you are aware
of that would strengthen this section or the conclusions of this
section?

I am not familiar with the literature on silicosis and PFT.  However,
based on the results of the studies as presented in the document, it
appears that the conclusions reasonably reflect the information that is
available, including the uncertainties associated with sometimes
inconsistent findings.

Questions on Section C. Carcinogenic Effects of Silica (Cancer of the
Lung and Other Sites)

C-1.	With such a large amount of epidemiological literature to review,
OSHA may be subjected to criticism of selection bias.  Is the selection
criteria used by OSHA clear?  Have we missed any important studies that
have good or at least reasonable exposure estimates?  Do you recommend
any alternate or additional selection criteria?

The selection criteria are fairly well laid out on pp. 237-238.  It
might have been better to establish the criteria “up front” in
Section C so that as the reader is considering what is written about
each study, s/he can have those criteria as a frame of reference.  Here,
as elsewhere in my comments, I suggest that a summary table showing how
the studies compare to one another with respect to these criteria would
be a useful addition.

C-2.	OSHA concludes that the studies involving four cohorts among the
20+ cohorts and case controls reviewed were the strongest in terms of
several criteria in record availability and data quality (exposure,
smoking, worker and facility history) as well as data treatment and
analyses?  Do you agree with this conclusion?  Would you add or delete
any of these studies?  

This question basically follows on from the question above.  And my
answer relates to two studies that were reviewed but excluded, for
reasons that are not entirely clear.  The Brown and Rushton study of
British industrial sand workers was apparently not selected because of
“the low cumulative exposure profile” (p. 179, line 8).  A similar
comment is made about the Ulm et al. study of German stone and ceramic
workers (p. 191, line 4).  There is some discussion of other limiting
factors for these two studies, but it is not clear how those limitations
relate to the selection criteria on pp. 237-238.  In particular, I would
argue that inclusion of studies with low exposures as part of a
dose-response analysis is very useful precisely because such studies
help constrain and define the low-dose (low-exposure) shapes that can be
considered to be consistent with the data and which are particularly
relevant for determining how low exposure must be to be consistent with
non-significant risk levels.  

It may be that these two studies have some fatal flaws that the selected
studies do not have, but that is not made clear enough.  In particular,
it needs to be shown that, based on the selection criteria listed, these
two studies are not adequate (or the selection criteria need to be
expanded or refined to truly reflect the selection process that OSHA
used).  Again, a summary tabulation of the studies cross-classified
against the criteria would facilitate such an assessment.

C-3.	Do the tables add to the text presentation?  Do you have any
recommendations to improve their content or presentation?  Would you add
or delete another table?  Provide an explanation for any proposed
deletion or a mock-up of any proposed addition.

The inclusion of tables appears to be rather hit-or-miss.  For some
study summaries a table is given, while for others the odds ratio
results are just presented in the text.  I would rather see a more
consistent presentation and would tend to prefer one that includes more
tables.  This should enhance the comparability of the assessments if the
tables are as consistent as possible across studies.  I have already
mentioned that summary tables of studies cross-categorized against
selection criteria would also be a useful addition.



C-4.	OSHA has reviewed a number of studies that examined the
relationships between exposure to silica and cancer at sites other than
the lung, including larynx and nasopharynx, stomach, and esophagus. 
Have we missed any important studies of cancer at other sites? If yes,
inform OSHA of those studies.

I am not aware of any other studies of cancer at other sites.

C-5.	OSHA has preliminarily concluded that an association has not been
established between silica exposure and excess mortality from cancer at
sites other than the lung.  Do you agree with this interpretation of the
data?  If not, please provide a detailed rationale for your different
interpretation.

From the data that have been presented, I would agree that the risk of
cancer at sites other than the lung appears to be much less
well-established and probably less in magnitude (if they exist at all),
compared to the risk of lung cancer.  For that reason I can support
OSHA’s implied decision not to pursue a risk assessment based on
cancers at other sites. 

Questions on Section D. Other Nonmalignant Respiratory Disease

D-1.	This draft section discusses COPD, NMRD, bronchitis, emphysema,
airways disease, etc.  Is it clear in the discussion how these health
endpoints are related or distinct entities?  If not, tell OSHA how to
improve the presentation.

It is not clear from the discussion exactly how these health endpoints
are related to one another.  I am not sure it really matters that much,
since OSHA’s conclusion is that silica exposure (probably) affects all
of them, but a simple flow-chart or other visual, graphical display of
the relationships should suffice.

Questions on Section E. Renal and Autoimmune Effects

E-1.	OSHA has preliminarily concluded that studies on renal and
autoimmune effects have not provided sufficient data (in quality or
amount) to support quantitative estimates of risk.  From your knowledge
and understanding of these studies, do you agree with OSHA’s
conclusion?  If you believe that there is sufficient data to support
quantitative estimates of risk, please tell OSHA specifically what data
you think could be used.

It is clear from the OSHA document itself that there are data that would
support a quantitative risk assessment for at least some renal or
autoimmune diseases (see pp, 320-321, for example).  The document even
states (p 323, lines 20-21) that some of the exposure-response analyses
that have been completed are “credible.”  Coupled with the general
tenor of the review of all the studies, including those for which no
quantitative analysis can be completed, I would conclude that the
“sufficiency” of the data (certainly in terms of amount) is
adequate.  I think it is not appropriate to base a determination of
whether or not there are sufficient data for a quantitative risk
assessment of renal or autoimmune disease on a comparison to what is
known about silicosis or lung cancer (p. 344, lines 14-16).  Certainly,
other compounds have been subjects of quantitative risk assessments with
much less data than what is available for silicosis and lung cancer, and
they have been considered adequate for some level of quantitative
evaluation.  One might ask oneself if, in the absence of the silicosis
or lung cancer data, one would use the renal and/or autoimmune data for
silica as the basis of a quantitative risk assessment and determination
of regulatory levels.  I suspect that one would.

Questions on Section F. Physical Factors that May Influence Toxicity of
Crystalline Silica

F-1.	The draft “Physical Factors…” section contains a discussion
of proposed mechanisms of action by which silica causes silicosis and
lung cancer.  Does the discussion of the mechanism of action reflect the
most recent thinking on this subject?  If not, tell OSHA what the most
recent thinking is, as you understand it.

I have no independent knowledge of what recent thinking has been on the
topic of proposed mechanisms, so I have no way of addressing this
question.

F-2.	The draft Mechanism of Action section contains a discussion on the
possible role of silicosis as a precursor or prerequisite for lung
cancer.  Does this discussion accurately reflect the current scientific
opinion on this issue?  If not, tell OSHA what the current scientific
opinion is on this issue, as you understand it.

I have no independent knowledge of what the current scientific opinion
is on the topic of the role of silicosis as a precursor for lung cancer,
so I have no way of addressing this question.



F-3.	In the draft “Physical Factors…” section OSHA preliminarily
concludes that the available information on the physical factors that
may influence crystalline silica toxicity cannot be used at this time to
refine quantitative estimates of the lung cancer and silicosis
mortality.  Does this seem to be an appropriate conclusion based on the
information presented?  Is it clear in the discussion how OSHA arrived
at this conclusion?  Would you arrive at a different conclusion based on
the information presented?  If so, tell OSHA what your conclusion would
be and explain how you arrived at it?  

It appears from the discussion that there are several and various
factors that could modify the toxicity of silica, both for lung cancer
and silicosis.  At this time, the manner in which they act (individually
or in combination) does not appear to be well understood (although there
are some ideas) and certainly does not appear to be quantifiable.  Thus,
I conclude that an overall (“average”) measure of risk is all that
can be derived and applied at this point in time.  Basing any regulatory
decisions or limits on a range of risk estimates (as OSHA appears to be
doing in the quantitative risk section) should reflect at least some of
the variation that could be associated with the physical
characteristics.  That appears to be a satisfactory approach at the
current time, pending accumulation of additional relevant information
that might allow a more context-specific set of risk estimates.



Murray Finkelstein

McMaster University, Thornhill, ON Canada

  SEQ CHAPTER \h \r 1 	Peer Review Comments on Silica Health Effects
Document

General Questions

1.	The draft health effects section reviews and summarizes a number of
studies.  Are the studies described in sufficient detail for the reader
to understand how the studies were done?  Are the descriptions of the
studies and results accurate?  Are the strengths and limitations of the
studies adequately discussed and appropriately evaluated?  Are OSHA’s
interpretations of the study results reasonable and explained
adequately?  Are discussions of studies clear and easy to follow?  If
not, let OSHA know what needs to be added to improve the information and
the manner in which it is presented.

I think that the studies are, in general, well described and that
OSHA’s interpretations are reasonable and adequately explained. The
mass of text makes it difficult, however, to locate specific studies and
results. I think that the presentation could be improved by including
summary tables of results. As an example of one possible layout, see the
tables in the Text and Appendices of the Institute of Medicine review
“Asbestos: Selected Cancers”	ISBN-10: 0-309-10169-7 and  ISBN-13:
978-0-309-10169-1

	The draft health effects section covers a selection of studies in each
topic area.  The discussion does not cover every study ever done on the
topics.  Do you think that OSHA should have discussed additional or
different studies on a specific health endpoint?  Are you aware of
significant studies that were omitted?  If yes, inform OSHA of those
studies.  

The literature on silica health effects is voluminous. I think that OSHA
has identified and discussed the major reports. I am aware of no
significant studies that were omitted.

3. The draft health effects section covers a number of health endpoints.
 Are you aware of additional significant health endpoints affected by
exposure to crystalline silica that should be discussed?  If so, inform
OSHA of the additional health endpoints and studies that address them.

I am not aware of any additional significant health endpoints.

4. The draft heath effects section contains conclusions for each topic
area as well as an overall conclusion.  Are these conclusions reasonable
in light of the available data?  Is it clear in the discussion how OSHA
arrived at those conclusions based on the studies discussed? 

The discussion is clear and the conclusions are reasonable.

Section A: Introduction

A-1.	Is OSHA’s review of the history of dust exposure monitoring
helpful to the reader in understanding the needs and limitations of
exposure conversions to a modern mass per volume metric in reference to
the exposure-response relationship?

OSHA’s review is helpful in providing a historic perspective and in
providing context for the retrospective-prospective epidemiologic
studies.

Section B: Silicosis

B-1.	OSHA presents background information on the use of radiography,
imaging techniques, pulmonary function studies, and other diagnostic
tools used in clinical settings and for in epidemiological studies. 
Does OSHA correctly present the benefits and shortcomings of these
techniques?  Are you aware of any additional information that would
either strengthen this discussion or contradict what is written?  

I think that the section on Diagnostic Tools requires revision. In
particular:

CT and HCRT: The study by Begin needs to be revisited by OSHA. There
were only 58 subjects and only 6 did not have radiographic silicosis.
The relevant question is: does CT find silicosis when X-ray does not?
There were 6 in Category 0 by X-ray and 12 in Category 0 by CT.

The Talini paper studies only 27 subjects. There was substantial
disagreement between CT and CXR at low profusions.

I think that the literature search needs to be updated for the
comparison of CT and CXR. The references in the document are old and my
own search on PubMed found additional references.

B-2 Surveillance

B-2.	OSHA attempted to describe the available silicosis surveillance
data.  In your opinion, were any important topics left uncovered?  Any
data sources not discussed?  Is the subject of under-reporting
sufficiently covered?  Are the conclusions supported by the discussion
and stated succinctly?  Are you aware of any important studies in this
area that OSHA did not cover?

I think that the section on surveillance was adequately done. I am not
aware of overlooked studies.

B-3 Progression

B-3.	For the discussion of progression of silicosis, OSHA did not review
every study that addressed this issue.  Are you aware of additional
studies that OSHA should consider, especially any studies that
illustrate progression in a cohort exposed to silica in the range of
OSHA’s current PEL (0.1 mg/m3)?   

I am not aware of additional studies that OSHA should consider.

B-4 PFT

B-4.	Results of studies covered in the PFT section are not always
consistent.  Did OSHA do a fair job of presenting these results and
drawing conclusions?  Are there any important studies that you are aware
of that would strengthen this section or the conclusions of this
section?

OSHA did a fair job of presenting the studies. However, most of these
studies were small, and of low statistical power. I suggest that OSHA
consider pooling the results (meta-analysis) of comparable studies in
order to increase the ability to draw conclusions from the data.

I am not aware of important omitted studies.

Comments with respect to: PRELIMINARY CONCLUSIONS: SILICOSIS AND
DISEASE PROGRESSION (Pg 107)

Pg 111: “Both of these studies report progression in terms of
percentage of workers whose chest x-ray categories progress, and their
results are close (2 percent to 6 percent). Based on these data, one can
estimate an annual progression rate of 2-6 percent for this exposure
range. This means that every year approximately 2 to 6 percent of
workers exposed to respirable silica dust in the range of 0.12 to 0.48
mg/m3 are expected to progress at least 1 sub-category on the ILO
scale.”

This statement omits to mention one crucial factor: latency. Progression
will depend upon the time from onset of exposure. It is almost certainly
not true, for example, that in the first 10 years of exposure every year
approximately 2 to 6 percent of workers exposed to respirable silica
dust in the range of 0.12 to 0.48 mg/m3 are expected to progress at
least 1 sub-category on the ILO scale.”. The OSHA authors need to
bring the time frame (latency) into the discussion.

Questions on Section C. Carcinogenic Effects of Silica (Cancer of the
Lung and Other Sites)

C-1.	With such a large amount of epidemiological literature to review,
OSHA may be subjected to criticism of selection bias.  Is the selection
criteria used by OSHA clear?  Have we missed any important studies that
have good or at least reasonable exposure estimates?  Do you recommend
any alternate or additional selection criteria?

I believe that the selection criteria used by OSHA are clear. I am not
aware of any important studies that have been missed.

C-2.	OSHA concludes that the studies involving four cohorts among the
20+ cohorts and case controls reviewed were the strongest in terms of
several criteria in record availability and data quality (exposure,
smoking, worker and facility history) as well as data treatment and
analyses?  Do you agree with this conclusion?  Would you add or delete
any of these studies?  

I believe that, given the abundance of studies, it is appropriate to
weight most highly those studies with the least confounding (due to the
presence of other known carcinogens or the absence of smoking
histories). There has been discussion of heterogeneity. Little mention
was made of exposure misclassification as a cause of heterogeneity. Of
all the silica studies, I have seen the raw exposure data only for the
Vermont granite sheds. In this data set there are many years in which
there was no sampling at all, and a number of jobs for which very few
dust measurements were obtained. It is my opinion that these data are
sparse and much interpolation of uncertain data is required. These data
can certainly be used for quantitative risk assessment, but the
resulting risk estimates will have a large measure of uncertainty
attached. I would not be surprised if the same were true of most of the
other cohorts. Nevertheless, I believe that it is important to make the
best use of whatever data are available. I thus support the pooling of
data from individual studies. I recognize that the pooling of poor
quality data will tend to obscure any relations between exposure and
response. Should an exposure-response relationship emerge from these
analyses, despite the weaknesses, then I believe that this supports the
existence of a real association between exposure and disease.	

C-3.	Do the tables add to the text presentation?  Do you have any
recommendations to improve their content or presentation?  Would you add
or delete another table?  Provide an explanation for any proposed
deletion or a mock-up of any proposed addition.

I like Table V-C-1.

Tables V-C-2, V-C-3, and V-C-4 are orphans. It is not clear why results
from only these studies should be presented in tabular format. I would
delete these tables.

Additional Tables:

1) I think that a summary Table describing the characteristics of each
study, such as:

Type of industry

Cohort or case-control

Other confounders?

Adjustment for smoking?

Quantitative exposure estimates?

would be very helpful as an overview.

2) I think that much of the material in the text could be usefully
summarized in results tables. These would replace the tables mentioned
above and summarize all the data in one place.

As an example of one possible layout, see the tables in the Text and
Appendices of the Institute of Medicine review “Asbestos: Selected
Cancers” ISBN-10: 0-309-10169-7 and  ISBN-13: 978-0-309-10169-1

(  HYPERLINK
"file:///\\\\ERG-LEX-B\\DATA3\\SHARED\\conf\\Projects\\OSHA%20Silica%202
014\\2014%20Peer%20Review%20Report\\(http:\\www.nap.edu\\catalog.php%3fr
ecord_id=11665)"  http://www.nap.edu/catalog.php?record_id=11665) 

C-4.	OSHA has reviewed a number of studies that examined the
relationships between exposure to silica and cancer at sites other than
the lung, including larynx and nasopharynx, stomach, and esophagus. 
Have we missed any important studies of cancer at other sites? If yes,
inform OSHA of those studies.

I am not aware of any important studies that have been missed.

C-5.	OSHA has preliminarily concluded that an association has not been
established between silica exposure and excess mortality from cancer at
sites other than the lung.  Do you agree with this interpretation of the
data?  If not, please provide a detailed rationale for your different
interpretation.

I agree that definitive associations between silica exposure and excess
mortality from cancer at other sites have not been established. I think
that a potential association between silica exposure and stomach cancer
is provocative, and additional work should be conducted exploring this
connection in a quantitative fashion.

Questions on Section D. Other Nonmalignant Respiratory Disease

D-1.	This draft section discusses COPD, NMRD, bronchitis, emphysema,
airways disease, etc.  Is it clear in the discussion how these health
endpoints are related or distinct entities?  If not, tell OSHA how to
improve the presentation.

Pg 264.  Chronic bronchitis and emphysema are NOT either/or.  Both
conditions may be present together.

Questions on Section E. Renal and Autoimmune Effects

E-1.	OSHA has preliminarily concluded that studies on renal and
autoimmune effects have not provided sufficient data (in quality or
amount) to support quantitative estimates of risk.  From your knowledge
and understanding of these studies, do you agree with OSHA’s
conclusion?  If you believe that there is sufficient data to support
quantitative estimates of risk, please tell OSHA specifically what data
you think could be used.

I agree that there is reasonable evidence to support an association
between silica exposure and renal and autoimmune effects. I also agree
that the data are insufficient in quality to support quantitative
estimates of risk.

Questions on Section F. Physical Factors that May Influence Toxicity of
Crystalline Silica

F-1.	The draft “Physical Factors…” section contains a discussion
of proposed mechanisms of action by which silica causes silicosis and
lung cancer.  Does the discussion of the mechanism of action reflect the
most recent thinking on this subject?  If not, tell OSHA what the most
recent thinking is, as you understand it.

I am not competent to discuss mechanisms.

F-2.	The draft Mechanism of Action section contains a discussion on the
possible role of silicosis as a precursor or prerequisite for lung
cancer.  Does this discussion accurately reflect the current scientific
opinion on this issue?  If not, tell OSHA what the current scientific
opinion is on this issue, as you understand it.

I believe that this discussion does accurately reflect the current
scientific opinion on this issue.

F-3.	In the draft “Physical Factors…” section OSHA preliminarily
concludes that the available information on the physical factors that
may influence crystalline silica toxicity cannot be used at this time to
refine quantitative estimates of the lung cancer and silicosis
mortality.  Does this seem to be an appropriate conclusion based on the
information presented?  Is it clear in the discussion how OSHA arrived
at this conclusion?  Would you arrive at a different conclusion based on
the information presented?  If so, tell OSHA what your conclusion would
be and explain how you arrived at it?  

I agree that it is not possible, at the present time, to include
information on physical factors in quantitative risk assessment. I think
that the OSHA discussion is clear.

Editorial Comments on the Text

1. Page 12 Paragraph 1: There is a typo in the last sentence... a phrase
is duplicated

2. The word “data” is plural. Thus, for example, Page 13, last
sentence paragraph 2, “The modern data was” should be “The modern
data were”

3. Pg 16 first sentence and Pg 21.  “Ends of the lung tissue” is a
poorly chosen phrase.

4. Pg 16 Para 4 line 2: “infrequently” should be replaced by “in
low density”.  “Frequency” is not an appropriate word in this
paragraph.

5. Pg 71: “In considering the exposure-response relationship with a
progressive condition like silica, the effect variable (in this case
silicosis) takes on a more continuous and less discrete nature.”......
 I don’t understand what this sentence means.



Gary Ginsberg

Connecticut Department of Public Health, Hartford, CT

Response to Charge Questions

OSHA’s Preliminary Health Effects Section for Silica

ERG Task # 0193.15.064.001

Reviewer:  Gary Ginsberg

November 4th 2009

The descriptions of studies are well developed and provide good
background information for the risk assessment.  The conclusions about
silica-induced diseases and the variables that can lead to inconsistency
or modify dose response are generally well founded.  However, this
document is disconnected from the risk assessment in that the study
descriptions are not tied to the overall purpose of developing a
quantitative analysis.  Most study descriptions lack critical evaluation
of whether that particular study has sufficient exposure assessment,
numbers of subjects, outcome measures and quality control for inclusion
in quantitative dose-response models.  OSHA’s identification of
industries which are least confounded and thus most useful is helpful
and seemingly accurate.  However, for the most part, the critical
sorting stops at that level.   For example, it is unclear why the 1989
Danish stone cutter study is not incorporated into quantitative models
while other studies are.  The stone cutting and sand industries are
identified as having little confounding and studies from these cohorts
are not included in Section VI.  There may be good reason for that, but
without those reasons explained, one does not know whether the
quantitative assessment could be more robust or is in some manner biased
by the exclusion of useful cohorts.  The description of the Italian
brick workers (pg 138) is an example where OSHA discusses study
limitations that help the reader understand why it might not have been a
good candidate for quantitative analysis.  However, this is the
exception rather than the rule and it appears that OSHA is relying upon
study selection made by others (e.g., NIOSH, 2002; IARC 1997; the
various meta analyses; pre-existing quantitative analyses).  A
study-by-study critique (not just review, as is now presented) is needed
to sort out the value of individual studies for quantitative analysis
and to document that the array of studies relied upon in Section VI is
reasonably complete.     

 

Better use of tables would greatly facilitate use of this document. 
V-C-1 is a good start in terms of laying out key studies, but a 2nd
(perhaps 3rd as well) table is/are needed which log key information for
each study in terms of study design (cohort size, measurements made,
exposure range, etc.), results and most importantly utility for
quantitative analysis.  Also, tables would be more useful if organized
in a discrete section in the back of the report and if they were listed
in the TOC.    

The selection of studies reviewed appears to be broad and extensive. 
Not starting from a background in silica toxicology/epidemiology, I
can’t really say whether important studies have been omitted from this
review.  As mentioned above, the selection of studies for quantitative
assessment needs better justification.  

The selection of health endpoints appears to be inclusive and it is hard
to imagine that other endpoints exist that would materially add to the
quantitative assessment.  However, my knowledge base in this area is
very limited. 

The conclusions for the various topic areas appear to be supported by
the data.  However, in some cases the conclusions are not complete
thoughts or do not tie to the quantitative assessment.  For example,
Page 396 states that many cases of silicosis may be missed with standard
radiographic and CT techniques.  This connects to the evidence I alluded
to above regarding false negatives being discovered when autopsies are
performed.  OSHA does not provide any perspective on how false negative
outcomes might influence their analysis.  On the next page, associations
between pulmonary function decrements and x-ray evidence of silicosis
are discussed with doubt raised about whether functional declines can be
seen at the earlier stages of pathology.  While the conclusion drawn may
be reasonable, how this makes a difference to the quantitative
assessment is unclear.  Does this affect OSHA’s case definition of
silicosis (e.g., only if there is associated functional decline)?  Does
it affect the way OSHA interprets the epidemiology? Does OSHA have a
degree of silicosis it is trying to prevent – i.e., does the minimal
ILO grade constitute an adverse effect that is in need of prevention
through regulation.  This decision affects how the epidemiology is
interpreted and risk assessment conducted.  For example, if the dose
response is from a study whose sensitivity was for mild to moderate
silicosis, OSHA may not want to weigh that study as heavily as one with
a more sensitive case definition.  It would appear that the effort OSHA
spends in Section V (page 18 and further) to define silicosis would be a
good place to discuss how it approaches the issue of case definition
when interpreting the epidemiology and in building the quantitative risk
assessment.  

A-1.  The introductory material regarding dust exposure monitoring is
useful for understanding the epidemiology, how the quality of studies
has changed over time, and how certain measurement variables can be a
source of uncertainty.

B-1.  I do not have the technical expertise to comment on whether OSHA
descriptions of radiological and pulmonary function tests are accurate. 


B-2.  The surveillance data review and underreporting issue appear to be
adequately described and of relevance to the remainder of the document
as general background information.  

B-3.  Not having background in this area, I cannot offer any additional
studies regarding progression of silicosis.  This section is of general
relevance but its unclear how it assists with the quantitative
assessment in Section VI.  

B-4.  The pulmonary function test results section appears to be a
reasonable discussion on the evidence surrounding PFT changes and
radiographic evidence of pathology.  I have no studies to add.  

C-1.  The selection of studies for this review is described on Page 115,
bottom paragraph.  This description is vague and doesn’t really
describe a selection process with defined criteria.  Rather, it just
states what OSHA reviewed and how this builds on previous reviews.  If
there were selection criteria for inclusion in the review or in the
quantitative analysis, this should be made explicit.  

C-2.  I agree that the 4 industries identified by IARC 1997 (Page 114)
and highlighted by OSHA appear to be the most useful for studying the
effects of crystalline silica on workers.  

C-3.  See my previous comment (under general comments) regarding
potential improvements to the tables.  

C-4.  I am not aware of any additional cancer sites or studies suitable
for this review. 

C-5.  I agree that the data describing cancer at other target sites are
too limited and in some cases too contradictory to allow OSHA to make
these sites a focus of the risk assessment.  

D-1.  The non-malignant respiratory disease (NMRD) section reads well
and makes a number of important points regarding the ability of silica
to induce these diseases without the need for silicosis to occur.  Once
again, this information is not tied to the risk assessment, in
particular, how the additional disease entities may add to the number of
silicosis cases or add to the cumulative respiratory disease burden.  In
this regard, the OSHA assessment has the potential to underestimate
silica-related non-cancer risks.  

D-1 – OSHA question to Ginsberg: You state that OSHA has not tied the
information on NMRD to the risk assessment, in particular, “how the
additional disease entities may add to the number of silicosis cases or
add to the cumulative respiratory disease burden.”  Could you clarify?

Response:  My intention with this comment is to point out the disconnect
in the two documents whereby respiratory disease not classified as
silicosis is understood to be related to cumulative silica exposure in
these worker cohorts in a dose response fashion, and yet there is little
analysis of how this additional respiratory disease burden could be
included in a quantitative risk assessment.  The Park et al. 2002
diatomaceous worker study is included in this section and is highlighted
in the quantitative risk assessment.  Its definition of lung diseases
other than cancer (LDOC) includes the silicosis cases as well as other
diseases logically related to silica exposure to derive a total lung
burden of disease.  The dose response for this endpoint is roughly 10
times more potent than for the pooled silicosis mortality estimate. 
This study (Park) attempts to correct for smoking and has other positive
features (high quality exposure assessment; reasonable sample size).  
In Table VI-12 and associated text, OSHA considers this as an upper
bound on silicosis-related mortality.  Perhaps the better way to view
the issue is that the Park et al. study represents a reasonable estimate
of silica-induced total respiratory mortality and keep that as a
separate outcome relative to the pooled analysis of silicosis mortality.
 This suggestion is also responsive to the followup question regarding
D-1 posed by OSHA.  

My relevant comments in response to the quantitative risk assessment
charge are pasted below.

“OSHA may want to evaluate whether the LDOC mortality (lung disease
other than cancer, as defined by Park et al., 2002) is the more
appropriate endpoint for silica-induced non-cancer mortality risk. 
Right now the focus is on silicosis, which as stated in the document,
can be an underreported condition, sometimes called emphysema or other
lung conditions as a cause of death.  Further, as documented in Section
V, silica can induce a wide variety of lung conditions in addition to,
and without the necessary occurrence of, silicosis.  Therefore, an
additive or more inclusive approach to silica-induced non-cancer risk
assessment (as suggested by the Park et al. 2002 approach) may be more
appropriate.  It is useful that Table VI-12 pulls that data point out
separately but its unclear how much weight will be given to this potency
value.”

E-1.  I agree with the OSHA analysis that the data are insufficient for
quantitative risk assessment on these non-cancer endpoints.  

F-1.  The proposed mechanisms of action for silicosis and lung cancer
are not spelled out in any detail and I would more refer to these as
“Mode of Action” descriptions (overall process rather than detailed
biochemical explanation).  As a MOA description, the current document
appears to adequately cover the potential pathways involved.  I would
have liked to have seen more about silica accumulation in the lung and
what governs deposition and retention.   

F-2.  This section makes a major point that silicosis is not a
prerequisite for lung cancer induction as there are mechanisms
independent of tissue damage that can instigate the oncogenic process. 
However, what is unclear from this description and the underlying cited
papers is the extent to which silicosis can enhance or synergize the DNA
damaging effect of silica.   Thus, while there is substantial evidence
that silicosis is not required for oncogenic potential, the degree of
enhancement and how this affects the dose response and risk assessment
is unclear.  

F-3.  The document spends considerable effort to explain various
physical factors and how they may influence crystalline silica potency
in the lung.  This would appear to create a substantial uncertainty in
the risk assessment as the most relevant study for a given occupation
may not be known.  It might be tempting to have a separate potency
factor for each industrial sector based upon prior studies of that
sector.  In theory, the surface properties at play in the published
study may still be present in today’s workers and may differ
substantially from other industrial sectors.  Therefore, to address this
issue, OSHA may want to either select the highest potency factor across
the board to cover the possibility that the workers might be exposed to
the most aggressive form in any given occupation.  Alternatively, OSHA
may apply a potency value specific to the industrial sector in the hopes
of capturing the silica properties specific to that industry.  It is
unclear how the OSHA risk range will in practice be applied, so I
don’t necessarily endorse this approach. 

F-3 – OSHA question to Ginsberg:  You suggest OSHA could assign
industry-specific potency values to account for differences in quartz
particle toxicities, but that you don’t necessarily endorse “this”
approach.  Do you mean the estimation of industry-specific potency
factors?

	Response:  Right, I don’t necessarily endorse industry-specific
potency factors.  Given how the physical state of crystalline silica may
vary across industries in a manner that could affect risk, ideally it
would be possible to make such distinctions.  However, this may not be
feasible for a number of reasons.  Rather, I raise the suggestion that
OSHA evaluate whether individual industries show a greater dose
response.  If so, they should decide whether it makes sense to identify
a plausible upper end of the potency range based upon a robust data set
from a specific industry which has defined silica physical
characteristics.  The discussion of this issue in the current draft
represents an uncertainty that is not really dealt with.   

 



Brian Miller

IOM Consulting Ltd., Scotland, UK

Response to Technical Charge to External Peer Reviewers

ERG Contract No. GS-10F-0125P

BPA No. DOLQ059622303

ERG Task No. 0193.15.064.001

Review of OSHA’s Preliminary Health Effects Section for Silica

General Questions

The draft health effects section reviews and summarizes a number of
studies.  Are the studies described in sufficient detail for the reader
to understand how the studies were done?  Are the descriptions of the
studies and results accurate?  Are the strengths and limitations of the
studies adequately discussed and appropriately evaluated?  Are OSHA’s
interpretations of the study results reasonable and explained
adequately?  Are discussions of studies clear and easy to follow?  If
not, let OSHA know what needs to be added to improve the information and
the manner in which it is presented.

In the main, the descriptions seem good.  I have annotated some
corrections, suggestions and comments by page number within each section
of this document.  There were some studies whose size made the lack of
statistically significant results almost a foregone conclusion:  the
concept of power of a study was not prominent, and could perhaps have
been emphasized more.

The draft health effects section covers a selection of studies in each
topic area.  The discussion does not cover every study ever done on the
topics.  Do you think that OSHA should have discussed additional or
different studies on a specific health endpoint?  Are you aware of
significant studies that were omitted?  If yes, inform OSHA of those
studies.  

No, I am not aware of other studies.  Those I was familiar with were
included, along with many others that were new to me.

The draft health effects section covers a number of health endpoints. 
Are you aware of additional significant health endpoints affected by
exposure to crystalline silica that should be discussed?  If so, inform
OSHA of the additional health endpoints and studies that address them.

This section covered all the effects that I was aware of as being
potentially related to silica.

The draft health effects section contains conclusions for each topic
area as well as an overall conclusion.  Are these conclusions reasonable
in light of the available data?  Is it clear in the discussion how OSHA
arrived at those conclusions based on the studies discussed? 

In the main, I believe the discussions balanced the strengths of the
studies convincingly, although as mentioned above some discussion of
statistical power might have helped here.

 

Questions on Section A. Introduction

A-1.	Is OSHA’s review of the history of dust exposure monitoring
helpful to the reader in understanding the needs and limitations of
exposure conversions to a modern mass per volume metric in reference to
the exposure-response relationship?

Given the disparity in measurement methods over the years, this is
necessary.  However, I’d suggest not describing these as exposures,
but as concentrations or intensities, and reserving the word
‘exposure’ for subjects’ interaction with those intensities, i.e.
including duration.  This is a subtle distinction, but a useful one to
maintain.  It is exposure, not concentration, that engenders risk.    

P5  para 2 Replace ‘measure of exposure’ with  ‘airborne
concentration’.  

P8 para 3 spans -> span.  Give ref(s) for methods?

P12 para 1 last sentence contains extra text.

P14 para 1 explain more fully what ‘respirable’ means? 

Questions on Section B. Silicosis

B-1.	OSHA presents background information on the use of radiography,
imaging techniques, pulmonary function studies, and other diagnostic
tools used in clinical settings and for in epidemiological studies. 
Does OSHA correctly present the benefits and shortcomings of these
techniques?  Are you aware of any additional information that would
either strengthen this discussion or contradict what is written?

  

I’d like to see the description of the ILO conventions for describing
the abnormalities on chest x-rays spelt out early, around PP14-16,
because that informs a large part of what we mean by ‘silicosis’. 
It appears too late at present.

I’d also suggest that this section might give more of an impression
that ‘chronic’silicosis, like all pneumoconiosis, is an evolving
damage process and that decisions about when a subject has the
‘disease’ involve arbitrary distinctions of severity along a
continuum.  This would also be a good place to have the (currently
later) material on reader variation. 

P18 para 4 The ILO system standardizes the description of chest x-rays,
not their ‘interpretation’.

P19 para 4 however -> but

P23 para 2.  The authors used multiple regression, not multivariate
analysis.

P32 para 1.  kappa  0.49 may be ‘better’ than 0.29, but it’s still
not very good.  Note kappa does not inform about the direction of
disagreement.  Might the relationship between HRCT and ILO be shown in a
table?

P34 para 1.  Note other lung function measurements can be very variable
too.

P35 para 2  Note that values below the 5th percentile may in fact be
perfectly normal?

P37 para 1 (and/or elsewhere in this section).  ?Make more explicit that
the high inter-individual variation in lung function values implies high
variation in %predicted, and that is why (1) %predicted should be used
only as an initial screening measurement and (2) rate of loss over time
in an individual is a much more sensitive marker.?

B-2.	OSHA attempted to describe the available silicosis surveillance
data.  In your opinion, were any important topics left uncovered?  Any
data sources not discussed?  Is the subject of under-reporting
sufficiently covered?  Are the conclusions supported by the discussion
and stated succinctly?  Are you aware of any important studies in this
area that OSHA did not cover?

I’m not close to these schemes, so I don’t have any others to
suggest.  The topic of ‘under-reporting’ is linked to the definition
of the minimum severity required to classify a subject as having
silicosis, and that might be brought out a little more clearly.

P48 para 1.  Presumably deaths in the young are from ‘acute’
silicosis?

P48 para 2.  I think YPLL is a misnomer, and capable of
misinterpretation.  Why not Years of Working Life Lost, YWLL, which is
what is meant.

P49 para 2.  Reword last sentence?  Workers in industries without silica
exposure are not likely to die of silicosis at all.

P52 para 2  Define ‘latency’?

P56 para 3.  If the three states are getting their silicosis cases from
hospital discharge records, presumably they are at the more serious end
of profusion or ‘acute’, or…, certainly not matching a definition
of ILO categories 1/0+.  There could be thousands of case that meet the
ILO definition that never go to hospitals. This highlights the necessity
of keeping it clear that not all uses of the word ‘silicosis’ mean
the same thing.     

P61 para 1 ‘Silicosis’ at ILO 1/0 is unlikely to be symptomatic. 
Such cases are only found if actively sought, e.g. with a radiographic
workplace surveillance program.  The conclusion in P62 para 3 is
therefore not at all surprising.

P68 para 3 Says there is no system that collects silicosis deaths, then
describes NORMS, which (I thought) does just that. But it’s true that
no system collects new live cases.

B-3.	For the discussion of progression of silicosis, OSHA did not review
every study that addressed this issue.  Are you aware of additional
studies that OSHA should consider, especially any studies that
illustrate progression in a cohort exposed to silica in the range of
OSHA’s current PEL (0.1 mg/m3)?   

I’m not aware of any other studies that should be considered.

P79 para 5  Is it worth pointing out that the problems in the Scottish
colliery were spotted as a rapid and atypical burst of radiological
progression?  If so, reference Seaton et al (Lancet, 1981)?  

P80 para 2 ‘occupational history since leaving the colliery’ was
particular to the 1990/91 follow-up survey.  (regular PFR surveys did
not include retirees.)

P81 para 1.  delete ‘, as well as the mean and maximum of the non-zero
exposure estimates’?  (That looks out of place.)

P83 para 4.  Should that say ‘it progresses quickly once
established’?

P84 para 4 silicosis defined how?

P102 para 3 presumably the incidence rate what was compared, not the
ratio.

P105 para 2 odd -> odds

P108 para 3 where part of the study cohort was exposed…for a period of
some years.

P109 What does ‘sufficient data were not provided’ mean?  Should
this be ‘there have been no follow-up surveys of this cohort since
1991’?  

P109 para 2 determinate -> determinant

B-4.	Results of studies covered in the PFT section are not always
consistent.  Did OSHA do a fair job of presenting these results and
drawing conclusions?  Are there any important studies that you are aware
of that would strengthen this section or the conclusions of this
section?

This section might go a little further to emphasise how the
inter-individual variation on lung function measurements between healthy
subjects of the same height and age makes it difficult to detect
abnormality cross-sectionally.  The idea of the lower 5% being
automatically accepted as ‘abnormal’ is reported rather
uncritically.

Questions on Section C. Carcinogenic Effects of Silica (Cancer of the
Lung and Other Sites)

C-1.	With such a large amount of epidemiological literature to review,
OSHA may be subjected to criticism of selection bias.  Is the selection
criteria used by OSHA clear?  Have we missed any important studies that
have good or at least reasonable exposure estimates?  Do you recommend
any alternate or additional selection criteria?

This section of the literature has been reviewed fairly extensively, and
the criteria for inclusion seem reasonable.  

C-2.	OSHA concludes that the studies involving four cohorts among the
20+ cohorts and case controls reviewed were the strongest in terms of
several criteria in record availability and data quality (exposure,
smoking, worker and facility history) as well as data treatment and
analyses?  Do you agree with this conclusion?  Would you add or delete
any of these studies?  

I agree that these appear the strongest at the time of review.  There is
a new paper on mortality in British coalminers that has results on
respirable quartz and lung cancer, just available online.  This has
among the best-characterised exposures of any studies.

Miller BG, MacCalman L. (2009). Cause-specific mortality in British coal
workers and exposure to respirable dust and quartz. Occup. Environ. Med.
(Published Online First: 9 October 2009. doi:10.1136/oem.2009.046151.  

The above paper is summarised from:

Miller BG, MacCalman L, Hutchison PA. (2007). Mortality over an extended
follow-up period in coal workers exposed to respirable dust and quartz.
Edinburgh: Institute of Occupational Medicine.  (IOM Report TM/07/06).  
 HYPERLINK "http://www.iom-world.org/pubs/IOM_TM0706rev.pdf" 
http://www.iom-world.org/pubs/IOM_TM0706rev.pdf 

C-3.	Do the tables add to the text presentation?  Do you have any
recommendations to improve their content or presentation?  Would you add
or delete another table?  Provide an explanation for any proposed
deletion or a mock-up of any proposed addition.

P113 para 1 delete ‘of’

P129 para 5 i.e. -> e.g.

P130 para 2 participants -> participants’

P141 para 2 base -> based

P141 para 2 Is it correct to describe an ‘elevated risk’ for gastric
cancer with a lower CL of 78?

P142 para 1 The cohort comprised white…

P144 para 1 …from regression models including …

P144 para 1 …exposure to dust could cause death from COPD only …

P156 para 5 …mineworkers who were employed…

P164 para 4  I don’t accept as valid a test of significance for a
pneumoconiosis SMR.  The general population does not contract any
pneumoconiosis, so arguably an SMR for this cause is meaningless;  the
implied null hypothesis is untenable.

P169 para 2 …quartile analysis of average respirable silica
concentration…

Table V-C-2 Average exposure -> Average concentration (throughout)

P183 para 4.  Confusing.  If ‘observed number of cases differed
significantly from expectation’, why is it that the ‘SMR based on
local rates barely missed being statistically significant’?  Are these
not the same, or is one using a different set of comparison rates?

P199 para 2 ‘decreased’ does not seem a good summary of the pattern
here.

P200 para 1 Should ‘a modest non-monotonic trend’ not read ‘no
trend’?  Presumably the ORs are in comparison with the lowest exposure
group, assumed at 1.0.  Did this really yield a p-value of 0.007?

P201 para 2 Follow-up was to 1996, not 1966.

P213 Table V-C-4  The column headed ‘Low/no’ puzzled me, because in
a case-control study it’s common to compare other exposure groups with
the lowest, and set this to an arbitrary level of 1.  However, the
column is simply mislabelled, following a typo in the original paper. In
Table 4 of Calvert et al (2003) the column is headed ‘Ever* low/no’,
and the asterisk leads to a footnote ‘*Ever =medium, high, and super
high categories combined’.  This makes it clear that this column is an
average OR for the exposed categories, and the heading in Table 4 should
have been rendered as ‘Ever* v low/no’.  Perhaps here the heading
should be ‘Ever v low/no’.

P213 Table V-C-4.  The ‘trend’ for lung cancer is not very
convincing, and the increases claimed are not large; a reminder that any
difference can be declared statistically significant with large enough
numbers.

P215 para 3  Variables are not ‘correlated in models’;  their
effects may be, i.e. demonstrate confounding.

P217 para 2.  ‘Gender’ (i.e. sex) was matched for. The analysis
method was unconditional; was gender not adjusted for?

P220 para 1 …range of relative risk estimates was…

P220 para 2 SIR is not an ‘incident rate’ but a ‘Standardized
Incidence Ratio’.

P223 para 1 …supplemented by stepwise multiple regression analysis
using proportional hazards modelling.

P223 para 2 SMRs are ‘Standardized Mortality Ratios’, not
‘standard mortality rates’.  The notions that SMRs were
‘elevated’, ‘there were trends’ that ‘became more
pronounced’, but then none of this was statistically significant,
seems like serious over-claiming.  Another way to put it is that there
was no convincing evidence of a relationship.  

P224 para 2 explain what the ‘computer simulations’ were doing here?
 Trying to compensate for an absence of data?

C-4.	OSHA has reviewed a number of studies that examined the
relationships between exposure to silica and cancer at sites other than
the lung, including larynx and nasopharynx, stomach, and esophagus. 
Have we missed any important studies of cancer at other sites? If yes,
inform OSHA of those studies.

I’m not aware of any additional studies regarding these or other
sites.

P252 Something wrong with the reference dates?

P258 para 3 …greater than one hundred times…

C-5.	OSHA has preliminarily concluded that an association has not been
established between silica exposure and excess mortality from cancer at
sites other than the lung.  Do you agree with this interpretation of the
data?  If not, please provide a detailed rationale for your different
interpretation.

I agree that the evidence is not convincing for increased risks at any
of these other sites.

Questions on Section D. Other Nonmalignant Respiratory Disease

D-1.	This draft section discusses COPD, NMRD, bronchitis, emphysema,
airways disease, etc.  Is it clear in the discussion how these health
endpoints are related or distinct entities?  If not, tell OSHA how to
improve the presentation.

It would be useful to define these terms clearly when they are first
used, including how they are usually defined in both clinical and
epidemiological contexts. It would then be clearer which are subsets of
others. Note that there are several different health endpoints here, and
it may not be appropriate to add together the numbers of cases of
different sorts.

P274 para 1.  Insert a definition of chronic bronchitis at beginning.

P275 para 2 ’reflected’ is vague. Chronic sputum production is a
defining symptom of bronchitis.

P275 para 3 explain “survivor” effect?

P286 para 2  Both ‘dust’ and ‘silica’ exposures are mentioned. 
Are these the same?   If so, maybe use ‘silica’ throughout?  If this
section is addressing both silica and mixed dust, don’t jump between
them so much, deal with them separately.

P289 para 1 multivariate analyses -> multiple regression 

P292 para 2 delete ‘In a bivariant analysis,’

P293 para 1 multivariate -> multiple regression

P298 para 2 spell out COPD

P308 para 2  Do the ‘overall conclusions’ need to distinguish those
findings related to silica exposure from those associated with
‘dust’?

Questions on Section E. Renal and Autoimmune Effects

E-1.	OSHA has preliminarily concluded that studies on renal and
autoimmune effects have not provided sufficient data (in quality or
amount) to support quantitative estimates of risk.  From your knowledge
and understanding of these studies, do you agree with OSHA’s
conclusion?  If you believe that there is sufficient data to support
quantitative estimates of risk, please tell OSHA specifically what data
you think could be used.

I agree that a full quantitative assessment is not yet possible.

Questions on Section F. Physical Factors that May Influence Toxicity of
Crystalline Silica

F-1.	The draft “Physical Factors…” section contains a discussion
of proposed mechanisms of action by which silica causes silicosis and
lung cancer.  Does the discussion of the mechanism of action reflect the
most recent thinking on this subject?  If not, tell OSHA what the most
recent thinking is, as you understand it.

My knowledge and experience do not qualify me to comment on the
biological mechanisms proposed.

P360 para 1 Other work has also shown agreement with surface area as a
metric.  Include references?

P360 para 2 then -> than

P366 para 2 then -> than

P378 para 2 quart -> quartz

P396 para 1 This refers to more severe silicosis, and not to the milder
definitions sometimes used elsewhere (e.g. ILO 1/0+)

F-2.	The draft Mechanism of Action section contains a discussion on the
possible role of silicosis as a precursor or prerequisite for lung
cancer.  Does this discussion accurately reflect the current scientific
opinion on this issue?  If not, tell OSHA what the current scientific
opinion is on this issue, as you understand it.

P401 para 2 Silicosis cannot be a ‘prerequisite for lung cancer’ -
many people contract lung cancer without either silicosis or silica
exposure - and so this phrase should never be used unqualified. Whether
any excess silica-related lung cancer risk operates through fibrosis or
other aspects of the silicotic response is a difficult question, and the
text reflects this.  

F-3.	In the draft “Physical Factors…” section OSHA preliminarily
concludes that the available information on the physical factors that
may influence crystalline silica toxicity cannot be used at this time to
refine quantitative estimates of the lung cancer and silicosis
mortality.  Does this seem to be an appropriate conclusion based on the
information presented?  Is it clear in the discussion how OSHA arrived
at this conclusion?  Would you arrive at a different conclusion based on
the information presented?  If so, tell OSHA what your conclusion would
be and explain how you arrived at it?  

Experiments in vitro and in vivo suggest that the toxicity of silica
varies between samples from different sources, and that some external
factors can modify the response to either reduce or intensify the
effects.  It’s difficult to see how those results could be generalized
from present knowledge.  The prudent approach is to regulate for those
situations where at is most toxic; I don’t see how you could adjust
the regulation across different industries or situations where the
toxicity differs.  



Andrew Salmon

Private Consultant, Lafayette, CA

Review of OSHA’s Preliminary Health Effects Section for Silica 

General Questions

1.	The draft health effects section reviews and summarizes a number of
studies.  Are the studies described in sufficient detail for the reader
to understand how the studies were done?  Are the descriptions of the
studies and results accurate?  Are the strengths and limitations of the
studies adequately discussed and appropriately evaluated?  Are OSHA’s
interpretations of the study results reasonable and explained
adequately?  Are discussions of studies clear and easy to follow?  If
not, let OSHA know what needs to be added to improve the information and
the manner in which it is presented.

Response:

The format used for descriptions of studies in the health effects
section is appropriate, and effectively lays out the content,
methodology, strengths and weaknesses of the studies described.  In
general the section is well written and clearly explains OSHA’s
interpretation of the available data.  There are a few cases where
studies have been either ignored completely, or given less consideration
than they deserve, (or, alternatively, over-emphasized), but these
difficulties are isolated and will be identified below where
appropriate.  The level of detail given in this section is sufficient
for its purpose, which is mainly a qualitative description of the health
effects of silica exposure: more details, including report of the
detailed quantitative evaluations of the studies by Toxichemica Inc.
(2004), are appropriately deferred to the section on quantitative risk
assessment.  The section represents a commendable attempt to reduce an
enormous literature to manageable proportions.  This is most successful
in the consideration of lung cancer, where the previously published
review by IARC (1997) and the meta-analysis by Steenland et al. (2001a)
provide an overall perspective.

2.	The draft health effects section covers a selection of studies in
each topic area.  The discussion does not cover every study ever done on
the topics.  Do you think that OSHA should have discussed additional or
different studies on a specific health endpoint?  Are you aware of
significant studies that were omitted?  If yes, inform OSHA of those
studies.  

Response:

The discussion of silicosis morbidity and mortality starts from the
premise that silicosis is the inevitable and unique result of exposure
to respirable crystalline silica particles, and provides only a very
brief mention of some of the historical reports underlying this
conclusion.  It then proceeds to review further only those studies which
the authors consider to provide information on specific topics
qualifying that association, such as the precision, repeatability and
sensitivity of the ILO radiographic method for diagnosis, the
association with other diseases such as tuberculosis, the probability
and rate of progression after initial diagnosis, etc.  While this is
somewhat reasonable in view of the large literature, it results in an
unbalanced view of the literature and produces some odd results.  Thus
this section does not describe some of the key studies used for
quantitative analysis in the following section, or only refers to them
in the context of some subsidiary point, without laying out the main
findings of a dose relationship between exposure to respirable dust
containing crystalline silica and observation of silicosis as defined by
radiography, lung function and/or pathological criteria.  It would
improve the overall perspective of this part of the narrative if these
important studies were described here, in addition to the analysis of
surveillance data.  I understand the logic of keeping a discussion of
the dose-response characteristics to the following section where they
are treated in detail.  However this section would be improved by a
demonstration that such a relationship does exist.  It would also help
the discussion of important qualitative differences such as the
distinction between acute silicosis (a high-dose effect) and the
longer-term responses.

The surveillance data which are reviewed certainly support the
assumption of a relationship between silicosis and exposure to silica,
but provide a somewhat strange perspective due to their individual and
clinical, rather than statistical, perspective.  Additionally they may
seriously understate the prevalence of the disease.  Not only is
mortality a severe endpoint which may not affect more than a minority of
those affected in less extreme ways, but as succinctly demonstrated in
the analysis presented here is prone to substantial under-reporting. 
Much of the surveillance data cited here provides a limited perspective.
 For a start it is largely confined to United States data which of
itself ignores the international dimension of the problem, and may
produce other weaknesses resulting from the peculiarities of the United
States health care system.  In general, apart from special projects
following pre-defined cohorts, United States data are significantly
lacking compared to those in other countries (especially Scandinavia and
the EU) where population-wide health care systems exist.  With regard to
morbidity statistics, in the case of silicosis these are mainly related
to workers’ compensation systems, which include an element of bias
towards a negative diagnosis (thus avoiding payment) unless the evidence
is overwhelming: thus these reports are likely to show a high degree of
accuracy but low sensitivity compared to an analysis without this bias,
even if the same nominal criteria are used.

3.	The draft health effects section covers a number of health endpoints.
 Are you aware of additional significant health endpoints affected by
exposure to crystalline silica that should be discussed?  If so, inform
OSHA of the additional health endpoints and studies that address them.

Response:

I am not aware of any additional endpoints which should be considered.

4.	The draft heath effects section contains conclusions for each topic
area as well as an overall conclusion.  Are these conclusions reasonable
in light of the available data?  Is it clear in the discussion how OSHA
arrived at those conclusions based on the studies discussed? 

Response:

The health effects summaries are clear and reasonable.

Questions on Section A. Introduction

A-1.	Is OSHA’s review of the history of dust exposure monitoring
helpful to the reader in understanding the needs and limitations of
exposure conversions to a modern mass per volume metric in reference to
the exposure-response relationship?

Response:

This section provides a summary of the major types of measurement
technology which have been used for estimating exposure to respirable
silica.  While accurate as far as it goes, this provides a rather
limited perspective.  OSHA decided some time ago that the ideal method
for measuring exposures to silica dust involves personal sampling using
a cyclone to remove non-respirable particles and a filter to collect the
respirable particles for gravimetric and X-ray crystallographic
analysis.  There are good reasons for choosing this methodology, but
this section does not do a very good job of explaining what these are. 
The section is somewhat dismissive of the various methods based on
particle counting equipment, characterizing these as “historical”
and mainly addressing how their results could be converted to the
OSHA-preferred gravimetric data.  It is not that the gravimetric measure
is intrinsically technically superior, but rather that the availability
of particle counts and size- or surface-area measurements as well as
gravimetric analyses made an important contribution to the eventual
conclusion that 1) long-term cumulative exposure was the chief
determinant of chronic silicosis and 2) the gravimetric measure is the
appropriate dose metric to be related to health effects.  It is curious
that the authors of this section chose to report conversion factors used
in the analysis of various US studies (Table V-A-1), but did not include
any quantitative analysis of the various reports on dust exposures in
the South African gold mines, which represent an important cohort for
study of health effects and for which there are a number of publications
presenting such comparative analysis (Beadle and Bradley, 1970;
Page-Shipp and Harris, 1972; DuToit 1991 and others).  I also do not see
any description of modern electronic particle-counting systems, some of
which are capable of providing a gravimetric description of the
distribution of particle sizes in an atmosphere, on a continuous basis. 
While these may be unsuitable for personal monitoring of workers, they
may provide important data especially for outdoor environments where
dust exposures are a concern.  More detailed consideration of particles
size and surface measures might further inform the later discussion of
factors potentially affecting the potency of specific types of silica
dust.

Questions on Section B. Silicosis

B-1.	OSHA presents background information on the use of radiography,
imaging techniques, pulmonary function studies, and other diagnostic
tools used in clinical settings and for in epidemiological studies. 
Does OSHA correctly present the benefits and shortcomings of these
techniques?  Are you aware of any additional information that would
either strengthen this discussion or contradict what is written?  

Response:

This chapter provides a good summary of the available methods for
recognition and evaluation of silicosis and related lung diseases.  It
makes an important point about the relatively low sensitivity of
radiography, as shown by Hnizdo et al. (1993), Craighead and Vallyathan
(1980) and others.  This point is important in evaluating the prevalence
of silicosis as reported in the various epidemiological studies
considered.  The discussion of the lung function test results in
patients affected by silicosis (however diagnosed or suspected) is also
useful.

B-2.	OSHA attempted to describe the available silicosis surveillance
data.  In your opinion, were any important topics left uncovered?  Any
data sources not discussed?  Is the subject of under-reporting
sufficiently covered?  Are the conclusions supported by the discussion
and stated succinctly?  Are you aware of any important studies in this
area that OSHA did not cover?

Response:

This description of surveillance data is sufficient as far as it goes in
describing the U.S. data.  There may be reasons why other sources of
this type of data, which certainly exist in various parts of the world
(U.K. in particular, many others), were not considered, but these
reasons are neither presented nor defended.  Specific problems relating
to under-reporting are identified and discussed.  However, as noted
previously in the general comments, the limitations in the surveillance
approach generally and in particular the concentration on only the most
severe effects (mortality, substantial disability and high-grade
radiographic diagnoses) limit the usefulness of this data source for
anything beyond the qualitative identification of disease end-points.

There is an additional, recently published study that OSHA should
consider, which is related to those using surveillance data by its
objective of examining the prevalence of silicosis in certain population
groups.  This is the study of pneumoconiosis in Californian agricultural
workers exposed to silica and silicate-containing dust, by Schenker et
al. (2009).  Instead of using reported illness or death certificates,
these authors used histological analysis and measurements of accumulated
particles, using lung specimens obtained at coroner’s autopsy. 
Significant degrees of pneumoconiosis and other pulmonary disease
indications were apparent in agricultural workers compared to others not
exposed to the mineral dusts prevalent in dry farming activities.  This
result is interesting both as an indicator of silicosis-type disease in
an occupational group not considered in the OSHA report and as a
possible contribution to the debate about aged vs. fresh silica
particles, surface occlusion by clays and so on.

Consideration of studies from outside the U.S. describing prevalence of
silicosis is entirely lacking: this is a substantial omission if the
objective was to provide a balanced view of the subject.  Some of the
key studies from the U.K., Asia, South Africa etc. are described in the
subsequent discussion of silicosis progression in this section, or in
the dose-response section, but their omission here is peculiar.  And
some important studies are omitted in both places.  It is a downright
eccentricity that the authors refer briefly to the study by Churchyard
et al. (2004) in the lung cancer chapter as a source of gravimetric
measurements of silica in South African mines, but this study is nowhere
mentioned either in this section or the dose-response section for its
presentation of data on the prevalence and dose-response of silicosis in
black South African miners, which is its primary subject.

B-3.	For the discussion of progression of silicosis, OSHA did not review
every study that addressed this issue.  Are you aware of additional
studies that OSHA should consider, especially any studies that
illustrate progression in a cohort exposed to silica in the range of
OSHA’s current PEL (0.1 mg/m3)?   

Response:

This part of the report does a good job of distilling what is certainly
a very large body of published data.  The key point, that progression
occurs both in the presence and absence of continuing exposure to
crystalline silica dust, is well made.  I do not have any suggestions
for additional studies that should be considered.

B-4.	Results of studies covered in the PFT section are not always
consistent.  Did OSHA do a fair job of presenting these results and
drawing conclusions?  Are there any important studies that you are aware
of that would strengthen this section or the conclusions of this
section?

Response:

I think this section is reasonable and well-considered.  The conclusions
drawn are cautious and defensible.  Given the complex and conflicting
nature of the data I think this chapter does an excellent job of the
analysis.  I am not aware of any additional relevant studies (although I
have to say that since I am not an expert in clinical lung function
testing I defer to others for a definitive judgment on that point).

Questions on Section C. Carcinogenic Effects of Silica (Cancer of the
Lung and Other Sites)

C-1.	With such a large amount of epidemiological literature to review,
OSHA may be subjected to criticism of selection bias.  Is the selection
criteria used by OSHA clear?  Have we missed any important studies that
have good or at least reasonable exposure estimates?  Do you recommend
any alternate or additional selection criteria?

Response:

This section does a good job of presenting an overview of the large and
complex topic, which is helped by the earlier reviews by IARC and NIOSH.
Unlike the previous section which dealt with silicosis, this section
does present an overview of the various studies not only in the U.S. but
worldwide.

I do not have any specific suggestions as to additional reports to
include.  OSHAs independent review is certainly comprehensive as regards
both the inclusion of relevant studies, and the breadth of different
industries with silica exposure which were considered.  In response to
the question about selection criteria it would appear that OSHA has
taken the inclusive approach of at least describing any possibly
relevant studies, although obviously giving weight in the final analysis
to those offering reasonable determinations of exposure to crystalline
silica, without complicating concurrent exposures to other potentially
carcinogenic materials.

C-2.	OSHA concludes that the studies involving four cohorts among the
20+ cohorts and case controls reviewed were the strongest in terms of
several criteria in record availability and data quality (exposure,
smoking, worker and facility history) as well as data treatment and
analyses?  Do you agree with this conclusion?  Would you add or delete
any of these studies?  

Response: 

 I’m not sure exactly which part of the report this charge question is
referring to – the chapter on lung cancer reviews about one hundred
studies in total, with 23 occupational cohort studies, 5 case-control
studies and 5 silicosis cohorts analyzed in detail.  The preliminary
conclusions (on page 237) cite five cohort studies as being “of the
best quality to evaluate exposure to crystalline silica as a risk factor
for lung cancer”.  These were 

Diatomaceous Earth Workers (Checkoway et al. 1993, 1996, 1997 and 1999;
Seixas et al. 1997)

British Pottery Workers (Burgess et al., 1997 and 1998; Vherry et al.,
1998; McDonald et al., 1995)

Vermont Granite Workers (Applebaum et al., 2007; Attfield and Costello,
2004; Costello and Graham, 1998; Davis et al., 1983)

North American Industrial Sand Workers (Hughes et al., 2001; McDonald et
al., 2001, 2005; Rando et al., 2001)

North American Industrial Sand Workers (Sanderson et al., 2000;
Steenland and Sanderson, 2001)

This list certainly identifies a selection of key studies with adequate
statistical power and reliability, and reasonable exposure and health
endpoint determination.  It seems to me that the omission of the South
African gold-mining studies is a mistake, particularly with regard to
the more recent reports (Hnizdo and Sluis-Cremer, 1991; Hnizdo et al.,
1997), about which OSHA comments somewhat favorably (page 236).  While
there are some questions about the exposure measurements used and
possible confounding by radon exposure it appear that most of these have
been addressed by related studies.  This group of miners has been an
important source of data on silica effects over the years and
historically these studies have contributed greatly to the worldwide
concern about, and understanding of, silica exposures and health
effects.  The fact that a modern epidemiological study finds a positive
association with lung cancer in this group is an important and
persuasive part of the whole jigsaw-puzzle.  In the last analysis
however the most persuasive part of the argument is not any one of the
single study reports, but the meta-analysis by Steenland et al. (2001a).

C-3.	Do the tables add to the text presentation?  Do you have any
recommendations to improve their content or presentation?  Would you add
or delete another table?  Provide an explanation for any proposed
deletion or a mock-up of any proposed addition.

Response:

The tables are fine as far as I am concerned

C-4.	OSHA has reviewed a number of studies that examined the
relationships between exposure to silica and cancer at sites other than
the lung, including larynx and nasopharynx, stomach, and esophagus. 
Have we missed any important studies of cancer at other sites? If yes,
inform OSHA of those studies.

Response:

I am not aware of any additional studies that should be considered.

C-5.	OSHA has preliminarily concluded that an association has not been
established between silica exposure and excess mortality from cancer at
sites other than the lung.  Do you agree with this interpretation of the
data?  If not, please provide a detailed rationale for your different
interpretation.

Response:

I think that is a reasonable conclusion.  There may be grounds for
suspicion of such effects, but they do not reach the level of certainty
to establish such a connection.

Questions on Section D. Other Nonmalignant Respiratory Disease

D-1.	This draft section discusses COPD, NMRD, bronchitis, emphysema,
airways disease, etc.  Is it clear in the discussion how these health
endpoints are related or distinct entities?  If not, tell OSHA how to
improve the presentation.

Response:

This section provides a clear and reasonable discussion of the presence
and relationship between these disease endpoints in those exposed to
silica dust.  It is important to emphasize that these different
pulmonary disease manifestations, plus the radiographic diagnosis of
silicosis, are all reflections of an underlying disease process related
to silica exposure, and no single diagnosis includes all possible
individual responses.  It is therefore important to look at all these
diseases entities in assessing the overall disease burden from exposure
to silica dust.

Questions on Section E. Renal and Autoimmune Effects

E-1.	OSHA has preliminarily concluded that studies on renal and
autoimmune effects have not provided sufficient data (in quality or
amount) to support quantitative estimates of risk.  From your knowledge
and understanding of these studies, do you agree with OSHA’s
conclusion?  If you believe that there is sufficient data to support
quantitative estimates of risk, please tell OSHA specifically what data
you think could be used.

Response:

OSHA comments in the description of the available studies that the
evidence is convincing for an association between renal disease and
exposure to silica.  This certainly seems to be the case, in that there
are a substantial number of positive epidemiological studies finding
such an effect, as well as a large number of case reports.  The
plausibility of a causal association is enhanced by various laboratory
studies, finding of silica particles in kidney tissues, etc., which lay
the foundation for a reasonable mechanistic explanation of how such an
effect might occur.  The existence of a few “negative” (i.e.
non-positive) reports is not a convincing counter-argument, since these
observations could easily be explained by low power, misclassification
of exposures, competing effects and various other features of those
studies.  In general, one or more good positive epidemiological studies
are much more convincing than any number of studies which failed to find
any clear result.

OSHA’s chapter on this topic also specifically describes at least two
recent and major studies (McDonald et al., 2005 and the pooled analysis
by Steenland et al., 2002a) which explicitly determine a quantitative
relationship between silica exposure and renal disease.  Although
McDonald et al. do not argue in favor of a causative explanation for
their observation, this may be simply a case of those authors confining
their comments to a conservative interpretation of their own data,
rather than attempting to consider the implications of the whole body of
published evidence on the topic.  OSHA’s reluctance to consider a
quantitative analysis of renal disease and mortality therefore seems
unjustified.  It would be entirely reasonable to endorse the existing
quantitative analysis by Steenland et al. (2002a), particularly as many
aspects of this cohort and the accompanying exposure data have been
extensively validated in the consideration of silicosis and lung cancer.
 OSHA’s conclusion on page 345 states that there are “only two
studies on renal disease provide quantitative exposure-response data”
– this in itself is something of an understatement, but even as it
stands it contradicts the following sentence which asserts that the data
are “insufficient”.  OSHA does not define what it means by a
“robust” estimate, but seems to be using this nebulous desideratum
to hide behind in this case.

The data on rheumatoid arthritis and other autoimmune conditions also
include some published quantitative analyses, so it is somewhat
disingenuous of OSHA to contend that the data will not “support
quantitative estimates of risk” when that is precisely what some
authors have successfully provided.  However, the issue is not whether
such estimates can be made, but what level of confidence should be
placed in them and whether they have any implications for setting
health-protective policy.  In the case of the arthritis and autoimmune
responses obviously the data are somewhat less convincing and
comprehensive, due at least in part to the complexity and variability of
this type of disease.  It could certainly be argued that the
quantitative estimates available for these health endpoints are somewhat
uncertain, and the links required to confidently assume causality are
not as well established as for the other endpoints discussed in this
section.  This is not in itself a good argument for dismissing the
effort to apply a quantitative analysis even where substantial
uncertainties are present, although obviously the associated
uncertainties would need to be addressed when considering the results.  
Indeed, it is particularly important to pursue an appropriate
quantitative analysis where possible in such a case, because it is only
by use of quantitative analytical techniques that the situation can be
properly defined, including identification of the sources and extent of
uncertainty, and estimation of confidence limits as well as measures of
central tendency for key dose-response parameters.

Questions on Section F. Physical Factors that May Influence Toxicity of
Crystalline Silica

F-1.	The draft “Physical Factors…” section contains a discussion
of proposed mechanisms of action by which silica causes silicosis and
lung cancer.  Does the discussion of the mechanism of action reflect the
most recent thinking on this subject?  If not, tell OSHA what the most
recent thinking is, as you understand it.

Response:

As far as I am aware this discussion reflects current thinking on this
topic.  The various identified markers of inflammatory response, and the
likely involvement of reactive oxygen toxicity are widely thought to
play a role in both silicosis and lung cancer, although obviously at
this point there is no absolute and complete description of the
mechanism of action for either endpoint.

F-2.	The draft Mechanism of Action section contains a discussion on the
possible role of silicosis as a precursor or prerequisite for lung
cancer.  Does this discussion accurately reflect the current scientific
opinion on this issue?  If not, tell OSHA what the current scientific
opinion is on this issue, as you understand it.

Response:

This discussion presents several key points on this question, which I
believe reflect current thinking.  It seems likely that both silicosis
and lung cancer share some common mechanistic steps at the beginning of
the processes which lead to these responses, but it is simply unknown
how far up the chain of events this commonality extends.  One key
difficulty with the underlying question is what initial stages can be
properly called “silicosis” in this context.  As has been eloquently
described elsewhere in this section, the standard indicators used to
diagnose silicosis are X-radiography and post-mortem histology, both of
which are “late” indicators and will certainly not serve to identify
initial stages of the disease which might, or might not, be required
steps in the eventual development of lung cancer.

F-3.	In the draft “Physical Factors…” section OSHA preliminarily
concludes that the available information on the physical factors that
may influence crystalline silica toxicity cannot be used at this time to
refine quantitative estimates of the lung cancer and silicosis
mortality.  Does this seem to be an appropriate conclusion based on the
information presented?  Is it clear in the discussion how OSHA arrived
at this conclusion?  Would you arrive at a different conclusion based on
the information presented?  If so, tell OSHA what your conclusion would
be and explain how you arrived at it?  

Response:

I consider that the description presented here is a clear, fair and
reasonable analysis of an admittedly complicated situation.  It seems to
be the case that different samples of crystalline silica dust do present
quantitatively (although probably not qualitatively) different hazards 
Differences between samples appear to be of basically unknown origin. 
Factors such as recent cleavage, occlusion by other minerals, surface
contaminants and so on may well influence potency, but it has not been
possible to consistently or quantitatively establish these connections. 
The only physical characteristic which does seem to have been
established as definitive is particle size – the particles need to be
respirable to cause the effect.

References

(other than those cited in the Health Effects document)

Beadle DG, Bradley AA. 1970. The composition of airborne dust in South
African gold mines. In: Shapiro HA (ed). Pneumoconiosis. Proceedings of
the International Conference. Johannesburg 1969. Cape Town: Oxford
University Press. pp. 462-6.

Page-Shipp RJ, Harris E. 1972. A study of dust exposure of South African
white gold miners. South Afr Inst Mining Metall. 73:10-24.

Schenker MB, Pinkerton KE, Mitchell D, Vallyathan V, Elvine-Kreis B,
Green FHY (2009).  Pneumoconiosis from Agricultural Dust Exposure among
Young California Farmworkers.  Environmental Health Perspectives 117(6):
988–994.



Noah Seixas

University of Washington, Seattle, WA

For Review of OSHA’s Preliminary Health Effects Section for Silica

Review Comments: Noah Seixas, PhD, CIH

December 13, 2009

General Questions

The draft health effects section reviews and summarizes a number of
studies.  Are the studies described in sufficient detail for the reader
to understand how the studies were done?  Are the descriptions of the
studies and results accurate?  Are the strengths and limitations of the
studies adequately discussed and appropriately evaluated?  Are OSHA’s
interpretations of the study results reasonable and explained
adequately?  Are discussions of studies clear and easy to follow?  If
not, let OSHA know what needs to be added to improve the information and
the manner in which it is presented.

In sum, yes.  OSHA has clearly and thoroughly evaluated the extensive
literature relating silica exposure to the key health effects known or
suspected to be related to this exposure.  The review is extensive,
covers all of the major studies to my knowledge, and provides an
appropriate level of discussion of each study.  In particular, the
methods, results, limitations and strength of the evidence are
considered for each study reviewed. Most of the document is clearly
written.

However, I think the document is lacking an adequate “Introduction.”
 The section labeled introduction is a detailed discussion of sampling
methods.  While this is relevant, I don’t believe it should be labeled
as introductory material.  In particular, I think the document as a
whole is lacking clear definitions of many of the concepts and
terminology used throughout the document.  There is even no clear
definition of the terms silica, crystalline silica, quartz, etc.  These
terms should be clearly defined at the beginning of the document, and
used consistently throughout.

The draft health effects section covers a selection of studies in each
topic area.  The discussion does not cover every study ever done on the
topics.  Do you think that OSHA should have discussed additional or
different studies on a specific health endpoint?  Are you aware of
significant studies that were omitted?  If yes, inform OSHA of those
studies.  

I am not aware of any significant studies on silica and health effects
that were not reviewed in this document. There are studies which
consider alternative dose metrics for silica, but these are more
appropriately addressed in the risk assessment document.  There is a
substantial literature on exposures, and exposure control techniques for
silica in various industries.  These studies are not reviewed here, I
think appropriately given the goals of the Health Effects document. 

The draft health effects section covers a number of health endpoints. 
Are you aware of additional significant health endpoints affected by
exposure to crystalline silica that should be discussed?  If so, inform
OSHA of the additional health endpoints and studies that address them.

The document addresses silicosis, lung function limitation including
obstructive lung disease, lung cancer, cancers other than lung, renal
and immunological endpoints.  These are known health effects associated
with silica exposure.  I am not aware of other significant health
effects of silica exposure.  

The draft heath effects section contains conclusions for each topic area
as well as an overall conclusion.  Are these conclusions reasonable in
light of the available data?  Is it clear in the discussion how OSHA
arrived at those conclusions based on the studies discussed? 

Yes.  I think OSHA has made very reasonable summaries and
interpretations of the various aspects of health effects associated with
silica.  The literature reviews are comprehensive, clear and adequate
for making the determinations provided in this section. 

Questions on Section A. Introduction

A-1.	Is OSHA’s review of the history of dust exposure monitoring
helpful to the reader in understanding the needs and limitations of
exposure conversions to a modern mass per volume metric in reference to
the exposure-response relationship?

This section of the document is important, given the historical changes
in dust monitoring techniques that have occurred, and the relevance of
these changes to estimating dose for chronic conditions associated with
silica.  I think that this discussion is important to the integrity and
interpretability of the document as a whole.  However, I find this
particular section poorly organized and presented. I think it could be
presented in both simpler and clearer terms.  

In particular, there are several issues within the context of
measurement technique and definitions, that are frequently confused. 
First, the duration of exposure monitoring should be discussed.  Grab
samples, or instantaneous samples should be distinguished from
short-term samples, from full-shift samples (called erroneously, TWA
samples).  Further, the duration of measurement is frequently confused
with the exposure metric – typically cumulative exposure.  These are
not the same issue and should not be confused.  Although some direct
reading instruments (DRI) are now available, they are not discussed. 
Although DRI are generally not relevant to historical exposure datasets,
they are increasingly in use and may be important in future studies –
their use and limitations should at least be addressed briefly.

Second, particle counts vs. mass concentrations should be discussed –
as it is.  The discussion of the conversion of mppcf to mg/m3 is useful
and thorough.  However, the rationale for the superiority of mass
concentration is not addressed adequately.  My understanding is that the
choice of mass fraction is primarily dependent on the relative
simplicity (and lower expense), rather than a biological
appropriateness.  While mass fraction may be a superior metric
biologically, the evidence for this is not presented. This issue should
be considered separately from how to convert between the two measures of
exposure.

Third, the rationale and definitions of particle size fractions are not
well described.  Because different studies, and different periods in
time have used different particulate size fraction measurement systems,
and because they may have different implications for different health
endpoints, this should be addressed more clearly.

Fourth, the differences in measurement of dust, vs. silica, vs.
crystalline silica should be more clearly explained.

Fifth, in the context of describing each of these issues, the use of
different sampling and analytic devices is also relevant and should be
described.  Sampling methods are substantially different over time and
between different countries.  Each of these sampling devices or
techniques have implications in terms of the issues discussed above, and
are therefore, relevant to the comparability of the data derived from
them. 

Finally, the choice of a standard based on crystalline silica
concentration in air, rather than the currently used formula combining
mass concentration and silica content of the dust should be explained. 
I believe this is a defensible approach, but given its shift from
historical practice, it would be useful to point out the limitations of
the earlier standard, and OSHA’s position on the superiority of the
new method of quantification.

  

Questions on Section B. Silicosis

B-1.	OSHA presents background information on the use of radiography,
imaging techniques, pulmonary function studies, and other diagnostic
tools used in clinical settings and for in epidemiological studies. 
Does OSHA correctly present the benefits and shortcomings of these
techniques?  Are you aware of any additional information that would
either strengthen this discussion or contradict what is written?  

The discussion of diagnostic tools, X-ray, CT, HRCT and pulmonary
function tests is appropriate, clear and adequately complete for
assessing their relations to silica-induced disease.  The discussion is
useful in terms of understanding the literature on silica risks,
however, there are no summary or conclusions with based on this
discussion that could be used for designing screening or surveillance
systems for silica-exposed workers.  Presumably, this discussion would
appear elsewhere before OSHA enters into recommendations for medical
surveillance requirements.

B-2.	OSHA attempted to describe the available silicosis surveillance
data.  In your opinion, were any important topics left uncovered?  Any
data sources not discussed?  Is the subject of under-reporting
sufficiently covered?  Are the conclusions supported by the discussion
and stated succinctly?  Are you aware of any important studies in this
area that OSHA did not cover?

The surveillance information presented is an excellent summary, covering
silicosis mortality and morbidity, silica exposure surveillance, and
estimates of under-reporting of these data.  Lessons learned from these
data systems, along with their limitations, are well discussed. 
However, one comes away with an important question.  Given the rapid and
consistent reduction in mortality, and the continuing exposure levels
over the current PEL, one must ask if the remaining disease observed is
not completely due to illegal (i.e., > PEL) exposures, rather than the
inadequacy of the current standard.  I think some additional discussion
of this, in particular the limitations in assessing non-silicosis
outcomes (e.g. COPD and Lung CA) in relation to silica exposures, is
warranted.  That is, just because there is a large reduction in
silicosis identified through the various surveillance systems, does not
suggest that there is little remaining risk, other than that where
exposures exceed the standard.

Given the limitations of the various surveillance data sources (both
exposure and outcomes), it should be pointed out that the best
information about the true risk of disease must come from well
constructed epidemiologic studies where the cohort can be adequately
defined, exposure can be sufficiently quantified, and outcomes can be
thoroughly ascertained.  

On p 63, I assume the quote from Windau should read…”pneumoconiosis
due to other inorganic dusts.” 

B-3.	For the discussion of progression of silicosis, OSHA did not review
every study that addressed this issue.  Are you aware of additional
studies that OSHA should consider, especially any studies that
illustrate progression in a cohort exposed to silica in the range of
OSHA’s current PEL (0.1 mg/m3)?   

No.  the discussion is well organized and to the extent of my knowledge,
accurate and complete.  The conclusions concerning progression, and
determinants of progression seem well founded.

B-4.	Results of studies covered in the PFT section are not always
consistent.  Did OSHA do a fair job of presenting these results and
drawing conclusions?  Are there any important studies that you are aware
of that would strengthen this section or the conclusions of this
section?

My reading of this presentation would suggest that OSHA has been overly
conservative in it’s conclusion concerning the association of
pulmonary function with the progression of silicosis.  Independent of
the role of emphysema, the literature supports a finding of decreased
lung function with progression of radiological silicosis.  The evidence
is most convincingly presented by the strongest study designs – those
with longitudinal follow-up of the cohorts.  The statement on page 101,
that, “it is plausible that exposure to silica impairs lung function
in at least some individuals at an earlier point in time than can be
detected on chest radiograph,” is unduly conservative.  

Questions on Section C. Carcinogenic Effects of Silica (Cancer of the
Lung and Other Sites)

C-1.	With such a large amount of epidemiological literature to review,
OSHA may be subjected to criticism of selection bias.  Is the selection
criteria used by OSHA clear?  Have we missed any important studies that
have good or at least reasonable exposure estimates?  Do you recommend
any alternate or additional selection criteria?

While I believe that the review conducted is comprehensive, and to the
degree that I am aware, has not missed any significant studies, I
don’t see a clear presentation of study selection criteria for the
studies reviewed.  I think it would be useful to state in the beginning
of this section how individual studies were selected.  Further, I would
be useful to identify any studies that were not included and provide the
rationale for why they were not.  

C-2.	OSHA concludes that the studies involving four cohorts among the
20+ cohorts and case controls reviewed were the strongest in terms of
several criteria in record availability and data quality (exposure,
smoking, worker and facility history) as well as data treatment and
analyses?  Do you agree with this conclusion?  Would you add or delete
any of these studies?  

I think the selected studies are among the most useful and complete
available. Again, I don’t think that OSHA has clearly stated the
criteria for inclusion, and demonstrated that the selected studies did
conform to these criteria, or that they were necessarily superior with
respect to the criteria than other studies which may not have been
included.  

Further, OSHA has selected 5 studies in four industries as the most
useful studies available.  While the attributes which are used to make
this judgment are provided, it is not until all the other studies are
reviewed that these criteria a stated (on page 237-8).  Stating these
criteria, and reviewing the literature with them as an explicit goal
could make a more convincing argument that the best studies were in fact
selected.

C-3.	Do the tables add to the text presentation?  Do you have any
recommendations to improve their content or presentation?  Would you add
or delete another table?  Provide an explanation for any proposed
deletion or a mock-up of any proposed addition.

Yes, however, there is no table summarizing the strength of the
evidence.  Although it might be more appropriate in the Risk Assessment
section, it might be possible to construct a table identifying the
strengths or weaknesses of each reviewed study (e.g., size of cohort,
strength of exposure characterization, number of cases, control of
confounding, etc.) and the summary of results in terms of RR or OR, etc.
Such a table could provide an easier way of summarizing the strength of
the evidence, unless the permutations become too unwieldy for a simple
table.

C-4.	OSHA has reviewed a number of studies that examined the
relationships between exposure to silica and cancer at sites other than
the lung, including larynx and nasopharynx, stomach, and esophagus. 
Have we missed any important studies of cancer at other sites? If yes,
inform OSHA of those studies.

I am not aware of any other cancer sites of importance with respect to
silica exposure.

C-5.	OSHA has preliminarily concluded that an association has not been
established between silica exposure and excess mortality from cancer at
sites other than the lung.  Do you agree with this interpretation of the
data?  If not, please provide a detailed rationale for your different
interpretation.

Given the studies reviewed by OSHA, I would agree with the assessment
that these other cancer sites are not clearly associated with silica
exposure, and if there is an underlying association, there are
insufficient data to provide any strong conclusions at this time.

Questions on Section D. Other Nonmalignant Respiratory Disease

D-1.	This draft section discusses COPD, NMRD, bronchitis, emphysema,
airways disease, etc.  Is it clear in the discussion how these health
endpoints are related or distinct entities?  If not, tell OSHA how to
improve the presentation.

This section has a high degree of overlap with section B3, in which
respiratory impairment associated with the appearance of silicosis is
discussed.  It is not clear to me why this section is presented
separately.  Nevertheless, the endpoints discussed a reasonable, well
defined and appropriately reviewed.  Integration of these studies with
those on impairment in silicosis would make more sense to me. 

Questions on Section E. Renal and Autoimmune Effects

E-1.	OSHA has preliminarily concluded that studies on renal and
autoimmune effects have not provided sufficient data (in quality or
amount) to support quantitative estimates of risk.  From your knowledge
and understanding of these studies, do you agree with OSHA’s
conclusion?  If you believe that there is sufficient data to support
quantitative estimates of risk, please tell OSHA specifically what data
you think could be used.

While the data appear to be sufficient to associate silica exposure with
these endpoints, I would agree that there are insufficient data for a
quantitative risk assessment at this time.  If one required quantitative
estimates of the risk of renal disease and silica exposure, it would be
feasible, given the available studies.  However, I don’t see a
compelling reason to do so, given the much more robust data resources on
silicosis and lung cancer.  Nevertheless, the clear association of renal
and autoimmune disease endpoints with silica exposure is an important
piece of the overall health effects assessment.

Questions on Section F. Physical Factors that May Influence Toxicity of
Crystalline Silica

F-1.	The draft “Physical Factors…” section contains a discussion
of proposed mechanisms of action by which silica causes silicosis and
lung cancer.  Does the discussion of the mechanism of action reflect the
most recent thinking on this subject?  If not, tell OSHA what the most
recent thinking is, as you understand it.

I am not qualified to discuss physical factors in any depth.  The data
presented by OSHA concerning these aspects of silica exposure and risk
appear to be well documented, thorough, and support the conclusions made
in the document.  

F-2.	The draft Mechanism of Action section contains a discussion on the
possible role of silicosis as a precursor or prerequisite for lung
cancer.  Does this discussion accurately reflect the current scientific
opinion on this issue?  If not, tell OSHA what the current scientific
opinion is on this issue, as you understand it.

I am not qualified to discuss the biological basis of silica toxicity in
any depth.  The data presented appear well documented, relevant to our
understanding of silica health effects, and appropriately interpreted.

F-3.	In the draft “Physical Factors…” section OSHA preliminarily
concludes that the available information on the physical factors that
may influence crystalline silica toxicity cannot be used at this time to
refine quantitative estimates of the lung cancer and silicosis
mortality.  Does this seem to be an appropriate conclusion based on the
information presented?  Is it clear in the discussion how OSHA arrived
at this conclusion?  Would you arrive at a different conclusion based on
the information presented?  If so, tell OSHA what your conclusion would
be and explain how you arrived at it?  

Given the historical suggestion that polymorphs other than quartz had
higher toxicity, this is an important discussion.  I agree that there is
insufficient evidence available to suggest that this is true, and in
fact, most of the evidence presented suggests that the polymorphs are of
comparable toxicity.  Further, the evidence concerning freshly fractured
silica and silica in mixed moiety (e.g. clay) have lower toxicity is
important and compelling.  However, I think OSHA has presented the
evidence as completely as possible, and made reasonable conclusions
based on the evidence.  I agree with OSHA that there is insufficient
evidence or consistency in the data to provide guidance on differential
control of exposure based on physical and surface characteristics. 

Technical Charge to External Peer Reviewers

Peer Review of OSHA’s Preliminary Quantitative Risk Assessment for
Silica

ERG Contract No. GS-10F-0125P

BPA No. DOLQ059622303

ERG Task No. 0193.15.064.001

Instructions to Peer Reviewers

You have been selected to participate in an external peer review of the
OSHA preliminary quantitative risk assessment for silica.  The risk
assessment section will be part of a proposed rule that would amend the
existing regulation for occupational exposure to silica.  The
quantitative risk assessment is the primary analysis upon which the
Agency must rely when making its determination that employees exposed to
silica at the current permissible exposure limit face a significant risk
of material impairment to their health and that the proposed standard
will substantially reduce that risk.  It is not meant to include all
adverse effects that result from silica exposure.  Other sections of the
proposed rule discuss these health hazards and the evidence that
supports their association with silica. 

The chosen data set(s) and dose metric(s) for exposure-response analysis
should be applicable to occupational exposures and adequately
demonstrate that silica influences the selected health effect.  The risk
model(s), assumptions and statistical treatments should be reasonable,
sound, and consistent with the underlying science.  

Keep in mind that OSHA’s responsibility is to protect worker health
and the Agency may choose conservative assumptions where appropriate. 
OSHA may also consider animal and experimental studies, where relevant. 
The overriding goal of the peer review is to ensure that the data and
methodologies used to estimate risk are clear and scientifically
credible.  OSHA has provided the following questions to guide and focus
your review on issues of importance to OSHA.  Your review should, at a
minimum, address each question, providing a discussion and rationale for
any “yes” or “no” responses, and providing other comprehensive
comments to clarify your comments and/or recommendations.

 

Charge Questions

The draft report evaluates exposure-response data from several
occupational studies as to their suitability for quantitative assessment
of lung cancer and silicosis risk resulting from airborne exposures to
silica.  Are the strengths and limitations of the selected data sets
adequately discussed and appropriately evaluated?  Are you aware of
other study data that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica?  If yes, inform
OSHA of the other data.

 The draft assessment relies on quantitative uncertainty analysis and
sensitivity analysis of exposure measurement error in key studies of
lung cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach?  If so, explain.

The draft assessment evaluates the use of a physiologically-based
pharmacokinetic model that predicts silica lung burdens in individuals
exposed to particulates through inhalation for estimating risk from lung
cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach? If so, explain.

The draft assessment summarizes several published risk assessments that
use a variety of models and statistical methodologies employed by the
various analysts to fit the featured data sets.  Are the models and
approaches used by these investigators adequately explained and
evaluated?  If not, let OSHA know what needs to be added to the
presentation.

The draft report identifies and describes several key areas of
uncertainty with regard to the estimates of risk.  Were the major
uncertainties in the risk assessment adequately characterized?  If not,
explain what should be included.

The draft assessment relies on the use of cumulative exposure (or a
transformation of cumulative exposure) as the principal dose metric on
which lung cancer and silicosis risks are evaluated.  The draft also
contains a discussion of studies that are suggestive of a dose-rate
effect.  Given this discussion, is cumulative exposure a reasonable
exposure metric for silica-induced lung cancer and silicosis in exposed
workers?  Is there a more appropriate exposure metric?  If so, explain.

OSHA’s estimates of lung cancer and silicosis mortality are
represented as a range of estimates with those derived from the
diatomaceous earth and U.S. granite studies considered to be plausible
upper bound risk estimates.  OSHA also acknowledges that the risk may be
lower such as suggested by the results of the IARC multi-centric study,
which relied on pooled data from several epidemiological studies.  Is
this a sound and reasonable estimation of risk from the available data? 
Is this decision adequately explained?  Should the risk estimates based
on one data set be preferred over the other?  Is there a more
appropriate representation of the lung cancer risk from silica exposure?
 If so, explain.

	The cumulative risk of silicosis morbidity is preliminarily considered
to be best represented by a range (approximately 3- to 4-fold) that is
based on results from four published exposure-response assessments. 
Does this range reflect a sound and reasonable estimation of risk from
the available data?  Is this decision adequately explained?  Should the
risk estimates based on one data set be preferred over the other?  Is
there a more appropriate representation of the silicosis morbidity risk
from silica exposure?  If so, explain.

OSHA estimated the risk of lung cancer mortality, silicosis mortality,
and silicosis morbidity for 45-year exposures to 0.25 and 0.5 mg/m3
respirable crystalline silica; these represent a range of exposures
consistent with OSHA’s current standards for construction and maritime
industry sectors, but are above OSHA’s current general industry
standards (approximately 0.1 mg/m3 for respirable quartz).  The
cumulative exposures corresponding to these higher exposure levels are
at or above the cumulative exposures of most of the more highly exposed
workers included in the studies that underlie the risk estimates.  Is
OSHA’s approach for extrapolating risk above the observed range
reasonable?  Is it clearly explained?  Is there a more appropriate
approach that should be considered to estimate risk at cumulative
exposures that are generally above the observed range of the underlying
epidemiological studies?



External Peer Review of OSHA’s

“Preliminary Quantitative Risk Assessment for Silica”

Peer Review Comments

Specific Questions

The draft report evaluates exposure-response data from several
occupational studies as to their suitability for quantitative assessment
of lung cancer and silicosis risk resulting from airborne exposures to
silica.  Are the strengths and limitations of the selected data sets
adequately discussed and appropriately evaluated?  Are you aware of
other study data that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica?  If yes, inform
OSHA of the other data.

Bruce Allen	In general, I found the discussion of the strengths and
weaknesses of the selected studies to be adequate.  Because such
discussions can be very discursive, I might suggest that a tabular
summary of the strengths and weaknesses be presented.  In such a summary
table, the criteria for evaluating strengths and weaknesses could be
laid out (say in the rows) and then for each study (one per column), a
brief point or two could be presented.  This would allow the more
in-depth discussions of the strengths and weaknesses in the text to be
tied back to this summary and would allow a clearer appreciation of how
the studies compared to one another with respect to the criteria under
consideration.

If anything, the weaknesses of all the studies with respect to
reconstruction of exposure histories (both with respect to the
atmospheric concentrations and the job-specific features that lead to
worker exposures to those concentrations) may not have been presented
with enough emphasis to convey just how limiting and problematic that
process can be.  There is some discussion of the “reasonableness of
the exposure assessment” on p. 10, where silicosis mortality odds
ratios are compared across exposure categories from the pooled cohorts
of Steenland et al.  The values presented do not give me a very strong
sense that exposure misclassification was negligible, since the odds
ratios presented hardly differ (the highest three are almost identical)
and I am not sure that a formal test would reject the hypothesis that
the odds ratios were the same, i.e., would not reject the hypothesis
that would hold if all individuals were randomly allocated to the four
groups regardless of their cumulative exposure level.  I have to wonder
if this is the best evidence that is available to OSHA to make the
desired point about impact of exposure misclassification.

I am not aware of any other data that are better suited for quantifying
the lung cancer and silicosis health risks.

Kenny Crump	The strengths and limitations of the selected data sets
appear to be adequately discussed and evaluated.  Areas that could be
improved in some relatively minor ways are discussed below.  I am not
aware of any other studies that are better suited for quantifying lung
cancer or silicosis risk from occupational exposure to silica.

Murray Finkelstein	I think that the strengths and limitations of the
selected data sets were adequately discussed and appropriately
evaluated. I am not aware of other study data that are better suited for
quantifying risk.

Gary Ginsberg

Gary Ginsberg	The variety of exposure-response studies are generally
well described in terms of numbers of subjects, exposure
characterization, statistical analysis, potential confounders, and where
applicable, dose response modeling.  The report however is lacking in
terms of providing a critical appraisal of each study and its relative
merit, power or weighting for the purposes of risk assessment.  The risk
assessment section relies upon 4 studies which conducted their own
quantitative dose response assessment, accepting those cohorts as the
basis for the OSHA evaluation. However, there are numerous other studies
which might lend themselves to quantitative assessment by virtue of
their exposure and outcome measurement.  For example, Section V.C.
describes many studies which might be suitable for quantitative
assessment (e.g., Guenel et al. 1989 of Danish stone cutters) but which
are not critically appraised for this purpose.  Further, the Cassidy
2007 European multi-center analysis is promoted in Section V.C. as
particularly useful (even “compelling” – numerous positive
qualities including exposure assessment) but yet is not included in the
Section VI quantitative risk assessment.  One wonders whether any bias
is introduced into the assessment by the omission of one or more
potentially useful studies.  The lack of a critical appraisal of the
quantitative utility of such studies or the potential bias introduced by
their omission, weakens the Section VI analysis.

Brian Miller	As far as I am aware, the studies are in the main well
described and evaluated.  It’s worth noting that the choice of
radiological opacities grade 2/1+ for analysis in the Buchanan et al.
(2003) paper was driven by the use of that level being required for
compensation in the UK systems.  The radiographs from the follow-up
study were all classified on the full ILO scale just as the earlier PFR
ones had been, and there are fuller descriptions of the studies and
descriptions and analyses of the distributions at lower levels e.g. 1/0+
in earlier publications.  The 1998 paper gives logistic regression
analyses only for 2/1+, but the 1995 research report (available to
download) also has detailed analyses of a 1/0+ response.  

Miller BG, Kinnear AG.  (1988).  Pneumoconiosis in coalminers and
exposure to dust of variable quartz content.  Edinburgh: Institute of
Occupational Medicine.  (IOM Report TM/88/17).
http://iom-world.org/pubs/IOM_TM8817.pdf

Miller BG, Hagen S, Love RG, Cowie HA, Kidd MW, Lorenzo S, Tielemans
ELJP, Robertson A, Soutar CA.  (1995).  A follow-up study of miners
exposed to unusual concentrations of quartz.  Edinburgh: Institute of
Occupational Medicine.  (IOM Report TM/95/03).                 HYPERLINK
"http://iom-world.org/pubs/IOM_TM9503.pdf" 
http://iom-world.org/pubs/IOM_TM9503.pdf 

Miller BG, Hagen S, Love RG, Soutar CA, Cowie HA, Kidd MW, Robertson A. 
(1998).  Risks of silicosis in coalworkers exposed to unusual
concentrations of respirable quartz.  Occupational and Environmental
Medicine;  55: 52-58.  

There are some new data on the risk of lung cancer with respect to
respirable silica exposure, from an extended follow-up of the British
PFR mortality study:  

Miller BG, MacCalman L. (2009). Cause-specific mortality in British coal
workers and exposure to respirable dust and quartz. Occup. Environ. Med.
(Published Online First: 9 October 2009. doi:10.1136/oem.2009.046151.  

The above paper is summarised from:  Miller BG, MacCalman L, Hutchison
PA. (2007). Mortality over an extended follow-up period in coal workers
exposed to respirable dust and quartz. Edinburgh: Institute of
Occupational Medicine.  (IOM Report TM/07/06).    HYPERLINK
"http://www.iom-world.org/pubs/IOM_TM0706rev.pdf" 
http://www.iom-world.org/pubs/IOM_TM0706rev.pdf 

Andrew Salmon	The studies considered are adequately described and
evaluated.  In particular, the U.S. studies and the IARC meta-analysis
for lung cancer are described in sufficient detail and carefully
evaluated.  However, the discussion of silicosis and choice of key
studies is somewhat less convincing.  The consideration of other
possible datasets, particularly the data on South African gold miners,
receive somewhat more cursory treatment.  The recent paper by Churchyard
et al. (2004) describing silicosis morbidity in black gold miners is
important both for the high level of effect seen in these workers and
the use of gravimetric exposure measures.  However, although this paper
was briefly noted elsewhere by OSHA, it is not considered in this
section.  The existence of analyses other than those performed by OSHA
and its regular collaborators is not consistently acknowledged.  This is
unfortunate since some of these analyses do to some extent address the
question of dose-response at lower levels, which is not adequately
discussed in this section (See for instance Collins et al., 2005 and
OEHHA, 2005), analyzing the data of Hnizdo and Sluis-Cremer, 1993;
Steenland and Brown, 1995; Chen et al., 2001 and Churchyard et al.,
2004).  The focus on extreme endpoints is illustrated by the extensive
coverage devoted to analysis of silicosis mortality, which is a good
technical analysis but from a public health standpoint feels like an
admission of failure!  Also, the choice of the Scottish mining study
(Miller et al., 1998) as a key study for silicosis dose-response
analysis is colored by this perspective.  This is undoubtedly an
excellent study from the technical standpoint, but its choice of a more
severe cut-point (ILO 2/1) for the radiographic diagnosis sets it apart
as being less sensitive than most of the other studies in the field, in
spite of evidence discussed here and in the section on health effects
showing that even an ILO 1/1 radiographic level substantially
under-diagnoses silicosis as assessed by later follow-up or other
methods e.g. histology.  There appears to be no discussion of the
complicating fact that these coal miners were also exposed to other
pneumoconiosis-inducing dusts, notably coal dust.  Miller and his
co-authors spend some time addressing this point in the paper, but it is
not referenced in OSHA’s narrative.

Evaluation of the dose-response relationships for disease endpoints
related to low-dose silica exposure will require evaluation of different
studies and dose-response models from those used in the existing
assessment, since those presented here (e.g. the study by Miller et al.,
1998 and linear or log-linear models) were selected for their
informative value at the pre-defined levels of interest rather than for
any information on dose-response at lower levels.  The appropriate
models would need to be selected and justified on the basis of the data
available in the lower exposure ranges, including the likelihood of a
threshold at least for non-cancer effects, as discussed above.  This
different approach is necessary to obtain a realistic evaluation of the
responses seen in the lower dose range.

Noah Seixas

Noah Seixas	The selected datasets represent as comprehensive a set as
available, to my knowledge.  The inclusion of the various cohorts is
appropriate, given the data quality needs for a robust assessment.  In
describing exclusions, (p8) it is noted that studies of coal workers and
foundry workers were excluded for good reasons.  However, it also notes
that 3 additional cohorts were excluded for ‘data unavailability or
incompatibility.’  These exclusions should be described in more
detail.  

The studies included in the risk assessment are very thoroughly
described and appropriately interpreted.  While limitations of each
study are described and considered, the strengths of the studies are not
described study by study.  The general strengths are evident from the
descriptions, so it may not be important to do so in any more detail.  A
brief summary of the strengths of the selected studies could help
support their use for this risk assessment.  Furthermore, published
critiques of the studies are also included in this discussion, and OSHA
has responded to these criticisms with thoughtful consideration and
clear conclusions. I am not aware of any key studies that were not
considered.



The draft assessment relies on quantitative uncertainty analysis and
sensitivity analysis of exposure measurement error in key studies of
lung cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach?  If so, explain.

Bruce Allen	In contrast to my comments above about the
less-than-convincing presentation related to exposure misclassification,
I did like the analysis that was presented to examine the quantitative
effects of uncertainties about the exposure reconstruction and the
individual worker exposures within that reconstruction.  The Monte Carlo
analysis approach that was applied is a reasonable way to address this
question.  My only comment related to that is that a typical Monte Carlo
analysis usually includes many more than 50 iterations (the maximum
number if simulations run in the uncertainty analyses for lung cancer
and silicosis mortality).

Note also that the analyses were performed by selecting some fixed
values for the variabilities considered.  Although the variability was
based on some observations (see bottom of p. 50 and bottom of p. 52 to
top of p. 53), those observations were themselves limited and may not be
the best reflections of the variability that is needed here.  It would
be informative to determine just how much variability is required to
result in a substantial change in the risk estimates.  If that amount of
variability can be deemed unreasonable, then there is stronger support
for the risk estimates that OSHA presents.  Or, another way of thinking
about it: if, even when there is very large variability (representing
uncertainty in the exposures), the calculated risks are still considered
substantial, then there is very strong support for the conclusion that
workers are at risk at the exposure levels under consideration.

Kenny Crump

Kenny Crump	This question refers to the analysis of exposure measurement
error in Toxichemica, Inc. (2004).  These analyses are well designed,
clearly explained and evaluated.  Evaluating and adjusting for exposure
measurement error is a very difficult issue.  The analysis of error
conducted by Toxichemica, Inc. is a very strong effort.  The assumptions
are clearly described and the data upon with they are based appear to be
appropriate and appropriately applied.  However, there are questions, as
there will always be with such an analysis, regarding the form(s)
assumed for the errors, the assumptions used in modeling the errors, and
in the completeness of the sources of errors modeled.  

For the Berkson component of the error, it would seem reasonable to
consider a multiplicative error model, i.e.,

		Exposuretrue = Exposureobserved * ε

where ε has a log-normal distribution.  In this form, both the mean and
the variance of the true exposure increases with the observed exposure,
which seems reasonable.  Also, unlike the additive error model, this
model does not produce negative exposures (to avoid this problem is
presumably why the distribution was truncated in the Toxichemica, Inc.
analysis).  This model might produce significantly different results
from the additive model, including modifying the shape of the dose
response curve (Carroll et al. 2006).   

It could also be assumed that silica concentrations differed randomly by
work area rather than by job.  Workers in same area of a facility will
breathe the same air even if they are doing different jobs.  Resampling
randomly by job within work area rather than by work area may tend to
average out variations by work areas and underestimate uncertainty. 

The method for evaluating error in converting from dust measurements to
quartz measurements appropriately based on data and well-designed and
implemented.  However, other important sources of error do not appear to
be accounted for.  

A major source of error that apparently was not accounted for is in
assuming that the average measure of exposure assigned to a job is the
true average.  But it is not always clear how representative the
underlying measurements were.  E.g., were some measurements taken at
times when it was believed there was a particular problem with high
exposures?  Moreover, some of these averages (particularly, those that
pertain to earlier times and which are likely the more influential since
they probably represent the highest exposures) were not based on direct
measurements, but were projected based on judgments of persons familiar
with the operations.  There is possibly considerable error in such
estimates.  Another source of uncertainty in the averages stems from the
need to convert from one measurement method to another (e.g., from
particle counts to gravimetric measurements).  

Based on these considerations, it seems likely that the effects of
exposure errors could have been larger than predicted by the
Toxichemica, Inc. analysis.  It would also be useful to evaluate the
effect of errors in exposure using other exposure metrics than log
cumulative exposure.  Risk models based on other exposure metrics are
likely to be more sensitive to exposure error.  (Also, see response to
Question 4 regarding the use of log cumulative exposure as the exposure
metric.)

Murray Finkelstein	The use of uncertainty and sensitivity analysis by
the Toxichemica team is well explained. I think that this is an
excellent approach for evaluating uncertainty.

Gary Ginsberg

Gary Ginsberg	I am not a statistician and don’t normally get involved
with quantitative uncertainty assessment.  Therefore, I cannot comment
directly on the detailed presentation of this work.  It does appear that
the major areas of measurement uncertainty have been identified and
dealt with in some manner by the extra effort to characterize these
uncertainties.   An area of concern however is the treatment of
potential confounding by smoking.  Page 44 assumes that the only issue
with smoking as an uncontrolled variable is the potential for smoking
rate to correlate with silica exposure in such manner to bias the
disease risk high.  The text goes on to assert that this is unlikely
from the available evidence and study design.  However, this rationale
excludes the possibility that when smoking is an uncontrolled variable
it will lead to elevated cancer risk in both the reference and
silica-exposed population leading to a high and variable baseline that
will tend to bias the results towards the null hypothesis and weaken
associations between silica and lung cancer.  I could not find any
explanation or substantiative evidence that such confounding by smoking
may not have occurred given the lack of smoking history and control for
smoking present in the underlying data.   The net result may be an
underestimation of silica-induced cancer risk relative to a non-smoking
population unless there is evidence that smoking potentiates
silica-induced lung cancer.    A similar potential exists for
underestimating cancer risk in diatomaceous earth workers simultaneously
exposed to asbestos and silica (page 46).  Once again, the concern
expressed in this document is the potential for additive risk across
these exposures to inflate the potency factor.  However, the rate of
mesothelioma in the asbestos exposed workers is not discussed.  If this
were a significant intercedent cause of mortality, it would curtail the
silica-induced cancer rate.   In addition, this report would be well
served to view radon and other uncontrolled exposure variables through
this lens of potential weakening of dose response relationships.

Brian Miller	Where there is error in exposure assessment, it is
important to assess the extent to which this may affect the estimation
of the exposure-response coefficient, and to correct for that if
necessary and possible.  Other sources of uncertainty may also be
important. The approach used is a reasonable one for assessing,
describing and allowing for such sources of uncertainty within the
approach taken for quantifying the expected or average risk, which needs
to be discussed separately (see below).

Andrew Salmon	The approach used is reasonable to address uncertainty in
exposure estimation.  Indeed, given OSHA’s recognized expertise in
this area it is unsurprising that they and their contractor
(Toxichemica) have done a thorough, indeed exhaustive, job on this
aspect.  The ultimate conclusion that there is some uncertainty, but
that it does not undermine the basic conclusions of the quantitative
analysis, is reassuring, although perhaps a little underwhelming given
the amount of effort that was applied to this topic.

Noah Seixas

Noah Seixas	Selection bias and confounding are addressed very well, and
convincingly suggest that these potential biases do not unduly affect
the risk estimates derived from these studies.  On p 42, it is stated
that the confidence intervals presented in Table VI-2 represent
uncertainty due to the sample of studies used for the analysis.  This
statement should be qualified by the uncertainty due to the systematic
biases that also may be present in the selected studies.  That is, the
confidence intervals represent uncertainty due to the sample of the
universe that the selected studies represent, given that the
observations in those studies are without error, themselves.

The simulation of exposure measurement error in assessing the degree of
bias that may have been present is a reasonable approach to assessing
this source of uncertainty. However, I have several comments about the
presentation of these results.  I will preface this by saying that
although I have thought about measurement error issues in occupational
exposure assessment and epidemiology for a long time, I am not an expert
in the statistical concepts, nor the application of these concepts to
assessing uncertainty.  Therefore, I raise these questions for
consideration, without claiming to know how this analysis should be
improved.

On p48, the statement that the metric conversions apply equally to
diseased and non-diseased subjects and therefore represent Berkson type
error, is incorrect.  Errors that are applied to cases and controls are
called ‘non-differential errors,’ but do not imply a Berkson error
structure. 

While the discussion of Berkson error models is generally correct, I
don’t believe the simulation has actually modeled a Berkson error, but
rather, a classical error model.  The typical Monte Carlo simulation,
which is what appears to have been done, would introduce classical
error.  I don’t think the report should repeatedly state that the
simulation produces coefficients “adjusted for Berkson error” as I
don’t believe this is Berkson, nor that the simulation actually
adjusts for it. 

I think that the introduction of random error to the assigned exposures
essentially adds error to that already present.  Thus, you are taking
the results of a model based on exposure estimates which include error,
and simulating additional error on top of it.  I’m not sure what the
results of this simulation then produces. In fact, the results produced
in Table VI-4- VI-6 almost all went down (toward the null), which is
what is generally expected by adding random (classical) error into the
exposure variable.

At the bottom of page 51, the changes in coefficients with introduction
of error are reported, however, no explanation or even speculation about
why some of these estimates were greatly reduced is provided.  Although
the associated table provides the SE associated with the coefficients
for the ‘adjusted’ effects, it should also present the SE for the
‘unadjusted’ values.

I remain unconvinced that the conclusion on page 52, that measurement
error “did not have a meaningful effect on the estimate of the
exposure-response coefficient” is well supported by this analysis.  I
would believe that a statement such as “addition of random error to
the estimated exposures in the pooled analysis did not appreciably
change the estimated exposure response relationship” would be correct.
 However, I don’t believe that the analyses conducted prove that
measurement error in the selected studies did not substantially affect
the observed relationships.  

There is no mention of measurement error generally reducing the effect
of exposure response relationships, and that this principle indicates
that the estimated risks are most likely to be underestimates, or
conservatively estimating risk. This is an important aspect of
measurement error with significant implications for risk assessment and
should not be overlooked.



The draft assessment evaluates the use of a physiologically-based
pharmacokinetic model that predicts silica lung burdens in individuals
exposed to particulates through inhalation for estimating risk from lung
cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach? If so, explain.

Bruce Allen

Bruce Allen	The use of a kinetic model to estimate lung burdens that
would then lead to inflammation, fibrosis, etc. is a very reasonable
approach.  However, the methods and approaches used by Kuempel et al.
for their assessment are not well explained or presented in the OSHA QRA
document.  

I am particularly troubled by the description starting in the middle of
p. 25 that says that Kuempel et al. used the dose coefficients that had
been previously derived for the two cohorts they analyzed.  Why would
they use those coefficients that were derived without consideration of
the lung burden metric, i.e., that considered simple cumulative
exposure, to estimate lifetime risks based on the lung burdens
associated with various exposure scenarios?  This is either an error in
their methodology or an incomplete and misleading reporting of their
analysis in the OSHA document.

The discussion on p. 27 (first full paragraph) describes the analysis
using the lung burden metric with the 10-cohort study data.  Although
the conclusion is that the lung burden metric does as good a job as
cumulative exposure, the risk estimates derived from that analysis are
not presented (neither in the text nor in Table VI-2, where the summary
of all the risk estimates is presented).  My impression is that OSHA has
not fully considered the lung burden-based risk estimates and perhaps
does not consider them to be “on a par” with the other estimates. 
The discussion of the Kuempel et al. study appears to be the most
cursory of the discussions of the various studies.  If it is the case
that this study is not considered as relevant or it is thought of as
secondary (perhaps because it was not an epidemiology study in and of
itself but relied on the response data of two of the other studies
cited), then there should be some discussion of that and the
inter-relationships between the estimates from that “study” and the
others.

Kenny Crump	This refers to an analysis in Toxichemica, Inc. (2004). 
Exposure metrics based on lung burdens can provide reasonable
alternatives to the more empirical metrics emphasized in the papers
reviewed by OSHA (lagged or unlagged cumulative exposure or log
cumulative exposure).  The lung cancer and silicosis analyses are
adequately explained as far as they go.  Whereas the silicosis analysis
considered several metrics based on lung burden, the lung cancer
analysis only considered one exposure metric – lung burden lagged 15
years.  However, both the lung cancer and silicosis analyses stopped
short of estimating risks using this approach.  Even if a fit to data
using a metric based on lung burdens is similar to the fit provided by
an empirical metric such as log cumulative exposure, that doesn’t
imply that the estimated risks from 45 years of exposure will be
similar.  Without estimating risk using the metrics based on lung
burden, the effect of these metrics upon the estimated risk will not be
known.  If workers in the epidemiological studies predominately were
exposed for only a short time (e.g., a few months), different exposure
metrics could predict significantly different risks from 45 years of
exposure even if they provided equivalent fits to the epidemiological
data.

Murray Finkelstein	The analyses are clearly explained. I think that this
would be a reasonable approach for evaluating silica-induced risk only
in the absence of human epidemiologic data. I think that the application
of dust doses and inflammatory effects in rats to the prediction of
mortality in humans is too great a leap. I think that the appropriate
approach is risk assessment based upon human epidemiologic data.

Gary Ginsberg

Gary Ginsberg	The “PBPK” lung burden model is not well explained and
pretty much just mentioned in passing as this line of evidence is not
given substantiative weight.  PBPK usually refers to a mathematical
description of chemical fate within systemic compartments that have
parameters for organ size, blood flow and clearance processes, with QC
check conducted to ensure mass balance.  I haven’t read Kuempel et al
(2001) but it sounds more like an inhalation deposition/dosimetry model
that a PBPK model.  The model description on Pages 24-27 could be
enhanced by showing the model structure, a table of parameter values for
both species (breathing rate, respiratory tract surface areas; clearance
rates in various lung compartments), how is the model affected by
breathing rate and particle size, whether clearance is a static
parameter or allowed to vary (decrease) with increased silica burden,
and how is the threshold for critical quartz lung burden (0.39 mg/g)
set.  Also, the description of the human model at the top of Page 25 is
vague – how many coal workers, are central estimates used or are all
the data used to create a statistical distribution, how was the rat
model adapted for a standard human adult?     

The lung burden model has potential utility as it defines a threshold
for lung cancer.  However, it is not explained well and this concept is
not used in the remainder of the report to assess risks from low dose
exposure.

Brian Miller	Where no epidemiology exists, it may be that the only
approach that can be used for predicting risk is extrapolation from an
animal model.  In the present context, much use is made of the concept
of lung burden.  This may be the correct target organ dose metric, but
epidemiological data on it is very sparse; hence the normal focus on
exposure experienced as a surrogate for dose.  The extrapolation
involved requires many more assumptions than assessing risk from
epidemiological studies, and it may not be prudent to place the same
degree of reliance on the model-based estimates as on those from
epidemiological observation.  Having said that, one advantage of the
animal model is that the exposure is usually rather better characterized
than in humans.

Andrew Salmon	The description of the lung-burden analysis by Kuempel et
al. (2001), and the further analysis by Toxichemica (2004), is quite
brief and limited mainly to describing the overall conclusions of this
work.  It is not sufficiently detailed to allow a reader to determine
exactly what was done, or whether the implicit assumptions are
reasonable.  What is presented sounds reasonable and interesting, but
Toxichemica’s eventual conclusion appears to be that it does not
improve the analysis of Steenland’s pooled cancer data relative to the
simpler cumulative exposure model.  It seems relatively unlikely that
the available data on lung cancer or silicosis are of sufficient
precision to directly support use of this or any other complex dose
metric.  Toxichemica (2004) also note that the effect of the lung burden
calculation is to build a lag time into the dose-response model.  While
this is a reasonable assumption, and there is limited evidence that this
is appropriate for the silicosis data, there are a number of
explanations besides dependence on silica lung burden for this effect.

Noah Seixas	It is not clear to me what the importance of this
presentation is.  The studies by Kuempel that adopted a toxicokinetic
model for lung deposition to estimate silica lung burden ‘were
similarly good predictors of lung cancer risk’ as cumulative exposure.
 It appears that these studies did not contribute to the risk analysis,
other than to note that these more complex models didn’t produce any
different results, and it is unclear to me that this is an important
observation.



The draft assessment summarizes several published risk assessments that
use a variety of models and statistical methodologies employed by the
various analysts to fit the featured data sets.  Are the models and
approaches used by these investigators adequately explained and
evaluated?  If not, let OSHA know what needs to be added to the
presentation.

Bruce Allen	I think that a summary table showing the approaches and
models used by the various investigators could improve the presentation
somewhat.  Table VI-2 does have some of that information (which model
was used for the risk estimation – although there appears to be an
error there in that the Attfield and Costello analysis appears to have
used a linear, not log-linear, relative risk model).

It is difficult to determine how much information to include in a
summary of the studies.  But seeking consistency in the presentation of
the methods and results across studies (as opposed to just reflecting
the perhaps-idiosyncratic information presented by the original authors)
is a desirable goal.  So, the presentation for each study could include
the same, “standardized” information: models considered, lag times
considered, how fitting was done and evaluated, special concerns about
model fit (e.g., nonmonotonicities in observed response rates or odds
ratios), etc.

Here are a few examples of questions that I had in reading the summaries
and evaluating what had been done.  

p. 20, line 7: it says Attfield and Costello used unlagged exposure. 
But how is that possible if they did not update work histories beyond
the original 1982 follow-up (p. 20, line 4) when they extended mortality
follow-up to 1994?

p. 21, line 6: how did Attfield and Costello evaluate “best-fitting”
among all their dose-response analyses, especially when some of them
were on different data (i.e., some with and some without the
high-exposure category)?

If Graham et al. (2004) “concluded that dust control measures were
shown to be effective in reducing lung disease and mortality” (p. 23,
lines 3-4), how could they conclude that the results “did not support
a causal relationship between exposure to quartz and lung cancer” (p.
23, lines 4-5)?

Kenny Crump

Kenny Crump

Kenny Crump

	The reasons given for emphasizing analyses of the pooled data on lung
cancer (Steenland et al. 2001a) that use log-transformed cumulative
exposures are faulty.  There are several reasons for preferring analyses
based on untransformed exposures.  Using untransformed exposure with a
15 year lag gave a considerably better fit to the data than
log-transformed exposure with a 15 year lag (log-likelihood of 21.4
versus 18.8—what is referred to as the likelihood in Steenland et al.
(2001a) is presumably the log-likelihood).  The fact that the model
based on log-transformed exposure did not show significant improvement
in fit when the interaction terms were added (i.e., did not show
significant heterogeneity) should be viewed as more of a disadvantage of
this model in comparison to the one based on untransformed exposures
rather than an advantage: The model based on untransformed exposures not
only fit better that the model based on untransformed exposures, but
adding interaction terms significantly improved the fit of the already
better-fitting model, but did not significantly improve the fit of the
poorer-fitting model.  

Figure 1 provides a simple illustration of this point.  This figure
depicts data from two studies, Study 1 (blue points) and Study 2 (red
points).  There is an apparent study effect as exposure shows a stronger
effect in Study 1 than in Study 2.  Two models are used to analyze these
data, Model 1 (orange lines) and Model 2 (green lines).  The solid lines
indicate the fits of the models to the combined data and the dashed
lines indicate the fit of the models to the data from the individual
studies.  The fit of Model 2 to the combined data (solid green line) is
clearly better (i.e., larger log-likelihood) than the fit of Model 1
(solid orange line).  However, there is a significant improvement in fit
when Model 2 is fit separately to the two studies (dotted green lines). 
Thus, this model detects significant heterogeneity.  However, with Model
1 the improvement in fit is not significant; hence there is no
significant heterogeneity identified by this model.  By the decision
logic described above, Model 2 would be ruled out in favor of Model 1
even though Model 2 is clearly superior.  This example has the same
characteristics as the fits to the lung cancer data in the pooled data,
with Model 2 playing the role of the model using untransformed exposure
and Model 1 playing the role of the model using log-transformed
exposure.      

There are several procedures for choosing the best model based on the
value of the likelihood (e.g., AIC (Akaike information criterion, Akaike
1974)).  Each of these procedures would identify the model based on
untransformed exposures as better than the one based on log-transformed
exposures.  The fact that the significant heterogeneity was found for
this model suggests that this model could be improved, but it does not
indicate that the model using log-transformed exposures is better.

Dose responses for the same effect (e.g., lung cancer from silica) would
not be expected to have fundamentally different forms in different
studies (e.g., derived from log-transformed exposure in one study and
untransformed exposure in another).  It would assist in comparing
results across studies to present risk estimates from the three studies
in Table VI-2 using a common exposure metric.  Similarly, it would
assist in comparing the effect of different exposure metrics, by
computing risk from single study using different exposure metrics.  Both
of these goals could be accomplished by adding to Table VI-2 risk
estimates derived from the analysis of the pooled cohorts (Steenland et
al. 2001) based on a linear relative risk model.  If the model is truly
linear in the exposure variable, use of the log-transformed exposures
will tend to underestimate risks at lower exposures and overestimate
risks at higher exposures.  

More consideration needs to be given in the document to model selection
and its effect upon risk estimates.  Although log-linear models are
relied upon extensively, such models are generally selected for
convenience (since this is the model generally used in Cox regression
(Cox 1972)) rather than for biological plausibility.  Such models are
supralinear when applied to untransformed exposure (increase at a rate
faster than linear with increasing exposure).  OSHA states that it does
not believe that a supralinear response “is biologically plausible and
believes that it is more reasonable to assume that the exposure-response
relationship remains linear …”  This argues for using a linear model
(like the linear relative rate model used in Rice et al. (2001)) instead
of a log-linear model.  It also argues against using a log-linear model
with log-transformed exposure, since such models are sublinear, as
illustrated by the very slow increase in risk with increasing exposure
exhibited by the lung cancer model implemented by Toxichemica Inc.
(2004) (Table VI-12).  To more fully evaluate the effect of model
selection, it would useful to reanalyze the data from one or more of the
key studies, such as the data from the pooled cohort, using a truly
linear model, and perhaps other models, and compare risk estimates
obtained from these models.  

It is claimed that the Attfield and Costello (2004) log linear model
becomes supralinear at cumulative exposures above 4.5 mg/m3-years.  How
OSHA arrived at this conclusion is not explained.  As noted above such a
model is actually supralinear at all values of cumulative exposure.

Murray Finkelstein	These models are adequately explained and evaluated.

Gary Ginsberg

Gary Ginsberg	The description of the published risk assessments is
generally adequate.  As described above, more information regarding why
certain studies were excluded from the meta analysis or not
quantitatively described would be helpful.  The issue of threshold is
not given proper attention in these descriptions.  The sense one gets is
that these studies and associated models generally find a dose response
in the range of current OSHA standards (0.05 to 0.1 mg/m3 and higher),
with a threshold possibly existing at lower concentrations:  e.g. $0.036
mg/m3 for 45 yrs based upon critical lung burden.   The issue of
possible thresholds should be systematically evaluated for each study
– was it statistically investigated by the study authors?  Were the
data sufficient to allow such an analysis?  Do the studies agree
regarding where a threshold might lie for lung cancer and silicosis and
how would such a threshold affect OSHA’s risk assessment? 

Also meriting greater description are the physical characteristics of
the silica in each study.  Given that features such as the manner in
which it was mined/produced/fractured, whether it has surface coatings,
its mineral content (quartz, cristobalite, tridymite) and its age all
appear to influence its reactivity, this should be described for each
industry studied.    

The 10 cohort study reported by Steenland et al. (2001a) shows general
agreement amongst 9 of the studies but one study (South African gold
miners) showed approximately 10 fold higher potency for lung cancer
(Page 15 and Table VI-1).  OSHA reports that Steenland reports that
exclusion of that dataset was not influential in the overall
meta-analysis (likely because it contained only 77 cases out of a pooled
analysis of over a thousand).  However, that is not the end of the
story.  In toxicology and risk assessment, one should consider whether
the most sensitive study and endpoint is of high enough quality to be
the source of risk calculations or whether the entire risk range should
be presented (e.g. the gold miners representing one end of a reported
potency range) or whether some central tendency potency estimate should
be used.  It may be that the South African gold miner study is more
potent for a good reason (e.g., least confounded by smoking, most potent
form of silica) and should be given serious consideration as a stand
alone data point for potency estimation and risk calculations.  Or
perhaps that study is no better or even inferior or less relevant than
the others.  The combination of Sections V and VI provide no analysis of
this but rather just accepts the Steenland treatment of the data. 

OSHA focuses on cancer potency in the range of current standards and
also on higher exposures (pp 28-32).  The rationale for this is unclear.
 Is OSHA contemplating raising the standard to make it less strict?  Why
not focus as well on the dose response at lower levels of exposure just
in case there might be a desire to lower the standards (more strict). 
It strikes me that the lower end of the dose response curve may be more
interesting given the potential for thresholds (e.g., the lung burden
model; potential that silicosis is needed for lung cancer to occur) and
the potential that lowering the standard might be desirable given the
high cancer risks associated with the current standards.

Brian Miller	The studies from which the risk coefficients are derived
are based on outcomes (mortality, diagnosis of silicosis) at different
lengths of follow-up from the exposures.  In a situation where the
effect of an exposure on the progression of risk may be time-dependent
(see 6 below), these coefficients may not be comparable.  There should
be more discussion of the possible effects of this disparity.

Andrew Salmon	The description of the models used and statistical methods
are in general adequately described.  OSHA relies on the models used by
the published analysts although these are in most cases very simplistic,
e.g. linear with cumulative dose.  These simple models are adequate to
fit the data in the observed ranges of the specific studies considered
(indeed, many of the studies lack the precision and/or cover an
insufficient range of different exposures to support anything more
complex).  OSHA describes these models well and confirms their validity
for prediction of response at the PELs, but makes no effort to determine
whether more informative models could be developed which would be useful
in critical exposure ranges such as those below these arbitrarily
pre-defined PELs.

Noah Seixas	The report discusses many alternative models including the
form of the linkage function, alternative exposure metrics, and
different latency periods.  These are well explained for each study, and
reasonable decisions are made as to which form of the model is most
useful and robust for the risk analysis.



The draft report identifies and describes several key areas of
uncertainty with regard to the estimates of risk.  Were the major
uncertainties in the risk assessment adequately characterized?  If not,
explain what should be included.

Bruce Allen	As stated above, although I think the uncertainty associated
with historical exposure reconstruction was under-characterized, I like
the Monte Carlo approach to addressing exposure uncertainties (with some
possible extensions suggested in response to question 2.)  The
discussions of possible selection bias, confounding, and bias in
conversion factors were adequate to dispel major concerns about their
having any negative impact on the findings.

Kenny Crump	The potential for selection bias and for bias due to
confounding are adequately addressed.  As explained in response to
Question 2, some aspects of uncertainty in risk due to uncertainty in
exposures are addressed but others probably are not.  In addition, the
effect upon risk of the exposure uncertainty that was quantified is
possibly underestimated due to the use of log-transformed exposures, as
risk estimates based upon log-transformed exposures tend to be relative
insensitive to changes in exposures.  As discussed in response to
Question 4, there needs to be additional evaluation of the uncertainty
due to the risk models used.

Murray Finkelstein	I think that the major uncertainties were adequately
characterized.

Gary Ginsberg	As described above, I believe the major uncertainties in
the estimate of risk not adequately dealt with are the potential for
confounding by smoking to decrease the strength of associations, and the
uncertainty regarding the existence of thresholds for cancer and
non-cancer endpoints.

Brian Miller	I have no other areas of uncertainty to suggest.  (We have
quite enough to deal with already.)

Andrew Salmon	Given that so much thoughtful analysis was devoted to
uncertainty in the exposure measurements, it is unfortunate that the
same level of investigation was not directed to determining the
uncertainty in determination of the other half of the dose-response
relationship.  There is some useful qualitative discussion of the
limitations of death certificates and the reliability, but low
sensitivity, of the ILO radiographic method of diagnosing silicosis in
the health effects section.  However, there is no quantitative analysis
of uncertainty in determination of health endpoints beyond the usual
statistical tests reported in epidemiological studies.  The overall
uncertainties in determining health endpoints could easily be as great
as, or greater than, the uncertainties in exposure assessment,
especially for silicosis morbidity and mortality.  An initial attempt at
quantifying uncertainties in health endpoints could be obtained by
reviewing confidence limits on measures such as SMRs for specific
endpoints, and the distribution of severity or response at different
dose levels. The latter consideration might be helpful in reviewing the
impact of reliance on measures of higher severity such as mortality,
higher-grade radiographic diagnosis, or eligibility for workers’
compensation, in some studies.   However, there is an underlying problem
in that although specific quantitative endpoints such as cancer
incidences or specific radiographic grades of silicosis are well
defined, the overall disease process is not well characterized by any
established integrative measures, so estimates of the overall disease
burden are subject to substantial qualitative as well as quantitative
uncertainties.

Noah Seixas	Yes and no – see answer to 2, above.



6. 	The draft assessment relies on the use of cumulative exposure (or a
transformation of cumulative exposure) as the principal dose metric on
which lung cancer and silicosis risks are evaluated.  The draft also
contains a discussion of studies that are suggestive of a dose-rate
effect.  Given this discussion, is cumulative exposure a reasonable
exposure metric for silica-induced lung cancer and silicosis in exposed
workers?  Is there a more appropriate exposure metric?  If so, explain.

Bruce Allen	It seems that the dose-rate effect may be related to the
lung burden metric predicted by the kinetic model of Kuempel et al.  If
that idea could be explored then there may be a way to provide further
support for the risk estimates that have been presented, especially
those based on that lung burden metric.  This is important given the
fact that, apparently, the lung burden metric did as well as cumulative
exposure (or log-cumulative exposure) in fitting the data.

On the other hand, I was not bothered by the fact that a dose-rate
effect might be present, given the indications that it may occur at
concentrations clearly above levels that OSHA needs to be concerned
about for reducing workers’ health risks.  Given the information about
where (at what concentrations) significant dose-rate effects appear to
be occurring, I am satisfied with the use of cumulative exposure for the
determination of concentration levels associated with the lower levels
of risk of concern for regulatory decision making.

Kenny Crump

Kenny Crump

	First it needs to be clear about what is meant by a “dose-rate
effect”.  My interpretation of that concept is that there is a
non-linearity in the exposure-response whereby a given increase in
intensity of exposure will cause a greater than proportional increase in
risk.  An example would be if risk varies as the integral over time of
the square of the exposure level.  Thus, there are dose metrics other
than cumulative exposure that do not incorporate a dose-rate effect. 
Since OSHA is relying on cumulative exposure to estimate risk, I
interpret this question as whether cumulative exposure is an appropriate
metric, or is some other metric that incorporates a dose-rate effect
more appropriate.  

Estimating risk based on an “incomplete” exposure metric like
average exposure is not recommended.  It is highly implausible that such
a metric could apply over an extended range of exposure patterns that
are likely to exist in a study.  E.g., exposure to a particular air
concentration for one week is unlikely to carry the same risk as
exposure to that concentration for 20 years, although the average
exposures are the same.  

Finding a significant relationship between average exposure and risk in
itself is not evidence of a dose-rate effect.  In general, even if risk
is truly a function of cumulative exposure, there with likely also be a
significant relationship between risk and average exposure, since
cumulative exposure and average exposure are likely to be correlated. 
For example, if all exposures are for the same duration, average
exposure is totally confounded with cumulative exposure.  On the other
hand, demonstration of a significant relationship with cumulative
exposure does not rule out a dose-rate effect, since cumulative exposure
will possibly be correlated with a metric that incorporates such an
effect.

To demonstrate evidence for a dose-rate effect that is not captured by
cumulative exposure, it would be most convincing to show some effect of
dose rate that is in addition to the effect of cumulative exposure.  To
demonstrate such an effect one would need to model both cumulative
exposure and some effect of dose rate, and show that adding the effect
of dose rate makes a statistically significant improvement to the model
over that provided by cumulative exposure alone.  I am not aware of any
such analysis that has been conducted for lung cancer from silica or for
silicosis mortality.  However, as discussed below, two studies conducted
such analyses for silicosis morbidity.

Most of the analyses of for silicosis morbidity did not examine exposure
metrics other than cumulative exposure.  Of the eight cumulative risk
studies summarized by OSHA (Table VI-11), only Hughes et al. (1998)
specifically tested for a dose-rate effect.  This study found that,
after controlling for cumulative exposure, workers exposed to ≥ 0.5
mg/m3 crystalline silica had a significantly higher risk than workers
exposed to lower levels.  A similar dose-rate effect was found by
Buchanan et al. (2003).  Although Steenland and Brown (1995a) found that
average exposure did not describe the morbidity in their cohort nearly
as well as cumulative exposure, this is not as specific a test for a
dose-rate effect as was conducted by Hughes et al. and Buchanan et al. 
The remaining six studies used to summarize risk of silicosis morbidity
(Table VI-11) considered only cumulative exposure as an exposure metric.
 So the two studies that looked carefully for a dose-rate effect found
evidence for such an effect.  In addition, Park et al. (2002) found a
13-fold higher incidence of silicosis during 1942-52 than in subsequent
years after controlling for cumulative exposure.  Although they
attributed this possibly to a change in surveillance, it could also be
due to high exposures during this early period and a dose-rate effect.  

The OSHA report states that a dose-rate effect for silicosis is
biologically plausible.  And if such an effect is present for silicosis
morbidity, it would presumably also be present for silicosis mortality. 
The report also states that even if there is such an effect it may not
exist at concentrations in the range of the current PEL.  However, even
if this is the case, risks in the range of the current PEL are being
estimated from studies that involved higher air concentrations.  Not
accounting for a dose-rate effect, if one exists, could overestimate
risk at lower concentrations.  

Having noted that there is evidence for a dose-rate effect for
silicosis, it may be difficult to account for it quantitatively.  The
data are likely to be limited by uncertainty in exposures at earlier
times, which were likely to be higher.  Perhaps rather than identifying
an alternative dose metric to cumulative exposure, it would be better to
retain cumulative exposure but use data from lower exposures where any
dose-rate effect would be less severe.  One possible approach that might
could be used to adjust existing estimates for a potential dose-rate
effect is to use either the coefficients estimated by Hughes et al.
and/or Buchanan et al. for low exposures (e.g., equation (1) of Hughes
et al 1998 with con = 0, or equation (2) of Buchanan et al. 2005). 
Another possibility is to reanalyze data from some of the better studies
using an approach similar to the approaches used by Hughes et al. (1998)
and Buchanan et al. (2003).    

The analysis by Toxichemica, Inc. (2004) that investigated several
exposure metrics based on lung burden, although it investigated exposure
metrics different from cumulative exposure, did not address a dose-rate
effect per se.

Murray Finkelstein	I am skeptical of the observation of a dose-rate
effect because I think that exposure assessment and assignment to
individual workers is too uncertain. I think that cumulative exposure is
reasonable for the risk assessment of chronic diseases such as silicosis
and lung cancer.

Gary Ginsberg	The studies suggesting that silica risk is driven by
exposure rate more so than cumulative exposure are not particularly
numerous or compelling and I agree with the current analysis.  Further,
the cumulative exposure metric ties into the critical lung burden I
mention above.  However, one cannot rule out the potential for dose-rate
to have an influence on disease rate and so it is appropriate to treat
this as an uncertainty.  In this regard, more could be done to describe
the possible ramifications of a dose-rate model on silica risk
assessment and more mechanistic information could be brought forth
regarding why dose-rate should (or should not) be considered relevant. 
For example, what non-linearities in the biological response might exist
that would be sensitive to dose rate (e.g., threshold phenomena such as
exceedance of defense mechanisms or damage requirement needed for
excitation of immune/inflammatory processes).  

Assuming cumulative exposure is the best metric for silica risk
assessment, it brings up an interesting assessment/intervention
question.  It raises the possibility that the OSHA standard should be
based upon worker cumulative exposure (in mg/m3-yrs) rather than a set
number (e.g. 0.05 ug/m3).  For example, it may be determined that to
reach a particular cancer risk target (e.g., 1 in a thousand), the
cumulative lifetime exposure should be no greater than xx mg/m3-years. 
Some workers may get this faster (higher exposure, shorter period) while
for others it may take longer.  Akin to radiological worker protection,
it may be possible to have a running tally of cumulative silica exposure
and to limit further exposure once the “acceptable risk” level has
been superceded.  This approach would seem compatible with the
cumulative exposure assessments in this document and allow the “bright
line” to be a particular risk target (e.g., 1 in a thousand).  From
the current document, it is unclear what OSHA’s target or acceptable
risk level is.  While the cumulative exposure approach would require
more record keeping, affected industries might find it attractive in
that it might allow short –term occupations to have higher exposure
levels while limiting the cumulative exposure for any given worker.

Brian Miller

Brian Miller

	Cumulative exposure has been widely used in studies of occupational
health risks, and it has been shown theoretically that it is an
appropriate metric in circumstances where the exposure acts over a short
period of time after it is experienced.   This is true of e.g. ionizing
radiation exposures, and has been found to be a reasonable fit for
exposure to respirable coal dust (in which the whole dust rather than
the silica component has usually appeared to drive the risk). 
Additionally, in most coalworker epidemiology, to a good approximation,
pneumoconiosis does not progress after exposure ceases.  

With silica exposure the responses are rather different.  High exposures
can precipitate severe or accelerated reactions, and silicosis, once
well established, is likely to progress even without further exposure. 
The data of Miller et al (1998), in the reanalysis of Buchanan et al
(2003), showed higher silicosis risks, mass for mass, from exposures
derived from higher concentrations.  This might be described as a
dose-rate effect, although that would not specifically address the
mechanism by which it derives, which is likely to be due to the
overloading of the human body’s defence and clearance mechanisms. 

Thus there is good reason to believe that cumulative exposure, with no
allowance for the rate at which it accumulated, and no consideration of
persistence of its effects over time, may not be the most appropriate
metric, and that a metric that allowed for these aspects would be
preferred.  However, to be useful for risk assessment, such a metric
would require a more complex formula for its calculation, and would need
quantitative estimates of the coefficients for that formula.  Such
estimation would need very good exposure data.  Buchanan et al (2003)
had some success in this area, notably in establishing the relative
increase in risk from exposures at higher concentrations, but were
unable to obtain a reliable coefficient for the effect of persistence.
This was assumed to be because the high silica exposures were
accumulated over a relatively short time period, so the follow-up time
varied little across the study group, and the power to estimate an
effect of elapsed time was accordingly very low.  It may be that other
studies with a wider range of follow-up lengths might be informative on
this aspect through re-analysis.

Andrew Salmon	It appears that cumulative exposure is the most
appropriate dose metric, and that this is adequate to correlate with the
observed response data in all the major studies examined.  It is
certainly possible that there are dose rate effects, especially at
extremely high (or low) doses, but the studies considered do not, it
seems, provide clear and unequivocal evidence to demonstrate or quantify
these effects.  Similarly, it is possible that lung residence time is a
factor in the development of silicosis and other health endpoints, but
the study which evaluated this metric did not show that it materially
improved the correlation with health effects beyond what was obtained
with the cumulative exposure metric.

Noah Seixas	Cumulative exposure, or the log of cumulative exposure is,
without doubt, the correct measure of exposure for the chronic diseases
under consideration in this risk assessment.  The alternative metrics
discussed, are appropriately considered; however both by the fact that
cumulative exposure is theoretically the most appropriate measure, and
because it fits the data best in most cases, it is selected
appropriately.  More complex models that are considered including
threshold models and dose-rate effects models are interesting
explorations which may shed light on mechanisms of disease, and may in
some cases provide better fits to the data.  However, such models rarely
outperform cumulative exposure to the extent that they should be adopted
for risk assessment purposes.  Even more importantly, policies based on
such models are extremely difficult to formulate as practical exposure
limits or guidelines.  Cumulative exposure is the correct measure from
various perspectives for this assessment.

Further, I would be very skeptical of adopting dose-response estimates
from the limited data we have looking at dose rate effects.  In
particular, the existing data on dose-rate effects are limited by likely
confounding between exposure level and time – that is, all high level
effects are derived from earlier time periods.  While the papers have
done a good job of trying to separate out these two potential aspects of
risk, they are insufficiently robust to allow adoption of differential
response rates in a quantitative manner.  I recommend using the average
risk estimate associated with cumulative exposure, without regard to
exposure level, for estimation of risk.



OSHA’s estimates of lung cancer and silicosis mortality are
represented as a range of estimates with those derived from the
diatomaceous earth and U.S. granite studies considered to be plausible
upper bound risk estimates.  OSHA also acknowledges that the risk may be
lower such as suggested by the results of the IARC multi-centric study,
which relied on pooled data from several epidemiological studies.  Is
this a sound and reasonable estimation of risk from the available data? 
Is this decision adequately explained?  Should the risk estimates based
on one data set be preferred over the other?  Is there a more
appropriate representation of the lung cancer risk from silica exposure?
 If so, explain.

Bruce Allen	I think it is appropriate to present a range of risk
estimates and to consider the differences in the estimates when making
decisions about regulatory limits. As discussed in the document, there
are various differences among the cohorts of individuals exposed to
silica, including the physical properties of the silica itself, that are
potentially important considerations and which may account for the range
of risk estimates. Given those differences, it is important to show the
range as a reflection of what those differences may entail (in addition
to the fact that no two sets of results will be exactly alike or give
exactly the same estimates, even if everything was identical).

The only reason that I can see to prefer one set of estimates to another
would be if that one set was based on a substantially superior study,
say in terms of sample size, relevance of exposure, measurement of
exposure, representativeness of the cohort members in relation to the
workers for whom regulatory limits are being set, etc.  The 10-cohort
study is larger than the other two cohorts alone, and its exposure
contexts clearly cover a wider spectrum.  It might be justified to base
decisions solely on the results of that study, but it is not clear that
those results should be presented to the exclusion of the others.

Kenny Crump	Other things being equal I would recommend emphasizing risks
estimated from the multi-centric study, since it incorporates data from
many studies and is less likely to be unduly influenced by exposure
errors from a single study.  However, I would not recommend emphasizing
risks from this study as currently estimated using log-transformed
exposures.  The lower risk estimated from the multi-centric study is
likely to be an artifact due to the use of log-normal transformed
exposure.  As pointed out earlier, untransformed exposure provided a
better fit to these data than log-transformed exposure.  The limited
increase in risk projected by this model to result from a 10-fold
increase in exposure (0.05 mg/m3 to 0.5 mg/m3) doesn’t seem plausible.
 OSHA could consider calculating risk from the multi-center using
untransformed exposures and a linear model.

Murray Finkelstein	I believe that the best estimate is that derived from
the IARC multi-centric study because it averages over study
uncertainties. In so doing, it is possible that the heterogeneity and
uncertainties blunt the dose response relationship. Nevertheless, I
think that individual study uncertainties are sufficiently large that
pooling is the best approach. I am particularly impressed by the
sensitivity analysis conducted by Steenland and colleague. I have
examined and worked with the raw exposure data from the US granite
study. I think that these data are much too weak to be used on a
stand-alone basis.



Gary Ginsberg	OSHA’s risk range is based upon the limited number of
studies which have received quantitative treatment.  As discussed above,
there may be others which would be suitable for quantitative assessment
if they were formally considered for this application, but this document
does not do so.  For example, as stated above, one study in the
Steenland et al 2001 analysis, South African gold miners, yielded a
potency that was approximately 10 times greater than the remainder of
the studies.  Perhaps that is an appropriate upper bound for lung cancer
risk.  Without further discussion of the merits of that study relative
to the others, one cannot determine how much weight to put on it.  

Further, the assumption is made that this risk range applies equally to
quartz and other polymorphs such as cristobalite, and with respect to
non-cancer endpoints, that it refers specifically to silicosis.  In
lumping quartz and cristobalite potency information, OSHA should provide
better support that this is appropriate.  The quartz/cristobalite
distinction is brought up in passing on Page 41 and more detailed
explanations are supposedly provided in Sections VI-B and VI-E.  I
couldn’t find this discussion in VI-B and in a later section of VI
titled “Physical Factors that may Influence ….. (page 97-98, I guess
VI-E but unclear)” two paragraphs are given to this distinction, with
rather unconvincing result.  For example, the statement is made that a
difference between these polymorphs has not been seen in epidemiologic
studies but only a limited dataset is being referred to here, and it
evidently does not include the Park et al. 2002 data which showed 8-9
times higher silica-related mortality in diatomaceous earth workers
(cristobalite) than in the multi-center study.  Clearly some of this
difference is due to the broader case definition in Park et al. 2002,
but that doesn’t seem to account for all of the difference as a large
number (percentage undefined) of the LDOC deaths were likely silicosis. 


A review by Bolsaitis and Wallace (1996) is cited as supporting no
difference in terms of hemolysis or fibrosis between cristobalite and
quartz but from the abstract that review appears to focus mostly on
surface properties rather than polymorph.  The OSHA document would be
benefited from further documentation along these lines especially since
there is at least some epidemiology of increased risk in association
with cristobalite.   

OSHA may want to evaluate whether the LDOC mortality (lung disease other
than cancer, as defined by Park et al., 2002) is the more appropriate
endpoint for silica-induced non-cancer mortality risk.  Right now the
focus is on silicosis, which as stated in the document, can be an
underreported condition, sometimes called emphysema or other lung
conditions as a cause of death.  Further, as documented in Section V,
silica can induce a wide variety of lung conditions in addition to, and
without the necessary occurrence of, silicosis.  Therefore, an additive
or more inclusive approach to silica-induced non-cancer risk assessment
(as suggested by the Park et al. 2002 approach) may be more appropriate.
 It is useful that Table VI-12 pulls that data point out separately but
its unclear how much weight will be given to this potency value.



Brian Miller	The calculation and presentation of risk estimates needs
further elucidation and discussion.

The section on risk assessment for lung cancer mortality displays
several different estimates of what it calls ‘lifetime’ risks of 45
years’ exposure to silica at chosen average concentrations, starting
from age 20.  These risks are based on predicting numbers of deaths,
allowing for intercurrent mortality from other causes, but it is not
clear exactly how this is done.  The commissioned report summarizing
these estimates (ToxaChemica, 2004) does not spell this out, merely
referring to a paper from 1974; the paper references Chiang, a standard
reference on life table methods.  Different estimates for mortality and
morbidity are quoted that predict over different periods, variously up
to and truncated at ages 65, 75 and 85.  None of these can claim to be
‘lifetime’ risks, although that to age 85 comes closest.  Certainly,
if silica exposure has an effect on lung cancer risk, it would be
expected to produce excess cases from the rising background risks way
beyond age 65, particularly since the data suggest no reduction of
excess risk on cessation of exposure.  So summation to 65 must produce a
considerable underestimate of the total impact.  To tabulate these
together as if they are comparable, as in Table VI-9, seems unwise
unless accompanied by carefully-worded warnings.

How large is the underestimation of lifetime risk depends on exactly how
the calculation was done.  If I interpret it correctly, the text
suggests that the risk is for 45 years’ exposure starting at age 20,
so follow-up from the end of exposure to age 65 has zero duration, and
therefore should have zero risk.  However, non-zero risks to age 65 are
shown in e.g. Table VI-9, suggesting that the life-table calculation is
done differently.  It is possible that the calculation proceeds stepwise
through the years, introducing excess risk as exposure cumulates, in
which case excess cases could be predicted to occur from 15 years after
exposure began (to allow for latency), but then those that die before
age 65 do not experience 40 years’ exposure.  

The process of predicting excess cases using life-table methods is a
purely arithmetic process, easily done with standard spreadsheets, but
the answers produced depend critically on various assumptions. Whether
to allow time-dependent exposure effects, and for how long to calculate
a follow-up are among these, and need to be described carefully and
accurately.  It is not difficult to produce estimates up to advanced
ages, which might truly be considered ‘lifetime’, so it is not clear
why any estimate so labeled should be truncated at age 65.

Andrew Salmon	In my opinion the presentation of the lung cancer and
silicosis mortality risk for the exposure range considered is well
handled in this section.  The IARC multi-center study probably
represents a reliable median estimate, but by its nature is likely to
underestimate the upper bound on risk in a situation where individual
cohorts show real variation in risks due to differences in the nature
and intensity of the exposure and the effectiveness in determining
health endpoints.  The description in this section describing and
characterizing the range of available estimates is therefore
appropriate.



Noah Seixas	OSHA has presented a thorough and well reasoned discussion
of risks from three substantial exposure-response studies for lung
cancer.  In general, the estimates made are reasonable, and remarkably
consistent in their results, especially within the exposure range
included in the original studies.  It is not clear why OSHA is
characterizing these estimates as ‘upper bound risk estimates.’ 
Because the estimates are based on the coefficient, and upper and lower
bounds are put on each estimate, these are ‘best estimates’ rather
than upper bounds. 

I would agree with putting more emphasis on the pooled study estimate
would make sense, given the size and heterogeneity of this study. 
Although somewhat more weight may be given to it, the analysis of the
two other cohorts, showing remarkably similar risk estimates, actually
adds to the overall weight of evidence presented.  Also, note that the
pooled study only presents lower estimates as exposure increases –
thus at the low end of exposure, the estimate is almost the same as the
Granite workers, and somewhat higher than the DE study results.  

It would be useful to clarify why the OSHA analyses provide different
estimates than the original study results, and are in fact consistently
higher than the original estimates.  In what ways were the analyses
different, and why did those differences result in higher estimates?



The cumulative risk of silicosis morbidity is preliminarily considered
to be best represented by a range (approximately 3- to 4-fold) that is
based on results from four published exposure-response assessments. 
Does this range reflect a sound and reasonable estimation of risk from
the available data?  Is this decision adequately explained?  Should the
risk estimates based on one data set be preferred over the other?  Is
there a more appropriate representation of the silicosis morbidity risk
from silica exposure?  If so, explain.

Bruce Allen	The decision to limit quantitative risk estimates to the
studies that included retirees as well as current workers was a good
one.  Cross-sectional studies are particularly problematic, and the
cumulative risk studies that excluded retired workers lose a segment of
the population that may have a different (higher) response rate than
those who were able to continue work.

The decision to present estimates from those five studies appears
reasonable.  OSHA has indicated a slight preference for the estimates
derived from Miller et al. based on the quality of the exposure indices.
 That is a reasonable position, but the variations across study are
informative and largely corroborative.  Again, given the differences
that might occur across current and future work places subject to
regulation, it is appropriate to consider as many different reasonable
estimates of risk as possible.

Kenny Crump	As discussed above, studies that have tested for a dose-rate
effect have found evidence of such an effect.  Given this, it would be
reasonable to take such an effect into account in the risk estimates. 
Some suggestions are made above about how this might be accomplished.

Murray Finkelstein	I agree with OSHA that the relevant cohorts are those
that include extended follow-up, including the experience of retirees. I
agree that it is important to acknowledge a sizeable range in the risk
estimates, due primarily to sampling variability, length of follow-up,
and exposure uncertainties. I think that the range presented reflects a
sound and reasonable estimation of risk from the available data. I think
that the decision is adequately explained. Because of uncertainties, I
favor the use of as many studies as are relevant, rather than relying on
single studies.

Gary Ginsberg	OSHA appropriately narrowed the field of potential
morbidity studies to those which employed retirement follow up and did a
good job of describing the pros and cons of each study.  Given the wide
disparity in exposures, exposure measures, and diagnostic procedures, it
is surprising that the potency estimates fall into a fairly narrow
range.  OSHA’s characterization of this range appears to be
appropriate.  However, the statement on Page 82 regarding autopsy
results in the 1993 study is troubling, suggesting a high rate of false
negatives.  This implies that autopsy results are more sensitive and
that there may be some systemic under-reporting of silicosis.  While the
uncertainty in this study regarding the possible underestimation of
exposure is mentioned in summary on Page 89, the possibility of
underestimation of disease is not.  The uncertainty created by this
possible source of underestimation should be further described.

Brian Miller	There is ample evidence that the morbidity risks from
silica vary widely across industries and across exposures that differ in
nature and in composition.  For that reason alone, different studies are
likely to produce a range of estimates.  All other things being equal,
for purposes of regulation one might want to focus on the highest risk
estimates.  However, all other things may not be equal.  For example, it
should be clear that much of the risk of silicosis from the Miller et al
(1998) study was from short exposures at concentrations well above the
target concentrations considered here. 

Although the Miller study is highlighted as having the
best-characterised exposures, the most recently published analyses
(Buchanan et al.) were of a response at ILO category 2/1+, and therefore
fewer than if all responses at 1/0+ had been taken.  It would of course
be possible to use from these data the results of analyses with a 1/0+
response, if that were required:  see the publications listed in answer
to Q1.  Whatever response is considered, interpretation needs to bear in
mind that the radiological responses were taken at a particular point in
time, and that it is both possible and likely that the abnormalities in
at least some of the workers studied would have progressed further, even
though no longer exposed.  With a progressive disease process,
quantitative lifetime risk estimates must take into account the rate of
progression over time.

Andrew Salmon

Andrew Salmon	A range of risk estimates is presented for the specific
concentrations previously identified as OSHA PEL values.  This is
obviously of interest in the specific context of this assessment, but it
does not provide an evaluation of the overall relationship between
exposure and effect, especially for an effect such as silicosis where
the dose-response relationship is not necessarily linear at low doses. 
The charge question refers to four studies chosen for principal
evaluation, but the summary description (page 112 and page 114) refers
to five studies (Chen et al., 2001; Chen et al., 2005; Hnizdo and
Sluis-Cremer, 1993; Miller et al., 1998 and Steenland and Brown, 1995b)
as useful in 

estimating long-term risk of silicosis.  This dichotomy needs to be
clarified as to exactly which sources OSHA is using as the basis of its
range of estimates.  

At several points in the discussion OSHA notes concerns with the
exposure measures used in the South African studies (specifically Hnizdo
and Sluis-Cremer, 1993 in this context).  It should be noted that
although obviously there is some uncertainty in any process of
conversion of exposure measures from one basis to another these data
have in this case been thoroughly explored (Beadle and Bradley, 1970;
Page-Shipp and Harris, 1972), providing reasonable confidence in the
results.  The assertion of Gibbs and duToit (2002) that, based on the
assumed silica content of the dust, exposure was underestimated by
Hnizdo and Sluis-Cremer (1993) has been shown to be incorrect (OEHHA,
2005).  Other sources (e.g. Kielblock et al., 1997) suggest that if
anything the uncertainty is in the other direction.  The more recent
report by Churchyard et al. (2004) is also useful to compare with the
earlier South African results, since these authors used gravimetric
exposure measurements in similar situations to the earlier
particle-based data.  This data set is also worth consideration in
providing additional estimates of silicosis morbidity, although the size
of the cohort is not as large as that examined by Hnizdo and
Sluis-Cremer (1993).  The large size and inclusion of job classes with a
substantial range of different exposure levels make the study by Hnizdo
and Sluis-Cremer (1993) important for determining the overall dose
response for silicosis.  Health endpoint determination used a relatively
sensitive endpoint (ILO 1/1) and exposure measures although not by the
latest preferred method appear to have been systematic and consistent. 
This is a significant contrast to the study (Miller et al., 1998) chosen
by OSHA as the “most reliable overall”, which although of excellent
quality in all respects used a less sensitive endpoint (ILO 2/1) and
thus is bound to provide a lower estimate of the risk.  This study also
involved a cohort size of little more than half that of Hnizdo and
Sluis-Cremer (1993) and apparently a narrower range of exposure
concentrations (most of the interindividual variation in cumulative
exposure within the two groups compared appears to be based on
differences in duration rather than exposure concentration.).  Thus,
although it has considerable power to determine risk at cumulative
exposures in the middle range comparable to the OSHA PELs, it is less
informative about the overall shape of the dose-response curve.  

If OSHA wishes to examine the overall dose-response relationship for
silicosis morbidity as opposed to merely single-point estimates they
would do well to examine analyses such as that published by Collins et
al. 2005) which attempt to address the response relationship in the
lower cumulative exposure range.

Noah Seixas	Yes – the approach and rationale for selecting the studies
used, and the interpretation of the range of results, is well explained
and sound.  It is of great importance that the range of findings is
presented, rather than relying on any one study or estimate. 
Identifying a study that OSHA thinks  perhaps provides the strongest
results, and the fact that that study (Miller, Coal workers) has risk
estimates near the middle of the range, provides good evidence of the
reasonableness of the presented range of risks.

As with the lung cancer risk estimates discussed below, OSHA must
consider how to address the clearly unacceptable level of risk estimated
to occur at 0.05 mg/m3.  Whether one chooses the Miller results (8 per
100) or the range presented by the other studies (2 – 13 per 100
workers), this risk level is clearly unacceptable.



OSHA estimated the risk of lung cancer mortality, silicosis mortality,
and silicosis morbidity for 45-year exposures to 0.25 and 0.5 mg/m3
respirable crystalline silica; these represent a range of exposures
consistent with OSHA’s current standards for construction and maritime
industry sectors, but are above OSHA’s current general industry
standards (approximately 0.1 mg/m3 for respirable quartz).  The
cumulative exposures corresponding to these higher exposure levels are
at or above the cumulative exposures of most of the more highly exposed
workers included in the studies that underlie the risk estimates.  Is
OSHA’s approach for extrapolating risk above the observed range
reasonable?  Is it clearly explained?  Is there a more appropriate
approach that should be considered to estimate risk at cumulative
exposures that are generally above the observed range of the underlying
epidemiological studies?

Bruce Allen	I consider these questions to be moot.  If the analyses show
substantial risk associated with concentrations lower than those in
question, ones that do not require extrapolation above the cumulative
exposures observed in the studies themselves, then it is entirely
reasonable to presume that higher concentrations (i.e., the 0.25 and 0.5
mg/m3 values) will be associated with even higher risks.  In that sense,
there need not be a detailed explanation of (nor angst about) methods
for extrapolation above the levels where there are observed increased
risks.

Kenny Crump	OSHA evaluated this issue by comparing cumulative exposures
corresponding to exposure levels of 0.25 and 0.5 mg/m3 to the median
exposures and third quartile exposures in various cohorts.  Such
comparisons are not particularly indicative of whether risks have been
extrapolated beyond the range of the data.  The important issue is
whether there are sufficient numbers of cohort members with exposures in
the range of interest.  Irrespective of the median of third quartile
exposures, there still could be significant numbers of cohort members
with exposures in the range of interest.  For example, adding many
additional cohort members with low exposures would lower both the median
and third quartile exposure, but would not reduce the ability to
quantify risk at higher exposures.  To explore this issue further, OSHA
could evaluate the potential numbers of cohort members in the various
studies with exposures in the range of interest.  My expectation is that
the pooled analysis will contain a significant number of cohort members
with exposures in this range.  (However, the pooled analysis may
underestimate risks in this range due to use of log-transformed
exposures – see response to Question 4.)

Murray Finkelstein	Extrapolation beyond the observed range depends on
the adoption of a risk model. The selection of the linear or log-linear
models is clearly explained. The approach adopted by OSHA is reasonable.
An alternative approach is simply to say that the risk is greater than
X, where X is the risk observed at the maximum exposure category in the
underlying epidemiologic studies.

Gary Ginsberg

Gary Ginsberg	The issue of extrapolation to cumulative exposures higher
than the majority of the empirical data is a concrete issue.  The lung
cancer high exposure analysis (pp 28-32) seems to be contradictory in
that OSHA argues that a linear relationship with cumulative dose is
likely (page 29) but then prefers the estimates derived from the log
cumulative exposure model which flattens out the dose response at high
dose.  OSHA argues that other forms of mortality or healthy worker
effect may be behind this flattening.  Given that the US granite worker
data show an apparent flattening at cumulative exposures below the range
of concern in this question (Attfield and Costello 2004 study described
on page 20), it would appear that a model that reflects a response
plateau may be appropriate.  OSHA focused on the high dose in Attfield
and Costello 2004 as a possible outlier.  I agree with this
interpretation but believe that a more appropriate focus would be on the
entire dose-response curve in this study and the others.  Is there
evidence from the other studies of flattening as well?  Are there cases
of a superlinear response at the high end of cumulative dose?  It would
appear that OSHA could make more complete use of the underlying data to
support model choice.

Brian Miller	The mechanism of the approach is clearly explained; the
appropriateness of the methodology depends on cumulative exposure being
the correct metric, and this is discussed above.  A different metric
would presumably produce a different set of differentials between the
average concentrations.

Andrew Salmon	Any extrapolation outside the range of exposures for which
sound quantitative data are available is subject to uncertainty, and
this is especially true here where there are suspicions of possible dose
rate effects and/or different pathological responses (e.g. acute
silicosis) at very high doses.  OSHA does not examine these
possibilities very thoroughly, but it is unclear what could be done
about them given the nature of the data for higher exposures (typically
including case studies and early investigations with uncertain or
non-existent exposure measurements, but no systematic large-scale
studies).  Overall the approach at least for these moderately higher
pre-defined exposure levels is reasonable, although there is no effort
to describe an overall dose-response curve which would characterize the
expected response at both low and high levels.  This alternative would
be useful in considering various questions about mechanism of causation
for various health endpoints, as well as informing a health-protective
standards-setting procedure.

Noah Seixas	Yes. The approach and rationale for this extrapolation is
well explained, and reasonable.  Further, it is not clear to me that the
precise estimates for risk in these higher levels of exposure are of any
consequence.  At the higher range of exposure, OSHA needs only to
demonstrate that a substantial risk exists, which is established using
any means of extrapolation chosen.  The precision of the risk estimates
at the lower end of exposure is much more important for the risk
analysis.



Additional Comments/References

Bruce Allen	None.

Kenny Crump

Kenny Crump

Kenny Crump

Kenny Crump

Kenny Crump

	Page 6

“OSHA’s current general industry PEL for quartz is the gravimetric
formula 10 mg/m3/(%quartz + 2) expressed as the concentration of
respirable dust.  For quartz concentrations exceeding 5 percent, the PEL
expressed as the concentration of respirable quartz exceeds 0.07 mg/m3
and approaches 0.1 mg/m3 as the quartz content increases.”

An example could perhaps make this clearer.

Page 12

“the log of cumulative exposure”

This should be the log of cumulative exposure plus 1.  This is an
important distinction that should be made throughout the document.  In a
log linear model that uses log-transformed exposure, the relative risk
approaches zero as exposure approaches zero, rather than approaching
1.0.  However, when 1 is added to cumulative exposure, the relative risk
approaches 1.0 as it should.

Page 15

“Steenland et al. (2001a) estimated excess lifetime risk of lung
cancer based on rate ratios from the spline and log-linear models
(15-year lagged log-cumulative exposure for both),”

“Spline” refers to how exposure is modeled.  “Log-linear refers to
how response is modeled as a function of exposure.  Therefore these
descriptions are incomplete and do not necessarily refer to separate
models.  The entire report needs to be carefully read for errors such as
this, and particularly to insure that “log-linear” and
“log-transformed exposure”  are not used incorrectly (e.g.,
interchangeably).  

A “log-linear model” based on cumulative exposure is a fundamentally
different model from a “log-linear model” based on the log-transform
of cumulative exposure.  For clarity the document should use different
terminology in referring to these different models (e.g., “cumulative
exposure log-linear” and “log-transformed cumulative exposure
log-linear”).   

Page 17

“Log quadratic: RR = (β1E + β2E2)”

This equation should read RR = EXP(β1E + β2E2).

 Page 19

Define REL the first time it is used.

Page 28

“For the model based on the pooled analysis, OSHA used the exposure
coefficient (β=0.060) that reflected corrections made by Steenland and
Bartell to the original data (see Table VI-1) and assumed 250 working
days per year.”

This section should state the exposure metric in every case.  E.g., the
coefficient of 0.060 is in what units and for which model?

Page 31-32

“It is also possible that the lung cancer response was diminished in
this group due to an exposure-related increase in competing causes of
death; in particular, tuberculosis, pneumoconiosis, and non-malignant
respiratory disease mortality rates were all 2- or 3-fold higher than
those for the remaining exposure groups.”

Is this really a problem, since the analysis controlled for competing
causes of death?

Page 35 and 36 and other places 

Mannetje et al. (2002a) should often be Mannetje et al. (2002b).

Pages 40-41

Another difference between methods of risk estimation by Mannetje et al.
(2002b) and Park et al. (2002) is that Park et al. apparently adjusted
for competing risks of death whereas Mannetje et al. did not.  This
would tend to make the disparity even greater.

Page 41

“study represent the most conservative estimates”

“Conservative” needs clarification.

Page 48

“Assuming that the errors resulting from the assignment of
job-specific mean exposures to individuals and the exposure metric
conversions affect diseased and nondiseased study participants equally
results in a type of error known as a Berkson error”

This definition is incorrect.  Berkson error has to do with the exposure
variable and is not defined in terms of the response.  

Page 56

“Part of the problem appears to have arisen from an erroneous footnote
in the Hnizdo et al. (1997) report in which they indicate they applied
the 30-percent conversion factor to respirable dust which had been
acid-washed (for which the 54% conversion rate would have been more
appropriate), whereas in fact the investigators applied the 30-percent
factor to untreated dust.”

This is taken from the Toxichemica Inc. report where it is stated
without any reference.  A reference is needed for this claim.

Page 57

“According to Toxichemica, Inc., (2004) the use of the
log-transformation (i.e., either the log-linear model with log
cumulative exposure or the spline model using log cumulative exposure)
ensures that any constant multiplicative bias in exposure has virtually
no effect on conditional logistic regression coefficients, and would
have no effect at all using a pure log-transformation (i.e., without
adding 1 to allow for exposures at 0 mg/m3-days).”

This is true, which argues against the biological reasonableness of such
a model.

Page 89

“Miller et al. (1998) study reflect the risk that nodular profusion
will progress such that a chest x-ray would be classified as ILO major
category 2 or higher.  Such estimates would understate the risk of
developing lesser degrees of profusion (i.e., ILO 1/0 or 1/1).”

It could also be noted that some of subjects in this study were still
fairly young (<= age 35) when the last evaluation was conducted, and it
seems likely that additional ones of these will develop opacities over
time.

Page 94  

“In addition, the results of this study are made uncertain by a lack
of follow up; for the cohort overall, the latest chest film was taken an
average of 11.5 years after hire, a period too short to ensure that all,
or nearly all, cases of silicosis that will develop could be
detected.”

But I don’t see how this detracts from their findings.  Their finding
was statistically significant with the amount of followup present in
their study.  Is OSHA suggesting that the effect would go away with more
followup?

Page 94

“several-fold above the current OSHA PEL (i.e., above …”

Change to “several-fold above the current OSHA PEL (i.e., several-fold
above …”.

~Page 95 

“freshly fractured silica”

Does freshly fractured silica contain more particles per mass?  Could
any different toxicity be due to this factor?  

Page 108

“The use of the log-transformed exposure metric in the pooled study
was necessary to reduce significant heterogeneity between the 10 worker
cohorts that was evident with a model based on untransformed cumulative
exposure.”

As pointed out in answers to questions, this is a faulty argument.

Page 108

“or a substantial increase in mortality from competing causes.”

The study controlled for mortality from competing causes, so this should
not be a factor.

Page 114

“OSHA believes that risk estimates based on the Miller et al. (1998)
study are the most reliable overall because of the availability of
high-quality exposure data, but they may understate the risk of
developing early signs of silicosis (i.e., corresponding to ILO x-ray
film classifications of 1/0 and 1/1).”

Also, this study may underestimate risk because it included a sizable
fraction of younger men who are at risk of developing silicosis in the
future.

Page 128, Footnotes to table 12

“Estimated by OSHA based on the Park et al. (2002) linear relative
rate model, RR = 0.5469*E where E is cumulative respirable crystalline
silica exposure in mg/m3-years.”

The model is stated incorrectly.  It should be RR = 0.5469*(1+E).  I
don’t know if this was a typographical error or an error in the
calculation.  If the latter, risk will be underestimated.  More
generally, these footnotes should be reviewed carefully and expanded, if
necessary, to make it clear exactly how these calculations were
conducted.

References not used in the document under review 

Akaike, H. (1974). A new look at the statistical model identification.
Automatic Control, IEEE Transactions on, 19(6), 716-723.  

Carroll, R. J., Ruppert, D., & Stefanski, L. A. (2006). Measurement
Error in Nonlinear Models (p. 455). Boca Raton: Chapman & Hall/CRC.  

Cox, D. R. (1972). Regression models and life tables. Journal of the
Royal Statistical Society, Series B (methodological), 34(2), 187-220.  

Figure 1

Illustration of why rejecting a model that shows significant
heterogeneity in favor one that doesn’t is not a good decision rule.  

This figure depicts data from two studies, Study 1 (blue points) and
Study 2 (red points).  Exposure to a toxicant is measured on the
horizontal axis and the corresponding biological response is plotted on
the vertical axis.  In both studies, increased exposure is associated
with a decrease in the biological effect being measured.  There is an
apparent study effect as exposure causes a stronger decrease in response
in Study 1 than in Study 2.  Two models are used to analyze these data:
Model 1 (orange lines) and Model 2 (green lines).  The solid lines
indicate the fits of the models to the combined data and the dashed
lines indicate the fit of the models to the data from the individual
studies.  The fit of Model 2 to the combined data (solid green line) is
clearly better (i.e., larger log-likelihood) than the fit of Model 1
(solid orange line).  However, there is a significant improvement in fit
when Model 2 is fit separately to the two studies (dotted green lines). 
Thus, this model detects significant heterogeneity.  However, with Model
1 the improvement in fit is not significant (in fact shows practically
no improvement); hence there is no significant heterogeneity detected by
this model.  By the decision logic used in the analysis of the lung
cancer data from the pooled analysis to select the model based on
log-transformed exposure over the model based on untransformed exposure,
Model 2 would be ruled out in favor of Model 1 even though Model 2 is
clearly superior.    



Murray Finkelstein	None.

Gary Ginsberg	None.

Brian Miller

Brian Miller	Edits, corrections and suggestions

P12 para 2 ?Make explicit that in each set of odds ratios the first is
set arbitrarily at 1.0 and therefore has a different meaning from the
others.

P13 para 2 The non-spline models…

P16 para 1 e.g. -> i.e.

P18 para 2 It should not be a surprise that results from Cox- type
analyses and Poisson regression are similar, since the underlying model
in both is the same.  Where they may differ is in the extent of grouping
applied.  If the same grouping strategy is applied, the results will be
identical.  So often these are compared as if they were different
models, when they’re not.  

Same again on P39

P32 para 1 line10 estimate -> estimates

P78 para 1 Poission -> Poisson

P103 para 2 difference -> differences

P107 para 1 OSHA estimates the risk *up to what age?* to be…

P114 para 2 may -> must

Andrew Salmon

Andrew Salmon

Andrew Salmon	General Comment:  Compared to the more usual kind of
quantitative risk assessment study, this is a very peculiar document. 
The usual risk assessment paradigm involves hazard identification,
exposure assessment, dose-response assessment for hazards identified as
being of concern, and finally a risk characterization which seeks to
identify either a slope factor for non-threshold effects, or a
health-protective level where the effect of concern is considered to
show a threshold.  These conclusions are then combined by the risk
manager with any other relevant considerations (including measurement
and feasibility) in setting suitable trigger levels for regulatory
action.  The section on health effects does a thorough job of
identifying and reviewing the information on health hazards.  This
section on the other hand starts with the previously chosen regulatory
levels, and seeks an analysis of various available data to provide
estimates of risk for key endpoints at these exposure levels.  

Given this perspective, which is the inverse of the normal approach, the
document makes a very good case that substantial adverse health effects
are anticipated in workers exposed for a working lifetime to these
predefined exposure levels.  The most thoroughly developed analyses are
for lung cancer incidence and mortality from silicosis, which are severe
endpoints, and the principal concern in selection of dose-response
models appears to have been the accuracy with which the data are fitted
in the range of the target exposures of interest.  It is hard not to
conclude from a prediction of a 2% lung cancer incidence in workers
exposed at the lower PEL that these regulatory levels are insufficient
to protect workers’ health.  Yet, because of the studies, models and
endpoints selected this report contributes almost nothing to
consideration of what might be more reasonable levels of exposure that
should be considered at least as goals, even if they prove difficult or
expensive to achieve.  In particular, there is no consideration of
dose-response modeling at the lower end of the observed dose ranges, and
no consideration of the possible existence and location of a threshold
in the dose-response curve for silicosis.  There is a tendency in the
current analysis to concentrate on severe endpoints such as mortality
and severe grades of silicosis (which are observed at the exposure
levels predetermined to be of interest), rather than to evaluate milder,
but still significant endpoints which are observed in some studies at
substantially lower exposure levels.

As in the case of most other non-cancer health effects, our
understanding of the mechanism for silicosis induction by quartz
particles, although far from complete, is certainly consistent with the
possibility of a threshold in the dose response relationship, although
this threshold might be well below the levels commonly encountered in
high-risk occupational settings.  The existence of such a threshold for
silicosis is supported by the fact that, although background exposure to
crystalline silica is not zero, this disease is not usually observed in
the general population (although a few special situations producing
“environmental silicosis” are known).  In more specific terms,
Schenker et al (2009) observed histologically identifiable
pneumoconiosis in agricultural workers from the dusty California Central
Valley, but not in individuals from the same area without the heavier
exposures characteristic of employment in agriculture.  Quantitative
analyses of silicosis dose-response which did consider the lower dose
ranges (e.g. Collins et al., 2005) have generally assumed a model which
does imply a threshold, rather than the linear, logistic or log-linear
relationships used to model the response in the range considered by
OSHA.

On the other hand, a threshold for lung cancer induction is not
necessarily expected on theoretical grounds.  There might nevertheless
be such a threshold, or at least a substantial increase in the dose
response slope at higher exposures, especially if the hypothesized link
between lung cancer and silicosis (or some precursor to that condition)
were to be established. The demonstration of any such thresholds would
have important implications for defining “safe” or at least low-risk
exposures in occupational settings.  Toxichemica (2004), citing
Steenland and Deddens (2002), note that such a threshold may in fact
exist at low levels –around 10 µg/m3, although they were unable to
confirm this from a purely statistical standpoint.  It is a significant
omission both in terms of defining a health protective standard, and in
analyzing possible mechanisms of silica lung carcinogenesis, that the
OSHA document does not further examine this possibility, and the
relationship between possible thresholds for lung carcinogenesis and for
silicosis.

References cited

(other than those appearing the bibliography for the OSHA QRA document)

Beadle DG, Bradley AA. 1970. The composition of airborne dust in South
African gold mines. In: Shapiro HA (ed). Pneumoconiosis. Proceedings of
the International Conference. Johannesburg 1969. Cape Town: Oxford
University Press. pp. 462-6.

Churchyard GJ, Ehrlich R, teWaterNaude JM, Pemba L, Dekker K, Vermeijs
M, White N, Myers J (2004). Silicosis prevalence and exposure-response
relations in South African goldminers. Occup Environ Med. 61(10):811-6.

Collins JF, Salmon AG, Brown JP, Marty MA, Alexeeff GV (2005). 
Development of a chronic inhalation reference level for respirable
crystalline silica.  Regul Toxicol Pharmacol.  43:292-300. 

OEHHA, 2005. Determination of Noncancer Chronic Reference Exposure
Levels.  Chronic Toxicity Summary for Silica (crystalline, respirable). 
  HYPERLINK
"http://www.oehha.ca.gov/air/hot_spots/2008/AppendixD3_final.pdf" \l
"page=486" 
http://www.oehha.ca.gov/air/hot_spots/2008/AppendixD3_final.pdf#page=486
.

Page-Shipp RJ, Harris E. 1972. A study of dust exposure of South African
white gold miners. South Afr Inst Mining Metall. 73:10-24.

Schenker MB, Pinkerton KE, Mitchell D, Vallyathan V, Elvine-Kreis B,
Green FHY (2009).  Pneumoconiosis from Agricultural Dust Exposure among
Young California Farmworkers.  Environmental Health Perspectives 117(6):
988–994.

Noah Seixas	One additional comment.

OSHA estimates risk of deaths due to silica exposure at 0.05 mg/m3 using
best estimates, in the range of 19-29 per 1000 workers, or 2-3 deaths
per 100 workers.  This is clearly an unacceptable risk of death.  The
risk analysis on which these estimates are given is very strong given
the underlying data.  There are three large studies based on eight
cohorts, each with a substantial degree of exposure quantification, and
the results are strikingly consistent. Given the strength of the risk
analysis, OSHA should set a standard with an acceptable level of
estimated risk.  Two to three deaths per 100 workers is clearly an
unacceptable level of risk.  A lower standard, consistent with legal and
societal norms of acceptable risk should be considered.



Individual Peer Review Comments

Bruce Allen

Bruce Allen Consulting, Chapel Hill, NCFrom: 	"Bruce Allen"
<bruce_allen@verizon.net>

To:	"'Kate Schalk'" <Kate.Schalk@erg.com>

Date: 	12/16/2009 1:52 PM

Subject: 	RE: OSHA SILICA - Final revised comments due to ERG 12/16

I have no additional comments or modifications to my earlier comments. 
The only part of the discussion that I contributed to that you hade
marked as perhaps needing input from me related to a comment about the
uncertainty related to exposure intensity measures. My comment was that
I did not think that they would be any more uncertain (and perhaps
slightly less uncertain) than cumulative exposure measures for any of
the cohorts used.  That would allow examination to see if adding
exposure intensity (related to the dose rate question) would
significantly improve the fit of the models to the data.  My comment was
in response to another reviewer who expressed the opinion that intensity
or rate data would be too uncertain to use; my response is that since
they are used now to estimate cumulative exposures, they can add no
other uncertainties than those that are already included.

Bruce 

Peer Review of OSHA’s Preliminary Quantitative Risk Assessment for
Silica

Comments from Bruce Allen

The draft report evaluates exposure-response data from several
occupational studies as to their suitability for quantitative assessment
of lung cancer and silicosis risk resulting from airborne exposures to
silica.  Are the strengths and limitations of the selected data sets
adequately discussed and appropriately evaluated?  Are you aware of
other study data that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica?  If yes, inform
OSHA of the other data.

In general, I found the discussion of the strengths and weaknesses of
the selected studies to be adequate.  Because such discussions can be
very discursive, I might suggest that a tabular summary of the strengths
and weaknesses be presented.  In such a summary table, the criteria for
evaluating strengths and weaknesses could be laid out (say in the rows)
and then for each study (one per column), a brief point or two could be
presented.  This would allow the more in-depth discussions of the
strengths and weaknesses in the text to be tied back to this summary and
would allow a clearer appreciation of how the studies compared to one
another with respect to the criteria under consideration.

If anything, the weaknesses of all the studies with respect to
reconstruction of exposure histories (both with respect to the
atmospheric concentrations and the job-specific features that lead to
worker exposures to those concentrations) may not have been presented
with enough emphasis to convey just how limiting and problematic that
process can be.  There is some discussion of the “reasonableness of
the exposure assessment” on p. 10, where silicosis mortality odds
ratios are compared across exposure categories from the pooled cohorts
of Steenland et al.  The values presented do not give me a very strong
sense that exposure misclassification was negligible, since the odds
ratios presented hardly differ (the highest three are almost identical)
and I am not sure that a formal test would reject the hypothesis that
the odds ratios were the same, i.e., would not reject the hypothesis
that would hold if all individuals were randomly allocated to the four
groups regardless of their cumulative exposure level.  I have to wonder
if this is the best evidence that is available to OSHA to make the
desired point about impact of exposure misclassification.

I am not aware of any other data that are better suited for quantifying
the lung cancer and silicosis health risks.  

The draft assessment relies on quantitative uncertainty analysis and
sensitivity analysis of exposure measurement error in key studies of
lung cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach?  If so, explain.

In contrast to my comments above about the less-than-convincing
presentation related to exposure misclassification, I did like the
analysis that was presented to examine the quantitative effects of
uncertainties about the exposure reconstruction and the individual
worker exposures within that reconstruction.  The Monte Carlo analysis
approach that was applied is a reasonable way to address this question. 
My only comment related to that is that a typical Monte Carlo analysis
usually includes many more than 50 iterations (the maximum number if
simulations run in the uncertainty analyses for lung cancer and
silicosis mortality).

Note also that the analyses were performed by selecting some fixed
values for the variabilities considered.  Although the variability was
based on some observations (see bottom of p. 50 and bottom of p. 52 to
top of p. 53), those observations were themselves limited and may not be
the best reflections of the variability that is needed here.  It would
be informative to determine just how much variability is required to
result in a substantial change in the risk estimates.  If that amount of
variability can be deemed unreasonable, then there is stronger support
for the risk estimates that OSHA presents.  Or, another way of thinking
about it: if, even when there is very large variability (representing
uncertainty in the exposures), the calculated risks are still considered
substantial, then there is very strong support for the conclusion that
workers are at risk at the exposure levels under consideration.

The draft assessment evaluates the use of a physiologically-based
pharmacokinetic model that predicts silica lung burdens in individuals
exposed to particulates through inhalation for estimating risk from lung
cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach? If so, explain.

The use of a kinetic model to estimate lung burdens that would then lead
to inflammation, fibrosis, etc. is a very reasonable approach.  However,
the methods and approaches used by Kuempel et al. for their assessment
are not well explained or presented in the OSHA QRA document.  

I am particularly troubled by the description starting in the middle of
p. 25 that says that Kuempel et al. used the dose coefficients that had
been previously derived for the two cohorts they analyzed.  Why would
they use those coefficients that were derived without consideration of
the lung burden metric, i.e., that considered simple cumulative
exposure, to estimate lifetime risks based on the lung burdens
associated with various exposure scenarios?  This is either an error in
their methodology or an incomplete and misleading reporting of their
analysis in the OSHA document.

The discussion on p. 27 (first full paragraph) describes the analysis
using the lung burden metric with the 10-cohort study data.  Although
the conclusion is that the lung burden metric does as good a job as
cumulative exposure, the risk estimates derived from that analysis are
not presented (neither in the text nor in Table VI-2, where the summary
of all the risk estimates is presented).  My impression is that OSHA has
not fully considered the lung burden-based risk estimates and perhaps
does not consider them to be “on a par” with the other estimates. 
The discussion of the Kuempel et al. study appears to be the most
cursory of the discussions of the various studies.  If it is the case
that this study is not considered as relevant or it is thought of as
secondary (perhaps because it was not an epidemiology study in and of
itself but relied on the response data of two of the other studies
cited), then there should be some discussion of that and the
inter-relationships between the estimates from that “study” and the
others.

The draft assessment summarizes several published risk assessments that
use a variety of models and statistical methodologies employed by the
various analysts to fit the featured data sets.  Are the models and
approaches used by these investigators adequately explained and
evaluated?  If not, let OSHA know what needs to be added to the
presentation.

I think that a summary table showing the approaches and models used by
the various investigators could improve the presentation somewhat. 
Table VI-2 does have some of that information (which model was used for
the risk estimation – although there appears to be an error there in
that the Attfield and Costello analysis appears to have used a linear,
not log-linear, relative risk model).

It is difficult to determine how much information to include in a
summary of the studies.  But seeking consistency in the presentation of
the methods and results across studies (as opposed to just reflecting
the perhaps-idiosyncratic information presented by the original authors)
is a desirable goal.  So, the presentation for each study could include
the same, “standardized” information: models considered, lag times
considered, how fitting was done and evaluated, special concerns about
model fit (e.g., nonmonotonicities in observed response rates or odds
ratios), etc.

Here are a few examples of questions that I had in reading the summaries
and evaluating what had been done.  

	a.  p. 20, line 7: it says Attfield and Costello used unlagged
exposure.  But how is that possible if they did not update work
histories beyond the original 1982 follow-up (p. 20, line 4) when they
extended mortality follow-up to 1994?

	b. p. 21, line 6: how did Attfield and Costello evaluate
“best-fitting” among all their dose-response analyses, especially
when some of them were on different data (i.e., some with and some
without the high-exposure category)?

	c. If Graham et al. (2004) “concluded that dust control measures were
shown to be effective in reducing lung disease and mortality” (p. 23,
lines 3-4), how could they conclude that the results “did not support
a causal relationship between exposure to quartz and lung cancer” (p.
23, lines 4-5)?

The draft report identifies and describes several key areas of
uncertainty with regard to the estimates of risk.  Were the major
uncertainties in the risk assessment adequately characterized?  If not,
explain what should be included.

As stated above, although I think the uncertainty associated with
historical exposure reconstruction was under-characterized, I like the
Monte Carlo approach to addressing exposure uncertainties (with some
possible extensions suggested in response to question 2.)  The
discussions of possible selection bias, confounding, and bias in
conversion factors were adequate to dispel major concerns about their
having any negative impact on the findings.

The draft assessment relies on the use of cumulative exposure (or a
transformation of cumulative exposure) as the principal dose metric on
which lung cancer and silicosis risks are evaluated.  The draft also
contains a discussion of studies that are suggestive of a dose-rate
effect.  Given this discussion, is cumulative exposure a reasonable
exposure metric for silica-induced lung cancer and silicosis in exposed
workers?  Is there a more appropriate exposure metric?  If so, explain.

It seems that the dose-rate effect may be related to the lung burden
metric predicted by the kinetic model of Kuempel et al.  If that idea
could be explored then there may be a way to provide further support for
the risk estimates that have been presented, especially those based on
that lung burden metric.  This is important given the fact that,
apparently, the lung burden metric did as well as cumulative exposure
(or log-cumulative exposure) in fitting the data.

On the other hand, I was not bothered by the fact that a dose-rate
effect might be present, given the indications that it may occur at
concentrations clearly above levels that OSHA needs to be concerned
about for reducing workers’ health risks.  Given the information about
where (at what concentrations) significant dose-rate effects appear to
be occurring, I am satisfied with the use of cumulative exposure for the
determination of concentration levels associated with the lower levels
of risk of concern for regulatory decision making.

OSHA’s estimates of lung cancer and silicosis mortality are
represented as a range of estimates with those derived from the
diatomaceous earth and U.S. granite studies considered to be plausible
upper bound risk estimates.  OSHA also acknowledges that the risk may be
lower such as suggested by the results of the IARC multi-centric study,
which relied on pooled data from several epidemiological studies.  Is
this a sound and reasonable estimation of risk from the available data? 
Is this decision adequately explained?  Should the risk estimates based
on one data set be preferred over the other?  Is there a more
appropriate representation of the lung cancer risk from silica exposure?
 If so, explain.

I think it is appropriate to present a range of risk estimates and to
consider the differences in the estimates when making decisions about
regulatory limits. As discussed in the document, there are various
differences among the cohorts of individuals exposed to silica,
including the physical properties of the silica itself, that are
potentially important considerations and which may account for the range
of risk estimates. Given those differences, it is important to show the
range as a reflection of what those differences may entail (in addition
to the fact that no two sets of results will be exactly alike or give
exactly the same estimates, even if everything was identical).

The only reason that I can see to prefer one set of estimates to another
would be if that one set was based on a substantially superior study,
say in terms of sample size, relevance of exposure, measurement of
exposure, representativeness of the cohort members in relation to the
workers for whom regulatory limits are being set, etc.  The 10-cohort
study is larger than the other two cohorts alone, and its exposure
contexts clearly cover a wider spectrum.  It might be justified to base
decisions solely on the results of that study, but it is not clear that
those results should be presented to the exclusion of the others.

The cumulative risk of silicosis morbidity is preliminarily considered
to be best represented by a range (approximately 3- to 4-fold) that is
based on results from four published exposure-response assessments. 
Does this range reflect a sound and reasonable estimation of risk from
the available data?  Is this decision adequately explained?  Should the
risk estimates based on one data set be preferred over the other?  Is
there a more appropriate representation of the silicosis morbidity risk
from silica exposure?  If so, explain.

The decision to limit quantitative risk estimates to the studies that
included retirees as well as current workers was a good one. 
Cross-sectional studies are particularly problematic, and the cumulative
risk studies that excluded retired workers lose a segment of the
population that may have a different (higher) response rate than those
who were able to continue work.

The decision to present estimates from those five studies appears
reasonable.  OSHA has indicated a slight preference for the estimates
derived from Miller et al. based on the quality of the exposure indices.
 That is a reasonable position, but the variations across study are
informative and largely corroborative.  Again, given the differences
that might occur across current and future work places subject to
regulation, it is appropriate to consider as many different reasonable
estimates of risk as possible.

OSHA estimated the risk of lung cancer mortality, silicosis mortality,
and silicosis morbidity for 45-year exposures to 0.25 and 0.5 mg/m3
respirable crystalline silica; these represent a range of exposures
consistent with OSHA’s current standards for construction and maritime
industry sectors, but are above OSHA’s current general industry
standards (approximately 0.1 mg/m3 for respirable quartz).  The
cumulative exposures corresponding to these higher exposure levels are
at or above the cumulative exposures of most of the more highly exposed
workers included in the studies that underlie the risk estimates.  Is
OSHA’s approach for extrapolating risk above the 

observed range reasonable?  Is it clearly explained?  Is there a more
appropriate approach that should be considered to estimate risk at
cumulative exposures that are generally above the observed range of the
underlying epidemiological studies?

I consider these questions to be moot.  If the analyses show substantial
risk associated with concentrations lower than those in question, ones
that do not require extrapolation above the cumulative exposures
observed in the studies themselves, then it is entirely reasonable to
presume that higher concentrations (i.e., the 0.25 and 0.5 mg/m3 values)
will be associated with even higher risks.  In that sense, there need
not be a detailed explanation of (nor angst about) methods for
extrapolation above the levels where there are observed increased risks.



Kenneth Crump

Louisiana Tech University Foundation, Inc., Ruston, LA

This review is organized around the questions provided to the peer
reviewers.

The draft report evaluates exposure-response data from several
occupational studies as to their suitability for quantitative assessment
of lung cancer and silicosis risk resulting from airborne exposures to
silica.  Are the strengths and limitations of the selected data sets
adequately discussed and appropriately evaluated?  Are you aware of
other study data that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica?  If yes, inform
OSHA of the other data.

Response

The strengths and limitations of the selected data sets appear to be
adequately discussed and evaluated.  Areas that could be improved in
some relatively minor ways are discussed below.  I am not aware of any
other studies that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica.

 The draft assessment relies on quantitative uncertainty analysis and
sensitivity analysis of exposure measurement error in key studies of
lung cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach?  If so, explain.

Response

This question refers to the analysis of exposure measurement error in
Toxichemica, Inc. (2004).  These analyses are well designed, clearly
explained and evaluated.  Evaluating and adjusting for exposure
measurement error is a very difficult issue.  The analysis of error
conducted by Toxichemica, Inc. is a very strong effort.  The assumptions
are clearly described and the data upon with they are based appear to be
appropriate and appropriately applied.  However, there are questions, as
there will always be with such an analysis, regarding the form(s)
assumed for the errors, the assumptions used in modeling the errors, and
in the completeness of the sources of errors modeled.  

For the Berkson component of the error, it would seem reasonable to
consider a multiplicative error model, i.e.,

		Exposuretrue = Exposureobserved * ε

where ε has a log-normal distribution.  In this form, both the mean and
the variance of the true exposure increases with the observed exposure,
which seems reasonable.  Also, unlike the additive error model, this
model does not produce negative exposures (to avoid this problem is
presumably why the distribution was truncated in the Toxichemica, Inc.
analysis).  This model might produce significantly different results
from the additive model, including modifying the shape of the dose
response curve (Carroll et al. 2006).   

It could also be assumed that silica concentrations differed randomly by
work area rather than by job.  Workers in same area of a facility will
breathe the same air even if they are doing different jobs.  Resampling
randomly by job within work area rather than by work area may tend to
average out variations by work areas and underestimate uncertainty. 

The method for evaluating error in converting from dust measurements to
quartz measurements appropriately based on data and well-designed and
implemented.  However, other important sources of error do not appear to
be accounted for.  

A major source of error that apparently was not accounted for is in
assuming that the average measure of exposure assigned to a job is the
true average.  But it is not always clear how representative the
underlying measurements were.  E.g., were some measurements taken at
times when it was believed there was a particular problem with high
exposures?  Moreover, some of these averages (particularly, those that
pertain to earlier times and which are likely the more influential since
they probably represent the highest exposures) were not based on direct
measurements, but were projected based on judgments of persons familiar
with the operations.  There is possibly considerable error in such
estimates.  Another source of uncertainty in the averages stems from the
need to convert from one measurement method to another (e.g., from
particle counts to gravimetric measurements).  

Based on these considerations, it seems likely that the effects of
exposure errors could have been larger than predicted by the
Toxichemica, Inc. analysis.  It would also be useful to evaluate the
effect of errors in exposure using other exposure metrics than log
cumulative exposure.  Risk models based on other exposure metrics are
likely to be more sensitive to exposure error.  (Also, see response to
Question 4 regarding the use of log cumulative exposure as the exposure
metric.)    

The draft assessment evaluates the use of a physiologically-based
pharmacokinetic model that predicts silica lung burdens in individuals
exposed to particulates through inhalation for estimating risk from lung
cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach? If so, explain.

Response

This refers to an analysis in Toxichemica, Inc. (2004).  Exposure
metrics based on lung burdens can provide reasonable alternatives to the
more empirical metrics emphasized in the papers reviewed by OSHA (lagged
or unlagged cumulative exposure or log cumulative exposure).  The lung
cancer and silicosis analyses are adequately explained as far as they
go.  Whereas the silicosis analysis considered several metrics based on
lung burden, the lung cancer analysis only considered one exposure
metric – lung burden lagged 15 years.  However, both the lung cancer
and silicosis analyses stopped short of estimating risks using this
approach.  Even if a fit to data using a metric based on lung burdens is
similar to the fit provided by an empirical metric such as log
cumulative exposure, that doesn’t imply that the estimated risks from
45 years of exposure will be similar.  Without estimating risk using the
metrics based on lung burden, the effect of these metrics upon the
estimated risk will not be known.  If workers in the epidemiological
studies predominately were exposed for only a short time (e.g., a few
months), different exposure metrics could predict significantly
different risks from 45 years of exposure even if they provided
equivalent fits to the epidemiological data. 

The draft assessment summarizes several published risk assessments that
use a variety of models and statistical methodologies employed by the
various analysts to fit the featured data sets.  Are the models and
approaches used by these investigators adequately explained and
evaluated?  If not, let OSHA know what needs to be added to the
presentation.

Response

The reasons given for emphasizing analyses of the pooled data on lung
cancer (Steenland et al. 2001a) that use log-transformed cumulative
exposures are faulty.  There are several reasons for preferring analyses
based on untransformed exposures.  Using untransformed exposure with a
15 year lag gave a considerably better fit to the data than
log-transformed exposure with a 15 year lag (log-likelihood of 21.4
versus 18.8—what is referred to as the likelihood in Steenland et al.
(2001a) is presumably the log-likelihood).  The fact that the model
based on log-transformed exposure did not show significant improvement
in fit when the interaction terms were added (i.e., did not show
significant heterogeneity) should be viewed as more of a disadvantage of
this model in comparison to the one based on untransformed exposures
rather than an advantage: The model based on untransformed exposures not
only fit better that the model based on untransformed exposures, but
adding interaction terms significantly improved the fit of the already
better-fitting model, but did not significantly improve the fit of the
poorer-fitting model.  

Figure 1 provides a simple illustration of this point.  This figure
depicts data from two studies, Study 1 (blue points) and Study 2 (red
points).  There is an apparent study effect as exposure shows a stronger
effect in Study 1 than in Study 2.  Two models are used to analyze these
data, Model 1 (orange lines) and Model 2 (green lines).  The solid lines
indicate the fits of the models to the combined data and the dashed
lines indicate the fit of the models to the data from the individual
studies.  The fit of Model 2 to the combined data (solid green line) is
clearly better (i.e., larger log-likelihood) than the fit of Model 1
(solid orange line).  However, there is a significant improvement in fit
when Model 2 is fit separately to the two studies (dotted green lines). 
Thus, this model detects significant heterogeneity.  However, with Model
1 the improvement in fit is not significant; hence there is no
significant heterogeneity identified by this model.  By the decision
logic described above, Model 2 would be ruled out in favor of Model 1
even though Model 2 is clearly superior.  This example has the same
characteristics as the fits to the lung cancer data in the pooled data,
with Model 2 playing the role of the model using untransformed exposure
and Model 1 playing the role of the model using log-transformed
exposure.      

There are several procedures for choosing the best model based on the
value of the likelihood (e.g., AIC (Akaike information criterion, Akaike
1974)).  Each of these procedures would identify the model based on
untransformed exposures as better than the one based on log-transformed
exposures.  The fact that the significant heterogeneity was found for
this model suggests that this model could be improved, but it does not
indicate that the model using log-transformed exposures is better.

Dose responses for the same effect (e.g., lung cancer from silica) would
not be expected to have fundamentally different forms in different
studies (e.g., derived from log-transformed exposure in one study and
untransformed exposure in another).  It would assist in comparing
results across studies to present risk estimates from the three studies
in Table VI-2 using a common exposure metric.  Similarly, it would
assist in comparing the effect of different exposure metrics, by
computing risk from single study using different exposure metrics.  Both
of these goals could be accomplished by adding to Table VI-2 risk
estimates derived from the analysis of the pooled cohorts (Steenland et
al. 2001) based on a linear relative risk model.  If the model is truly
linear in the exposure variable, use of the log-transformed exposures
will tend to underestimate risks at lower exposures and overestimate
risks at higher exposures.  

More consideration needs to be given in the document to model selection
and its effect upon risk estimates.  Although log-linear models are
relied upon extensively, such models are generally selected for
convenience (since this is the model generally used in Cox regression
(Cox 1972)) rather than for biological plausibility.  Such models are
supralinear when applied to untransformed exposure (increase at a rate
faster than linear with increasing exposure).  OSHA states that it does
not believe that a supralinear response “is biologically plausible and
believes that it is more reasonable to assume that the exposure-response
relationship remains linear …”  This argues for using a linear model
(like the linear relative rate model used in Rice et al. (2001)) instead
of a log-linear model.  It also argues against using a log-linear model
with log-transformed exposure, since such models are sublinear, as
illustrated by the very slow increase in risk with increasing exposure
exhibited by the lung cancer model implemented by Toxichemica Inc.
(2004) (Table VI-12).  To more fully evaluate the effect of model
selection, it would useful to reanalyze the data from one or more of the
key studies, such as the data from the pooled cohort, using a truly
linear model, and perhaps other models, and compare risk estimates
obtained from these models.  

It is claimed that the Attfield and Costello (2004) log linear model
becomes supralinear at cumulative exposures above 4.5 mg/m3-years.  How
OSHA arrived at this conclusion is not explained.  As noted above such a
model is actually supralinear at all values of cumulative exposure.

The draft report identifies and describes several key areas of
uncertainty with regard to the estimates of risk.  Were the major
uncertainties in the risk assessment adequately characterized?  If not,
explain what should be included.

Response

The potential for selection bias and for bias due to confounding are
adequately addressed.  As explained in response to Question 2, some
aspects of uncertainty in risk due to uncertainty in exposures are
addressed but others probably are not.  In addition, the effect upon
risk of the exposure uncertainty that was quantified is possibly
underestimated due to the use of log-transformed exposures, as risk
estimates based upon log-transformed exposures tend to be relative
insensitive to changes in exposures.  As discussed in response to
Question 4, there needs to be additional evaluation of the uncertainty
due to the risk models used.     

The draft assessment relies on the use of cumulative exposure (or a
transformation of cumulative exposure) as the principal dose metric on
which lung cancer and silicosis risks are evaluated.  The draft also
contains a discussion of studies that are suggestive of a dose-rate
effect.  Given this discussion, is cumulative exposure a reasonable
exposure metric for silica-induced lung cancer and silicosis in exposed
workers?  Is there a more appropriate exposure metric?  If so, explain.

Response

First it needs to be clear about what is meant by a “dose-rate
effect”.  My interpretation of that concept is that there is a
non-linearity in the exposure-response whereby a given increase in
intensity of exposure will cause a greater than proportional increase in
risk.  An example would be if risk varies as the integral over time of
the square of the exposure level.  Thus, there are dose metrics other
than cumulative exposure that do not incorporate a dose-rate effect. 
Since OSHA is relying on cumulative exposure to estimate risk, I
interpret this question as whether cumulative exposure is an appropriate
metric, or is some other metric that incorporates a dose-rate effect
more appropriate.  

Estimating risk based on an “incomplete” exposure metric like
average exposure is not recommended.  It is highly implausible that such
a metric could apply over an extended range of exposure patterns that
are likely to exist in a study.  E.g., exposure to a particular air
concentration for one week is unlikely to carry the same risk as
exposure to that concentration for 20 years, although the average
exposures are the same.  

Finding a significant relationship between average exposure and risk in
itself is not evidence of a dose-rate effect.  In general, even if risk
is truly a function of cumulative exposure, there with likely also be a
significant relationship between risk and average exposure, since
cumulative exposure and average exposure are likely to be correlated. 
For example, if all exposures are for the same duration, average
exposure is totally confounded with cumulative exposure.  On the other
hand, demonstration of a significant relationship with cumulative
exposure does not rule out a dose-rate effect, since cumulative exposure
will possibly be correlated with a metric that incorporates such an
effect.

To demonstrate evidence for a dose-rate effect that is not captured by
cumulative exposure, it would be most convincing to show some effect of
dose rate that is in addition to the effect of cumulative exposure.  To
demonstrate such an effect one would need to model both cumulative
exposure and some effect of dose rate, and show that adding the effect
of dose rate makes a statistically significant improvement to the model
over that provided by cumulative exposure alone.  I am not aware of any
such analysis that has been conducted for lung cancer from silica or for
silicosis mortality.  However, as discussed below, two studies conducted
such analyses for silicosis morbidity.

Most of the analyses of for silicosis morbidity did not examine exposure
metrics other than cumulative exposure.  Of the eight cumulative risk
studies summarized by OSHA (Table VI-11), only Hughes et al. (1998)
specifically tested for a dose-rate effect.  This study found that,
after controlling for cumulative exposure, workers exposed to ≥ 0.5
mg/m3 crystalline silica had a significantly higher risk than workers
exposed to lower levels.  A similar dose-rate effect was found by
Buchanan et al. (2003).  Although Steenland and Brown (1995a) found that
average exposure did not describe the morbidity in their cohort nearly
as well as cumulative exposure, this is not as specific a test for a
dose-rate effect as was conducted by Hughes et al. and Buchanan et al. 
The remaining six studies used to summarize risk of silicosis morbidity
(Table VI-11) considered only cumulative exposure as an exposure metric.
 So the two studies that looked carefully for a dose-rate effect found
evidence for such an effect.  In addition, Park et al. (2002) found a
13-fold higher incidence of silicosis during 1942-52 than in subsequent
years after controlling for cumulative exposure.  Although they
attributed this possibly to a change in surveillance, it could also be
due to high exposures during this early period and a dose-rate effect.  

The OSHA report states that a dose-rate effect for silicosis is
biologically plausible.  And if such an effect is present for silicosis
morbidity, it would presumably also be present for silicosis mortality. 
The report also states that even if there is such an effect it may not
exist at concentrations in the range of the current PEL.  However, even
if this is the case, risks in the range of the current PEL are being
estimated from studies that involved higher air concentrations.  Not
accounting for a dose-rate effect, if one exists, could overestimate
risk at lower concentrations.  

Having noted that there is evidence for a dose-rate effect for
silicosis, it may be difficult to account for it quantitatively.  The
data are likely to be limited by uncertainty in exposures at earlier
times, which were likely to be higher.  Perhaps rather than identifying
an alternative dose metric to cumulative exposure, it would be better to
retain cumulative exposure but use data from lower exposures where any
dose-rate effect would be less severe.  One possible approach that might
could be used to adjust existing estimates for a potential dose-rate
effect is to use either the coefficients estimated by Hughes et al.
and/or Buchanan et al. for low exposures (e.g., equation (1) of Hughes
et al 1998 with con = 0, or equation (2) of Buchanan et al. 2005). 
Another possibility is to reanalyze data from some of the better studies
using an approach similar to the approaches used by Hughes et al. (1998)
and Buchanan et al. (2003).    

The analysis by Toxichemica, Inc. (2004) that investigated several
exposure metrics based on lung burden, although it investigated exposure
metrics different from cumulative exposure, did not address a dose-rate
effect per se.  

OSHA’s estimates of lung cancer and silicosis mortality are
represented as a range of estimates with those derived from the
diatomaceous earth and U.S. granite studies considered to be plausible
upper bound risk estimates.  OSHA also acknowledges that the risk may be
lower such as suggested by the results of the IARC multi-centric study,
which relied on pooled data from several epidemiological studies.  Is
this a sound and reasonable estimation of risk from the available data? 
Is this decision adequately explained?  Should the risk estimates based
on one data set be preferred over the other?  Is there a more
appropriate representation of the lung cancer risk from silica exposure?
 If so, explain.

Response

Other things being equal I would recommend emphasizing risks estimated
from the multi-centric study, since it incorporates data from many
studies and is less likely to be unduly influenced by exposure errors
from a single study.  However, I would not recommend emphasizing risks
from this study as currently estimated using log-transformed exposures. 
The lower risk estimated from the multi-centric study is likely to be an
artifact due to the use of log-normal transformed exposure.  As pointed
out earlier, untransformed exposure provided a better fit to these data
than log-transformed exposure.  The limited increase in risk projected
by this model to result from a 10-fold increase in exposure (0.05 mg/m3
to 0.5 mg/m3) doesn’t seem plausible.  OSHA could consider calculating
risk from the multi-center using untransformed exposures and a linear
model.   

	The cumulative risk of silicosis morbidity is preliminarily considered
to be best represented by a range (approximately 3- to 4-fold) that is
based on results from four published exposure-response assessments. 
Does this range reflect a sound and reasonable estimation of risk from
the available data?  Is this decision adequately explained?  Should the
risk estimates based on one data set be preferred over the other?  Is
there a more appropriate representation of the silicosis morbidity risk
from silica exposure?  If so, explain.

Response

As discussed above, studies that have tested for a dose-rate effect have
found evidence of such an effect.  Given this, it would be reasonable to
take such an effect into account in the risk estimates.  Some
suggestions are made above about how this might be accomplished.

OSHA estimated the risk of lung cancer mortality, silicosis mortality,
and silicosis morbidity for 45-year exposures to 0.25 and 0.5 mg/m3
respirable crystalline silica; these represent a range of exposures
consistent with OSHA’s current standards for construction and maritime
industry sectors, but are above OSHA’s current general industry
standards (approximately 0.1 mg/m3 for respirable quartz).  The
cumulative exposures corresponding to these higher exposure levels are
at or above the cumulative exposures of most of the more highly exposed
workers included in the studies that underlie the risk estimates.  Is
OSHA’s approach for extrapolating risk above the observed range
reasonable?  Is it clearly explained?  Is there a more appropriate
approach that should be considered to estimate risk at cumulative
exposures that are generally above the observed range of the underlying
epidemiological studies?

Response

OSHA evaluated this issue by comparing cumulative exposures
corresponding to exposure levels of 0.25 and 0.5 mg/m3 to the median
exposures and third quartile exposures in various cohorts.  Such
comparisons are not particularly indicative of whether risks have been
extrapolated beyond the range of the data.  The important issue is
whether there are sufficient numbers of cohort members with exposures in
the range of interest.  Irrespective of the median of third quartile
exposures, there still could be significant numbers of cohort members
with exposures in the range of interest.  For example, adding many
additional cohort members with low exposures would lower both the median
and third quartile exposure, but would not reduce the ability to
quantify risk at higher exposures.  To explore this issue further, OSHA
could evaluate the potential numbers of cohort members in the various
studies with exposures in the range of interest.  My expectation is that
the pooled analysis will contain a significant number of cohort members
with exposures in this range.  (However, the pooled analysis may
underestimate risks in this range due to use of log-transformed
exposures – see response to Question 4.)

Additional comments

Page 6

“OSHA’s current general industry PEL for quartz is the gravimetric
formula 10 mg/m3/(%quartz + 2) expressed as the concentration of
respirable dust.  For quartz concentrations exceeding 5 percent, the PEL
expressed as the concentration of respirable quartz exceeds 0.07 mg/m3
and approaches 0.1 mg/m3 as the quartz content increases.”

An example could perhaps make this clearer.

Page 12

“the log of cumulative exposure”

This should be the log of cumulative exposure plus 1.  This is an
important distinction that should be made throughout the document.  In a
log linear model that uses log-transformed exposure, the relative risk
approaches zero as exposure approaches zero, rather than approaching
1.0.  However, when 1 is added to cumulative exposure, the relative risk
approaches 1.0 as it should.

Page 15

“Steenland et al. (2001a) estimated excess lifetime risk of lung
cancer based on rate ratios from the spline and log-linear models
(15-year lagged log-cumulative exposure for both),”

“Spline” refers to how exposure is modeled.  “Log-linear refers to
how response is modeled as a function of exposure.  Therefore these
descriptions are incomplete and do not necessarily refer to separate
models.  The entire report needs to be carefully read for errors such as
this, and particularly to insure that “log-linear” and
“log-transformed exposure”  are not used incorrectly (e.g.,
interchangeably).  

A “log-linear model” based on cumulative exposure is a fundamentally
different model from a “log-linear model” based on the log-transform
of cumulative exposure.  For clarity the document should use different
terminology in referring to these different models (e.g., “cumulative
exposure log-linear” and “log-transformed cumulative exposure
log-linear”).   

Page 17

β1E + β2E2)”

This equation should read RR = EXP(β1E + β2E2).

 Page 19

Define REL the first time it is used.

Page 28

“For the model based on the pooled analysis, OSHA used the exposure
coefficient (β=0.060) that reflected corrections made by Steenland and
Bartell to the original data (see Table VI-1) and assumed 250 working
days per year.”

This section should state the exposure metric in every case.  E.g., the
coefficient of 0.060 is in what units and for which model?

Page 31-32

“It is also possible that the lung cancer response was diminished in
this group due to an exposure-related increase in competing causes of
death; in particular, tuberculosis, pneumoconiosis, and non-malignant
respiratory disease mortality rates were all 2- or 3-fold higher than
those for the remaining exposure groups.”

Is this really a problem, since the analysis controlled for competing
causes of death?

Page 35 and 36 and other places 

Mannetje et al. (2002a) should often be Mannetje et al. (2002b).

Pages 40-41

Another difference between methods of risk estimation by Mannetje et al.
(2002b) and Park et al. (2002) is that Park et al. apparently adjusted
for competing risks of death whereas Mannetje et al. did not.  This
would tend to make the disparity even greater.

Page 41

“study represent the most conservative estimates”

“Conservative” needs clarification.

Page 48

“Assuming that the errors resulting from the assignment of
job-specific mean exposures to individuals and the exposure metric
conversions affect diseased and nondiseased study participants equally
results in a type of error known as a Berkson error”

This definition is incorrect.  Berkson error has to do with the exposure
variable and is not defined in terms of the response.  

Page 56

“Part of the problem appears to have arisen from an erroneous footnote
in the Hnizdo et al. (1997) report in which they indicate they applied
the 30-percent conversion factor to respirable dust which had been
acid-washed (for which the 54% conversion rate would have been more
appropriate), whereas in fact the investigators applied the 30-percent
factor to untreated dust.”

This is taken from the Toxichemica Inc. report where it is stated
without any reference.  A reference is needed for this claim.

Page 57

“According to Toxichemica, Inc., (2004) the use of the
log-transformation (i.e., either the log-linear model with log
cumulative exposure or the spline model using log cumulative exposure)
ensures that any constant multiplicative bias in exposure has virtually
no effect on conditional logistic regression coefficients, and would
have no effect at all using a pure log-transformation (i.e., without
adding 1 to allow for exposures at 0 mg/m3-days).”

This is true, which argues against the biological reasonableness of such
a model.

Page 89

“Miller et al. (1998) study reflect the risk that nodular profusion
will progress such that a chest x-ray would be classified as ILO major
category 2 or higher.  Such estimates would understate the risk of
developing lesser degrees of profusion (i.e., ILO 1/0 or 1/1).”

It could also be noted that some of subjects in this study were still
fairly young (<= age 35) when the last evaluation was conducted, and it
seems likely that additional ones of these will develop opacities over
time.

Page 94  

“In addition, the results of this study are made uncertain by a lack
of follow up; for the cohort overall, the latest chest film was taken an
average of 11.5 years after hire, a period too short to ensure that all,
or nearly all, cases of silicosis that will develop could be
detected.”

But I don’t see how this detracts from their findings.  Their finding
was statistically significant with the amount of followup present in
their study.  Is OSHA suggesting that the effect would go away with more
followup?

Page 94

“several-fold above the current OSHA PEL (i.e., above …”

Change to “several-fold above the current OSHA PEL (i.e., several-fold
above …”.

~Page 95 

“freshly fractured silica”

Does freshly fractured silica contain more particles per mass?  Could
any different toxicity be due to this factor?  

Page 108

“The use of the log-transformed exposure metric in the pooled study
was necessary to reduce significant heterogeneity between the 10 worker
cohorts that was evident with a model based on untransformed cumulative
exposure.”

As pointed out in answers to questions, this is a faulty argument.

Page 108

“or a substantial increase in mortality from competing causes.”

The study controlled for mortality from competing causes, so this should
not be a factor.

Page 114

“OSHA believes that risk estimates based on the Miller et al. (1998)
study are the most reliable overall because of the availability of
high-quality exposure data, but they may understate the risk of
developing early signs of silicosis (i.e., corresponding to ILO x-ray
film classifications of 1/0 and 1/1).”

Also, this study may underestimate risk because it included a sizable
fraction of younger men who are at risk of developing silicosis in the
future.

Page 128, Footnotes to table 12

“Estimated by OSHA based on the Park et al. (2002) linear relative
rate model, RR = 0.5469*E where E is cumulative respirable crystalline
silica exposure in mg/m3-years.”

The model is stated incorrectly.  It should be RR = 0.5469*(1+E).  I
don’t know if this was a typographical error or an error in the
calculation.  If the latter, risk will be underestimated.  More
generally, these footnotes should be reviewed carefully and expanded, if
necessary, to make it clear exactly how these calculations were
conducted.

References not used in the document under review 

Akaike, H. (1974). A new look at the statistical model identification.
Automatic Control, IEEE Transactions on, 19(6), 716-723.  

Carroll, R. J., Ruppert, D., & Stefanski, L. A. (2006). Measurement
Error in Nonlinear Models (p. 455). Boca Raton: Chapman & Hall/CRC.  

Cox, D. R. (1972). Regression models and life tables. Journal of the
Royal Statistical Society, Series B (methodological), 34(2), 187-220.  

Figure 1

Illustration of why rejecting a model that shows significant
heterogeneity in favor one that doesn’t is not a good decision rule.  

This figure depicts data from two studies, Study 1 (blue points) and
Study 2 (red points).  Exposure to a toxicant is measured on the
horizontal axis and the corresponding biological response is plotted on
the vertical axis.  In both studies, increased exposure is associated
with a decrease in the biological effect being measured.  There is an
apparent study effect as exposure causes a stronger decrease in response
in Study 1 than in Study 2.  Two models are used to analyze these data:
Model 1 (orange lines) and Model 2 (green lines).  The solid lines
indicate the fits of the models to the combined data and the dashed
lines indicate the fit of the models to the data from the individual
studies.  The fit of Model 2 to the combined data (solid green line) is
clearly better (i.e., larger log-likelihood) than the fit of Model 1
(solid orange line).  However, there is a significant improvement in fit
when Model 2 is fit separately to the two studies (dotted green lines). 
Thus, this model detects significant heterogeneity.  However, with Model
1 the improvement in fit is not significant (in fact shows practically
no improvement); hence there is no significant heterogeneity detected by
this model.  By the decision logic used in the analysis of the lung
cancer data from the pooled analysis to select the model based on
log-transformed exposure over the model based on untransformed exposure,
Model 2 would be ruled out in favor of Model 1 even though Model 2 is
clearly superior.    



Murray Finkelstein

McMaster University, Thornhill, ON Canada

  SEQ CHAPTER \h \r 1 Peer Review of OSHA’s Preliminary Quantitative
Risk Assessment for Silica

	The draft report evaluates exposure-response data from several
occupational studies as to their suitability for quantitative assessment
of lung cancer and silicosis risk resulting from airborne exposures to
silica.  Are the strengths and limitations of the selected data sets
adequately discussed and appropriately evaluated?  Are you aware of
other study data that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica?  If yes, inform
OSHA of the other data.

I think that the strengths and limitations of the selected data sets
were adequately discussed and appropriately evaluated. I am not aware of
other study data that are better suited for quantifying risk.

	 The draft assessment relies on quantitative uncertainty analysis and
sensitivity analysis of exposure measurement error in key studies of
lung cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach?  If so, explain.

The use of uncertainty and sensitivity analysis by the Toxichemica team
is well explained. I think that this is an excellent approach for
evaluating uncertainty.

	The draft assessment evaluates the use of a physiologically-based
pharmacokinetic model that predicts silica lung burdens in individuals
exposed to particulates through inhalation for estimating risk from lung
cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach? If so, explain.

The analyses are clearly explained. I think that this would be a
reasonable approach for evaluating silica-induced risk only in the
absence of human epidemiologic data. I think that the application of
dust doses and inflammatory effects in rats to the prediction of
mortality in humans is too great a leap. I think that the appropriate
approach is risk assessment based upon human epidemiologic data.

	The draft assessment summarizes several published risk assessments that
use a variety of models and statistical methodologies employed by the
various analysts to fit the featured data sets.  Are the models and
approaches used by these investigators adequately explained and
evaluated?  If not, let OSHA know what needs to be added to the
presentation.

These models are adequately explained and evaluated.

	The draft report identifies and describes several key areas of
uncertainty with regard to the estimates of risk.  Were the major
uncertainties in the risk assessment adequately characterized?  If not,
explain what should be included.

I think that the major uncertainties were adequately characterized.

	The draft assessment relies on the use of cumulative exposure (or a
transformation of cumulative exposure) as the principal dose metric on
which lung cancer and silicosis risks are evaluated.  The draft also
contains a discussion of studies that are suggestive of a dose-rate
effect.  Given this discussion, is cumulative exposure a reasonable
exposure metric for silica-induced lung cancer and silicosis in exposed
workers?  Is there a more appropriate exposure metric?  If so, explain.

I am skeptical of the observation of a dose-rate effect because I think
that exposure assessment and assignment to individual workers is too
uncertain. I think that cumulative exposure is reasonable for the risk
assessment of chronic diseases such as silicosis and lung cancer.

	OSHA’s estimates of lung cancer and silicosis mortality are
represented as a range of estimates with those derived from the
diatomaceous earth and U.S. granite studies considered to be plausible
upper bound risk estimates.  OSHA also acknowledges that the risk may be
lower such as suggested by the results of the IARC multi-centric study,
which relied on pooled data from several epidemiological studies.  Is
this a sound and reasonable estimation of risk from the available data? 
Is this decision adequately explained?  Should the risk estimates based
on one data set be preferred over the other?  Is there a more
appropriate representation of the lung cancer risk from silica exposure?
 If so, explain.

I believe that the best estimate is that derived from the IARC
multi-centric study because it averages over study uncertainties. In so
doing, it is possible that the heterogeneity and uncertainties blunt the
dose response relationship. Nevertheless, I think that individual study
uncertainties are sufficiently large that pooling is the best approach.
I am particularly impressed by the sensitivity analysis conducted by
Steenland and colleague. I have examined and worked with the raw
exposure data from the US granite study. I think that these data are
much too weak to be used on a stand-alone basis.

	The cumulative risk of silicosis morbidity is preliminarily considered
to be best represented by a range (approximately 3- to 4-fold) that is
based on results from four published exposure-response assessments. 
Does this range reflect a sound and reasonable estimation of risk from
the available data?  Is this decision adequately explained?  Should the
risk estimates based on one data set be preferred over the other?  Is
there a more appropriate representation of the silicosis morbidity risk
from silica exposure?  If so, explain.

I agree with OSHA that the relevant cohorts are those that include
extended follow-up, including the experience of retirees. I agree that
it is important to acknowledge a sizeable range in the risk estimates,
due primarily to sampling variability, length of follow-up, and exposure
uncertainties. I think that the range presented reflects a sound and
reasonable estimation of risk from the available data. I think that the
decision is adequately explained. Because of uncertainties, I favor the
use of as many studies as are relevant, rather than relying on single
studies. 

	OSHA estimated the risk of lung cancer mortality, silicosis mortality,
and silicosis morbidity for 45-year exposures to 0.25 and 0.5 mg/m3
respirable crystalline silica; these represent a range of exposures
consistent with OSHA’s current standards for construction and maritime
industry sectors, but are above OSHA’s current general industry
standards (approximately 0.1 mg/m3 for respirable quartz).  The
cumulative exposures corresponding to these higher exposure levels are
at or above the cumulative exposures of most of the more highly exposed
workers included in the studies that underlie the risk estimates.  Is
OSHA’s approach for extrapolating risk above the observed range
reasonable?  Is it clearly explained?  Is there a more appropriate
approach that should be considered to estimate risk at cumulative
exposures that are generally above the observed range of the underlying
epidemiological studies?

Extrapolation beyond the observed range depends on the adoption of a
risk model. The selection of the linear or log-linear models is clearly
explained. The approach adopted by OSHA is reasonable. An alternative
approach is simply to say that the risk is greater than X, where X is
the risk observed at the maximum exposure category in the underlying
epidemiologic studies.



Gary Ginsberg

Connecticut Department of Public Health, Hartford, CTResponse to Charge
Questions

OSHA’s Preliminary Quantitative Risk Assessment for Silica

ERG Task # 0193.15.064.001

Reviewer:  Gary Ginsberg

November 4th 2009

1.  The variety of exposure-response studies are generally well
described in terms of numbers of subjects, exposure characterization,
statistical analysis, potential confounders, and where applicable, dose
response modeling.  The report however is lacking in terms of providing
a critical appraisal of each study and its relative merit, power or
weighting for the purposes of risk assessment.  The risk assessment
section relies upon 4 studies which conducted their own quantitative
dose response assessment, accepting those cohorts as the basis for the
OSHA evaluation. However, there are numerous other studies which might
lend themselves to quantitative assessment by virtue of their exposure
and outcome measurement.  For example, Section V.C. describes many
studies which might be suitable for quantitative assessment (e.g.,
Guenel et al. 1989 of Danish stone cutters) but which are not critically
appraised for this purpose.  Further, the Cassidy 2007 European
multi-center analysis is promoted in Section V.C. as particularly useful
(even “compelling” – numerous positive qualities including
exposure assessment) but yet is not included in the Section VI
quantitative risk assessment.  One wonders whether any bias is
introduced into the assessment by the omission of one or more
potentially useful studies.  The lack of a critical appraisal of the
quantitative utility of such studies or the potential bias introduced by
their omission, weakens the Section VI analysis.

2.   I am not a statistician and don’t normally get involved with
quantitative uncertainty assessment.  Therefore, I cannot comment
directly on the detailed presentation of this work.  It does appear that
the major areas of measurement uncertainty have been identified and
dealt with in some manner by the extra effort to characterize these
uncertainties.   An area of concern however is the treatment of
potential confounding by smoking.  Page 44 assumes that the only issue
with smoking as an uncontrolled variable is the potential for smoking
rate to correlate with silica exposure in such manner to bias the
disease risk high.  The text goes on to assert that this is unlikely
from the available evidence and study design.  However, this rationale
excludes the possibility that when smoking is an uncontrolled variable
it will lead to elevated cancer risk in both the reference and
silica-exposed population leading to a high and variable baseline that
will tend to bias the results towards the null hypothesis and weaken
associations between silica and lung cancer.  I could not find any
explanation or substantiative evidence that such confounding by smoking
may not have occurred given the lack of smoking history and control for
smoking present in the underlying data.   The net result may be an
underestimation of silica-induced cancer risk relative to a non-smoking
population unless there is evidence that smoking potentiates
silica-induced lung cancer.    A similar potential exists for
underestimating cancer risk in diatomaceous earth workers simultaneously
exposed to asbestos and silica (page 46).  Once again, the concern
expressed in this document is the potential for additive risk across
these exposures to inflate the potency factor.  However, the rate of
mesothelioma in the asbestos exposed workers is not discussed.  If this
were a significant intercedent cause of mortality, it would curtail the
silica-induced cancer rate.   In addition, this report would be well
served to view radon and other uncontrolled exposure variables through
this lens of potential weakening of dose response relationships.      

3.  The “PBPK” lung burden model is not well explained and pretty
much just mentioned in passing as this line of evidence is not given
substantiative weight.  PBPK usually refers to a mathematical
description of chemical fate within systemic compartments that have
parameters for organ size, blood flow and clearance processes, with QC
check conducted to ensure mass balance.  I haven’t read Kuempel et al
(2001) but it sounds more like an inhalation deposition/dosimetry model
that a PBPK model.  The model description on Pages 24-27 could be
enhanced by showing the model structure, a table of parameter values for
both species (breathing rate, respiratory tract surface areas; clearance
rates in various lung compartments), how is the model affected by
breathing rate and particle size, whether clearance is a static
parameter or allowed to vary (decrease) with increased silica burden,
and how is the threshold for critical quartz lung burden (0.39 mg/g)
set.  Also, the description of the human model at the top of Page 25 is
vague – how many coal workers, are central estimates used or are all
the data used to create a statistical distribution, how was the rat
model adapted for a standard human adult?     

The lung burden model has potential utility as it defines a threshold
for lung cancer.  However, it is not explained well and this concept is
not used in the remainder of the report to assess risks from low dose
exposure. 

The description of the published risk assessments is generally adequate.
 As described above, more information regarding why certain studies were
excluded from the meta analysis or not quantitatively described would be
helpful.  The issue of threshold is not given proper attention in these
descriptions.  The sense one gets is that these studies and associated
models generally find a dose response in the range of current OSHA
standards (0.05 to 0.1 mg/m3 and higher), with a threshold possibly
existing at lower concentrations:  e.g. $0.036 mg/m3 for 45 yrs based
upon critical lung burden.   The issue of possible thresholds should be
systematically evaluated for each study – was it statistically
investigated by the study authors?  Were the data sufficient to allow
such an analysis?  Do the studies agree regarding where a threshold
might lie for lung cancer and silicosis and how would such a threshold
affect OSHA’s risk assessment? 

Also meriting greater description are the physical characteristics of
the silica in each study.  Given that features such as the manner in
which it was mined/produced/fractured, whether it has surface coatings,
its mineral content (quartz, cristobalite, tridymite) and its age all
appear to influence its reactivity, this should be described for each
industry studied.    

The 10 cohort study reported by Steenland et al. (2001a) shows general
agreement amongst 9 of the studies but one study (South African gold
miners) showed approximately 10 fold higher potency for lung cancer
(Page 15 and Table VI-1).  OSHA reports that Steenland reports that
exclusion of that dataset was not influential in the overall
meta-analysis (likely because it contained only 77 cases out of a pooled
analysis of over a thousand).  However, that is not the end of the
story.  In toxicology and risk assessment, one should consider whether
the most sensitive study and endpoint is of high enough quality to be
the source of risk calculations or whether the entire risk range should
be presented (e.g. the gold miners representing one end of a reported
potency range) or whether some central tendency potency estimate should
be used.  It may be that the South African gold miner study is more
potent for a good reason (e.g., least confounded by smoking, most potent
form of silica) and should be given serious consideration as a stand
alone data point for potency estimation and risk calculations.  Or
perhaps that study is no better or even inferior or less relevant than
the others.  The combination of Sections V and VI provide no analysis of
this but rather just accepts the Steenland treatment of the data. 

OSHA focuses on cancer potency in the range of current standards and
also on higher exposures (pp 28-32).  The rationale for this is unclear.
 Is OSHA contemplating raising the standard to make it less strict?  Why
not focus as well on the dose response at lower levels of exposure just
in case there might be a desire to lower the standards (more strict). 
It strikes me that the lower end of the dose response curve may be more
interesting given the potential for thresholds (e.g., the lung burden
model; potential that silicosis is needed for lung cancer to occur) and
the potential that lowering the standard might be desirable given the
high cancer risks associated with the current standards.

As described above, I believe the major uncertainties in the estimate of
risk not adequately dealt with are the potential for confounding by
smoking to decrease the strength of associations, and the uncertainty
regarding the existence of thresholds for cancer and non-cancer
endpoints.  

The studies suggesting that silica risk is driven by exposure rate more
so than cumulative exposure are not particularly numerous or compelling
and I agree with the current analysis.  Further, the cumulative exposure
metric ties into the critical lung burden I mention above.  However, one
cannot rule out the potential for dose-rate to have an influence on
disease rate and so it is appropriate to treat this as an uncertainty. 
In this regard, more could be done to describe the possible
ramifications of a dose-rate model on silica risk assessment and more
mechanistic information could be brought forth regarding why dose-rate
should (or should not) be considered relevant.  For example, what
non-linearities in the biological response might exist that would be
sensitive to dose rate (e.g., threshold phenomena such as exceedance of
defense mechanisms or damage requirement needed for excitation of
immune/inflammatory processes).  

Assuming cumulative exposure is the best metric for silica risk
assessment, it brings up an interesting assessment/intervention
question.  It raises the possibility that the OSHA standard should be
based upon worker cumulative exposure (in mg/m3-yrs) rather than a set
number (e.g. 0.05 ug/m3).  For example, it may be determined that to
reach a particular cancer risk target (e.g., 1 in a thousand), the
cumulative lifetime exposure should be no greater than xx mg/m3-years. 
Some workers may get this faster (higher exposure, shorter period) while
for others it may take longer.  Akin to radiological worker protection,
it may be possible to have a running tally of cumulative silica exposure
and to limit further exposure once the “acceptable risk” level has
been superceded.  This approach would seem compatible with the
cumulative exposure assessments in this document and allow the “bright
line” to be a particular risk target (e.g., 1 in a thousand).  From
the current document, it is unclear what OSHA’s target or acceptable
risk level is.  While the cumulative exposure approach would require
more record keeping, affected industries might find it attractive in
that it might allow short –term occupations to have higher exposure
levels while limiting the cumulative exposure for any given worker.   

7.  OSHA’s risk range is based upon the limited number of studies
which have received quantitative treatment.  As discussed above, there
may be others which would be suitable for quantitative assessment if
they were formally considered for this application, but this document
does not do so.  For example, as stated above, one study in the
Steenland et al 2001 analysis, South African gold miners, yielded a
potency that was approximately 10 times greater than the remainder of
the studies.  Perhaps that is an appropriate upper bound for lung cancer
risk.  Without further discussion of the merits of that study relative
to the others, one cannot determine how much weight to put on it.  

Further, the assumption is made that this risk range applies equally to
quartz and other polymorphs such as cristobalite, and with respect to
non-cancer endpoints, that it refers specifically to silicosis.  In
lumping quartz and cristobalite potency information, OSHA should provide
better support that this is appropriate.  The quartz/cristobalite
distinction is brought up in passing on Page 41 and more detailed
explanations are supposedly provided in Sections VI-B and VI-E.  I
couldn’t find this discussion in VI-B and in a later section of VI
titled “Physical Factors that may Influence ….. (page 97-98, I guess
VI-E but unclear)” two paragraphs are given to this distinction, with
rather unconvincing result.  For example, the statement is made that a
difference between these polymorphs has not been seen in epidemiologic
studies but only a limited dataset is being referred to here, and it
evidently does not include the Park et al. 2002 data which showed 8-9
times higher silica-related mortality in diatomaceous earth workers
(cristobalite) than in the multi-center study.  Clearly some of this
difference is due to the broader case definition in Park et al. 2002,
but that doesn’t seem to account for all of the difference as a large
number (percentage undefined) of the LDOC deaths were likely silicosis. 


A review by Bolsaitis and Wallace (1996) is cited as supporting no
difference in terms of hemolysis or fibrosis between cristobalite and
quartz but from the abstract that review appears to focus mostly on
surface properties rather than polymorph.  The OSHA document would be
benefited from further documentation along these lines especially since
there is at least some epidemiology of increased risk in association
with cristobalite.   

OSHA may want to evaluate whether the LDOC mortality (lung disease other
than cancer, as defined by Park et al., 2002) is the more appropriate
endpoint for silica-induced non-cancer mortality risk.  Right now the
focus is on silicosis, which as stated in the document, can be an
underreported condition, sometimes called emphysema or other lung
conditions as a cause of death.  Further, as documented in Section V,
silica can induce a wide variety of lung conditions in addition to, and
without the necessary occurrence of, silicosis.  Therefore, an additive
or more inclusive approach to silica-induced non-cancer risk assessment
(as suggested by the Park et al. 2002 approach) may be more appropriate.
 It is useful that Table VI-12 pulls that data point out separately but
its unclear how much weight will be given to this potency value.  

8.  OSHA appropriately narrowed the field of potential morbidity studies
to those which employed retirement followup and did a good job of
describing the pros and cons of each study.  Given the wide disparity in
exposures, exposure measures, and diagnostic procedures, it is
surprising that the potency estimates fall into a fairly narrow range. 
OSHA’s characterization of this range appears to be appropriate. 
However, the statement on Page 82 regarding autopsy results in the 1993
study is troubling, suggesting a high rate of false negatives.  This
implies that autopsy results are more sensitive and that there may be
some systemic under-reporting of silicosis.  While the uncertainty in
this study regarding the possible underestimation of exposure is
mentioned in summary on Page 89, the possibility of underestimation of
disease is not.  The uncertainty created by this possible source of
underestimation should be further described. 

9.  The issue of extrapolation to cumulative exposures higher than the
majority of the empirical data is a concrete issue.  The lung cancer
high exposure analysis (pp 28-32) seems to be contradictory in that OSHA
argues that a linear relationship with cumulative dose is likely (page
29) but then prefers the estimates derived from the log cumulative
exposure model which flattens out the dose response at high dose.  OSHA
argues that other forms of mortality or healthy worker effect may be
behind this flattening.  Given that the US granite worker data show an
apparent flattening at cumulative exposures below the range of concern
in this question (Attfield and Costello 2004 study described on page
20), it would appear that a model that reflects a response plateau may
be appropriate.  OSHA focused on the high dose in Attfield and Costello
2004 as a possible outlier.  I agree with this interpretation but
believe that a more appropriate focus would be on the entire
dose-response curve in this study and the others.  Is there evidence
from the other studies of flattening as well?  Are there cases of a
superlinear response at the high end of cumulative dose?  It would
appear that OSHA could make more complete use of the underlying data to
support model choice.    



Brian Miller

IOM Consulting Ltd., Scotland, UK

Response to Technical Charge to External Peer Reviewers

ERG Contract No. GS-10F-0125P

BPA No. DOLQ059622303

ERG Task No. 0193.15.064.001

Review of OSHA’s Preliminary Quantitative Risk Assessment for Silica

Charge Questions

The draft report evaluates exposure-response data from several
occupational studies as to their suitability for quantitative assessment
of lung cancer and silicosis risk resulting from airborne exposures to
silica.  Are the strengths and limitations of the selected data sets
adequately discussed and appropriately evaluated?  Are you aware of
other study data that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica?  If yes, inform
OSHA of the other data.

As far as I am aware, the studies are in the main well described and
evaluated.  It’s worth noting that the choice of radiological
opacities grade 2/1+ for analysis in the Buchanan et al. (2003) paper
was driven by the use of that level being required for compensation in
the UK systems.  The radiographs from the follow-up study were all
classified on the full ILO scale just as the earlier PFR ones had been,
and there are fuller descriptions of the studies and descriptions and
analyses of the distributions at lower levels e.g. 1/0+ in earlier
publications.  The 1998 paper gives logistic regression analyses only
for 2/1+, but the 1995 research report (available to download) also has
detailed analyses of a 1/0+ response.  

Miller BG, Kinnear AG.  (1988).  Pneumoconiosis in coalminers and
exposure to dust of variable quartz content.  Edinburgh: Institute of
Occupational Medicine.  (IOM Report TM/88/17).
http://iom-world.org/pubs/IOM_TM8817.pdf

Miller BG, Hagen S, Love RG, Cowie HA, Kidd MW, Lorenzo S, Tielemans
ELJP, Robertson A, Soutar CA.  (1995).  A follow-up study of miners
exposed to unusual concentrations of quartz.  Edinburgh: Institute of
Occupational Medicine.  (IOM Report TM/95/03).               

  HYPERLINK "http://iom-world.org/pubs/IOM_TM9503.pdf" 
http://iom-world.org/pubs/IOM_TM9503.pdf 

Miller BG, Hagen S, Love RG, Soutar CA, Cowie HA, Kidd MW, Robertson A. 
(1998).  Risks of silicosis in coalworkers exposed to unusual
concentrations of respirable quartz.  Occupational and Environmental
Medicine;  55: 52-58.  

There are some new data on the risk of lung cancer with respect to
respirable silica exposure, from an extended follow-up of the British
PFR mortality study:  

Miller BG, MacCalman L. (2009). Cause-specific mortality in British coal
workers and exposure to respirable dust and quartz. Occup. Environ. Med.
(Published Online First: 9 October 2009. doi:10.1136/oem.2009.046151.  

The above paper is summarised from:

Miller BG, MacCalman L, Hutchison PA. (2007). Mortality over an extended
follow-up period in coal workers exposed to respirable dust and quartz.
Edinburgh: Institute of Occupational Medicine.  (IOM Report TM/07/06).  
 HYPERLINK "http://www.iom-world.org/pubs/IOM_TM0706rev.pdf" 
http://www.iom-world.org/pubs/IOM_TM0706rev.pdf 

 

 The draft assessment relies on quantitative uncertainty analysis and
sensitivity analysis of exposure measurement error in key studies of
lung cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach?  If so, explain.

Where there is error in exposure assessment, it is important to assess
the extent to which this may affect the estimation of the
exposure-response coefficient, and to correct for that if necessary and
possible.  Other sources of uncertainty may also be important. The
approach used is a reasonable one for assessing, describing and allowing
for such sources of uncertainty within the approach taken for
quantifying the expected or average risk, which needs to be discussed
separately (see below). 

The draft assessment evaluates the use of a physiologically-based
pharmacokinetic model that predicts silica lung burdens in individuals
exposed to particulates through inhalation for estimating risk from lung
cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach? If so, explain.

Where no epidemiology exists, it may be that the only approach that can
be used for predicting risk is extrapolation from an animal model.  In
the present context, much use is made of the concept of lung burden. 
This may be the correct target organ dose metric, but epidemiological
data on it is very sparse; hence the normal focus on exposure
experienced as a surrogate for dose.  The extrapolation involved
requires many more assumptions than assessing risk from epidemiological
studies, and it may not be prudent to place the same degree of reliance
on the model-based estimates as on those from epidemiological
observation.  Having said that, one advantage of the animal model is
that the exposure is usually rather better characterized than in humans.

The draft assessment summarizes several published risk assessments that
use a variety of models and statistical methodologies employed by the
various analysts to fit the featured data sets.  Are the models and
approaches used by these investigators adequately explained and
evaluated?  If not, let OSHA know what needs to be added to the
presentation.

The studies from which the risk coefficients are derived are based on
outcomes (mortality, diagnosis of silicosis) at different lengths of
follow-up from the exposures.  In a situation where the effect of an
exposure on the progression of risk may be time-dependent (see 6 below),
these coefficients may not be comparable.  There should be more
discussion of the possible effects of this disparity. 

The draft report identifies and describes several key areas of
uncertainty with regard to the estimates of risk.  Were the major
uncertainties in the risk assessment adequately characterized?  If not,
explain what should be included.

I have no other areas of uncertainty to suggest.  (We have quite enough
to deal with already.)

The draft assessment relies on the use of cumulative exposure (or a
transformation of cumulative exposure) as the principal dose metric on
which lung cancer and silicosis risks are evaluated.  The draft also
contains a discussion of studies that are suggestive of a dose-rate
effect.  Given this discussion, is cumulative exposure a reasonable
exposure metric for silica-induced lung cancer and silicosis in exposed
workers?  Is there a more appropriate exposure metric?  If so, explain.

Cumulative exposure has been widely used in studies of occupational
health risks, and it has been shown theoretically that it is an
appropriate metric in circumstances where the exposure acts over a short
period of time after it is experienced.   This is true of e.g. ionizing
radiation exposures, and has been found to be a reasonable fit for
exposure to respirable coal dust (in which the whole dust rather than
the silica component has usually appeared to drive the risk). 
Additionally, in most coalworker epidemiology, to a good approximation,
pneumoconiosis does not progress after exposure ceases.  

With silica exposure the responses are rather different.  High exposures
can precipitate severe or accelerated reactions, and silicosis, once
well established, is likely to progress even without further exposure. 
The data of Miller et al (1998), in the reanalysis of Buchanan et al
(2003), showed higher silicosis risks, mass for mass, from exposures
derived from higher concentrations.  This might be described as a
dose-rate effect, although that would not specifically address the
mechanism by which it derives, which is likely to be due to the
overloading of the human body’s defence and clearance mechanisms. 

Thus there is good reason to believe that cumulative exposure, with no
allowance for the rate at which it accumulated, and no consideration of
persistence of its effects over time, may not be the most appropriate
metric, and that a metric that allowed for these aspects would be
preferred.  However, to be useful for risk assessment, such a metric
would require a more complex formula for its calculation, and would need
quantitative estimates of the coefficients for that formula.  Such
estimation would need very good exposure data.  Buchanan et al (2003)
had some success in this area, notably in establishing the relative
increase in risk from exposures at higher concentrations, but were
unable to obtain a reliable coefficient for the effect of persistence.
This was assumed to be because the high silica exposures were
accumulated over a relatively short time period, so the follow-up time
varied little across the study group, and the power to estimate an
effect of elapsed time was accordingly very low.  It may be that other
studies with a wider range of follow-up lengths might be informative on
this aspect through re-analysis. 

OSHA’s estimates of lung cancer and silicosis mortality are
represented as a range of estimates with those derived from the
diatomaceous earth and U.S. granite studies considered to be plausible
upper bound risk estimates.  OSHA also acknowledges that the risk may be
lower such as suggested by the results of the IARC multi-centric study,
which relied on pooled data from several epidemiological studies.  Is
this a sound and reasonable estimation of risk from the available data? 
Is this decision adequately explained?  Should the risk estimates based
on one data set be preferred over the other?  Is there a more
appropriate representation of the lung cancer risk from silica exposure?
 If so, explain.

The calculation and presentation of risk estimates needs further
elucidation and discussion.

The section on risk assessment for lung cancer mortality displays
several different estimates of what it calls ‘lifetime’ risks of 45
years’ exposure to silica at chosen average concentrations, starting
from age 20.  These risks are based on predicting numbers of deaths,
allowing for intercurrent mortality from other causes, but it is not
clear exactly how this is done.  The commissioned report summarizing
these estimates (ToxaChemica, 2004) does not spell this out, merely
referring to a paper from 1974; the paper references Chiang, a standard
reference on life table methods.  Different estimates for mortality and
morbidity are quoted that predict over different periods, variously up
to and truncated at ages 65, 75 and 85.  None of these can claim to be
‘lifetime’ risks, although that to age 85 comes closest.  Certainly,
if silica exposure has an effect on lung cancer risk, it would be
expected to produce excess cases from the rising background risks way
beyond age 65, particularly since the data suggest no reduction of
excess risk on cessation of exposure.  So summation to 65 must produce a
considerable underestimate of the total impact.  To tabulate these
together as if they are comparable, as in Table VI-9, seems unwise
unless accompanied by carefully-worded warnings.

How large is the underestimation of lifetime risk depends on exactly how
the calculation was done.  If I interpret it correctly, the text
suggests that the risk is for 45 years’ exposure starting at age 20,
so follow-up from the end of exposure to age 65 has zero duration, and
therefore should have zero risk.  However, non-zero risks to age 65 are
shown in e.g. Table VI-9, suggesting that the life-table calculation is
done differently.  It is possible that the calculation proceeds stepwise
through the years, introducing excess risk as exposure cumulates, in
which case excess cases could be predicted to occur from 15 years after
exposure began (to allow for latency), but then those that die before
age 65 do not experience 40 years’ exposure.  

The process of predicting excess cases using life-table methods is a
purely arithmetic process, easily done with standard spreadsheets, but
the answers produced depend critically on various assumptions. Whether
to allow time-dependent exposure effects, and for how long to calculate
a follow-up are among these, and need to be described carefully and
accurately.  It is not difficult to produce estimates up to advanced
ages, which might truly be considered ‘lifetime’, so it is not clear
why any estimate so labeled should be truncated at age 65.   

	The cumulative risk of silicosis morbidity is preliminarily considered
to be best represented by a range (approximately 3- to 4-fold) that is
based on results from four published exposure-response assessments. 
Does this range reflect a sound and reasonable estimation of risk from
the available data?  Is this decision adequately explained?  Should the
risk estimates based on one data set be preferred over the other?  Is
there a more appropriate representation of the silicosis morbidity risk
from silica exposure?  If so, explain.

There is ample evidence that the morbidity risks from silica vary widely
across industries and across exposures that differ in nature and in
composition.  For that reason alone, different studies are likely to
produce a range of estimates.  All other things being equal, for
purposes of regulation one might want to focus on the highest risk
estimates.  However, all other things may not be equal.  For example, it
should be clear that much of the risk of silicosis from the Miller et al
(1998) study was from short exposures at concentrations well above the
target concentrations considered here. 

Although the Miller study is highlighted as having the
best-characterised exposures, the most recently published analyses
(Buchanan et al.) were of a response at ILO category 2/1+, and therefore
fewer than if all responses at 1/0+ had been taken.  It would of course
be possible to use from these data the results of analyses with a 1/0+
response, if that were required:  see the publications listed in answer
to Q1.  Whatever response is considered, interpretation needs to bear in
mind that the radiological responses were taken at a particular point in
time, and that it is both possible and likely that the abnormalities in
at least some of the workers studied would have progressed further, even
though no longer exposed.  With a progressive disease process,
quantitative lifetime risk estimates must take into account the rate of
progression over time.

OSHA estimated the risk of lung cancer mortality, silicosis mortality,
and silicosis morbidity for 45-year exposures to 0.25 and 0.5 mg/m3
respirable crystalline silica; these represent a range of exposures
consistent with OSHA’s current standards for construction and maritime
industry sectors, but are above OSHA’s current general industry
standards (approximately 0.1 mg/m3 for respirable quartz).  The
cumulative exposures corresponding to these higher exposure levels are
at or above the cumulative exposures of most of the more highly exposed
workers included in the studies that underlie the risk estimates.  Is
OSHA’s approach for extrapolating risk above the observed range
reasonable?  Is it clearly explained?  Is there a more appropriate
approach that should be considered to estimate risk at cumulative
exposures that are generally above the observed range of the underlying
epidemiological studies?

The mechanism of the approach is clearly explained;  the appropriateness
of the methodology depends on cumulative exposure being the correct
metric, and this is discussed above.  A different metric would
presumably produce a different set of differentials between the average
concentrations.

Edits, corrections and suggestions

P12 para 2 ?Make explicit that in each set of odds ratios the first is
set arbitrarily at 1.0 and therefore has a different meaning from the
others.

P13 para 2 The non-spline models…

P16 para 1 e.g. -> i.e.

P18 para 2 It should not be a surprise that results from Cox- type
analyses and Poisson regression are similar, since the underlying model
in both is the same.  Where they may differ is in the extent of grouping
applied.  If the same grouping strategy is applied, the results will be
identical.  So often these are compared as if they were different
models, when they’re not.  

Same again on P39

P32 para 1 line10 estimate -> estimates

P78 para 1 Poission -> Poisson

P103 para 2 difference -> differences

P107 para 1 OSHA estimates the risk *up to what age?* to be…

P114 para 2 may -> must



Andrew Salmon

Private Consultant, Lafayette, CAReview of OSHA’s Preliminary
Quantitative Risk Assessment for Silica

General Comment

Compared to the more usual kind of quantitative risk assessment study,
this is a very peculiar document.  The usual risk assessment paradigm
involves hazard identification, exposure assessment, dose-response
assessment for hazards identified as being of concern, and finally a
risk characterization which seeks to identify either a slope factor for
non-threshold effects, or a health-protective level where the effect of
concern is considered to show a threshold.  These conclusions are then
combined by the risk manager with any other relevant considerations
(including measurement and feasibility) in setting suitable trigger
levels for regulatory action.  The section on health effects does a
thorough job of identifying and reviewing the information on health
hazards.  This section on the other hand starts with the previously
chosen regulatory levels, and seeks an analysis of various available
data to provide estimates of risk for key endpoints at these exposure
levels. 

 

Given this perspective, which is the inverse of the normal approach, the
document makes a very good case that substantial adverse health effects
are anticipated in workers exposed for a working lifetime to these
predefined exposure levels.  The most thoroughly developed analyses are
for lung cancer incidence and mortality from silicosis, which are severe
endpoints, and the principal concern in selection of dose-response
models appears to have been the accuracy with which the data are fitted
in the range of the target exposures of interest.  It is hard not to
conclude from a prediction of a 2% lung cancer incidence in workers
exposed at the lower PEL that these regulatory levels are insufficient
to protect workers’ health.  Yet, because of the studies, models and
endpoints selected this report contributes almost nothing to
consideration of what might be more reasonable levels of exposure that
should be considered at least as goals, even if they prove difficult or
expensive to achieve.  In particular, there is no consideration of
dose-response modeling at the lower end of the observed dose ranges, and
no consideration of the possible existence and location of a threshold
in the dose-response curve for silicosis.  There is a tendency in the
current analysis to concentrate on severe endpoints such as mortality
and severe grades of silicosis (which are observed at the exposure
levels predetermined to be of interest), rather than to evaluate milder,
but still significant endpoints which are observed in some studies at
substantially lower exposure levels.

As in the case of most other non-cancer health effects, our
understanding of the mechanism for silicosis induction by quartz
particles, although far from complete, is certainly consistent with the
possibility of a threshold in the dose response relationship, although
this threshold might be well below the levels commonly encountered in
high-risk occupational settings.  The existence of such a threshold for
silicosis is supported by the fact that, although background exposure to
crystalline silica is not zero, this disease is not usually observed in
the general population (although a few special situations producing
“environmental silicosis” are known).  In more specific terms,
Schenker et al (2009) observed histologically identifiable
pneumoconiosis in agricultural workers from the dusty California Central
Valley, but not in individuals from the same area without the heavier
exposures characteristic of employment in agriculture.  Quantitative
analyses of silicosis dose-response which did consider the lower dose
ranges (e.g. Collins et al., 2005) have generally assumed a model which
does imply a threshold, rather than the linear, logistic or log-linear
relationships used to model the response in the range considered by
OSHA.

On the other hand, a threshold for lung cancer induction is not
necessarily expected on theoretical grounds.  There might nevertheless
be such a threshold, or at least a substantial increase in the dose
response slope at higher exposures, especially if the hypothesized link
between lung cancer and silicosis (or some precursor to that condition)
were to be established. The demonstration of any such thresholds would
have important implications for defining “safe” or at least low-risk
exposures in occupational settings.  Toxichemica (2004), citing
Steenland and Deddens (2002), note that such a threshold may in fact
exist at low levels –around 10 µg/m3, although they were unable to
confirm this from a purely statistical standpoint.  It is a significant
omission both in terms of defining a health protective standard, and in
analyzing possible mechanisms of silica lung carcinogenesis, that the
OSHA document does not further examine this possibility, and the
relationship between possible thresholds for lung carcinogenesis and for
silicosis.

Charge Questions

The draft report evaluates exposure-response data from several
occupational studies as to their suitability for quantitative assessment
of lung cancer and silicosis risk resulting from airborne exposures to
silica.  Are the strengths and limitations of the selected data sets
adequately discussed and appropriately evaluated?  Are you aware of
other study data that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica?  If yes, inform
OSHA of the other data.

Response:

The studies considered are adequately described and evaluated.  In
particular, the U.S. studies and the IARC meta-analysis for lung cancer
are described in sufficient detail and carefully evaluated.  However,
the discussion of silicosis and choice of key studies is somewhat less
convincing.  The consideration of other possible datasets, particularly
the data on South African gold miners, receive somewhat more cursory
treatment.  The recent paper by Churchyard et al. (2004) describing
silicosis morbidity in black gold miners is important both for the high
level of effect seen in these workers and the use of gravimetric
exposure measures.  However, although this paper was briefly noted
elsewhere by OSHA, it is not considered in this section.  The existence
of analyses other than those performed by OSHA and its regular
collaborators is not consistently acknowledged.  This is unfortunate
since some of these analyses do to some extent address the question of
dose-response at lower levels, which is not adequately discussed in this
section (See for instance Collins et al., 2005 and OEHHA, 2005),
analyzing the data of Hnizdo and Sluis-Cremer, 1993; Steenland and
Brown, 1995; Chen et al., 2001 and Churchyard et al., 2004).  The focus
on extreme endpoints is illustrated by the extensive coverage devoted to
analysis of silicosis mortality, which is a good technical analysis but
from a public health standpoint feels like an admission of failure! 
Also, the choice of the Scottish mining study (Miller et al., 1998) as a
key study for silicosis dose-response analysis is colored by this
perspective.  This is undoubtedly an excellent study from the technical
standpoint, but its choice of a more severe cut-point (ILO 2/1) for the
radiographic diagnosis sets it apart as being less sensitive than most
of the other studies in the field, in spite of evidence discussed here
and in the section on health effects showing that even an ILO 1/1
radiographic level substantially under-diagnoses silicosis as assessed
by later follow-up or other methods e.g. histology.  There appears to be
no discussion of the complicating fact that these coal miners were also
exposed to other pneumoconiosis-inducing dusts, notably coal dust. 
Miller and his co-authors spend some time addressing this point in the
paper, but it is not referenced in OSHA’s narrative.

Evaluation of the dose-response relationships for disease endpoints
related to low-dose silica exposure will require evaluation of different
studies and dose-response models from those used in the existing
assessment, since those presented here (e.g. the study by Miller et al.,
1998 and linear or log-linear models) were selected for their
informative value at the pre-defined levels of interest rather than for
any information on dose-response at lower levels.  The appropriate
models would need to be selected and justified on the basis of the data
available in the lower exposure ranges, including the likelihood of a
threshold at least for non-cancer effects, as discussed above.  This
different approach is necessary to obtain a realistic evaluation of the
responses seen in the lower dose range.

 The draft assessment relies on quantitative uncertainty analysis and
sensitivity analysis of exposure measurement error in key studies of
lung cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach?  If so, explain.

Response:

The approach used is reasonable to address uncertainty in exposure
estimation.  Indeed, given OSHA’s recognized expertise in this area it
is unsurprising that they and their contractor (Toxichemica) have done a
thorough, indeed exhaustive, job on this aspect.  The ultimate
conclusion that there is some uncertainty, but that it does not
undermine the basic conclusions of the quantitative analysis, is
reassuring, although perhaps a little underwhelming given the amount of
effort that was applied to this topic.

The draft assessment evaluates the use of a physiologically-based
pharmacokinetic model that predicts silica lung burdens in individuals
exposed to particulates through inhalation for estimating risk from lung
cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach? If so, explain.

Response:

The description of the lung-burden analysis by Kuempel et al. (2001),
and the further analysis by Toxichemica (2004), is quite brief and
limited mainly to describing the overall conclusions of this work.  It
is not sufficiently detailed to allow a reader to determine exactly what
was done, or whether the implicit assumptions are reasonable.  What is
presented sounds reasonable and interesting, but Toxichemica’s
eventual conclusion appears to be that it does not improve the analysis
of Steenland’s pooled cancer data relative to the simpler cumulative
exposure model.  It seems relatively unlikely that the available data on
lung cancer or silicosis are of sufficient precision to directly support
use of this or any other complex dose metric.  Toxichemica (2004) also
note that the effect of the lung burden calculation is to build a lag
time into the dose-response model.  While this is a reasonable
assumption, and there is limited evidence that this is appropriate for
the silicosis data, there are a number of explanations besides
dependence on silica lung burden for this effect.

The draft assessment summarizes several published risk assessments that
use a variety of models and statistical methodologies employed by the
various analysts to fit the featured data sets.  Are the models and
approaches used by these investigators adequately explained and
evaluated?  If not, let OSHA know what needs to be added to the
presentation.

Response:

The description of the models used and statistical methods are in
general adequately described.  OSHA relies on the models used by the
published analysts although these are in most cases very simplistic,
e.g. linear with cumulative dose.  These simple models are adequate to
fit the data in the observed ranges of the specific studies considered
(indeed, many of the studies lack the precision and/or cover an
insufficient range of different exposures to support anything more
complex).  OSHA describes these models well and confirms their validity
for prediction of response at the PELs, but makes no effort to determine
whether more informative models could be developed which would be useful
in critical exposure ranges such as those below these arbitrarily
pre-defined PELs. 

The draft report identifies and describes several key areas of
uncertainty with regard to the estimates of risk.  Were the major
uncertainties in the risk assessment adequately characterized?  If not,
explain what should be included.

Response:

Given that so much thoughtful analysis was devoted to uncertainty in the
exposure measurements, it is unfortunate that the same level of
investigation was not directed to determining the uncertainty in
determination of the other half of the dose-response relationship. 
There is some useful qualitative discussion of the limitations of death
certificates and the reliability, but low sensitivity, of the ILO
radiographic method of diagnosing silicosis in the health effects
section.  However, there is no quantitative analysis of uncertainty in
determination of health endpoints beyond the usual statistical tests
reported in epidemiological studies.  The overall uncertainties in
determining health endpoints could easily be as great as, or greater
than, the uncertainties in exposure assessment, especially for silicosis
morbidity and mortality.  An initial attempt at quantifying
uncertainties in health endpoints could be obtained by reviewing
confidence limits on measures such as SMRs for specific endpoints, and
the distribution of severity or response at different dose levels. The
latter consideration might be helpful in reviewing the impact of
reliance on measures of higher severity such as mortality, higher-grade
radiographic diagnosis, or eligibility for workers’ compensation, in
some studies.   However, there is an underlying problem in that although
specific quantitative endpoints such as cancer incidences or specific
radiographic grades of silicosis are well defined, the overall disease
process is not well characterized by any established integrative
measures, so estimates of the overall disease burden are subject to
substantial qualitative as well as quantitative uncertainties.

The draft assessment relies on the use of cumulative exposure (or a
transformation of cumulative exposure) as the principal dose metric on
which lung cancer and silicosis risks are evaluated.  The draft also
contains a discussion of studies that are suggestive of a dose-rate
effect.  Given this discussion, is cumulative exposure a reasonable
exposure metric for silica-induced lung cancer and silicosis in exposed
workers?  Is there a more appropriate exposure metric?  If so, explain.

Response:

It appears that cumulative exposure is the most appropriate dose metric,
and that this is adequate to correlate with the observed response data
in all the major studies examined.  It is certainly possible that there
are dose rate effects, especially at extremely high (or low) doses, but
the studies considered do not, it seems, provide clear and unequivocal
evidence to demonstrate or quantify these effects.  Similarly, it is
possible that lung residence time is a factor in the development of
silicosis and other health endpoints, but the study which evaluated this
metric did not show that it materially improved the correlation with
health effects beyond what was obtained with the cumulative exposure
metric.

OSHA’s estimates of lung cancer and silicosis mortality are
represented as a range of estimates with those derived from the
diatomaceous earth and U.S. granite studies considered to be plausible
upper bound risk estimates.  OSHA also acknowledges that the risk may be
lower such as suggested by the results of the IARC multi-centric study,
which relied on pooled data from several epidemiological studies.  Is
this a sound and reasonable estimation of risk from the available data? 
Is this decision adequately explained?  Should the risk estimates based
on one data set be preferred over the other?  Is there a more
appropriate representation of the lung cancer risk from silica exposure?
 If so, explain.

Response:

In my opinion the presentation of the lung cancer and silicosis
mortality risk for the exposure range considered is well handled in this
section.  The IARC multi-center study probably represents a reliable
median estimate, but by its nature is likely to underestimate the upper
bound on risk in a situation where individual cohorts show real
variation in risks due to differences in the nature and intensity of the
exposure and the effectiveness in determining health endpoints.  The
description in this section describing and characterizing the range of
available estimates is therefore appropriate.

The cumulative risk of silicosis morbidity is preliminarily considered
to be best represented by a range (approximately 3- to 4-fold) that is
based on results from four published exposure-response assessments. 
Does this range reflect a sound and reasonable estimation of risk from
the available data?  Is this decision adequately explained?  Should the
risk estimates based on one data set be preferred over the other?  Is
there a more appropriate representation of the silicosis morbidity risk
from silica exposure?  If so, explain.

Response:

A range of risk estimates is presented for the specific concentrations
previously identified as OSHA PEL values.  This is obviously of interest
in the specific context of this assessment, but it does not provide an
evaluation of the overall relationship between exposure and effect,
especially for an effect such as silicosis where the dose-response
relationship is not necessarily linear at low doses.  The charge
question refers to four studies chosen for principal evaluation, but the
summary description (page 112 and page 114) refers to five studies (Chen
et al., 2001; Chen et al., 2005; Hnizdo and Sluis-Cremer, 1993; Miller
et al., 1998 and Steenland and Brown, 1995b) as useful in estimating
long-term risk of silicosis.  This dichotomy needs to be clarified as to
exactly which sources OSHA is using as the basis of its range of
estimates.  

At several points in the discussion OSHA notes concerns with the
exposure measures used in the South African studies (specifically Hnizdo
and Sluis-Cremer, 1993 in this context).  It should be noted that
although obviously there is some uncertainty in any process of
conversion of exposure measures from one basis to another these data
have in this case been thoroughly explored (Beadle and Bradley, 1970;
Page-Shipp and Harris, 1972), providing reasonable confidence in the
results.  The assertion of Gibbs and duToit (2002) that, based on the
assumed silica content of the dust, exposure was underestimated by
Hnizdo and Sluis-Cremer (1993) has been shown to be incorrect (OEHHA,
2005).  Other sources (e.g. Kielblock et al., 1997) suggest that if
anything the uncertainty is in the other direction.  The more recent
report by Churchyard et al. (2004) is also useful to compare with the
earlier South African results, since these authors used gravimetric
exposure measurements in similar situations to the earlier
particle-based data.  This data set is also worth consideration in
providing additional estimates of silicosis morbidity, although the size
of the cohort is not as large as that examined by Hnizdo and
Sluis-Cremer (1993).  The large size and inclusion of job classes with a
substantial range of different exposure levels make the study by Hnizdo
and Sluis-Cremer (1993) important for determining the overall dose
response for silicosis.  Health endpoint determination used a relatively
sensitive endpoint (ILO 1/1) and exposure measures although not by the
latest preferred method appear to have been systematic and consistent. 
This is a significant contrast to the study (Miller et al., 1998) chosen
by OSHA as the “most reliable overall”, which although of excellent
quality in all respects used a less sensitive endpoint (ILO 2/1) and
thus is bound to provide a lower estimate of the risk.  This study also
involved a cohort size of little more than half that of Hnizdo and
Sluis-Cremer (1993) and apparently a narrower range of exposure
concentrations (most of the interindividual variation in cumulative
exposure within the two groups compared appears to be based on
differences in duration rather than exposure concentration.).  Thus,
although it has considerable power to determine risk at cumulative
exposures in the middle range comparable to the OSHA PELs, it is less
informative about the overall shape of the dose-response curve.  

If OSHA wishes to examine the overall dose-response relationship for
silicosis morbidity as opposed to merely single-point estimates they
would do well to examine analyses such as that published by Collins et
al. 2005) which attempt to address the response relationship in the
lower cumulative exposure range.

OSHA estimated the risk of lung cancer mortality, silicosis mortality,
and silicosis morbidity for 45-year exposures to 0.25 and 0.5 mg/m3
respirable crystalline silica; these represent a range of exposures
consistent with OSHA’s current standards for construction and maritime
industry sectors, but are above OSHA’s current general industry
standards (approximately 0.1 mg/m3 for respirable quartz).  The
cumulative exposures corresponding to these higher exposure levels are
at or above the cumulative exposures of most of the more highly exposed
workers included in the studies that underlie the risk estimates.  Is
OSHA’s approach for extrapolating risk above the observed range
reasonable?  Is it clearly explained?  Is there a more appropriate
approach that should be considered to estimate risk at cumulative
exposures that are generally above the observed range of the underlying
epidemiological studies?

Response:

Any extrapolation outside the range of exposures for which sound
quantitative data are available is subject to uncertainty, and this is
especially true here where there are suspicions of possible dose rate
effects and/or different pathological responses (e.g. acute silicosis)
at very high doses.  OSHA does not examine these possibilities very
thoroughly, but it is unclear what could be done about them given the
nature of the data for higher exposures (typically including case
studies and early investigations with uncertain or non-existent exposure
measurements, but no systematic large-scale studies).  Overall the
approach at least for these moderately higher pre-defined exposure
levels is reasonable, although there is no effort to describe an overall
dose-response curve which would characterize the expected response at
both low and high levels.  This alternative would be useful in
considering various questions about mechanism of causation for various
health endpoints, as well as informing a health-protective
standards-setting procedure.

References cited

(other than those appearing the bibliography for the OSHA QRA document)

Beadle DG, Bradley AA. 1970. The composition of airborne dust in South
African gold mines. In: Shapiro HA (ed). Pneumoconiosis. Proceedings of
the International Conference. Johannesburg 1969. Cape Town: Oxford
University Press. pp. 462-6.

Churchyard GJ, Ehrlich R, teWaterNaude JM, Pemba L, Dekker K, Vermeijs
M, White N, Myers J (2004). Silicosis prevalence and exposure-response
relations in South African goldminers. Occup Environ Med. 61(10):811-6.

Collins JF, Salmon AG, Brown JP, Marty MA, Alexeeff GV (2005). 
Development of a chronic inhalation reference level for respirable
crystalline silica.  Regul Toxicol Pharmacol.  43:292-300. 

OEHHA, 2005. Determination of Noncancer Chronic Reference Exposure
Levels.  Chronic Toxicity Summary for Silica (crystalline, respirable). 
  HYPERLINK
"http://www.oehha.ca.gov/air/hot_spots/2008/AppendixD3_final.pdf" \l
"page=486" 
http://www.oehha.ca.gov/air/hot_spots/2008/AppendixD3_final.pdf#page=486
.

Page-Shipp RJ, Harris E. 1972. A study of dust exposure of South African
white gold miners. South Afr Inst Mining Metall. 73:10-24.

Schenker MB, Pinkerton KE, Mitchell D, Vallyathan V, Elvine-Kreis B,
Green FHY (2009).  Pneumoconiosis from Agricultural Dust Exposure among
Young California Farmworkers.  Environmental Health Perspectives 117(6):
988–994.



Noah Seixas

University of Washington, Seattle, WAPeer Review of OSHA’s
Preliminary Quantitative Risk Assessment for Silica

Review Comments: Noah Seixas, PhD, CIH

December 13, 2009

Charge Questions

The draft report evaluates exposure-response data from several
occupational studies as to their suitability for quantitative assessment
of lung cancer and silicosis risk resulting from airborne exposures to
silica.  Are the strengths and limitations of the selected data sets
adequately discussed and appropriately evaluated?  Are you aware of
other study data that are better suited for quantifying lung cancer or
silicosis risk from occupational exposure to silica?  If yes, inform
OSHA of the other data.

The selected datasets represent as comprehensive a set as available, to
my knowledge.  The inclusion of the various cohorts is appropriate,
given the data quality needs for a robust assessment.  In describing
exclusions, (p8) it is noted that studies of coal workers and foundry
workers were excluded for good reasons.  However, it also notes that 3
additional cohorts were excluded for ‘data unavailability or
incompatibility.’  These exclusions should be described in more
detail.  

The studies included in the risk assessment are very thoroughly
described and appropriately interpreted.  While limitations of each
study are described and considered, the strengths of the studies are not
described study by study.  The general strengths are evident from the
descriptions, so it may not be important to do so in any more detail.  A
brief summary of the strengths of the selected studies could help
support their use for this risk assessment.  Furthermore, published
critiques of the studies are also included in this discussion, and OSHA
has responded to these criticisms with thoughtful consideration and
clear conclusions. I am not aware of any key studies that were not
considered.  

 The draft assessment relies on quantitative uncertainty analysis and
sensitivity analysis of exposure measurement error in key studies of
lung cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach?  If so, explain.

Selection bias and confounding are addressed very well, and convincingly
suggest that these potential biases do not unduly affect the risk
estimates derived from these studies.  On p 42, it is stated that the
confidence intervals presented in Table VI-2 represent uncertainty due
to the sample of studies used for the analysis.  This statement should
be qualified by the uncertainty due to the systematic biases that also
may be present in the selected studies.  That is, the confidence
intervals represent uncertainty due to the sample of the universe that
the selected studies represent, given that the observations in those
studies are without error, themselves.

The simulation of exposure measurement error in assessing the degree of
bias that may have been present is a reasonable approach to assessing
this source of uncertainty. However, I have several comments about the
presentation of these results.  I will preface this by saying that
although I have thought about measurement error issues in occupational
exposure assessment and epidemiology for a long time, I am not an expert
in the statistical concepts, nor the application of these concepts to
assessing uncertainty.  Therefore, I raise these questions for
consideration, without claiming to know how this analysis should be
improved.

On p48, the statement that the metric conversions apply equally to
diseased and non-diseased subjects and therefore represent Berkson type
error, is incorrect.  Errors that are applied to cases and controls are
called ‘non-differential errors,’ but do not imply a Berkson error
structure. 

While the discussion of Berkson error models is generally correct, I
don’t believe the simulation has actually modeled a Berkson error, but
rather, a classical error model.  The typical Monte Carlo simulation,
which is what appears to have been done, would introduce classical
error.  I don’t think the report should repeatedly state that the
simulation produces coefficients “adjusted for Berkson error” as I
don’t believe this is Berkson, nor that the simulation actually
adjusts for it. 

I think that the introduction of random error to the assigned exposures
essentially adds error to that already present.  Thus, you are taking
the results of a model based on exposure estimates which include error,
and simulating additional error on top of it.  I’m not sure what the
results of this simulation then produces. In fact, the results produced
in Table VI-4- VI-6 almost all went down (toward the null), which is
what is generally expected by adding random (classical) error into the
exposure variable.

At the bottom of page 51, the changes in coefficients with introduction
of error are reported, however, no explanation or even speculation about
why some of these estimates were greatly reduced is provided.  Although
the associated table provides the SE associated with the coefficients
for the ‘adjusted’ effects, it should also present the SE for the
‘unadjusted’ values.

I remain unconvinced that the conclusion on page 52, that measurement
error “did not have a meaningful effect on the estimate of the
exposure-response coefficient” is well supported by this analysis.  I
would believe that a statement such as “addition of random error to
the estimated exposures in the pooled analysis did not appreciably
change the estimated exposure response relationship” would be correct.
 However, I don’t believe that the analyses conducted prove that
measurement error in the selected studies did not substantially affect
the observed relationships.  

There is no mention of measurement error generally reducing the effect
of exposure response relationships, and that this principle indicates
that the estimated risks are most likely to be underestimates, or
conservatively estimating risk. This is an important aspect of
measurement error with significant implications for risk assessment and
should not be overlooked.

The draft assessment evaluates the use of a physiologically-based
pharmacokinetic model that predicts silica lung burdens in individuals
exposed to particulates through inhalation for estimating risk from lung
cancer and silicosis.  Are these analyses clearly explained and
evaluated?  Is this a reasonable approach for evaluating silica-induced
lung cancer and silicosis in exposed workers?  Is there a more
appropriate approach? If so, explain.

It is not clear to me what the importance of this presentation is.  The
studies by Kuempel that adopted a toxicokinetic model for lung
deposition to estimate silica lung burden ‘were similarly good
predictors of lung cancer risk’ as cumulative exposure.  It appears
that these studies did not contribute to the risk analysis, other than
to note that these more complex models didn’t produce any different
results, and it is unclear to me that this is an important observation. 


The draft assessment summarizes several published risk assessments that
use a variety of models and statistical methodologies employed by the
various analysts to fit the featured data sets.  Are the models and
approaches used by these investigators adequately explained and
evaluated?  If not, let OSHA know what needs to be added to the
presentation.

The report discusses many alternative models including the form of the
linkage function, alternative exposure metrics, and different latency
periods.  These are well explained for each study, and reasonable
decisions are made as to which form of the model is most useful and
robust for the risk analysis.  

The draft report identifies and describes several key areas of
uncertainty with regard to the estimates of risk.  Were the major
uncertainties in the risk assessment adequately characterized?  If not,
explain what should be included.

Yes and no – see answer to 2, above.

The draft assessment relies on the use of cumulative exposure (or a
transformation of cumulative exposure) as the principal dose metric on
which lung cancer and silicosis risks are evaluated.  The draft also
contains a discussion of studies that are suggestive of a dose-rate
effect.  Given this discussion, is cumulative exposure a reasonable
exposure metric for silica-induced lung cancer and silicosis in exposed
workers?  Is there a more appropriate exposure metric?  If so, explain.

Cumulative exposure, or the log of cumulative exposure is, without
doubt, the correct measure of exposure for the chronic diseases under
consideration in this risk assessment.  The alternative metrics
discussed, are appropriately considered; however both by the fact that
cumulative exposure is theoretically the most appropriate measure, and
because it fits the data best in most cases, it is selected
appropriately.  More complex models that are considered including
threshold models and dose-rate effects models are interesting
explorations which may shed light on mechanisms of disease, and may in
some cases provide better fits to the data.  However, such models rarely
outperform cumulative exposure to the extent that they should be adopted
for risk assessment purposes.  Even more importantly, policies based on
such models are extremely difficult to formulate as practical exposure
limits or guidelines.  Cumulative exposure is the correct measure from
various perspectives for this assessment.

Further, I would be very skeptical of adopting dose-response estimates
from the limited data we have looking at dose rate effects.  In
particular, the existing data on dose-rate effects are limited by likely
confounding between exposure level and time – that is, all high level
effects are derived from earlier time periods.  While the papers have
done a good job of trying to separate out these two potential aspects of
risk, they are insufficiently robust to allow adoption of differential
response rates in a quantitative manner.  I recommend using the average
risk estimate associated with cumulative exposure, without regard to
exposure level, for estimation of risk.

OSHA’s estimates of lung cancer and silicosis mortality are
represented as a range of estimates with those derived from the
diatomaceous earth and U.S. granite studies considered to be plausible
upper bound risk estimates.  OSHA also acknowledges that the risk may be
lower such as suggested by the results of the IARC multi-centric study,
which relied on pooled data from several epidemiological studies.  Is
this a sound and reasonable estimation of risk from the available data? 
Is this decision adequately explained?  Should the risk estimates based
on one data set be preferred over the other?  Is there a more
appropriate representation of the lung cancer risk from silica exposure?
 If so, explain.

OSHA has presented a thorough and well reasoned discussion of risks from
three substantial exposure-response studies for lung cancer.  In
general, the estimates made are reasonable, and remarkably consistent in
their results, especially within the exposure range included in the
original studies.  It is not clear why OSHA is characterizing these
estimates as ‘upper bound risk estimates.’  Because the estimates
are based on the coefficient, and upper and lower bounds are put on each
estimate, these are ‘best estimates’ rather than upper bounds. 

I would agree with putting more emphasis on the pooled study estimate
would make sense, given the size and heterogeneity of this study. 
Although somewhat more weight may be given to it, the analysis of the
two other cohorts, showing remarkably similar risk estimates, actually
adds to the overall weight of evidence presented.  Also, note that the
pooled study only presents lower estimates as exposure increases –
thus at the low end of exposure, the estimate is almost the same as the
Granite workers, and somewhat higher than the DE study results.  

It would be useful to clarify why the OSHA analyses provide different
estimates than the original study results, and are in fact consistently
higher than the original estimates.  In what ways were the analyses
different, and why did those differences result in higher estimates?  

	The cumulative risk of silicosis morbidity is preliminarily considered
to be best represented by a range (approximately 3- to 4-fold) that is
based on results from four published exposure-response assessments. 
Does this range reflect a sound and reasonable estimation of risk from
the available data?  Is this decision adequately explained?  Should the
risk estimates based on one data set be preferred over the other?  Is
there a more appropriate representation of the silicosis morbidity risk
from silica exposure?  If so, explain.

 h>

!

"

‹

Œ

à

á

ᘆ镨긓─B

O

P

†

‡

ˆ

Š

‹

Œ

«

¬

¸

¹

Û

Ü

Ý

ß

à

á

â

$

(

1

f

耀Ȇ

B*

juL

jð™

kdª˜

 hŽ9

 hŽ9

jŒš

 h>

j°©

hó

j3©

jªª

hó

j-ª

j!¬

j¤«

hó

j'«



hó

jž¬

j ®

hó

j˜

hó

¯

hó

j’®

j	°

hó

jŒ¯

j€±

±

hó

j†°

jz²

hó

jý±

hó

j÷²

耀䀆

kd|´

Ðô

@

摫딝

@

摫떾

@

摫뙟

@

摫뜀

@

摫랡

@

摫롂

@

h½

 h

 hé

 h½

 h½

 hé

6

7

8

E

F

[

]

 h½

 h½

 h½

摧晱Ö

葞Ũ摧晱Ö

옍)

‘

÷

5

6

<

}

†

‡

옍)

옍)

ሀ桤ā㄀$摧緟`฀}

~



‡

옍)

옍)

 h

퀁萏ː萑﷤搒Ũ葞ː葠﷤摧緟`

ä

å

æ

å

æ

 hé

 h½

 h½

K

Ž

×

o

K

Ž

—

×

o

萎ﺘ搒Ũ葝ﺘ摧᭤H

搒Ũ摧᭤H

਀&䘋+搒Ũꐔð摧緟`

ꐔð摧緟`

옍)

萏ʔ萑ﵬ葞ʔ葠ﵬ摧灟&

ø

ù

ú

û

Z

[

‘

’

ø

ù

ú

[

‘

옍)

l

_o

`o

Œm

m

Õn

Ön

hr

Pt

Qt

jt

kt

\w

葞:摧緟`᠀xplained and sound.  It is of great importance that the
range of findings is presented, rather than relying on any one study or
estimate.  Identifying a study that OSHA thinks  perhaps provides the
strongest results, and the fact that that study (Miller, Coal workers)
has risk estimates near the middle of the range, provides good evidence
of the reasonableness of the presented range of risks.

As with the lung cancer risk estimates discussed below, OSHA must
consider how to address the clearly unacceptable level of risk estimated
to occur at 0.05 mg/m3.  Whether one chooses the Miller results (8 per
100) or the range presented by the other studies (2 – 13 per 100
workers), this risk level is clearly unacceptable.  

OSHA estimated the risk of lung cancer mortality, silicosis mortality,
and silicosis morbidity for 45-year exposures to 0.25 and 0.5 mg/m3
respirable crystalline silica; these represent a range of exposures
consistent with OSHA’s current standards for construction and maritime
industry sectors, but are above OSHA’s current general industry
standards (approximately 0.1 mg/m3 for respirable quartz).  The
cumulative exposures corresponding to these higher exposure levels are
at or above the cumulative exposures of most of the more highly exposed
workers included in the studies that underlie the risk estimates.  Is
OSHA’s approach for extrapolating risk above the observed range
reasonable?  Is it clearly explained?  Is there a more appropriate
approach that should be considered to estimate risk at cumulative
exposures that are generally above the observed range of the underlying
epidemiological studies?

Yes. The approach and rationale for this extrapolation is well
explained, and reasonable.  Further, it is not clear to me that the
precise estimates for risk in these higher levels of exposure are of any
consequence.  At the higher range of exposure, OSHA needs only to
demonstrate that a substantial risk exists, which is established using
any means of extrapolation chosen.  The precision of the risk estimates
at the lower end of exposure is much more important for the risk
analysis.

One additional comment.

OSHA estimates risk of deaths due to silica exposure at 0.05 mg/m3 using
best estimates, in the range of 19-29 per 1000 workers, or 2-3 deaths
per 100 workers.  This is clearly an unacceptable risk of death.  The
risk analysis on which these estimates are given is very strong given
the underlying data.  There are three large studies based on eight
cohorts, each with a substantial degree of exposure quantification, and
the results are strikingly consistent. Given the strength of the risk
analysis, OSHA should set a standard with an acceptable level of
estimated risk.  Two to three deaths per 100 workers is clearly an
unacceptable level of risk.  A lower standard, consistent with legal and
societal norms of acceptable risk should be considered.

 BPA No. DOLQ059622303, Order No. DOLU099F28863, Contract No. GS10F0125P

 Task Order No.16, Contract No. DOLJ129F32859 

 See Lutz, W. (2001).  Susceptibility differences in chemical
carcinogenesis linearize the dose–response relationship: threshold
doses can be defined only for individuals. Mutation Research/Fundamental
and Molecular Mechanisms of Mutagenesis 482: 71-76.

 All these illustrative calculations use the exposure level rather than
cumulative exposure.  If we are interested, as OSHA is, in a standard 45
year occupational exposure duration, then there is no difference between
these calculations and alternatives that used cumulative measures. 
Furthermore, we assume that the probability of silicosis in the absence
of silica exposure is zero, hence the lack of a background term in Eq.
1.

 See Figures 1b and 1c in Gollapudi, B.B., et al. (2013).  Quantitative
Approaches for Assessing Dose-Response Relationships in Genetic
Toxicology Studies.  Environmental and Molecular Mutagenesis 54:8-18.

  this problem disappears.

 Task Order No. 059622303/099P28863, Contract No. GS10F0125P, with a
period of performance from May 22, 2009 to May 22, 2010.

  PAGE   \* MERGEFORMAT  4 

  PAGE   \* MERGEFORMAT  i 

  PAGE   \* MERGEFORMAT  1 

  PAGE   \* MERGEFORMAT  3 

Bruce Allen

  PAGE   \* MERGEFORMAT  5 

Bruce Allen

Bruce Allen

  PAGE   \* MERGEFORMAT  2 

  PAGE   \* MERGEFORMAT  13 

  PAGE   \* MERGEFORMAT  15 

Kenny Crump

Kenny Crump

  PAGE   \* MERGEFORMAT  17 

  PAGE   \* MERGEFORMAT  19 

Gary Ginsberg

Gary Ginsberg

  PAGE   \* MERGEFORMAT  39 

Brian Miller

Brian Miller

Andrew Salmon

Andrew Salmon

  PAGE  1 

  PAGE  6 

  PAGE  5 

  PAGE  103 

  PAGE  104 

Bruce Allen

Bruce Allen

Murray Finkelstein

Murray Finkelstein

Brian Miller

Gary Ginsberg

Gary Ginsberg

Brian Miller

Brian Miller

Andrew Salmon

Andrew Salmon

Noah Seixas

Noah Seixas

  PAGE  218 

  PAGE  217 

Bruce Allen

Bruce Allen

Kenneth Crump

Kenneth Crump

Murray Finkelstein

Murray Finkelstein

Gary Ginsberg

Gary Ginsberg

Brian Miller

Brian Miller

Andrew Salmon

Andrew Salmon

Noah Seixas

Noah Seixas

