1
of
113
UNITED
STATES
ENVIRONMENTAL
PROTECTION
AGENCY
WASHINGTON,
D.
C.
20460
OFFICE
OF
PREVENTION,
PESTICIDES,
AND
TOXIC
SUBSTANCES
April
15,
2005
MEMORANDUM
SUBJECT:
Transmittal
of
Meeting
Minutes
of
the
FIFRA
Scientific
Advisory
Panel
Meeting
Held
February
15
­
18,
2005
on
the
N­
methyl
Carbamate
Cumulative
Risk
Assessment:
Pilot
Cumulative
Analysis
TO:
James
J.
Jones,
Director
Office
of
Pesticide
Programs
FROM:
Myrta
R.
Christian,
Designated
Federal
Official
Joseph
E.
Bailey,
Designated
Federal
Official
FIFRA
Scientific
Advisory
Panel
Office
of
Science
Coordination
and
Policy
THRU:
Larry
C.
Dorsey,
Executive
Secretary
FIFRA
Scientific
Advisory
Panel
Office
of
Science
Coordination
and
Policy
Clifford
J.
Gabriel,
Director
Office
of
Science
Coordination
and
Policy
Attached,
please
find
the
meeting
minutes
of
the
FIFRA
Scientific
Advisory
Panel
open
meeting
held
in
Arlington,
Virginia
on
February
15
­
18,
2005.
This
report
addresses
a
set
of
scientific
issues
being
considered
by
the
Environmental
Protection
Agency
pertaining
to
the
Nmethyl
carbamate
cumulative
risk
assessment:
pilot
cumulative
analysis.

Attachment
2
of
113
cc:

Susan
Hazen
Margaret
Schneider
Anne
Lindsay
Margie
Fehrenbach
Janet
Andersen
Debbie
Edwards
Steven
Bradbury
William
Diamond
Arnold
Layne
Tina
Levine
Lois
Rossi
Frank
Sanders
Richard
Keigwin
Randolph
Perfetti
William
Jordan
Douglas
Parsons
Enesta
Jones
Vanessa
Vu
(
SAB)
Anna
Lowit
David
J.
Miller
Nelson
Thurman
Dirk
Young
David
Hrdy
Jeff
Evans
Steve
Nako
Stephanie
Padilla
R.
Woodrow
Setzer
Ginger
Moser
Miles
Okino
Jerry
Blancato
Fred
Power
Curtis
Dary
Tom
Nolan,
USGS
OPP
Docket
3
of
113
FIFRA
Scientific
Advisory
Panel
Members
Stephen
M.
Roberts,
Ph.
D.
(
Chair
of
the
FIFRA
SAP)
Janice
E.
Chambers,
Ph.
D.
H.
Christopher
Frey,
Ph.
D.
Stuart
Handwerger,
M.
D.
Steven
G.
Heeringa,
Ph.
D.
Gary
Isom,
Ph.
D.
Kenneth
M.
Portier,
Ph.
D.

FQPA
Science
Review
Board
Members
John
Adgate,
Ph.
D.
William
Brimijoin,
Ph.
D.
George
B.
Corcoran,
Ph.
D.
Lutz
Edler,
Ph.
D.
Bernard
Engel,
Ph.
D.
Scott
Ferson,
Ph.
D.
Lawrence
Fischer,
Ph.
D.
Natalie
Freeman,
Ph.
D.
Gaylia
Jean
Harry,
Ph.
D.
Dale
Hattis,
Ph.
D.
James
P.
Kehrer,
Ph.
D.
Chensheng
Lu,
Ph.
D.
Peter
Macdonald,
D.
Phil.
David
MacIntosh,
Ph.
D.
Robert
Malone,
Ph.
D.
Christopher
Portier,
Ph.
D.
Nu­
May
Ruby
Reed,
Ph.
D.,
D.
A.
B.
T.
Barry
Ryan,
Ph.
D.
Michael
Sohn,
Ph.
D.
Tammo
Steenhuis,
Ph.
D.
Michael
D.
Wheeler,
Ph.
D.
4
of
113
SAP
Minutes
No.
2005­
01
A
Set
of
Scientific
Issues
Being
Considered
by
the
Environmental
Protection
Agency
Regarding:

N­
METHYL
CARBAMATE
PESTICIDE
CUMULATIVE
RISK
ASSESSMENT:
PILOT
CUMULATIVE
ANALYSIS
Session
1:
Issues
Related
to
Cumulative
Hazard
Assessment
Session
2:
Physiologically
Based
Pharmacokinetic/
Pharmacodynamic
(
PBPK/
PD)
Modeling
for
Carbaryl
Session
3:
Drinking
Water
Exposure
Analysis
Session
4:
N­
methyl
Carbamate
Exposure
Assessment:
A
Pilot
Case
Study.

FEBRUARY
15
­
18,
2005
FIFRA
Scientific
Advisory
Panel
Meeting,
held
at
the
Holiday
Inn
­
National
Airport,
Arlington,
Virginia
5
of
113
TABLE
OF
CONTENTS
N­
Methyl
Carbamate
Cumulative
Risk
Assessment:
Pilot
Cumulative
Analysis
Notice....................................................................................................................
..
6
Introduction..........................................................................................................
.7
Public
Commenters...............................................................................................
8
Session
1:
Issues
Related
to
Cumulative
Hazard
Assessment..........................
9
Session
2:
Physiologically
Based
Pharmacokinetic/
Pharmacodynamic
Modeling
for
Carbaryl......................................................................
36
Session
3:
Drinking
Water
Exposure
Analysis................................................
68
Session
4:
N­
methyl
Carbamate
Exposure
Assessment:
A
Pilot
Case
Study..........................................................................................
92
6
of
113
NOTICE
These
meeting
minutes
have
been
written
as
part
of
the
activities
of
the
Federal
Insecticide,
Fungicide,
and
Rodenticide
Act
(
FIFRA),
Scientific
Advisory
Panel
(
SAP).
The
meeting
minutes
represent
the
views
and
recommendations
of
the
FIFRA
SAP,
not
the
United
States
Environmental
Protection
Agency
(
Agency).
The
contents
of
the
meeting
minutes
do
not
represent
information
approved
or
disseminated
by
the
Agency.
The
meeting
minutes
have
not
been
reviewed
for
approval
by
the
Agency
and,
hence,
the
contents
of
these
meeting
minutes
do
not
necessarily
represent
the
views
and
policies
of
the
Agency,
nor
of
other
agencies
in
the
Executive
Branch
of
the
Federal
government.
Nor
does
mention
of
trade
names
or
commercial
products
constitute
a
recommendation
for
use.

The
FIFRA
SAP
is
a
Federal
advisory
committee
operating
in
accordance
with
the
Federal
Advisory
Committee
Act
and
established
under
the
provisions
of
FIFRA
as
amended
by
the
Food
Quality
Protection
Act
(
FQPA)
of
1996.
The
FIFRA
SAP
provides
advice,
information,
and
recommendations
to
the
Agency
Administrator
on
pesticides
and
pesticide­
related
issues
regarding
the
impact
of
regulatory
actions
on
health
and
the
environment.
The
Panel
serves
as
the
primary
scientific
peer
review
mechanism
of
the
EPA,
Office
of
Pesticide
Programs
(
OPP),
and
is
structured
to
provide
balanced
expert
assessment
of
pesticide
and
pesticide­
related
matters
facing
the
Agency.
Food
Quality
Protection
Act
Science
Review
Board
members
serve
the
FIFRA
SAP
on
an
ad
hoc
basis
to
assist
in
reviews
conducted
by
the
FIFRA
SAP.
Further
information
about
FIFRA
SAP
reports
and
activities
can
be
obtained
from
its
website
at
http://
www.
epa.
gov/
scipoly/
sap/
or
the
OPP
Docket
at
(
703)
305­
5805.
Interested
persons
are
invited
to
contact
Myrta
R.
Christian
or
Joseph
E.
Bailey,
SAP
Designated
Federal
Officials,
via
e­
mail
at
christian.
myrta@.
epa.
gov
or
bailey.
joseph@
epa.
gov.

In
preparing
the
meeting
minutes,
the
Panels
carefully
considered
all
information
provided
and
presented
by
the
Agency,
as
well
as
information
presented
by
public
commenters.
This
document
addresses
the
information
provided
and
presented
by
the
Agency
within
the
structure
of
the
charge.
7
of
113
INTRODUCTION
The
FIFRA
SAP
has
completed
its
review
of
the
set
of
scientific
issues
being
considered
by
the
Agency
pertaining
to
the
N­
methyl
carbamate
pesticide
cumulative
risk
assessment:
pilot
cumulative
analysis.
Advance
notice
of
the
meeting
was
published
in
the
Federal
Register
on
January
6,
2005.
The
review
was
conducted
in
an
open
Panel
meeting
held
in
Arlington,
Virginia,
on
February
15
­
18,
2005.
The
meeting
included
four
separate
sessions
as
follows.

 
Session
1
­
Issues
Related
to
Cumulative
Hazard
Assessment
 
Session
2
­
Physiologically
Based
Pharmacokinetic/
Pharmacodynamic
(
PBPK/
PD)
Modeling
for
Carbaryl
 
Session
3
­
Drinking
Water
Exposure
Analysis
 
Session
4
­
N­
methyl
Carbamate
Exposure
Assessment:
A
Pilot
Case
Study
Sessions
1
and
2
were
chaired
by
Dr.
Gary
Isom
and
Myrta
R.
Christian
served
as
the
Designated
Federal
Official.
Sessions
3
and
4
were
chaired
by
Dr.
Steven
Heeringa
and
Joseph
E.
Bailey
served
as
Designated
Federal
Official.
A
public
comment
period
was
held
at
the
beginning
of
the
4­
day
meeting
on
February
15,
2005.
8
of
113
PUBLIC
COMMENTERS
Oral
statements
were
presented
as
follows:

On
behalf
of
the
Natural
Resources
Defense
Council:
Jennifer
Sass,
Ph.
D.,
Natural
Resources
Defense
Council
On
behalf
of
Bayer
Crop
Science:
Abraham
Tobias,
Ph.
D.
9
of
113
SAP
Minutes
No.
2005­
01
A
Set
of
Scientific
Issues
Being
Considered
by
the
Environmental
Protection
Agency
Regarding:

N­
METHYL
CARBAMATE
PESTICIDE
CUMULATIVE
RISK
ASSESSMENT:
PILOT
CUMULATIVE
ANALYSIS
SESSION
1:
ISSUES
RELATED
TO
CUMULATIVE
HAZARD
ASSESSMENT
FEBRUARY
15,
2005
FIFRA
Scientific
Advisory
Panel
Meeting,
held
at
the
Holiday
Inn
­
National
Airport,
Arlington,
Virginia
Myrta
R.
Christian,
M.
S.
Gary
Isom,
Ph.
D.
Designated
Federal
Official
FIFRA
SAP
Session
Chair
FIFRA
Scientific
Advisory
Panel
FIFRA
Scientific
Advisory
Panel
Date:
April
15,
2005
Date:
April
15,
2005
10
of
113
Federal
Insecticide,
Fungicide,
and
Rodenticide
Act
Scientific
Advisory
Panel
Meeting
February
15,
2005
N­
METHYL
CARBAMATE
PESTICIDE
CUMULATIVE
RISK
ASSESSMENT:
PILOT
CUMULATIVE
ANALYSIS
SESSION
1:
ISSUES
RELATED
TO
CUMULATIVE
HAZARD
ASSESSMENT
PARTICIPANTS
FIFRA
SAP
Session
Chair
Gary
Isom,
Ph.
D.,
Professor
of
Toxicology,
School
of
Pharmacy
&
Pharmacal
Sciences,
Purdue
University,
West
Lafayette,
IN
Designated
Federal
Official
Myrta
R.
Christian,
M.
S.,
FIFRA
Scientific
Advisory
Panel
Staff,
Office
of
Science
Coordination
and
Policy,
EPA
FIFRA
Scientific
Advisory
Panel
Members
Stuart
Handwerger,
M.
D.,
Professor
of
Pediatrics,
University
of
Cincinnati,
Children's
Hospital
Medical
Center,
Cincinnati,
OH
S
teven
G.
Heeringa,
Ph.
D.,
Research
Scientist
&
Director
for
Statistical
Design,
University
of
Michigan,
Institute
for
Social
Research,
Ann
Arbor,
MI
Kenneth
M.
Portier,
Ph.
D.,
Associate
Professor,
Statistics,
Institute
of
Food
and
Agricultural
Sciences,
University
of
Florida,
Gainesville,
FL
FQPA
Science
Review
Board
Members
William
Brimijoin,
Ph.
D.,
Chair,
Pharmacology,
Mayo
Clinic
and
Medical
School,
Rochester,
MN
George
B.
Corcoran,
Ph.
D.,
Professor
&
Chairman,
Department
of
Pharmaceutical
Sciences,
Eugene
Applebaum
College
of
Pharmacy
&
Health
Sciences,
Wayne
State
University,
Detroit,
MI
Lutz
Edler,
Ph.
D.,
Head,
Biostatistics
Unit
C060,
German
Cancer
Research
Center,
Heidelberg,
Germany
11
of
113
Scott
Ferson,
Ph.
D.,
Senior
Scientist,
Applied
Biomathematics,
Setauket,
NY
Lawrence
Fischer,
Ph.
D.,
Center
for
Integrative
Toxicology,
Michigan
State
University,
East
Lansing,
MI
Gaylia
Jean
Harry,
Ph.
D.,
Neurotoxicology
Group
Leader,
National
Institute
of
Environmental
Health
Sciences,
Research
Triangle
Park,
NC
Dale
Hattis,
Ph.
D.,
Research
Professor,
Center
for
Technology,
Environment
&
Development
(
CENTED),
George
Perkins
Marsh
Institute,
Clark
University,
Worcester,
MA
James
P.
Kehrer,
Ph.
D.,
Director,
Center
for
Molecular
&
Cellular
Toxicology,
College
of
Pharmacy,
The
University
of
Texas
at
Austin,
Austin,
TX
Chensheng
Lu,
Ph.
D.,
Assistant
Professor,
Department
of
Environmental
&
Occupational
Health,
Rollins
School
of
Public
Health,
Emory
University,
Atlanta,
GA
Peter
Macdonald,
D.
Phil.,
Professor
of
Mathematics
and
Statistics,
McMaster
University,
Hamilton,
Ontario,
Canada
Christopher
J.
Portier,
Ph.
D.,
Director,
Environmental
Toxicology
Program,
National
Institute
of
Environmental
Health
Sciences,
Research
Triangle
Park,
NC
Nu­
May
Ruby
Reed,
Ph.
D.,
D.
A.
B.
T.,
Staff
Toxicologist,
Pesticide
Regulation,
California
Environmental
Protection
Agency,
Sacramento,
CA
P.
Barry
Ryan,
Ph.
D.,
Professor,
Environmental
&
Occupational
Health,
Rollins
School
of
Public
Health,
Emory
University,
Atlanta,
GA
Michael
D.
Sohn,
Ph.
D.,
Scientist,
Environmental
Energy
Technologies
Division,
Lawrence
Berkeley
National
Laboratory,
University
of
California,
Berkeley,
CA
Michael
D.
Wheeler,
Ph.
D.,
Assistant
Professor,
Departments
of
Pharmacology
&
Medicine,
University
of
North
Carolina,
Skipper
Bowles
Center
for
Alcohol
Studies,
Chapel
Hill,
NC
12
of
113
INTRODUCTION
In
Session
1
of
this
meeting,
the
FIFRA
SAP
met
to
consider
and
review
N­
methyl
carbamate
pesticide
cumulative
risk
assessment:
pilot
cumulative
analysis,
issues
related
to
cumulative
hazard
assessment.
EPA
acknowledges
that
there
are
toxicological
characteristics
unique
to
the
N­
methyl
carbamates
which
need
to
be
considered
in
a
cumulative
risk
assessment
for
this
group.
Specifically,
the
mechanism
of
action
for
this
group
of
pesticides
is
carbamylation
of
the
acetylcholinesterase
(
AChE)
active
site.
This
chemical
change
is
reversible,
allowing
for
relatively
rapid
recovery
from
inhibition.
OPP
is
collaborating
with
laboratory
scientists
and
statisticians
from
EPA's
National
Health
and
Environmental
Effects
Research
Laboratory
(
NHEERL)
to
evaluate
biological
and
empirical
aspects
of
recovery.
EPA
solicited
comment
from
the
SAP
on
specific
issues
related
to
dose­
response
modeling
of
AChE
data,
empirical
estimation
of
time
to
recovery,
and
the
impact
of
the
laboratory
method
used
to
measure
AChE
inhibition
on
estimates
of
relative
potency.
The
agenda
for
this
SAP
meeting
session
involved
an
introduction,
background,
and
detailed
presentations
of
the
issues
related
to
cumulative
hazard
assessment
provided
by
Dr.
Anna
Lowit
(
Health
Effects
Division,
Office
of
Pesticide
Programs,
EPA),
Dr.
Stephanie
Padilla,
Dr.
R.
Woodrow
Setzer,
and
Dr.
Ginger
Moser,
(
Office
of
Research
and
Development,
National
Health
and
Environmental
Effects
Research
Laboratory,
EPA).
Dr.
Clifford
Gabriel
(
Director,
Office
of
Science
Coordination
and
Policy,
EPA),
Mr.
Jim
Jones
(
Director,
Office
of
Pesticides
Programs,
EPA),
Mr.
Joseph
J.
Merenda,
Jr.
(
Office
of
Prevention,
Pesticides
and
Toxic
Substances,
EPA),
and
Dr.
Tina
Levine
(
Acting
Director,
Health
Effects
Division,
Office
of
Pesticide
Programs,
EPA)
offered
opening
remarks
at
the
meeting.
Dr.
William
Brimijoin
and
Dr.
Gaylia
Jean
Harry
presented
comments
submitted
by
Dr.
Janice
Chambers.
13
of
113
SUMMARY
OF
PANEL
DISCUSSION
AND
RECOMMENDATIONS
The
overall
issue
posed
to
the
panel
for
the
Day
1
discussions
was
the
suitability
of
the
Agency's
near­
term
plans
for
quantitative
estimation
of
Relative
Potency
Factors
(
RPFs)
for
use
in
the
carbamate
cumulative
risk
assessment.
An
initial
question
was
the
reliability
and
usability
of
the
basic
cholinesterase
inhibition
data
supplied
to
the
EPA
by
pesticide
registrants.
The
second
generic
question
called
on
the
Panel
to
evaluate
three
aspects
of
the
Agency's
plans
for
quantitative
analysis
of
those
data:
(
a)
the
basic
biological/
toxicological
assumptions
underlying
the
analyses,
(
b)
the
statistical
methodology
incorporating
a
simplified
pharmacokinetic
approach
planned
for
estimating
RPFs,
and
(
c)
the
later
potential
for
use
of
probabilistic
methods
to
characterize
uncertainty
distributions
related
to
standard
point­
estimate
uncertainty
factors
that
have
traditionally
been
used
in
Agency
risk
analyses
for
non­
cancer
effects.

The
Panel's
principal
response
to
the
first
issue
was
based
on
a
qualitative
graphical
comparison
of
cholinesterase
inhibition
observations
for
several
carbamates
from
an
extensive
set
of
Agency­
conducted
experiments
using
a
modern
radiometric
method
with
results
from
(
a)
Agency­
conducted
experiments
using
an
older
version
of
a
colorimetric
method,
and
(
b)
results
from
Agency
files
of
experiments
reported
by
pesticide
registrants
using
an
updated
modification
of
the
colorimetric
method.
Briefly,
these
comparisons
revealed
considerable
discrepancies
between
the
results
of
the
radiometric
assay
and
the
results
from
the
older
colorimetric
method.
However
discrepancies
were
much
less
apparent,
though
not
completely
absent,
between
the
radiometric
results
and
the
results
of
the
modified
colorimetric
procedure
that
has
evidently
been
used
by
registrants
in
recent
years.
The
committee
stressed
that
ultimate
conclusions
on
this
issue
must
be
based
on
a
full
quantitative
analysis
of
the
dose­
timeresponse
relationships
seen
for
the
radiometric
data
in
relation
to
the
registrant
data.
If
a
full
statistical
analysis
reveals
anomalies
in
the
comparisons
of
key
parameters
for
particular
compounds,
then
some
adjustment
of
registrant
data
for
individual
compounds
may
be
needed.
In
the
meantime,
on
the
assumption
that
proper
care
was
taken
by
the
registrants
in
the
conduct
of
the
assays,
it
appears
that
the
results
of
the
studies
submitted
for
pesticide
registration
provide
a
fairly
reliable
database
of
AChE
inhibition
in
relation
to
carbamate
exposure.
Thus,
on
balance,
the
Panel
supports
EPA's
position
that
the
data
are
suitable
to
support
the
cumulative
risk
assessment.

On
the
second
question,
in
general,
the
Panel
approves
the
way
the
Agency
is
approaching
the
empirical
analysis
of
AChE
data
for
N­
methyl
carbamates.
The
empirical
model
being
used
should
be
flexible
and
is
likely
to
fit
the
data,
although
the
Panel
also
offered
several
suggestions
on
details
of
the
primary
modeling
approach
and
modest
variants
that
should
be
considered
by
the
Agency.
Five
simplifying
assumptions
articulated
by
the
Agency
were
all
generally
considered
reasonable,
although
there
were
reservations
on
some,
and
the
Panel
re­
emphasizes
the
need
for
quantitative
statistical
analyses
of
the
uncertainties
resulting
from
departures
from
these
assumptions
where
possible.
Finally,
the
Panel
welcomed
the
openness
on
the
part
of
EPA
staff
to
explore
the
14
of
113
potential
for
using
distributional
approaches
in
noncancer
risk
assessments
to
the
extent
possible.
The
Panel
offers
several
suggestions
in
the
spirit
of
encouraging
the
Agency
to
help
develop
and
use
existing
and
emerging
probabilistic
techniques.
The
Panel
cautions,
however,
that
the
process
of
replacing
the
uncertainty
factors
with
distributions
based
on
empirical
data
is
far
from
straightforward.
In
addition
to
distributional
techniques
for
analysis,
this
process
will
involve
new
articulation
of
risk
management
goals,
which
are,
of
course,
beyond
the
scope
of
the
Panel's
charge
to
offer
advice
on
technical
issues.
For
example,
although
technical
methods
can
in
theory
provide
a
description
of
how
confident
we
should
be
that
the
population
incidence
of
harm
of
a
given
severity
level
for
a
given
exposure
is
less
than
a
series
of
defined
values,
there
has
as
yet
been
no
articulation
of
how
to
interpret
such
a
distributional
description
in
the
derivation
of
an
RfD
or
an
"
Acceptable
Daily
Intake"
(
ADI).
What
fraction
of
the
population
is
intended
to
be
"
covered"
with
what
confidence
by
the
current
system
of
single­
point
"
uncertainty"
factors?
15
of
113
PANEL
DELIBERATIONS
AND
RESPONSE
TO
QUESTIONS
The
specific
issues
addressed
by
the
Panel
are
keyed
to
the
Agency's
background
documents,
references,
and
the
Agency's
charge
questions.

Questions
Issue
1.1:
Laboratory
method
for
measuring
cholinesterase
inhibition
As
discussed
in
the
paper
(
Cumulative
Hazard
Assessment:
Issues
for
the
FIFRA
SAP),
in
toxicology
studies
performed
for
pesticide
registration,
typically,
acetylcholinesterase
(
AChE)
inhibition
is
measured
using
modified
Ellman,
spectrophotometric
methods
(
Ellman
et
al,
1961).
Scientists
at
EPA's
National
Health
and
Environmental
Effects
Research
Laboratory
(
NHEERL)
have
performed
a
series
of
time
course
and
dose­
response
studies
(
Appendices
1
and
2)
with
seven
N­
methyl
carbamate
pesticides.
These
studies
have
compared
radiometric
and
spectrophotometric
methods
following
acute
dosing
in
rats.
EPA
has
provided
plots
of
these
studies
along
with
plots
of
dose­
response
data
from
single
dosing
rat
studies
submitted
for
pesticide
registration
(
Appendix
2).
Prior
to
the
completion
of
these
studies,
there
was
a
concern
that
studies
submitted
to
EPA
for
pesticide
registration
may
underestimate
relative
potency.
Specifically,
using
spectrophotometric
methods,
recovery
of
inhibition
can
occur
prior
to
analysis
if
the
proper
precautions
are
not
taken
in
the
laboratory.

Statistical
analyses
evaluating
results
of
the
radiometric
data
generated
by
EPA
and
the
spectrophotometric
data
reported
by
registration
studies
have
not
been
performed.
However,
in
general,
based
on
visual
observation
of
these
plots
(
Appendix
2),
there
appears
to
be
good
concordance,
particularly
at
doses
at
or
near
10%
inhibition,
between
the
results
of
studies
submitted
for
registration
purposes
where
spectrophotometric
methods
were
used
and
the
results
of
studies
performed
by
EPA
where
radiometric
analyses
were
performed.
Nostrandt
et
al.
(
1993)
have
previously
shown
that
modified
Ellman
assay
gave
answers
comparable
to
the
radiometric
assay
if
some
special
precautions
were
taken.
EPA
does
not
know
the
exact
conditions
used
in
various
laboratories
performing
registration
studies.
However,
it
appears
that
the
AChE
data
provided
in
the
registration
studies
are
of
sufficient
quality
for
evaluating
relative
potency.

Question
1.1
Please
comment
on
EPA's
observations
regarding
the
results
of
radiometric
studies
conducted
by
NHEERL
and
the
results
of
studies
submitted
for
pesticide
registration.
16
of
113
Panel
Response
The
Panel
debated
whether
the
comparative
studies
of
seven
N­
methyl
carbamates
at
NHEERL
give
clear
evidence
that
recent
AChE
data
provided
by
the
registrants
to
EPA
are
of
sufficient
quality
to
support
a
cumulative
risk
assessment.
The
Panel
commended
the
EPA
scientists
for
their
well­
designed
and
valuable
study
and
reached
a
consensus
that
the
results
demonstrated
"
reasonable
reliability"
in
the
registrant
data.
Several
unresolved
issues
were
noted,
however,
and
qualifications
and
reservations
were
expressed
as
indicated
below.

The
Agency
presented
dose­
response
data
generated
by
two
different
methods
for
a
number
of
carbamates
(
coded
A
to
H)
of
varying
potency.
One
set
of
data
was
based
on
radiometric
assays
involving
minimal
sample
dilution
and
minimal
assay
duration.
These
data
were
expected
to
approximate
the
true
levels
of
AChE
inhibition.
A
second
set
of
data
was
based
on
automated
spectrophotometric
assays,
with
no
attempt
to
limit
sample
dilution
or
assay
time.
The
data
from
these
"
conventional
Ellman
assays"
were
expected
to
be
more
variable
and
to
underestimate
true
levels
of
AChE
inhibition
by
an
unknown
amount.
The
purpose
of
this
comparison
was
to
gain
perspective
on
the
quality
of
data
previously
supplied
to
EPA
by
the
registrants
of
the
same
seven
carbamates,
who
carried
out
their
assays
with
a
version
of
the
spectrophotometric
method.

The
time­
course
and
dose­
response
plots
for
the
seven
selected
N­
methyl
carbamates
in
Appendix
2
of
the
EPA
document
differed
between
the
radiometric
and
the
spectrophotometric
methods.
The
differences
did
not
follow
a
consistent
pattern
and
therefore
did
not
lend
themselves
to
easy
correction
(
such
as
a
simple
multiplicative
adjustment
in
the
percentage
of
inhibition
measured
by
one
method
to
convert
to
percentage
of
inhibition
measured
by
the
other
method.)
It
would
be
interesting
to
search
systematically
for
the
factors
that
could
account
for
the
differences
observed.
Beyond
the
factors
discussed
by
the
Agency
(
dilution,
temperature)
there
may
be
others
(
e.
g.
experimental
design,
type
of
carbamate,
other
assay
details)
that
could
explain
at
least
some
of
the
discrepancies
between
assays.
This
information
could
help
in
judging
the
value
and
limitations
of
spectrophotometric
methods
for
determining
cholinesterase
inhibition.

For
objective
comparison
of
the
two
assay
approaches,
one
should
probably
parameterize
the
time­
response
and
dose­
response
data
from
each
method
and
use
derived
parameters
to
describe
the
differences.
The
time
course
of
AChE
inhibition
in
both
data
sets
consisted
of
a
sharp
drop
followed
by
a
recovery
to
100%
of
control
activity.
One
Panelist
suggested
that
the
recovery
phase
could
be
fitted
to
a
two­
parameter
Michaelis­
Menten
type
of
curve.
A
one­
parameter
exponential
recovery
model,
as
used
in
the
current
proposal,
is
also
an
option.
Similarly,
the
dose­
response
data
could
be
described
by
a
slope
parameter
obtained
with
regression
techniques.
Because
low­
dose
effects
are
important
for
risk
assessment,
one
should
explore
how
to
calculate
best
the
low­
dose
slopes
of
the
different
dose­
inhibition
curves.
For
compounds
where
shoulder
effects
17
of
113
appear,
e.
g.
owing
to
the
normalization
against
control
values,
data
in
the
shoulder
region
could
be
dropped
from
the
regressions
or
modeled
with
the
same
kind
of
Michaelis­
Menten
approach
used
for
the
organophosphates.
That
is,
the
Agency
could
model
the
dose­
response
curves
and
express
potencies
as
reciprocals
of
the
benchmark
doses
associated
with
a
predetermined
percentage
of
inhibition
(
e.
g.,
the
reciprocal
of
the
ED10,
or
BMD10,
at
which
cholinesterase
is
inhibited
by
10%).
Differences
between
compounds
could
so
be
described
­
possibly
quantitatively­
more
objectively
than
when
viewing
and
interpreting
the
plots
only.

Without
prejudging
the
outcome
of
a
rigorous
methodological/
statistical
comparison,
the
Panel
discussed
the
reliability
of
EPA's
radiometric
approach
to
assessing
AChE
inhibition
by
carbamates.
A
point
not
addressed
in
the
submitted
assay
protocols
was
how
long
samples
waited
between
homogenization
and
assay.
The
Panel
heard
in
session
that
the
delay
was
up
to
90
minutes.
Such
a
delay
might
have
allowed
substantial
reactivation
if
samples
had
been
highly
diluted.
However,
the
Panel
was
informed
that
tissues
for
these
assays
were
homogenized
in
less
than
5
volumes
of
buffer
and
were
kept
at
ice
temperature.
Although
the
Panel
would
prefer
direct
evidence
that
reactivation
did
not
occur
before
the
assay,
it
is
reasonable
to
conclude
that
the
NHEERL
radiometric
data
reflect
the
true
levels
of
carbamate­
induced
AChE
inhibition.
This
accuracy
contrasts
with
the
data
from
the
same
sample
set
obtained
with
a
conventional
Ellman
procedure.
In
the
NHEERL
study,
the
conventional
data
were
much
more
variable,
especially
with
red
cells,
and
in
most
cases
they
underestimated
AChE
inhibition
in
samples
exposed
to
higher
levels
of
carbamate.

Turning
to
the
registrant
data,
all
based
on
spectrophotometric
methods,
the
Panel
saw
generally
good
agreement
with
the
EPA/
NHEERL
results
obtained
using
the
radiometric
method.
In
many
cases,
the
levels
of
inhibition
reported
by
the
registrants
were
similar
to
those
in
NHEERL's
radiometric
data.
This
agreement
would
be
surprising
if
the
registrants
had
used
standard
automated
Ellman
assays.
In
public
comment,
however,
an
industry
representative
stated
that
the
registrants
as
a
group
have
made
a
concerted
shift
to
a
more
rigorous
modified
method.
The
Panel
recognizes
that
a
modified
Ellman
procedure
will
reliably
measure
AChE
inhibition
if
tissue
dilution,
assay
time,
and
assay
temperature
are
minimized
as
described
by
Nostrandt
et
al.
(
1993).
One
Panel
member
pointed
out
that
running
such
assays
in
a
real­
time
or
"
kinetic
mode"
offered
the
important
advantage
of
being
able
to
detect
reactivation
as
it
occurred,
which
is
not
possible
with
the
end­
point
determinations
in
a
radiometric
assay.
Taking
the
data
at
face
value,
one
can
conclude
that
the
registrants'
results
do
give
a
reliable
index
of
cholinesterase
inhibition
with
most
of
the
tested
carbamates.
Assuming
similar
quality
in
the
studies
on
the
carbamates
not
tested
by
NHEERL,
then
most
if
not
all
of
those
registrant
data
should
also
be
useable
for
risk
assessment.

The
agreement
between
the
data
sets
from
NHEERL
and
registrants
should
not
be
overstated,
however.
In
the
first
place,
as
mentioned
above,
firm
conclusions
await
a
rigorous
statistical
analysis.
Second,
the
graphs
in
Appendix
2
are
somewhat
misleading
18
of
113
because,
in
standardizing
for
the
control,
the
curves
are
forced
together
on
the
left
(
low
dose
region).
This
squeezing
effect
can
account
for
the
apparent
"
optical
agreement"
at
low
levels
of
AChE
inhibition.
Other
options
that
could
be
explored
by
a
statistician
with
access
to
the
raw
data
would
be
expressing
the
comparative
results
in
terms
of
ratios
of
the
observed
percentage
of
inhibition
 
either
in
aggregated
form
for
all
animals
studied,
or
on
the
basis
of
distributions
of
individual
animal
results
in
a
form
such
as
comparative
box
plots.
Of
more
concern
were
specific
instances
in
which
the
NHEERL
and
registrant
data
diverged
systematically.
Sometimes
the
divergence
was
conservative
for
risk
assessment;
i.
e.,
the
registrant
data
predicted
greater
inhibition
than
the
NHEERL
data,
but
in
other
cases
the
divergence
was
non­
conservative.
With
compound
D,
for
example,
the
registrant
data
underestimated
relative
potency
in
the
brain
by
a
factor
of
about
two.
With
compound
C
the
underestimation
of
potency
in
red
blood
cells
(
RBC)
was
equally
great.
The
formal
analysis
should
quantify
the
uncertainty
in
comparative
potency
estimates
arising
from
such
cases.

In
light
of
the
available
information,
the
Panel
was
open
to
the
possibility
that
these
individual
discrepancies
merely
reflect
random
effects
in,
for
example,
statistical
mean
control
for
AChE
activity.
Variations
in
the
control
for
AChE
activity
are
likely
since
individual
animals
can
differ
by
50%
or
more.
As
one
Panel
member
pointed
out,
it
would
be
advantageous
in
the
future
to
collect
a
series
of
timed
blood
samples
from
the
same
animal,
starting
before
the
toxicant
is
administered.
A
modified
study
design
with
sameanimal
controls
will
reduce
variability
and
increase
statistical
power.
Unfortunately,
but
obviously,
repeated
sampling
cannot
be
extended
to
tissues
like
brain
tissues.
While
brain
AChE
remains
a
very
important
target
site,
future
efforts
should
be
devoted
to
develop
models
for
RBC
data
as
well,
considering
AChE
activities
in
RBC
are
significantly
correlated
with
activities
in
brain
tissues
as
presented
by
the
Agency's
scientists.
The
concern
about
using
brain
AChE
alone
was
raised
by
several
SAP
panel
members
in
previous
meetings,
and
for
the
purpose
of
validating
the
PBPK/
PD
models
and
cumulative
risk
assessments,
an
accessible
specimen
sample,
such
as
red
blood
cells,
is
essential.
In
this
case
brain
sampling
from
humans
is
not
feasible.
Additionally,
issues
of
statistical
power
come
to
the
fore
in
attempts
to
determine
low
levels
of
AChE
inhibition.
Previous
SAP
meetings
have
emphasized
that
precise
determinations
of
BMD10,
critical
for
estimates
of
relative
potency,
require
accurate
measures
of
inhibition
at
or
below
the
level
of
10%.
Using
present
methods,
this
goal
is
unachievable
unless
group
sizes
are
very
large,
even
with
a
time­
series
study
design.
Novel
approaches
to
address
this
problem
are
worth
seeking.

Returning
to
the
question
of
why
some
registrant
data
appeared
to
underestimate
the
levels
of
AChE
inhibition,
the
Panel
could
not
exclude
the
possibility
that
certain
studies
suffered
from
shortcomings
in
experimental
procedure.
In
order
to
evaluate
those
concerns
it
is
important
to
review
the
exact
conditions
used,
and
what
Good
Laboratory
Practice
(
GLP)
quality
standards
were
implemented
in
the
various
laboratories.
The
standard
operating
procedures
of
the
registrants
should
be
made
available
to
the
Agency
together
with
the
original
data
if
they
are
to
be
considered
for
the
risk
assessment.
One
19
of
113
Panel
member
suggested
that
a
Round
Robin
test
using
known
activities
of
AChE
in
human
or
animal
tissue
samples
should
be
adapted
so
the
variations
between
assays
or
between
labs
can
be
quantified
within
a
pre­
determined
percentage.
Similar
activities
are
being
undertaken
in
the
State
of
California.
The
Panel
also
considered
the
possibility
of
compound­
specific
effects
on
enzyme
activity
under
different
assay
conditions.
Data
assessing
these
possibilities
would
be
of
valuable.
Ultimate
conclusions
should
be
based
on
a
full
quantitative
analysis
of
the
absorption
rate
constant,
recovery
rate
constant
and
AUC
of
the
dose­
response
curves.
Of
these,
the
parameter
most
likely
to
vary
widely
across
laboratories/
animals
is
the
absorption
rate
constant.
However
the
recovery
rates
and
the
AUCs
should
agree.
If
these
metrics
show
anomalies
for
particular
chemicals
after
a
full
statistical
analysis,
then
some
adjustment
of
registrant
data
for
individual
compounds
may
be
needed.
In
the
meantime,
on
the
assumption
that
proper
care
was
taken
by
the
registrants,
it
appears
that
the
results
of
the
studies
submitted
for
pesticide
registration
provide
a
fairly
reliable
database
of
AChE
inhibition.
Thus,
on
balance,
the
Panel
supports
EPA's
position
that
the
data
are
suitable
to
support
the
cumulative
risk
assessment.

Issue
1.2:
Empirical
modeling
of
AChE
Data
Part
A.
Benchmark
dose
calculations:

In
the
EPA's
cumulative
risk
assessment
of
the
organophosphorus
pesticides
(
OPs),
a
decreasing
exponential
model
was
used
to
develop
benchmark
dose
estimates.
The
FIFRA
SAP
previously
endorsed
this
approach
(
FIFRA
SAP,
2001
&
2002).
EPA
plans
to
use
again
a
decreasing
exponential
model
in
its
benchmark
dose
estimates
for
the
N­
methyl
carbamate
pesticides,
with
the
addition
of
a
component
to
model
the
time
course
of
AChE
inhibition.
This
model
was
provided
in
the
Eqs.
1
 
4
and
the
associated
text
(
See
Cumulative
Hazard
Assessment:
Issues
for
the
FIFRA
SAP
and
Appendices
3­
4).

Question
1.2a
Please
comment
on
the
appropriateness
of
using
the
model
provided
in
Equations
1
 
4
to
calculate
benchmark
dose
estimates
based
on
cholinesterase
inhibition
for
the
N­
methyl
carbamate
pesticides.

Panel
Response
In
general,
the
Panel
approves
of
the
way
the
EPA
is
approaching
the
empirical
analysis
of
AChE
data
for
N­
methyl
carbamates.
The
empirical
model
being
used
should
be
flexible
and
is
likely
to
fit
the
data.
The
multiplicative
model
f(
t,
d)
=
A[
1­
g(
d)
h(
t)],
is
a
straightforward
empirical
modeling
approach
that
has
the
advantage
of
being
both
practical
and
transparent.
However,
when
proceeding
with
this
modeling,
the
Agency
should
be
aware
that
the
multiplication
of
the
dose
function
with
the
time
function
could
yield
a
dose­
time­
response
surface
that
could,
in
some
occasions,
lack
sufficient
flexibility
20
of
113
for
fitting
all
of
the
carbamate
data
sets
(
see
e.
g.
Figure
1
in
Appendix
3).
The
inclusion
of
a
term
combining
time
t
and
dose
d
may
need
to
be
considered.

On
the
dose
scale,
the
model
is
a
bit
over­
parameterized
and
it
will
prove
difficult
to
find
an
optimal
solution.
Constraints
on
the
model
parameters
and/
or
Bayesian
methods
could
reduce
this
problem.
However,
the
SAP
feels
a
more
complicated
model
is
probably
not
justified.

There
could
be
greater
emphasis
on
statistical
hypothesis
testing
(
e.
g.
does
the
ED10
change
from
one
strain
to
the
next;
across
sexes).
In
addition,
because
this
is
an
empirical
model,
goodness­
of­
fit
needs
to
be
carefully
assessed
when
using
this
to
make
risk
projections.
The
Agency
is
encouraged
to
continue
exploration
of
the
Aikake
Information
Criterion
(
AIC)
for
comparative
assessment
of
the
quality
of
model
fits
to
the
data.

The
time
component
of
the
dose­
response
function
is
the
classic
solution
to
a
simple
PK
process
and
adequate
as
a
first
look
at
the
time
course
effects
on
doseresponse
The
dismissal
of
the
human
data
for
direct
estimation
and
its
projected
use
as
a
"
validation"
of
an
allometric
scaling
of
the
rat
model
seems
premature
(
especially
in
light
of
the
limited
analysis
used
by
the
EPA
to
review
the
agreement
between
the
two
methods
for
evaluating
AChE).
The
Panel
encourages
the
Agency
to
either
directly
estimate
parameters
using
the
human
data
or
try
to
fit
the
human
and
rat
data
simultaneously
with
the
species
extrapolation
built
into
the
modeling.

The
use
of
the
Benchmark
Dose
(
BMD)
concept
and
terminology
for
the
cumulative
risk
assessment
of
carbamates
needs
clarification.
Otherwise
it
could
be
confused
with
the
BMD
approach
used
in
standard
risk
assessment
(
RA)
of
chemicals.
There,
the
BMD
is
used
for
the
definition
of
a
starting
point
(
point
of
departure)
for
the
low
dose
extrapolation
and
the
definition
of
risk
specific
doses.
In
contrast,
in
the
present
context
of
cumulative
RA
of
carbamates,
the
BMD
approach
has
been
implemented
to
derive
relative
potency
factors
(
RPFs).
In
addition,
this
BMD
method
does
not
incorporate
a
process
to
estimate
a
BMD
lower
confidence
level
(
BMDL)
which
is
usually
calculated
in
standard
RA.
Possible
later
Monte
Carlo
uncertainty
or
sensitivity
analyses
might
benefit
from
a
series
of
upper
and
lower
confidence
limits
on
the
estimates
of
mean
RPFs.

If
estimates
of
the
relative
potency
factors
are
to
be
obtained
for
individual
animals,
the
solution
of
averaged
parameter
values
will
not
be
the
same
as
the
average
solution.
In
addition
to
the
uncertainty
distributions
for
the
central
estimate
RPFs
suggested
in
the
previous
paragraph,
the
Agency
may
need
to
generate
solutions
for
random
sets
of
animals
to
look
at
a
variability
distribution
of
relative
potency
factors.
The
RPFs
will
be
calculated
as
ratios
of
two
BMDs,
say
DR
/
DR­
index
,
where
the
denominator
will
be
the
BMD
of
the
index
chemical,
DR­
index
and
DR
the
BMD
of
the
compound
to
be
considered
for
an
inclusion
in
the
cumulative
RA.
From
a
statistical
point
of
view,
the
21
of
113
RPF
is
then
a
ratio
of
two
quantities
where
each
was
estimated
from
dose­
response
data
and
possesses
therefore
statistical
variability.
Consequently,
the
variability
of
the
RPF
can
be
calculated
and
is
a
function
of
the
statistical
variability
of
DR
and
DR­
index.
In
order
to
calculate
the
standard
error
of
the
RPF
=
DR
/
DR­
index
according
to
a
well
known
statistical
method,
the
appropriate
standard
error
estimates­­
s.
e.(
DR)
and
s.
e.(
DR­
index)­­
are
required.

A
source
of
uncertainty
in
this
BMD
approach
stems
from
the
choice
of
R=
10%
inhibition
as
the
critical
effect
size.
When
inspecting
the
figures
of
Appendix
2
one
realizes
that
the
slope
of
the
dose­
response
curve
is
very
shallow
as
one
approaches
10%
inhibition.
The
shallow
slope
translates
into
a
large
variability
in
estimates
of
the
dose
that
causes
this
level
of
inhibition.
Therefore,
one
may
question
whether
the
choice
of
R=
10%
is
optimal
or
whether
it
should
be
replaced
or
compared
to
other
choices
of
the
critical
effect
size.
Furthermore,
since
the
BMD
for
the
cumulative
RA
is
derived
using
a
dosetime
response
function
f(
t,
d)
the
DR
is
also
a
function
of
time,
better
denoted
DR(
t).
When
using
the
simple
PK
model,
the
RPF
is
also
a
function
of
time
t
after
exposure
better
denoted
RPF(
t)=
DR
(
t)/
DR­
index(
t).
When
representing
the
response
as
a
surface
over
the
dose­
time
coordinate
system,
the
BMD
is
similar
to
an
isobole,
a
parameter
well
known
in
combination
experiments.
RPFs
are
then
simply
ratios
of
isoboles
from
different
experiments
and
should
be
presented
as
such.
If
there
is
stability
over
time,
then
that
can
be
noted
and
used.

The
multiplicative
model
f(
t,
d)
uses
the
time
of
the
peak
effect,
which
is
usually
not
known,
as
a
model
parameter
to
be
estimated
from
the
dose­
time
data.
Since
measurements
near
the
time
of
the
peak
effect
are
the
exception
rather
than
the
rule
in
experiments
like
the
present
one,
it
is
often
difficult
to
obtain
precise
estimates
of
that
time.
This
difficulty
reduces
the
interpretability
of
what
is
otherwise
an
important
model
parameter
in
the
simple
PK
model.
The
data
presented
by
the
Agency
came
from
experiments
that
measured
AChE
inhibition
no
earlier
than
15
minutes
after
exposure.
The
Agency
should
attempt
to
obtain
reliable
measurements
of
AChE
inhibition
at
earlier
time
points.
If
such
data
cannot
be
obtained,
it
may
be
possible
to
use
a
simpler
model
for
the
time
component
that
does
not
require
a
biphasic
curve
but
instead
uses
an
instantaneous
drop
at
exposure.

On
the
other
hand,
one
Panelist
expressed
a
preference
for
the
Agency
to
adapt
the
simple
pharmacokinetic
model
in
a
direction
that
is
more
in
line
with
a
mechanistic
interpretation
of
the
original
equation
for
the
time
course
of
cholinesterase
inhibition.
The
current
bi­
exponential
equation
implies
that
all
of
the
inhibition
that
will
ultimately
occur
is
available
immediately
at
the
outset
of
the
AChE
regeneration
process.
This
is
not
strictly
correct
because
none
of
the
AChE
enzyme
molecules
can
undergo
regeneration
before
they
have
actually
been
inhibited.
In
the
view
of
this
Panelist,
a
more
mechanistically
faithful
model/
equation
would
be
a
solution
to
the
following
differential
equation:

d[
Inhibition]/
dT
=
k1[
unabsorbed]
 
k2[
Inhibition],
where
[
unabsorbed]
is
the
maximum
possible
inhibition
if
all
the
carbamate
were
delivered
to
the
site
of
action
in
the
brain
instantaneously.
22
of
113
This
equation
does
not
appear
to
have
a
closed­
form
solution,
but
it
can
be
solved
numerically
using
appropriate
software.
It
will
avoid
the
need
in
the
present
equation
for
the
statistical
constraint
that
k1
is
arbitrarily
set
to
be
less
than
k2
(
which
will
likely
be
violated
for
dermal
exposure
experiments).

A
concerted
effort
is
needed
to
find
data
on
intrinsic
clearance
rate
for
any
of
these
chemicals
since
there
may
be
compound
specific
changes
that
could
affect
the
regeneration
rate.
The
Agency
needs
additional
guidance
and
could
possibly
use
a
simple
biochemical
assay
to
get
the
theoretical
recovery
rates.

Finally,
the
SAP
was
pleased
that
the
Agency
chose
to
perform
the
calculations
using
the
R
software
package.
The
equations
were
nicely
coded
and
commented.
Transparency
in
EPA's
model
development
and
application
will
continue
to
improve
the
public's
confidence
in
the
EPA's
work.

Part
B.
Simple
pharmacokinetic
model
As
discussed
in
the
background
document
prepared
for
the
FIFRA
SAP,
EPA
is
committed
to
improving
methodologies
and
approaches
for
conducting
cumulative
risk
assessments.
To
this
end,
EPA
has
begun
development
of
a
simple,
pharmacokinetic
(
PK)
based
approach
for
incorporating
recovery
of
cholinesterase
inhibition
in
risk
estimates.
The
simple
PK
model
is
more
sophisticated
than
conventional
relative
potency
approaches
but
less
data­
intensive
than
physiologically­
based
pharmacokinetic/
pharmacodynamic
approaches
and
thus
provides
a
pragmatic
method
for
considering
PK
and/
or
mechanistic
information
in
risk
estimates.
There
are
still,
however,
limitations
to
the
application
of
this
approach
for
the
N­
methyl
carbamate
cumulative
risk
assessment
 
namely,
the
capability
for
cumulative
exposure
models
to
output
distributions
of
exposure
(
in
mg/
kg
or
similar
units)
to
individuals.
Given
this
limitation,
EPA
continues
to
pursue
practical
methods
for
improving
risk
assessment
methods.
It
is
unclear
at
this
time
the
degree
to
which
this
simple
PK
approach
may
be
used
in
the
cumulative
risk
assessment
for
the
N­
methyl
carbamate
pesticides.
However,
as
EPA
continues
to
work
towards
improving
its
risk
assessment
methods,
EPA
is
requesting
comment
from
the
FIFRA
SAP
regarding
aspects
of
the
development
and
application
of
this
simple
PK
approach.
23
of
113
Question
1.2.
b
Please
comment
on
the
simplifying
assumptions
used
in
the
simple
PK
approach
to
predicting
cholinesterase
inhibition.
Please
include
in
your
comments
whether
these
assumptions
tend
to
underestimate
or
overestimate
potential
risk.
These
assumptions
are:

 
The
inhibitor
is
cleared
quickly
from
the
target
tissue,
so
that
recovery
time
mostly
depends
upon
the
rate
of
decarbamylation
of
AChE.
 
Competition
among
multiple
inhibitors
for
AChE
or
clearance
pathways
is
quantitatively
insignificant.
 
Inhibitors
do
not
modify
the
affinity
of
AChE
for
other
inhibitors
(
e.
g.,
by
binding
to
a
site
on
the
AChE
molecule
that
has
allosteric
effects),
or
such
effects
are
quantitatively
insignificant.
 
It
is
appropriate
to
ignore
resynthesis
of
new
AChE
molecules
in
the
time­
frame
of
interest
(
1
 
6
hours).
 
The
model
for
effects
in
humans
can
be
calibrated
by
scaling
parameters
of
models
fit
to
rodent
data.

Panel
Response
The
consensus
of
the
Panel
is
that
at
low
dose
levels
the
assumptions
listed
below
for
the
current
model
are
probable
valid.
There
are,
however,
limited
actual
data
to
address
or
support
these
assumptions
for
carbamate
pesticides.
As
with
any
modeling
exercise,
the
accuracy
of
the
model
is
limited
by
the
quality
of
the
original
data.
Thus,
the
Agency
is
encouraged
to
examine
all
available
data
sets
including
the
human
data
comparison
mentioned
in
the
document,
EPA
studies
on
mixtures,
and
data
provided
by
industry
both
in
and
outside
of
the
registration
packages
for
individual
pesticides.
When
evaluating
these
and
other
assumptions,
the
Agency
is
encouraged
to
consider
how
the
model
would
need
to
be
adjusted
to
account
for
any
assumptions.
The
adaptations
needed
to
alter
or
remove
some
assumptions
might
not
appreciably
change
the
fundamental
model
structure
or
implementing
software,
while
accounting
for
other
assumptions
may
require
a
new
model
or
software.

Assumption
1.
The
inhibitor
is
cleared
quickly
from
the
target
tissue,
so
that
recovery
time
mostly
depends
upon
the
rate
of
decarbamylation
of
AChE.

The
assumption
that
recovery
of
AChE
activity
mainly
reflects
the
rate
of
decarbamylation
is
sound,
given
the
half­
life
values
of
current
members
of
the
cumulative
assessment
group
and
the
rate
of
exchange
between
fat
and
blood
that
can
be
inferred
24
of
113
from
simple
pharmacokinetic
information.
1
This
assumption
would
be
true
at
low
concentrations.
However,
with
one
compound,
inhibition
lasted
twice
as
long
as
with
most
of
the
others.
The
intrinsic
rate
of
decarbamylation
with
this
compound
might
be
atypically
slow,
but
that
would
be
surprising
since
all
of
the
tested
carbamates
should
generate
identical
enzyme
adducts.
Therefore,
assumption
one
may
not
be
universally
true.
In
other
words,
the
rate
of
recovery
of
enzyme
activity
in
rats
treated
with
certain
N­
methyl
carbamates
might
be
driven
by
extrinsic
factors,
other
than
intrinsic
rates
of
decarbamylation.
Should
new
carbamate
pesticides
with
much
higher
degrees
of
lipophilicity
and
longer
half­
life
values
become
available,
and
appear
to
a
significant
extent
in
the
environment,
the
validity
of
this
assumption
should
be
revisited.
Some
reevaluation
also
might
be
in
order
if
elimination
rates
from
human
fat
are
much
slower
than
the
observed
regeneration
half
lives
for
acetylcholinesterase
activity
in
blood.
That
may
be
the
case
because,
as
indicated
by
the
human
model
parameters
in
ERDEM,
the
ratio
of
blood
flow
to
fat
volume
is
much
lower
in
humans
than
in
rats.

Assumption
2.
Competition
among
multiple
inhibitors
for
AChE
or
clearance
pathways
is
quantitatively
insignificant.

The
assumption
that
competition
among
multiple
inhibitors
for
AChE
or
for
clearance
pathways
is
quantitatively
insignificant
should
hold
across
much
of
the
1
For
example,
the
ERDEM
pharmacokinetic
model
incorporates
an
estimated
fat/
blood
partition
coefficient
of
17.1.
For
rats,
the
blood
flow
incorporated
into
this
model
is
0.51
liters/
minute
and
the
tissue
volume
is
approximately
.015
Liters
for
a
0.25
kg
rat.
Given
this,
the
rate
constant
for
elimination
of
carbaryl
from
fat
is
expected
to
be
[
blood
flow
/(
tissue
volume
*
fat//
blood
partition
coefficient
=
.51/(.
015*
17.1)
=
2.0/
hour.
This
rate
constant
directly
implies
a
half
life
in
fat
of
[
ln(
2)/
2.0
=
0.35
hours],
or
about
21
minutes.
Thus,
for
the
rat
experiments,
release
from
the
fat
seems
unlikely
to
appreciably
distort
observations
of
the
half
life
of
about
1.7
hours
attributed
to
enzyme
regeneration.
This
would
continue
to
be
true
even
if
the
actual
elimination
half
life
for
rat
were
about
twice
as
long
(
41
minutes)
as
implied
by
an
alternative
rat
tissue/
blood
partition
coefficient
estimation
model
suggested
by
one
Panel
member
(
Ginsberg,
G.
L.,
Pepelko,
W.
E.,
Goble,
R.
L.,
and
Hattis,
D.
B.
"
Comparison
of
Contact
Site
Cancer
Potency
Across
Dose
Routes:
Case
Study
with
Epichlorohydrin,"
Risk
Analysis
Vol.
16,
pp.
667­
681,
1996;
Walker,
K.,
Hattis,
D.,
Russ,
A.,
and
Ginsberg,
G.
"
Physiologically­
Based
Toxicokinetic
Modeling
for
Acrylamide
 
Risk
Implications
of
Polymorphisms
and
Developmental
Changes
in
Selected
Metabolic
Enzymes,"
Report
from
the
George
Perkins
Marsh
Institute,
Clark
University,
and
the
Connecticut
Department
of
Public
Health
to
the
U.
S.
Environmental
Protection
Agency
under
Cooperative
Agreement
#
827195­
0,
December
2004;
Ginsberg,
G.
L.,
Goble,
R.
L.,
and
Hattis,
D.
B.
"
Slope
Factor
Comparison
Across
Dose
Routes:
Case
Study
with
Epichlorohydrin,"
Report
to
the
U.
S.
Environmental
Protection
Agency
by
TRC
Environmental
Corporation,
April,
1994.
Spreadsheets
incorporating
the
alternative
assumptions
for
estimating
partition
coefficients
in
rats
and
humans
will
be
supplied
to
EPA
on
request.
25
of
113
anticipated
human
exposure
range.
However,
this
simplifying
assumption
may
not
hold
under
conditions
of
relatively
high
exposure
to
two
or
more
carbamate
pesticides,
where
competition
for
the
AChE
protein
and
for
clearance
pathways
is
more
likely.
Limited
data
with
carbaryl
and
malathion
indicate
that
malathion
decreased
the
rate
constant
of
absorption
and
beta­
phase
elimination
of
radiolabeled
carbaryl
when
both
were
administered
orally
at
10
mg/
kg
to
rat
(
Waldron
and
Abdel­
Rahman,
1986).
This
study
suggests
that
at
higher
exposures,
competition
for
the
target
as
well
as
for
elimination
pathways
would
be
expected.

Assumption
3.
Inhibitors
do
not
modify
the
affinity
of
AChE
for
other
inhibitors
(
e.
g.,
by
binding
to
a
site
on
the
AChE
molecule
that
has
allosteric
effects),
or
such
effects
are
quantitatively
insignificant.

The
assumption
that
inhibitors
do
not
modify
the
affinity
of
AChE
for
other
inhibitors
is
sound,
based
on
existing
knowledge.
This
is
again
especially
true
at
low
concentrations.
The
potential
interaction
of
carbamates
should
be
acknowledged,
and
accounted
for
when
such
interaction
significantly
influences
risk
analysis.
Data
to
be
obtained
from
the
ongoing
EPA
studies
of
the
combined
7
compounds
should
provide
additional
information
on
possible
interactions.
However,
the
current
design
of
these
mixture
studies
may
not
be
sufficient
to
fully
resolve
critical
questions
of
effect
additivity.
The
inclusion
of
neural
cell
cultures
as
a
complementary
model
system
to
assess
specific
interactions
should
be
considered
by
the
Agency.
The
validity
of
this
assumption
should
remain
under
surveillance
for
new
carbamates
that
may
interact
by
allosteric
means
with
AChE.
If
allosteric
interactions
are
discovered,
their
effect
on
enzyme
sensitivity
will
have
to
be
defined
before
one
can
determine
whether
the
operating
assumption
will
tend
to
overestimate
or
underestimate
risk.

Assumption
4.
It
is
appropriate
to
ignore
resynthesis
of
new
AChE
molecules
in
the
time­
frame
of
interest
(
1
 
6
hours).

Simplification
of
any
model
is
desirable
especially
in
reducing
the
number
of
parameters
for
which
limited
or
no
data
are
available.
AChE
protein
is
normally
produced
in
excess
with
the
majority
being
degraded.
The
half­
life
of
AChE
synthesis
in
rodents
is
approximate
7
days,
providing
a
net
turnover
of
approximately
10%/
day
or
0.5%/
hr.
This
turnover
leads
to
the
expectation
that
the
maximum
impact
of
new
enzyme
synthesis
on
the
overall
rate
of
reversal
of
AChE
inhibition
should
be
approximately
3%.
The
relative
impact
of
resynthesis
of
new
AChE
molecules
is
greater
at
higher
levels
of
inhibition
and
becomes
less
important
as
activity
returns
to
normal
levels.
Re­
synthesis
of
new
AChE
molecules
is
therefore
unlikely
to
contribute
much
to
the
process
of
recovery
during
the
first
6
hours.

Assumption
5.
The
model
for
effects
in
humans
can
be
calibrated
by
scaling
parameters
of
models
fit
to
rodent
data.
26
of
113
Body
weight3/
4
scaling
is
the
usual
assumption
for
model
extrapolation
of
bulk
metabolic
capacity
among
species,
in
parallel
with
observations
of
the
scaling
of
metabolic
rates.
There
are
no
obvious
data
to
indicate
that
the
model
for
effects
in
humans
should
not
be
initially
calibrated
by
scaling
parameters
of
the
models
fit
to
rodent
data.
If
these
parameters
are
identified
as
key
contributing
factors
by
sensitivity
analysis
the
Agency
should
provide
some
guidance
for
targeted
efforts
to
test
the
scaling
assumptions
for
key
parameters
with
limited
human
studies.

As
indicated
earlier
in
the
response
to
Question
1.1,
ultimate
conclusions
should
be
based
on
a
full
quantitative
analysis
of
the
absorption
rate
constant,
recovery
rate
constant
and
AUCs/
dose.
Of
these
parameters,
the
most
likely
to
vary
substantially
across
laboratories/
animals
is
the
absorption
rate
constant.
In
contrast,
the
recovery
rates
and
the
AUCs
per
dose
should
match
from
experiment
to
experiment.
If
these
parameters
show
discrepancies
for
particular
chemicals,
given
a
full
statistical
analysis,
some
adjustment
of
registrant
data
for
individual
compounds
should
be
considered.

Question
1.2c
EPA
historically
has
utilized
(
default)
uncertainty
factors
for
interspecies
and
intraspecies
extrapolation.
EPA's
issue
paper
(
and
related
appendices
3­
4)
suggests
that
application
of
the
simple,
PK
approach
to
estimation
of
risk
provides
an
opportunity
to
consider
probabilistic
methods
in
uncertainty
analysis
for
cumulative
hazard
assessment.
Please
comment
on
biological
and
quantitative
factors
which
may
be
important
for
consideration
in
the
event
probabilistic
methods
were
to
be
used
to
perform
uncertainty
analysis
in
cumulative
hazard
assessment.

Panel
Response
This
question
reflects
a
welcome
openness
on
the
part
of
the
EPA
staff
to
explore
the
potential
for
using
distributional
approaches
in
noncancer
risk
assessments
to
the
extent
possible.
The
presenters
exhibited
a
commendable
dedication
to
achieve
congressionally
mandated
public
health
protection
goals
with
the
aid
of
the
best
inputs
of
technical
understanding
that
can
be
mustered
in
a
reasonable
time.
The
Panel
offers
the
suggestions
below
in
the
spirit
of
encouraging
the
Agency
to
help
develop
and
use
emerging
probabilistic
techniques.
However,
the
Panel
is
not
able
to
offer
a
straightforward
incremental
path
to
utilize
the
insights
gleaned
from
those
techniques
in
the
current
system
of
single­
point
uncertainty
factors
that
has
developed
over
the
decades
since
the
original
1954
paper
of
Lehman
and
Fitzhugh
(
Lehman
and
Fitzhugh,
1954).
In
part,
this
is
because
there
has
only
been
a
beginning
to
public
discussion
of
how
to
separate
the
risk
assessment
and
risk
management
elements
embedded
in
the
current
RfD
procedure
by
specifying
quantitative
objectives
that
should
be
met
by
RfDs
in
terms
of
variability
and
uncertainty
(
Hattis
et
al.,
2002).
That
is,
for
exposure
at
an
RfD
level,
27
of
113
what
degree
of
confidence
should
there
be
that
the
population
incidence
of
some
specified
degree
of
biological
effect
is
below
a
particular
value?

The
discussion
below
will
first
cover
one
Panelist's
perspective
on
some
issues
in
the
potential
use
of
probabilistic
techniques
to
help
inform
choices
of
the
values
of
specific
uncertainty
factors
(
those
for
interspecies
projection
and
human
inter­
individual
variability,
as
supplemented
with
the
additional
FQPA
factor).
Then
there
will
be
some
more
general
comments
on
opportunities
and
pitfalls
in
the
use
of
probabilistic
techniques
in
the
analysis
of
currently
available
data
for
the
carbamate
cholinesterase
inhibition
agents.

Interspecies
Factor:

The
goal
of
this
factor
is
to
convert
the
external
doses
associated
with
a
sensitive
but
still
"
adverse"
toxicological
endpoint
in
a
sensitive
species
of
animals
to
doses
that
would
be
expected
to
cause
the
same
"
benchmark"
incidence
and
severity
of
effects
in
humans.
Distributional
analyses
of
empirical
data
on
interspecies
projection
are
potentially
useful
in
the
following
way:
By
assembling
data
bases
of
putatively
analogous
cases
for
relatively
well
studied
chemicals,
the
"
uncertainty"
in
interspecies
projections
for
a
chemical
for
which
there
are
inadequate
direct
test
information
are
estimated
by
observing
the
variability
among
a
set
of
analogous
chemicals
with
better
information.
There
is,
of
course,
some
need
for
judgment
in
the
choice
of
a
particular
set
of
chemicals
that
can
be
considered
"
analogous"
cases
for
a
specific
untested
chemical.

The
most
recent
distributional
analyses
relevant
to
this
factor
(
Hattis
et
al.,
2002)
for
multi­
dose
exposures
are
based
on
data
for
putatively
analogous
indices
of
systemic
subchronic
toxicity
for
61
anti­
cancer
agents
in
animals
and
people
assembled
by
Price
et
al.,
2002.
[
The
endpoints
used
as
benchmarks
for
toxicity
in
animals
and
people
were
not
identical
but
were
those
historically
used
for
prediction
of
human
potencies
from
animal
data
by
the
National
Cancer
Institute
 
usually
LD10
levels
were
used
for
rodents
and
compared
with
"
maximum
tolerated
doses"
2
inferred
from
human
studies
of
the
anticancer
agents.
For
dogs
and
monkeys,
the
benchmarks
were
typically
TDL
(
Toxic
Dose
Low)
values.
3]
The
results
of
this
analysis,
in
brief,
are
that
where
data
from
single
species
are
used
to
project
the
potency
of
the
same
chemical
in
humans,
the
geometric
mean
of
the
estimates
are
close
to
what
would
be
expected
for
a
body
weight3/
4
projection.

2
"
Maximum
Tolerated
Doses"
are
operationally
defined
as
the
dose
level
at
which
none
of
six
or
one
of
six
patients
experience
dose
limiting
toxicity
with
the
next
higher
dose
level
having
two
or
more
patients
experiencing
dose
limiting
toxicity
[
Storer
B.
E
(
1989),
Design
and
analysis
of
phase
I
clinical
trials.
Biometrics.
45(
3),
25­
37;
and
Edler
L
(
2001)
Overview
of
Phase
I
Trials.
In
Handbook
of
Statistics
in
Clinical
Oncology
(
ed.
J
Crowley)
Marcel
Dekker,
1­
34.]
3
This
is
defined
as
the
lowest
dose
to
produce
pathological
alterations
in
hematological,
chemical,
clinical,
or
morphological
parameters,
the
doubling
of
which
produces
no
lethality.
28
of
113
There
is,
however,
a
considerable
spread
in
the
relationship
between
the
human
toxic
potencies
calculated
from
this
scaling
assumption
and
the
potencies
estimated
from
human
data.
For
example,
for
the
18
chemicals
for
which
there
are
rat
and
human
potency
estimates,
the
geometric
mean
human
potency
for
systemic
toxicity
is
about
90%
of
the
value
projected
from
the
rat
data;
however
the
95th
percentile
of
the
distribution
corresponds
to
more
than
four
times
the
value
projected
from
the
a
body
weight3/
4
scaling
rule.

A
similar
distributional
analysis
is
possible
for
acute
toxicity
using
a
database
of
thousands
of
LD50
observations
in
different
species
of
animals
assembled
by
Rhomberg
and
Wolff
(
1998).
Unfortunately,
there
are
only
four
chemicals
in
this
data
base
where
estimates
of
human
LD50s
are
provided.
However,
analyses
have
been
made
of
the
distributions
of
interspecies
differences
for
the
diverse
species
represented
in
the
collection.
On
this
basis,
the
central
tendency
of
interspecies
projections
using
these
acute
toxicity
data
is
close
to
dose/
body
weight1
rather
than
the
body
weight3/
4
scaling
rule
observed
for
the
subchronic
data
discussed
above.
The
preliminary
explanation
for
the
difference
in
the
central
tendency
scaling
rules
for
subchronic
versus
acute
toxicity
is
that
the
key
causally
relevant
internal
dosimeter
for
systemic
subchronic
toxicity
tends
to
be
an
AUC
(
integrated
Area
Under
the
Curve
of
concentration
X
time)
for
the
toxicant,
for
which
energy­
dependent
elimination
rates
are
important.
By
contrast,
it
is
thought
that
the
key
pharmacokinetic
determinant
of
acute
toxicity
following
a
single
dose
of
a
typical
toxicant
is
an
internal
maximum
concentration
of
the
toxicant,
which
is
likely
to
be
more
strongly
influenced
by
simple
dilution
of
the
administered
dose
in
an
appropriate
distribution
volume­­
which
tends
to
scale
across
species
in
direct
proportion
to
body
weight.
This
comparison
has
not,
however,
been
done
within
mode­
of­
action
groups.
It
might
be
informative
to
select
the
anticholinesterase
agents
from
the
existing
Rhomberg
and
Wolff
(
1998)
data
base
and
evaluate
the
interspecies
scaling
of
that
subset.
The
full
Rhomberg
and
Wolff
(
1998)
data
base
is
available
from
the
first
author.

In
the
context
of
the
analysis
of
carbamate
toxicity,
for
the
chemicals
where
doses
causing
comparable
degrees
of
cholinesterase
inhibition
can
be
determined,
it
would
be
useful
to
examine
the
distribution
of
human/
animal
toxic
potencies
for
whatever
degree
of
inhibition
is
selected
as
the
"
benchmark"
response.

Human
Inter­
individual
Variability
Factor
It
is
important
to
begin
this
discussion
by
mentioning
the
distinction
between
"
dose­
effect
relationships"
(
where
the
dependent
variable
is
a
continuous
response,
as
in
a
cholinesterase
level)
and
"
dose­
response
relationships"
(
where
the
dependent
variable
is
the
fraction
of
individuals
who
show
a
quantal,
or
binary,
response
such
as
death
or
a
cleft
palate.)
In
the
case
of
a
quantal
response
that
is
caused
by
threshold
mechanism,
a
"
benchmark
dose"
(
BMD)
is
interpretable
as
the
fraction
of
individual
animals
(
or
people,
if
human
data
are
used)
who
have
thresholds
for
response
that
are
lower
than
the
defined
"
benchmark"
response
frequency
(
often
10%).
Therefore,
for
a
quantal
effect,
the
BMD
29
of
113
contains
some
element
of
inter­
individual
variability,
at
least
within
the
group
of
animals
(
or
humans)
that
was
directly
studied.

By
contrast,
where
the
biological
response
variable
studied
is
a
continuous
parameter,
as
in
the
case
of
a
cholinesterase
level,
the
BMD
is
usually4
defined
in
terms
of
a
change
(
such
as
a
10%
reduction)
in
the
group
mean
of
the
parameter
under
study.
The
chosen
extent
of
change
in
the
continuous
variable
is
sometimes
referred
to
as
the
"
benchmark
response"
(
BMR).
In
this
case,
there
is
no
component
of
individual
variability
in
the
BMD
 
essentially
the
BMD
is
an
estimate
of
the
dose
that
causes
the
designated
percentage
change
in
the
continuous
variable
(
BMR)
in
the
typical
or
average
member
of
the
experimental
animal
group
tested.
Therefore,
for
a
continuous
variable
like
a
particular
degree
of
acetylcholinesterase
inhibition,
the
purpose
of
the
human
interindividual
variability
factor
should
be
seen
as
the
multiple
of
dose
needed
to
convert
the
external
doses
associated
with
a
particular
percentage
of
inhibition
(
or
other
associated
biological
effects)
in
a
typical
or
average
person,
to
the
dose
that
can
be
expected
with
high
confidence
to
produce
a
corresponding
cholinesterase
inhibition
(
or
other
associated
effects)
in
an
acceptably
small
proportion
of
exposed
humans.
Stated
in
this
way,
questions
are
raised
as
to
how
small
is
an
"
acceptably
small
proportion
of
exposed
humans"
and
how
high
the
confidence
level
should
be
that
the
incidence
of
cholinesterase
inhibition
at
the
BMR
level
is
less
than
the
incidence
deemed
"
acceptable."
These
are
policy
determinations
that
are
beyond
the
scope
of
the
technical
advice
requested
from
the
Panel,
although
it
can
be
noted
that
there
are
preliminary
proposals
for
operational
answers
to
these
questions
in
the
existing
literature
(
Hattis
et
al.,
2002).

With
this
background
it
also
should
be
noted
that
inter­
individual
variability
in
the
dose
producing
a
given
amount
of
cholinesterase
inhibition
in
people
does
not
cover
all
of
the
pharmacodynamic
variability.
In
the
taxonomy
of
human
variability
proposed
in
the
past
work
of
one
Panelist
(
Hattis
et
al.,
1999),
pharmacodynamic
inter­
individual
variability
has
two
components:

 
Differences
among
people
in
the
amount
of
the
active
agent
at
the
active
site
needed
to
produce
a
particular
change
in
a
physiological
parameter
(
e.
g.
cholinesterase
inhibition),
and
4
Alternatives
to
this
have
been
suggested
in
some
cases.
For
example
some
have
advocated
defining
an
"
effect"
for
a
continuous
parameter
in
terms
of
the
change
in
the
number
of
individuals
who
are
some
number
of
standard
deviations
(
or
log
standard
deviations
for
a
lognormal
distribution)
away
from
the
group
mean
based
on
the
distributional
statistics
of
an
unexposed
control
population.
However
this
"
quantalization"
approach
discards
some
of
the
inherent
statistical
strength
of
a
continuous
biological
response
parameter,
and,
unless
there
is
some
basis
for
choosing
the
number
of
standard
deviations
used
to
define
abnormality,
appears
no
less
arbitrary
than
picking
some
percentage
change
in
the
group
mean
level
for
defining
a
benchmark
response.
30
of
113
 
Differences
among
people
in
the
amount
of
change
in
the
physiological
parameter
needed
to
cause
an
adverse
response
(
e.
g.,
an
acute
change
in
cholinesterase
function
needed
to
cause
short
term
symptoms,
or
developmental
effects
during
possible
"
windows
of
vulnerability",
or
a
longer
term
adaptation
needed
to
produce
possible
adaptive
synaptic
changes
that
could
mediate
longer
term
adverse
alterations
in
neural
or
neuromuscular
signaling
as
has
recently
been
reported
for
some
organophosphate
agents)
(
van
Helden
et
al.,
2004;
Delgado
et
al.,
2004;
Sánchez­
Santed
et
al.,
2004;
Kassa
et
al.,
2001a;
Kassa
et
al.,
2001b;
Kassa
et
al.,
2001c).
Subtle
developmental
effects
are
theoretically
possible,
for
example,
an
early
life
marginal
strengthening
of
signaling
for
some
cholinergic
pathways
relative
to
other
non­
cholinergic
pathways.
This
is
because,
during
some
sensitive
times,
initially
formed
synaptic
connections
may
be
lost
if
there
is
few
signaling
traffic
along
them.
One
Panelist
noted
that
identifying
sensitive
window(
s)
for
this
kind
of
effect
may
be
complex
because
different
portions
of
the
human
nervous
system
develop
(
and
perhaps
undergo
this
synaptic
"
pruning")
at
different
times/
ages
(
Johnston,
1995;
Lichtman
et
al.,
2000;
Lowel
and
Singer,
1992;
Walsh
and
Lichtman,
2003).

Because
the
types
of
variability
mentioned
in
the
second
bullet
would
not
be
captured
even
if
there
were
perfect
measurements
of
the
human
inter­
individual
variability
in
changes
in
cholinesterase
activities
as
a
function
of
dose,
the
Agency
should
be
cautious
in
its
consideration
for
possible
changes
in
the
normal
10­
fold
human
inter­
individual
variability
factor
for
anticholinesterase
agents
in
the
light
of
human
clinical
studies
of
cholinesterase
inhibition
from
carbamates
that
have
been
conducted
by
or
for
the
pesticide
registrants.
The
variability
indicated
by
those
human
studies
should
be
analyzed
and
compared
with
variability
information
for
other
general
toxicants,
but
possible
changes
in
the
uncertainty
factor
for
inter­
individual
variability
should
be
limited
by
the
fact
that
the
full
pharmacodynamic
pathway
to
functional
response
is
not
captured
in
cholinesterase
inhibition
measurements
standing
by
themselves.
The
FQPA
10­
fold
factor
can
be
thought
of
as
a
policy­
based
recognition
of
the
possibility
of
developmental
effects
mentioned
in
the
second
bullet
above.

Other
General
Comments
Panelists
expressed
general
support
for
exploration
of
probabilistic
approaches
that
would
depart
from
the
routine
application
of
multiple
factors
of
10
for
the
various
uncertainties
in
interspecies
projection
and
the
extent
of
human
inter­
individual
variability.
However,
there
was
skepticism
that
the
results
of
probabilistic
uncertainty
analysis
could,
in
the
near
term,
be
used
to
completely
replace
the
traditional
interspecies
uncertainty
factors
without
both
PK
and
PD
data
for
validation.
Furthermore,
while
there
was
optimism
that
pharmacokinetic
scaling
for
interspecies
projection
is
likely
to
be
approachable,
the
Panel
had
reservations
on
the
current
capability
to
treat
the
PD
portion
of
the
pathway,
especially
with
the
relatively
sparse
data
specific
to
carbamate
31
of
113
pharmacodynamics.
For
the
inter­
individual
variability
factor,
the
Panel
expressed
a
need
for
explicit
consideration
of
the
possible
effects
of
human
genetic
differences
leading
to
differences
in
absorption
or
metabolic
elimination.
One
Panelist
expressed
a
strong
preference
that
most
of
the
analytical
resources
available
for
this
effort
be
devoted
to
statistical
analysis
of
the
cholinesterase
inhibition
dose­
response
data,
with
no
more
than
10%
of
the
available
analytical
resources
devoted
to
a
possible
probabilistic
analysis.

The
Panel
expressed
its
support
of
the
suggestion
made
by
the
public
presenter
(
Dr.
Sass)
to
fully
document
and
expose
to
independent
verification
the
various
models
used
for
analysis
of
the
cholinesterase
inhibition
information,
as
called
for
in
the
EPA's
Council
for
Regulatory
Environmental
Modeling
(
CREM)
guidelines
(
http://
epa.
gov/
osp/
crem.
htm).

Another
Panelist,
with
some
explicit
support
from
other
members,
made
four
distinct
methodological
points
for
EPA
to
consider
in
framing
its
probabilistic
analyses:

 
First,
uncertainties
other
than
those
arising
from
simple
sampling
errors
should
be
included.
Conventional
uncertainty
statistics
such
as
standard
errors
are
based
solely
on
observable
fluctuations
in
the
available
data.
However
empirical
experience
has
shown
that
even
in
such
mathematically
sophisticated
fields
as
physics,
as
improved
experimental
techniques
allow
better
measurements,
there
is
a
strong
tendency
for
the
updated
measurements
for
such
parameters
as
fundamental
particle
constants
to
be
found
to
fall
outside
of
the
bounds
of
previously
stated
statistical
confidence
limits
more
often
than
would
be
expected
by
chance.
The
likely
generic
explanation
for
this
is
that
there
are
generally
unsuspected
sources
of
systematic
error
(
miscalibration
of
instruments;
unrepresentativeness
of
statistical
samples
in
surveys,
etc.)
that
mean
that
existing
standard
errors
should
generally
be
expanded
beyond
those
conventionally
calculated
from
internal
differences
within
data
sets.
Techniques
have
been
suggested
to
do
this
(
Shlyakhter,
1994;
Hattis
and
Burmaster,
1994),
but
these
have
not
yet
been
widely
adopted.

 
Second,
the
issue
of
dependence
should
be
thought
through
for
any
probabilistic
analysis.
Dependency
issues
are
of
two
kinds.
First,
one
parameter
may
depend
on
another
because
of
real
causal
mechanisms
that
connect
them.
For
example,
when
fat
is
absorbed
into
the
blood
following
a
fatty
meal,
this
will
be
likely
to
both
increase
the
fat/
blood
partition
coefficient
for
a
volatile
lipophilic
chemical
in
a
physiologically
based
pharmacokinetic
(
PBPK)
model,
leading
to
somewhat
greater
short
term
absorption
from
an
inhalation
exposure,
and,
in
parallel,
decrease
the
tissue/
blood
partition
coefficients,
and
hence
the
rate
of
elimination
of
the
chemical
from
a
variety
of
lipid­
containing
model
compartments
(
fat,
brain)
with
venous
blood
exiting
the
organs.
A
second
distinct
source
of
dependency
is
induced
correlations
in
estimation
errors
for
different
parameters
in
the
same
regression
or
other
statistical
modeling
analysis.
For
example,
in
the
simple
pharmacokinetic
model
presented
it
is
quite
likely
that
the
simultaneous
estimation
32
of
113
of
cholinesterase
inhibition
half
life
and
"
absorption"
half
life"
induces
crosscorrelations
in
estimates
of
these
parameters.
Appropriate
use
of
the
calculated
estimation
errors
for
these
parameters
in
a
probabilistic
uncertainty
analysis
requires
direct
or
indirect
consideration
of
the
dependencies
of
the
error
distributions
for
one
of
these
parameters
on
the
others.

 
Third,
although
transformations
can
be
convenient
ways
to
enforce
some
constraints
on
parameter
values
in
an
estimation
procedure
(
as
was
mentioned
in
the
discussion
of
the
simple
pharmacokinetic
model),
such
transformations
may
not
always
be
ideal
for
deriving
uncertainty
distributions.

 
Finally,
although
it
is
prudent
to
use
the
guidance
mentioned
in
the
simple
pharmacokinetic
model
justification
to
keep
the
model
simple,
there
may
be
some
merit
in
exploring
even
simpler
models
for
sensitivity
analysis
(
for
example,
eliminating
the
"
absorption"
half
life).

A
further
concern
that
was
mentioned
by
one
Panelist
(
but
was
shared
by
other
Panelist)
was
for
more
analysis
of
errors
caused
by
model
uncertainty
or
misspecification.
What
uncertainty
arises
from
possible
errors
in
the
mathematical
forms
chosen
for
representing
resynthesis
or
reactivation
of
inhibited
acetylcholinesterase,
or
specifying
the
relationships
between
cholinesterase
inhibition
kinetics
in
rats
versus
people?
Although
the
quantitative
information
for
carbaryl
and
other
carbamates
are
likely
to
be
among
the
better
data
sets,
they
should
not
be
expected
to
yield
very
precise
estimates
of
the
values
of
model
parameters.
This
reinforces
the
need
for
careful
thought,
sensitivity
and
uncertainty
analyses.
33
of
113
REFERENCES
Delgado
E.,
McConnell
R.,
Miranda
J.,
Keifer
M.,
Lundberg
I.,
Partanen
T.,
Wesseling
C.
(
2004).
Central
nervous
system
effects
of
acute
organophosphate
poisoning
in
a
two­
year
follow­
up.
Scand
J
Work
Environ
Health.
30(
5):
362­
70.

Edler
L
(
2001).
Overview
of
Phase
I
Trials.
In
Handbook
of
Statistics
in
Clinical
Oncology
(
ed.
J
Crowley)
Marcel
Dekker,
1­
34.

Ginsberg,
G.
L.,
Goble,
R.
L.,
and
Hattis,
D.
B.
(
1994).
Slope
Factor
Comparison
Across
Dose
Routes:
Case
Study
with
Epichlorohydrin.
Report
to
the
U.
S.
Environmental
Protection
Agency
by
TRC
Environmental
Corporation.

Ginsberg,
G.
L.,
Pepelko,
W.
E.,
Goble,
R.
L.,
and
Hattis,
D.
B.
(
1996).
Comparison
of
Contact
Site
Cancer
Potency
Across
Dose
Routes:
Case
Study
with
Epichlorohydrin.
Risk
Analysis
Vol.
16:
667­
681.

Hattis,
D.
and
Burmaster,
D.
E.
(
1994).
Assessment
of
Variability
and
Uncertainty
Distributions
for
Practical
Risk
Analyses.
Risk
Analysis
14:
713­
730.

Hattis,
D.,
Banati,
P.,
Goble,
R.,
and
Burmaster,
D.
(
1999).
Human
Inter­
individual
Variability
in
Parameters
Related
to
Health
Risks.
Risk
Analysis
19:
705­
720.

Hattis,
D.,
Baird,
S.,
and
Goble,
R.
(
2002).
A
"
straw
man"
proposal
for
a
quantitative
definition
of
the
RfD.
Drug
and
Chemical
Toxicology
25:
403­
436.

Hattis,
D.,
Walker,
K.,
Russ,
A.,
and
Goble,
R.
(
2004).
Role
of
Dosimetric
Scaling
and
Species
Extrapolation
in
Evaluating
Risks
Across
Life
Stages.
I.
Dosimetric
Considerations
in
Childhood.
Draft
Report
to
the
U.
S.
Environmental
Protection
Agency
Under
RFQ
No.
DC­
03­
00009,
October,
2004;
also
see
website
http://
www2.
clarku.
edu/
faculty/
dhattis
Jager,
T.,
Vermeire
T.
G.,
Rikken
M.
G.,
van
der
Poel
P.
(
2001).
Opportunities
for
a
probabilistic
risk
assessment
of
chemicals
in
the
European
Union.
Chemosphere
43(
2):
257­
64.

Johnston,
M.
V.
(
1995).
Neurotransmitters
and
vulnerability
of
the
developing
brain.
Brain
Dev
17(
5):
301­
6.

Kassa,
J.,
Koupilova,
M.,
Vachek,
J.
2001a.
Long­
term
effects
of
low­
level
sarin
inhalation
exposure
on
the
spatial
memory
of
rats
in
a
T­
maze.
Acta
Medica
(
Hradec
Kralove).
44(
3):
93­
96.
34
of
113
Kassa,
J.,
Koupilova,
M.,
Vachek,
J.
(
2001b).
The
influence
of
low­
level
sarin
inhalation
exposure
on
spatial
memory
in
rats.
Pharmacol
Biochem
Behav.
70(
1):
175­
179.

Kassa,
J.,
Koupilova,
M.,
Herink,
J.,
Vachek,
J.
(
2001c).
The
long­
term
influence
of
lowlevel
sarin
exposure
on
behavioral
and
neurophysiological
functions
in
rats.
Acta
Medica
(
Hradec
Kralove).
44(
1):
21­
27.

Lehman,
A.
J.
and
O.
G.
Fitzhugh
(
1954).
100­
fold
margin
of
safety.
Assoc.
Food
Drug
Off.
U.
S.
Q.
Bull.
18:
33­
35.

Lichtman,
J.
W.
and
Colman,
H.
(
2000).
Neuron.
Synapse
elimination
and
indelible
memory
25(
2):
269­
78.

Lowel,
S.
and
Singer,
W.,
(
1992).
Selection
of
intrinsic
horizontal
connections
in
the
visual
cortex
by
correlated
neuronal
activity.
Science
255:
209­
212.

Nostrandt,
A.
C.,
Duncan,
J.
A.,
Padilla,
S.
(
1993).
A
modified
spectrophotometric
method
appropriate
for
measuring
cholinesterase
activity
in
tissues
from
carbaryl­
treated
animals.
Fundam.
Appl.
Toxicol.
21:
196­
203.

Pieters,
M.
N.,
Kramer,
H.
J.,
Slob,
W.
(
1998).
Evaluation
of
the
uncertainty
factor
for
subchronic­
to­
chronic
extrapolation:
statistical
analysis
of
toxicity
data.
Regul
Toxicol
Pharmacol.
27(
2):
108­
11.

Pieters,
M.
N.,
Freijer,
J.,
Baars,
B.
J.,
Fiolet,
D.
C.,
van
Klaveren,
J.,
Slob,
W.
(
2002).
Risk
assessment
of
deoxynivalenol
in
food:
concentration
limits,
exposure
and
effects.
Adv
Exp
Med
Biol.
504:
235­
48
Price,
P.
S.,
Swartout,
J.
C.,
Schmidt,
C.,
and
Keenan,
R.
E.
(
2002).
Characterizing
interspecies
uncertainty
using
data
from
studies
of
anti­
neoplastic
agents
in
animals
and
humans.
Human
and
Ecological
Risk
Assessment
in
preparation,
­­
paper
and
data
available
on
the
web
at
http://
www2.
clarku.
edu/
faculty/
dhattis.

Rhomberg,
L.
R.
and
Wolff,
S.
K.
(
1998).
Empirical
scaling
of
single
oral
lethal
doses
across
mammalian
species
based
on
a
large
database.
Risk
Anal.
18(
6):
741­
753.

Sánchez­
Santed,
F.,
Cañadas,
F.,
Flores,
P.,
López­
Grancha,
M.,
Cardona,
D.
(
2004).
Long­
term
functional
neurotoxicity
of
paraoxon
and
chlorpyrifos:
behavioural
and
pharmacological
evidence.
Neurotoxicol
Teratol.
26(
2):
305­
17.

Shlyakhter,
A.
I.
(
1994).
An
improved
framework
for
uncertainty
analysis:
Accounting
for
unsuspected
errors.
Risk
Analysis
14:
441­
447.
35
of
113
Slob,
W.,
Pieters,
M.
N.
(
1998).
A
probabilistic
approach
for
deriving
acceptable
human
intake
limits
and
human
health
risks
from
toxicological
studies:
general
framework.
Risk
Anal.
18(
6):
787­
98.

Storer,
B.
E.
(
1989).
Design
and
analysis
of
phase
I
clinical
trials.
Biometrics
45(
3):
25­
37.

van
Helden,
H.
P.,
Vanwersch,
R.
A.,
Kuijpers,
W.
C.,
Trap,
H.
C.,
Philippens,
I.
H.,
Benschop,
H.
P.
(
2004).
Low
levels
of
sarin
affect
the
EEG
in
marmoset
monkeys:
a
pilot
study.
J
Appl
Toxicol.
24(
6):
475­
83.

Vermeire,
T.,
Stevenson,
H.,
Peiters,
M.
N.,
Rennen,
M.,
Slob,
W.,
Hakkert,
B.
C.
(
1999).
Assessment
factors
for
human
health
risk
assessment:
a
discussion
paper.
Crit
Rev
Toxicol.
29(
5):
439­
90.
Review.

Waldron
Lechner,
D.,
Abdel­
Rahman,
M.
S.
(
1986).
Kinetics
of
carbaryl
and
malathion
in
combination
in
the
rat.
J
Toxicol
Environ
Health.
18(
2):
241­
56.

Walker,
K.,
Hattis,
D.,
Russ,
A.,
and
Ginsberg,
G.
(
2004).
Physiologically­
Based
Toxicokinetic
Modeling
for
Acrylamide
 
Risk
Implications
of
Polymorphisms
and
Developmental
Changes
in
Selected
Metabolic
Enzymes.
Report
from
the
George
Perkins
Marsh
Institute,
Clark
University,
and
the
Connecticut
Department
of
Public
Health
to
the
U.
S.
Environmental
Protection
Agency
under
Cooperative
Agreement
#
827195­
0,
December
2004
Walsh,
M.
K.,
Lichtman,
J.
W.
(
2003).
In
vivo
time­
lapse
imaging
of
synaptic
takeover
associated
with
naturally
occurring
synapse
elimination.
Neuron.
37(
1):
67­
73.
36
of
113
SAP
Minutes
No.
2005­
01
A
Set
of
Scientific
Issues
Being
Considered
by
the
Environmental
Protection
Agency
Regarding:

N­
METHYL
CARBAMATE
PESTICIDE
CUMULATIVE
RISK
ASSESSMENT:
PILOT
CUMULATIVE
ANALYSIS
SESSION
2:
PHYSIOLOGICALLY
BASED
PHARMACOKINETIC/
PHARMACODYNAMIC
(
PBPK/
PD)
MODELING
FOR
CARBARYL
FEBRUARY
16,
2005
FIFRA
Scientific
Advisory
Panel
Meeting,
held
at
the
Holiday
Inn
­
National
Airport,
Arlington,
Virginia
Myrta
R.
Christian,
M.
S.
Gary
Isom,
Ph.
D.
Designated
Federal
Official
FIFRA
SAP
Session
Chair
FIFRA
Scientific
Advisory
Panel
FIFRA
Scientific
Advisory
Panel
Date:
April
15,
2005
Date:
April
15,
2005
37
of
113
Federal
Insecticide,
Fungicide,
and
Rodenticide
Act
Scientific
Advisory
Panel
Meeting
February
16,
2005
N­
METHYL
CARBAMATE
PESTICIDE
CUMULATIVE
RISK
ASSESSMENT:
PILOT
CUMULATIVE
ANALYSIS
SESSION
2:
PHYSIOLOGICALLY
BASED
PHARMACOKINETIC/
PHARMACODYNAMIC
(
PBPK/
PD)
MODELING
FOR
CARBARYL
PARTICIPANTS
FIFRA
SAP
Session
Chair
Gary
Isom,
Ph.
D.,
Professor
of
Toxicology,
School
of
Pharmacy
&
Pharmacal
Sciences,
Purdue
University,
West
Lafayette,
IN
Designated
Federal
Official
Myrta
R.
Christian,
M.
S.,
FIFRA
Scientific
Advisory
Panel
Staff,
Office
of
Science
Coordination
and
Policy,
EPA
FIFRA
Scientific
Advisory
Panel
Members
Stuart
Handwerger,
M.
D.,
Professor
of
Pediatrics,
University
of
Cincinnati,
Children's
Hospital
Medical
Center,
Cincinnati,
OH
S
teven
G.
Heeringa,
Ph.
D.,
Research
Scientist
&
Director
for
Statistical
Design,
University
of
Michigan,
Institute
for
Social
Research,
Ann
Arbor,
MI
Kenneth
M.
Portier,
Ph.
D.,
Associate
Professor,
Statistics,
Institute
of
Food
and
Agricultural
Sciences,
University
of
Florida,
Gainesville,
FL
FQPA
Science
Review
Board
Members
John
Adgate,
Ph.
D.,
Assistant
Professor,
Division
of
Environmental
&
Occupational
Health,
University
of
Minnesota,
School
of
Public
Health,
Minneapolis,
MN
William
Brimijoin,
Ph.
D.,
Chair,
Pharmacology,
Mayo
Clinic
and
Medical
School,
Rochester,
MN
38
of
113
George
B.
Corcoran,
Ph.
D.,
Professor
&
Chairman,
Department
of
Pharmaceutical
Sciences,
Eugene
Applebaum
College
of
Pharmacy
&
Health
Sciences,
Wayne
State
University,
Detroit,
MI
Lutz
Edler,
Ph.
D.,
Head,
Biostatistics
Unit
C060,
German
Cancer
Research
Center,
Heidelberg,
Germany
Scott
Ferson,
Ph.
D.,
Senior
Scientist,
Applied
Biomathematics,
Setauket,
NY
Lawrence
Fischer,
Ph.
D.,
Center
for
Integrative
Toxicology,
Michigan
State
University,
East
Lansing,
MI
Gaylia
Jean
Harry,
Ph.
D.,
Neurotoxicology
Group
Leader,
National
Institute
of
Environmental
Health
Sciences,
Research
Triangle
Park,
NC
Dale
Hattis,
Ph.
D.,
Research
Professor,
Center
for
Technology,
Environment
&
Development
(
CENTED),
George
Perkins
Marsh
Institute,
Clark
University,
Worcester,
MA
James
P.
Kehrer,
Ph.
D.,
Director,
Center
for
Molecular
&
Cellular
Toxicology,
College
of
Pharmacy,
The
University
of
Texas
at
Austin,
Austin,
TX
Chensheng
Lu,
Ph.
D.,
Assistant
Professor,
Department
of
Environmental
&
Occupational
Health,
Rollins
School
of
Public
Health,
Emory
University,
Atlanta,
GA
Peter
Macdonald,
D.
Phil.,
Professor
of
Mathematics
and
Statistics,
McMaster
University,
Hamilton,
Ontario,
Canada
David
MacIntosh,
Sc.
D.,
Senior
Associate,
Environmental
Health
&
Engineering,
Inc.,
Newton,
MA
Christopher
J.
Portier,
Ph.
D.,
Director,
Environmental
Toxicology
Program,
National
Institute
of
Environmental
Health
Sciences,
Research
Triangle
Park,
NC
Nu­
May
Ruby
Reed,
Ph.
D.,
D.
A.
B.
T.,
Staff
Toxicologist,
Pesticide
Regulation,
California
Environmental
Protection
Agency,
Sacramento,
CA
P.
Barry
Ryan,
Ph.
D.,
Professor,
Environmental
&
Occupational
Health,
Rollins
School
of
Public
Health,
Emory
University,
Atlanta,
GA
Michael
D.
Sohn,
Ph.
D.,
Scientist,
Environmental
Energy
Technologies
Division,
Lawrence
Berkeley
National
Laboratory,
University
of
California,
Berkeley,
CA
39
of
113
Michael
D.
Wheeler,
Ph.
D.,
Assistant
Professor,
Departments
of
Pharmacology
&
Medicine,
University
of
North
Carolina,
Skipper
Bowles
Center
for
Alcohol
Studies,
Chapel
Hill,
NC
40
of
113
INTRODUCTION
In
Session
2
of
this
meeting,
the
FIFRA
SAP
met
to
consider
and
review
N­
methyl
carbamate
pesticide
cumulative
risk
assessment:
pilot
cumulative
analysis,
physiologically
based
pharmacokinetic/
pharmacodynamic
(
PBPK/
PD)
modeling
for
carbaryl.
OPP
is
collaborating
with
scientists
from
EPA's
National
Exposure
Research
Laboratory
(
NERL)
to
develop
a
PBPK/
PD
model
for
carbaryl
within
the
Exposure
Related
Dose
Estimating
Model
(
ERDEM)
Platform
(
Blancato
et
al.,
2002;
Okino
et
al.
2004).
The
carbaryl
model
will
form
the
basic
structure
of
a
generalized
model
for
the
N­
methyl
carbamates.
A
Quantitative
Structure
Activity
Relationship
(
QSAR)
database
of
physicochemical
descriptors
and
provisional
PK
and
PD
parameter
values
has
been
assembled
for
selected
N­
methyl
carbamates.
The
completeness
and
representativeness
of
the
QSAR
database
will
influence
the
application
of
the
PBPK/
PD
model
for
use
in
the
cumulative
risk
assessment
of
the
N­
methyl
carbamates.
EPA
solicited
comment
on
specific
aspects
of
the
appropriate
use
of
ERDEM
for
this
Risk
Assessment.
The
agenda
for
this
SAP
meeting
involved
an
introduction,
background,
and
detailed
presentations
of
the
issues
related
to
PBPK/
PD
modeling
for
carbaryl
provided
by
Dr.
Anna
Lowit
(
Health
Effects
Division,
Office
of
Pesticide
Programs,
EPA),
Dr.
Miles
Okino,
Dr.
Jerry
Blancato,
Mr.
Fred
Power,
and
Dr.
Curtis
Dary
(
Office
of
Research
and
Development,
National
Exposure
Research
Laboratory,
EPA).
41
of
113
SUMMARY
OF
PANEL
DISCUSSION
AND
RECOMMENDATIONS
The
Panel
commends
and
supports
the
Agency's
effort
to
develop
a
PBPK/
PD
model
for
carbaryl
within
the
Exposure
Related
Dose
Estimating
model
(
ERDEM)
platform
and
encourages
its
continuing
improvement
as
the
basic
structure
applicable
for
N­
methyl
carbamates
(
NMC).

Comments
were
made
regarding
the
coverage
in
the
Agency
background
document.
Specifically,
the
given
information
is
insufficient
for
a
comprehensive
review
regarding
the
completeness
and
availability
of
database
for
model
development.
Greater
transparency
and
details
also
are
needed
for
tracking
model
building
decisions
and
calibration.

Comments
and
suggestions
were
made
on
several
areas
of
model
construction,
parameter
estimation,
and
model
fitting.
The
Panel
expressed
concerns
about
the
complexity
of
the
model
that
appears
to
go
beyond
what
the
available
data
can
support.
Several
suggestions
for
its
simplification
are
given.
On
the
other
hand,
there
are
areas
that
may
be
inadequately
modeled.
These
include
considering
the
inactivation
of
carbaryl
when
it
is
released
from
binding
to
cholinesterase,
and
the
bioavailability
of
carbaryl
through
dermal
uptake.
If
necessary,
it
may
be
possible
to
place
greater
emphasis
on
the
lower
dose
range
pertinent
to
human
exposures
when
calibrating
model
fit
through
evaluating
the
discrepancies
between
model
predictions
and
experimental
data.
Thoughtful
considerations
should
be
given
to
the
decision
to
focus
the
model
fit
on
the
blood:
brain
partition
coefficient.
A
set
of
fat:
blood
partition
coefficients
are
provided
for
initiating
a
sensitivity
analysis.
With
respect
to
the
PD
portion
of
the
model,
the
general
lack
of
PD
data
is
noted.

The
Agency's
initial
effort
is
focused
on
group
average
parameters.
Future
efforts
can
explore
and
assess
parameter
range
and
variability.
Modeling
data
from
individual
animals
may
reveal
a
unique
set
of
coefficients
for
model
parameters.
When
aggregated
data
are
applied
to
the
model,
the
pertinence
of
prediction
at
the
99.9th
percentile
warrants
careful
consideration.
It
is
important
to
consider
the
model's
utility
when
inputting
biomonitoring
data
collected
in
humans.
Instead
of
using
only
the
portion
of
data
collected
in
Missouri,
biomonitoring
data
collected
in
California
also
should
be
used
to
characterize
human
exposures.
Finally,
the
Panel
provides
detailed
suggestions
on
statistical
evaluation,
specifically
for
determining
model
complexity
and
parameter
estimation,
assessing
model's
goodness­
of­
fit,
and
for
sensitivity
and
uncertainty
analyses.

PBPK/
PD
modeling
opens
up
a
variety
of
possible
endpoints
for
risk
assessment.
The
Panel
provides
perspectives
for
each
of
the
five
endpoints
proposed
by
the
Agency
in
the
context
of
duration
and
level
of
exposure,
toxicity
implication,
and
the
expression
of
risk.
The
use
of
more
than
one
endpoint
is
encouraged
because
it
can
enhance
the
understanding
of
how
risk
is
defined
and
facilitates
risk
communication.
42
of
113
It
is
too
early
in
the
model
development
to
determine
its
application
in
cumulative
risk
assessment
for
the
NMC.
The
Agency
is
encouraged
to
continue
its
development.
General
considerations
for
the
future
development
of
the
ERDEM
model
include
ensuring
that
the
sizable
model
output
is
organized
with
options
for
ease
in
data
and
file
management
and
export,
and
broadening
the
model
application
by
adding
the
capability
to
model
endpoints
at
target
fetal
tissues.
43
of
113
PANEL
DELIBERATIONS
AND
RESPONSE
TO
QUESTIONS
The
specific
issues
addressed
by
the
Panel
are
keyed
to
the
Agency's
background
documents,
references,
and
the
Agency's
charge
questions.

Questions
Issue
2.1:
Data
Requirements
for
PBPK/
PD
Models
The
document
"
Assessment
of
carbaryl
exposure
following
turf
application
using
a
physiologically
based
pharmacokinetic
model"
describes
the
application
of
a
carbaryl
specific
physiologically
based
pharmacokinetic/
pharmacodynamic
(
PBPK/
PD)
model
to
a
case
study
of
exposure
for
selected
exposure
scenarios
involving
broadcast
applications
of
a
liquid
formulation
of
carbaryl
to
turf.
A
PBPK/
PD
model
was
developed
based
on
available
laboratory
studies,
and
applied
to
post­
application
exposure
scenarios.

The
PBPK/
PD
model
provided
a
structure
to
evaluate
the
pharmacokinetic
(
PK)
and
pharmacodynamic
(
PD)
data
available
for
carbaryl.
This
case
study
was
instructive
in
discovering
data
gaps,
such
as
blood:
brain
partition
coefficient
values,
isolated
metabolism
rates,
and
identification
of
specific
metabolites.
Data
published
in
the
open
literature
were
generally
incomplete
for
the
purposes
of
PBPK/
PD
modeling,
where
simultaneous
tissue
concentration,
excretion
and
effect
data
are
ideal
for
model
evaluation.
Registrant
data
provided
some
constraints
for
PBPK/
PD
parameter
values,
but
uncertainty
remains
in
those
values
due
to
the
dependence
on
fitting
in
vivo
data
and
structure­
activity
methods
(
Poulin
and
Theil
2000;
J
Pharm
Sci
89:
16­
35).

Question
2.1a
Please
comment
on
the
completeness
of
the
data
used
to
develop
the
PBPK/
PD
model.

Panel
Response
The
documents
provided
to
the
Panel
do
not
contain
a
comprehensive
review
of
the
data
available
for
potential
use
in
model
calibration,
and
so
it
is
difficult
to
respond
to
this
question
fairly.
Future
documentation
of
the
model
and
its
derivation
should
include
such
potentially
usable
information.
Table
1
in
the
background
document
is
far
from
sufficient
for
this
purpose.
This
is
because
(
1)
it
evidently
omits
some
data
sets
that
the
authors
decided
not
to
use,
(
in
some
cases
for
good
reasons­­
e.
g.
because
they
were
based
on
experiments
with
non­
ring
labeled
carbaryl)
and
(
2)
because
the
observations
themselves
are
not
presented
in
tabular
form.
Presentation
of
the
useable
data
in
tabular
form,
with
available
information
on
observed
variation
and
measurement
errors
in
the
data
44
of
113
in
an
appendix,
facilitates
independent
assessment
of
the
correspondence
between
observations
and
model
predictions.

Question
2.1b
Please
comment
on
the
way
the
data
sets
were
used
to
estimate
and
constrain
parameter
values.

Panel
Response
In
general,
the
Panel
is
supportive
of
the
ERDEM
effort
but
considers
the
system
to
be
in
a
relatively
early
stage
of
development.
This
is
particularly
evident
in
the
application
to
a
complex
modeling
problem
such
as
that
presented
by
the
many
metabolic/
inactivation
pathways
and
multiple
physiological
compartments
of
the
current
PBPK/
PD
models
for
carbaryl
in
rats,
adult
humans,
and
children
of
various
ages.
The
potential
power
of
the
model
is
that
it
can
project
across
dose
levels,
species,
and
age
groups.
However,
there
need
to
be
caveats
alerting
the
reader
to
the
limitations
and
uncertainties
in
the
model
as
currently
developed
and
calibrated.
The
Agency
indicated
that
the
model
will
continue
to
be
improved
as
more
data
become
available,
and
this
is
encouraged
by
the
Panel.
Such
improvements
increase
the
likelihood
that
this
model
can
be
successfully
extended
to
other
carbamates.

This
question
elicited
a
wide
variety
of
comments
on
different
aspects
of
model
development,
calibration,
and
testing.
The
responses
below
are
organized
into
four
major
headings:

 
Model
documentation
and
disclosure
of
the
history
of
model
development
and
calibration
with
observational
data
 
Fundamental
model
structure
 
General
conception
and
execution
of
the
"
fitting"
problem.
This
includes
o
The
problem
of
"
overparameterization"
and
the
balance
of
considerations
in
basing
estimates
of
various
model
parameters
on
fitting
results
within
the
model
versus
exogenous
information
(
such
as
organ
volumes
and
blood
flows)
o
The
need
for
more
formal
statistical
criteria
for
model
fitting,
instead
of
the
current
relatively
informal
procedure
o
The
issue
of
model
"
validation"
o
The
issue
of
group
average
versus
individual
values
of
model
parameters
 
Evaluation
of
the
current
ERDEM
model
fits
and
implications
of
these
early
results
with
respect
to
o
Which
observations
should
be
emphasized
in
model
calibration?
45
of
113
o
Anomalies
in
the
current
fitting,
and
implications
for
adapting
model
inputs
and
refitting
Model
Documentation
and
Development
History
The
existing
report
provides
only
the
most
basic
outline
of
the
processes
for
using
the
data
to
calibrate
the
model.
There
is
no
presentation
of
what
fitting
techniques
were
applied,
which
parameters
were
fixed
or
fitted,
or
what
data
were
used
to
evaluate
the
fits.
Several
Panelists
also
indicated
that
additional
documentation
of
the
modelers'
decisions
in
response
to
inadequacies
in
first­
generation
model
fits
is
desirable.
The
oral
presentation
indicated
a
relatively
informal
process
of
adjusting
the
brain/
blood
partition
coefficient.
Some
Panelists
suggested
that
the
EPA
should
undertake
a
more
detailed
and
statistically
formal
effort
to
resolve
the
uncertainties
in
the
calibration
for
the
rat
model
before
the
projection
is
seriously
used
for
estimating
human
delivered
doses
and
inferring
brain
cholinesterase
inhibition
from
the
few
urinary
metabolite
data
available
for
human
children.

Fundamental
Model
Structure
One
Panel
member
made
an
observation
that
has
significant
implications
for
the
structure
of
the
carbaryl
model.
This
was
that
carbaryl
binding
to
the
active
site
of
acetylcholinesterase
(
AChE)
is,
essentially,
a
metabolic
destruction
step
for
the
carbaryl
molecule.
Even
if
the
cholinesterase
enzyme
molecule
itself
is
regenerated
by
hydrolysis
of
the
methylcarbamoyl
moiety
bound
to
the
active­
site
serine,
the
original
carbaryl
is
not
regenerated.
Simple
calculations
by
this
Panelist
suggested
that
this
might
not
be
the
primary
mode
of
metabolism
of
carbaryl,
but
it
could
make
a
significant
contribution
to
overall
metabolism.
If
the
contribution
is
appreciable
then
the
model
will
need
to
be
adapted.
Instead
of
locating
all
metabolism
of
carbaryl
in
the
liver
(
following
a
convention
in
PBPK
modeling
that
may
often
be
an
imperfect
reflection
of
reality)
the
model
should
provide
for
catalyzed
hydrolysis
of
carbaryl
in
all
compartments
where
AChE
and
butyrylcholinesterase
enzyme
activities
are
known
to
be
present­­
particularly
the
brain,
blood,
and
liver.

General
Conception
and
Execution
of
the
"
Fitting"
Problem
a)
The
issue
of
"
overparameterization"

There
is
a
well­
known
adage
that
represents
the
deceptively
high
quality
of
fits
where
there
are
few
data
and
many
parameters:
"
with
four
parameters
I
can
fit
an
elephant;
with
five
I
can
make
it
wave."
Actually,
a
formal
paper
on
this
subject
referred
to
in
the
Panel
discussion
suggests
that
many
more
than
four
parameters
are
required
to
create
a
mathematical
function
that
will
represent
a
drawing
of
an
elephant,
but
the
point
is
clear.
If
a
relatively
limited
data
set
is
used
to
estimate
the
values
of
many
parameters,
it
will
usually
be
possible
to
arrive
at
a
"
fit"
that
appears
excellent,
but
which
in
reality
has
46
of
113
no
predictive
power
because
the
curve
has
simply
been
made
to
arbitrarily
follow
the
data
points,
including
all
the
experimental
errors
in
the
data,
regardless
of
the
accuracy
of
the
implied
mechanistic
structure
of
the
mathematical
relationships
in
the
model.

PBPK/
PD
models
generally
have
many
more
parameters
than
can
be
usefully
estimated
with
available
data
sets
on
the
time
course
of
the
concentrations
of
parent
chemical
and
metabolites
in
different
places
in
the
body
and
excreta.
Therefore,
the
tradition
is
to
divide
parameters
into
two
subsets:

 
The
vast
majority
of
the
parameters
are
estimated
using
data
that
are
exogenous
to
the
model
and
are
thought
to
be
relatively
robust
whatever
conclusions
are
drawn
about
the
values
of
other
model
parameters.
These
exogenous
inputs
to
the
model
typically
include
organ
sizes,
blood
flow
rates,
alveolar
ventilation,
and
tissue/
blood
partition
coefficients
based
on
direct
measurements.
The
inputs
may
also
include
such
relatively
distinctive
elements
of
the
ERDEM
model
as
transit
rates
of
food
through
portions
of
the
gastrointestinal
system,
bile
and
urinary
flow
rates.

 
By
contrast,
a
few
types
of
parameters
are
usually
estimated
by
"
fitting"
or
"
calibrating"
those
model
parameters
so
that
model
outputs
correspond
to
some
set
of
time
course
concentration
data
collected
following
a
defined
in
vivo
experimental
administration
of
the
compound
to
animals
or
people.
This
second
set
of
parameters
typically
includes
Michaelis­
Menten
enzyme
kinetic
parameters
(
Vmax
and
Km
 
the
asymptotic
maximum
velocity
of
the
reaction
and
the
concentration
of
substrate
that
elicits
half
the
maximum
reaction
rate)
and
dermal
absorption
rates.

The
ERDEM
carbaryl
model
contains
more
parameters
of
the
fitted
category
than
usual,
in
part
because
the
modelers
tried
to
follow
the
whole
ensemble
of
carbaryl
metabolites
in
order
to
(
1)
use
calibrating
information
in
the
form
of
total
14C
levels
in
some
compartments,
and
urinary
outputs
of
specific
metabolites
for
rats,
and
(
2)
draw
inferences
about
internal
exposure
and,
indirectly,
brain
cholinesterase
inhibition
from
urinary
metabolite
data
in
exposed
children.
Several
Panelists
were
critical
of
the
ERDEM
modelers'
choice
to
adjust
the
brain/
blood
partition
coefficient
(
and
only
that
one
partition
coefficient)
to
achieve
better
agreement
between
observed
and
predicted
brain
concentrations,
in
part
because
this
converts
a
parameter
that
was
estimated
on
the
basis
of
exogenous
physical
chemical
information
into
one
that
is
fitted
via
the
model
to
specific
data
(
see
below
for
a
more
extended
discussion).

b)
The
need
for
more
formal
statistical
criteria
for
model
fitting
Many
of
the
Panelists
strongly
favored
a
more
formal
statistical
approach
to
model
fitting/
calibration
using
the
existing
data.
This
does
not
necessarily
imply
a
completely
equivalent
(
unweighted)
treatment
of
all
data
 
some
types
of
data
were
judged
by
different
Panelists
to
be
more
salient
for
risk­
related
predictions
than
others
(
see
below),
47
of
113
but
development
of
some
objective
and
statistically
defensible
procedure
was
widely
considered
to
be
an
important
aspect
that
should
be
considered
for
the
next
iteration
of
the
ERDEM
system.

c)
Model
"
validation"

"
Validation"
is
the
term
used
to
describe
a
process
whereby
some
predictions
from
a
model
are
compared
to
observed
data
that
were
not
used
in
the
original
calibration
of
model
parameters.
This
process
is
most
useful
when
there
are
relatively
large
amounts
of
data
available
to
estimate
relatively
few
parameters,
unlike
the
present
case
where
the
available
data
are
relatively
sparse
in
relation
to
the
number
of
parameters.

Some
Panelists
with
experience
in
PBPK/
PD
modeling
were
critical
of
"
validation"
as
a
term.
They
believe
that
it
implicitly
suggests
that
a
model
that
survives
a
comparison
with
independent
data
is
therefore
"
valid"
or
"
true".
In
fact,
while
such
a
successful
test
can
increase
confidence
in
model
predictions,
in
general
there
may
be
many
other
but
untested
models
that
would
have
equal
mechanistic
plausibility
and
not
be
incompatible
with
the
same
data.

At
least
one
of
these
Panelists
suggested
that
instead
of
"
validation",
modelers
describe
the
process
as
"
juxtaposition
of
model
predictions
and
data."
This
treats
the
data
and
the
model
predictions
on
a
more
even
footing.
In
fact,
when
model
predictions
and
data
are
incompatible,
the
data
can
be
wrong,
the
model
can
be
wrong,
or
both
can
be
wrong
in
varying
degrees.
Discrepancies
between
model
predictions
and
observations
present
the
modeler
with
a
need/
opportunity
to
reflect,
and
create
different
hypotheses
about
the
likely
source(
s)
of
the
discrepancies.
In
this
process
it
is
helpful
to
look
for
patterns
in
the
observed/
predicted
differences
that
can
suggest
the
mechanistic
origins
of
systematic
errors
in
model
structure
or
data.
Are
the
discrepancies
more
prominent
with
one
strain
or
gender
of
animals
than
another?
At
early
versus
late
time
points?
At
high
versus
low
doses?
In
one
tissue
more
than
another?
Each
of
these
types
of
patterns
can
suggest
hypotheses
that
can
be
the
starting
points
for
further
examination
with
other
information.
Such
hypotheses
also
can
be
used
in
sensitivity
analyses
to
assess
how
much
the
ultimate
risk­
related
result
might
be
changed
if
the
model
was
restructured
in
various
ways,
or
data
sources
were
weighted
in
ways
that
emphasized
the
least
problematic
or
most
likely
relevant
information
for
environmental
exposures.

d)
Group
average
versus
individual
values
of
model
parameters
One
Panelist
pointed
out
that
the
ERDEM
carbaryl
model
parameter
values
result
from
fitting
one
set
of
coefficients
to
an
average
response
from
replicate
animals
or
people.
However,
if
each
individual
has
a
unique
set
of
coefficients
for
model
parameters,
the
solution
of
averaged
parameter
values
will
not
generally
be
the
same
as
the
average
solution.
To
assess
this,
the
modeling
team
should
obtain,
present,
and
analyze
the
raw
data
for
individual
animals
if
possible.
The
modeling
team
may
eventually
need
to
generate
solutions
at
random
sets
of
correlated
coefficient
values
and
look
at
the
48
of
113
distribution
of
the
results
to
obtain
better
insights
into
the
effects
of
individual
variability
on
both
averages
and
variability
distributions
for
risk
related
outcomes.

Evaluation
of
the
Current
ERDEM
Model
Fits
and
Implications
of
the
Current
Results
a)
Which
observations
should
be
emphasized
in
model
calibration?

The
apparent
widespread
use
of
total
14C
levels
in
various
organs
for
the
purpose
of
calibration
drew
critical
comments
from
several,
but
not
all,
Panelists
because
this
represents
an
uncertain
and
varying
proportion
of
unchanged
carbaryl
and
metabolites
at
different
times
and
places
in
the
body.
Virtually
all
who
commented
preferred
to
emphasize
the
data
on
unchanged
carbaryl
for
model
calibration
as
being
more
causally
relevant
for
the
endpoint
of
cholinesterase
inhibition.
However,
one
Panelist
indicated
that
measurements
of
total
14C
were
likely
to
have
been
more
accurately
measured.

One
Panelist
also
emphasized
the
need
to
use
some
urinary
metabolite
information
to
help
calibrate
the
rat
model,
because
urinary
metabolite
data
for
human
children
represents
essentially
all
the
data
that
are
available
for
calibration
of
the
human
model.
In
response,
EPA
staff
members
indicated
that
information
on
the
total
metabolite
distribution
cumulated
at
long
times
after
exposure
were
an
important
part
of
the
calibration
of
the
relative
values
for
the
metabolism
components
for
specific
metabolites
in
the
rat
model.

Another
preference
that
was
widely
shared
among
the
Panelists
was
for
calibration
of
the
model
to
make
predictions
that
most
closely
correspond
with
observations
at
relatively
low
doses
that
are
closest
to
those
likely
to
be
experienced
in
people.
The
EPA
staff
presenter
for
the
ERDEM
model
expressed
a
similar
preference
but
the
system
he
showed
for
summarizing
the
model/
data
discrepancies
with
dose
as
a
linear
regression
fit
may
tend
to
implicitly
emphasize
discrepancies
at
relatively
high
exposure
levels.
Expression
of
discrepancies
as
the
weighted
sum
of
squares
of
log
deviations
of
observations
from
predictions
would
avoid
this
inadvertent
emphasis
on
the
fits
at
high
dose
levels.
More
sophisticated
likelihood
methods
(
again,
most
likely
based
on
a
log
transformation
of
observations
and
predictions)
could
similarly
be
used
to
specifically
evaluate
discrepancies
at
lower
dose
levels.

A
final
issue
was
the
choice
of
the
registrant's
Missouri
rather
than
the
California
data
for
drawing
modeling
inferences
about
human
exposures.
The
reason
given
for
this
was
that
the
Missouri
data
reflected
adherence
to
the
label
directions
for
application
of
the
carbaryl­
containing
products.
This
may
be
a
reasonable
policy­
related
consideration
in
the
evaluation
of
the
data
for
purposes
of
re­
registration
decisions.
However
for
purposes
of
creating
a
model
descriptive
of
actual
human
exposures
the
Panel
suggested
including
the
California
data.
If
use
at
greater
than
recommended
rates
occurred
within
the
registrant's
49
of
113
own
controlled
experiment,
it
is
at
least
plausible
that
similar
departures
from
ideal
application
occur
in
the
practical
use
of
carbaryl­
containing
products
in
the
much
less
controlled
settings
of
residential
use
by
consumers
and
perhaps
also
commercial
applicators.
Moreover,
the
California
data
could
provide
quantitative
information
on
the
responses
of
urinary
biomarkers
to
exposures
over
a
much
broader
range
of
exposures
 
allowing
better
calibration
and
testing
of
model
predictions
and
inputs
such
as
assumed
human
metabolism
"
Km"
s
that
would
otherwise
be
assumed
to
be
identical
to
estimates
from
the
animal
data.

b)
Anomalies
in
the
current
fitting,
and
implications
for
adapting
model
inputs
and
refitting
The
Panel
agreed
that
the
model
developed
has
promise,
as
outlined
in
a
summary
slide,
to
 
Provide
structure
to
evaluate
the
effects
of
diverse
data
on
inferred
risk
 
Efficiently
evaluate
different
exposure
scenarios
 
Identify
factors
affecting
dose
However,
the
Panel
had
difficulty
evaluating
the
quality
of
such
outputs
without
a
better
understanding
of
the
source
of
all
the
parameters
put
into
the
model,
and
improvements
to
reconcile
many
of
the
model/
data
departures
that
are
apparent
in
the
Appendix.

Discussion
in
an
earlier
subsection
emphasized
the
need
to
reflect
on
patterns
in
model/
data
discrepancies
to
form
mechanistic
hypotheses
about
the
source(
s)
of
possible
systematic
errors.
In
doing
this,
the
modeler
also
asks:
(
1)
which
parameter
estimates
are
more
vulnerable
to
larger
changes
because
of
experimental
variability
or
alternative
interpretations
(
e.
g.
alternative
metabolite
compositions
of
total
14C)
and
(
2)
which
parameters
are
key
to
the
determination
of
the
ultimate
assessment
parameters
that
affect
risk
(
e.
g.
unchanged
carbaryl
versus
metabolite
concentrations).
The
conclusions
from
the
first
reflection
process
help
identify
which
parameters
should
be
allowed
to
vary,
if
they
are
influential
in
affecting
the
model
fit
to
observed
information.
The
decision
to
radically
adjust
the
brain
partition
coefficient,
and
only
that
one,
is
one
of
the
more
questionable
choices
in
model
development
in
the
opinion
of
several
Panelists.
The
results
of
the
current
data
contain
interesting
indications
that
perhaps
errors
in
estimating
other
partition
coefficients
may
be
contributing
to
the
excessive
model
predictions
of
brain
carbaryl
concentrations
that
caused
the
modelers
to
drastically
reduce
the
predicted
brain/
blood
partition
coefficient
from
what
was
predicted
on
physiochemical
grounds.

Many
of
the
data/
model
fits
are
not
as
good
as
they
should
be
for
a
final
model.
In
particular,
the
model
seriously
overestimates
brain
14C
concentrations
and
somewhat
overestimates
brain
and
blood
14C
(
Figures
E2
and
E3,
p.
130
in
EPA
draft
document
"
Assessment
of
Carbaryl
Exposure
Following
Turf
Application
Using
a
Physiologically
50
of
113
Based
Pharmacokinetic
Model").
On
the
other
hand,
the
model
(
after
the
suspect
adjustment
of
the
brain/
blood
partition
coefficient)
does
better
at
estimating
the
brain
carbaryl
concentration
but
if
anything
underestimates
brain
carbaryl
at
short
times
after
exposure.
There
are
huge
discrepancies
between
data
and
model
predictions
for
blood
naphthol
and
naphthol
sulfate
concentrations
(
Figures
E10
and
E11,
p.
134).
Brain
and
fat
carbaryl
concentration
kinetics
also
show
appreciable
differences
between
data
and
model
predictions
(
Figures
E18
and
E19,
p.
138).
Peak
brain
carbaryl
(
Figure
E18)
is
under
predicted
by
two­
fold.
In
the
case
of
fat,
the
time
of
occurrence
of
peak
levels
(
Figure
E19)
indicates
some
fundamental
problem
with
the
partitioning
or
effective
flow
rates
or
blood
brain
transfer
kinetics.
Similarly,
in
Figure
E22
(
p.
140)
it
can
be
seen
that
the
predicted
brain
napthol
concentrations
underestimate
the
observed
data
by
about
sevenfold
Liver
and
blood
napthol
comparisons
are
off
by
something
like
ten­
fold
(
E24,
p.
141).
Therefore,
improvements
in
the
model
for
rats
are
needed
before
extending
it
to
make
inferences
about
cholinesterase
inhibition
from
urinary
metabolite
data
in
humans.

Of
these
anomalies,
the
one
that
is
particularly
likely
to
be
salient
for
alternative
model
development,
in
the
view
of
one
Panel
member,
is
the
apparent
pattern
of
delay
in
the
observed
time
to
peak
levels
of
carbaryl
in
the
fat
relative
to
that
predicted
by
the
current
model
(
Figure
E19,
p.
138).
This
suggests
one
or
two
possible
underlying
differences
between
reality
and
the
model
inputs:
(
a)
The
estimated
fat/
blood
partition
coefficient
is
too
low
and/
or
(
b)
the
rate
of
transfer
from
the
blood
to
the
fat
is
slower
than
expected
from
perfusion­
limited
transport
due
to
a
relatively
slowly­
reversible
binding
of
carbaryl
in
the
blood.
Binding
in
the
blood
would
tend
to
lower
the
effective
brain/
blood
partition
coefficient
at
longer
times
after
oral
exposure
and
also
might
contribute
to
an
explanation
of
the
failure
(
commented
on
by
several
Panelists)
to
measure
detectable
levels
of
carbaryl
in
the
blood.

To
begin
to
explore
this
issue,
one
Panelist
utilized
his
own
data
base
of
rat
and
human
tissue/
blood
partition
coefficients
and
regression
model
formulae
(
previously
used
for
a
few
different
published
PBPK
models
(
Ginsberg
et
al.,
1996;
Ginsberg
et
al.,
2004;
Walker
et
al.,
2004))
to
estimate
the
fat/
blood
partition
coefficients
and
the
half­
life
for
perfusion
based
removal
of
carbaryl
from
fat
in
rats
and
humans
based
on
the
same
octanol/
water
partition
coefficient
for
carbaryl
used
by
the
ERDEM
authors.
(
Spreadsheets
containing
these
data
bases
and
the
formulae
for
tissue/
blood
partition
coefficient
estimation
are
available
in
the
EPA
OPP
Docket
(
Docket
Number:
OPP
2005­
0405).
The
results
are
presented
in
Table
1.
It
can
be
seen
that
the
alternative
estimates
of
the
fat/
blood
partition
coefficients
are
about
twice
as
large
as
the
estimates
currently
used
in
the
rat
model
in
ERDEM,
and
about
three
times
as
large
in
the
case
of
humans.
This
leads
to
expectations
of
considerably
longer
elimination
half
lives
for
carbaryl
in
fat
 
in
the
direction
needed
to
at
least
partially
account
for
the
model/
data
difference
for
the
fat
compartment
seen
in
the
rat
model
used
in
ERDEM.
It
is
suggested
that
the
set
of
alternative
partition
coefficients
could
be
used
as
an
alternative
starting
point
for
sensitivity
analysis
for
the
ERDEM
carbaryl
modeling
exercise.
51
of
113
Table
1
Comparative
Estimates
of
Fat/
Blood
Partition
Coefficients
and
Elimination
Half
Lives
for
Carbaryl
in
Fat
in
the
Current
ERDEM
Model
versus
an
Alternative
Regression­
Based
Method
Species
Body
Weight
(
kg)
Fat
Volume
(
L)
Fat
Perfusion
(
L/
Hour)
Fat/
Blood
Partition
Coefficient
Source
of
Fat/
Blood
Partition
Coefficient
Elim.
Rate
Constant
from
Fat
(
1/
hr)
by
Perfusion
Half
Life
of
Carbaryl
in
Fat
(
hr)
Half
life
of
Carbaryl
in
Fat
(
min)

Rats
0.25
0.015
0.51
17.1
ERDEM
2.0
0.35
21
Rats
0.25
0.015
0.51
33.9
Hattis
rat
regression
model
1.0
0.69
41
Humans­­
9
yr
20.4
2.35
16.9
17.1
ERDEM
0.42
1.7
99
Humans­­
9
yr
20.4
2.35
16.9
63.6
Hattis
rat
regression
model
0.11
6.1
368
Issue
2.2:
PBPK/
PD
Model
Fidelity
The
PBPK/
PD
model
was
developed
in
the
Exposure
Related
Dose
Estimating
Model
(
ERDEM)
platform.
The
ERDEM
platform
is,
by
design,
highly
structured
and
flexible
for
adaptation
to
new
or
emerging
exposure
and
risk
assessment
needs.
In
PBPK/
PD
modeling,
there
is
a
need
to
balance
completeness
regarding
anatomical/
physiological
pathways/
routes
with
the
desire
for
model
simplicity.
The
model
is
required
to
simulate
the
relevant
dose
metrics
and
provide
the
capability
to
extrapolate
from
the
laboratory
setting
to
exposure
scenarios
of
interest.
Modeling
runs
with
the
ERDEM
platform
are
typically
short;
thus
computational
time
is
not
an
issue.
The
important
consideration
is
the
in
silico
representation
of
the
species
and
the
connection
with
pathways
and
routes
of
exposure.

Question
2.2a
Please
comment
on
the
carbaryl
PBPK/
PD
model
structure
for
evaluating
diverse
exposure
scenarios,
including
the
exposure
to
children
on
the
turf
described
in
the
report.
Please
include
in
your
comments
a
consideration
of
the
degree
to
which
the
compartments
included
in
this
model
reasonably
describe
the
PK
and
PD
characteristics
of
carbaryl
and
52
of
113
provide
the
ability
to
extrapolate
the
model
across
species
and
scenarios,
balanced
against
model
simplicity.

Question
2.2b
As
more
PK
and
PD
data
become
available,
the
model
structure
from
this
application
may
be
applied
to
other
N­
methyl
carbamates,
including
mixtures.
Please
comment
on
the
suitability
of
the
carbaryl
specific
PBPK/
PD
model
structure
as
developed
in
the
ERDEM
platform
for
expansion
to
include
other
N­
methyl
carbamates.

Panel
Response
(
to
Question
2.2
a
&
b)

The
Panel
commends
the
Agency
and
its
scientists
who
are
building
this
comprehensive
PBPK/
PD
model
for
their
work
which
has
lead
to
an
important
milestone
in
the
utility
of
mechanism­
based
models.
This
model
is
relevant
in
the
present
context
for
cumulative
risk
assessment
and
will
be
important
for
doing
biologically­
based
risk
assessments
for
single
compounds.
The
model
has
achieved
a
level
of
maturation
where
all
of
the
relevant
parts
and
modules
are
publicly
available
and
ready
for
open
scientific
scrutiny.
This
model
can
be
applied
to
the
estimation
of
biomarkers
of
exposure
and,
to
a
lesser
extent,
biomarkers
of
effect.

At
this
stage
of
model
development,
it
is
important
to
consider
the
applicability
and
acceptability
of
the
model.
The
model
is
extremely
complex
and
the
Panel
expressed
concern
that
this
level
of
complexity
may
not
have
sufficient
support
in
the
available
data.
Some
modules
of
the
model
are
structurally
identical
(
e.
g.
those
used
as
simple
storage
compartments)
and
could
be
considered
as
sufficiently
similar
to
be
treated
with
one
generic
differential
equation
thereby
reducing
the
apparent
complexity.
A
second
means
of
reducing
apparent
complexity
would
be
to
group
modules
of
similar
influence
on
the
AChE
levels
(
e.
g.
in
response
to
the
results
of
a
sensitivity
analysis).
Finally,
the
Agency
may
want
to
consider
grouping
some
of
the
different
metabolic
rates
to
reduce
the
need
for
2
parameters
for
each
metabolite.

However,
some
changes
are
needed
that
might
increase
complexity.
Dermal
absorption
seems
to
be
inadequately
modeled.
A
portion
of
carbaryl
absorbed
by
the
skin
and
stored
in
the
dermis
may
not
be
readily
available
for
distribution
in
the
circulation.
This
is
evident
by
the
results
in
Figure
12
on
page
36
of
the
draft
document
"
Assessment
of
carbaryl
exposure
following
turf
application
using
a
physiologically
based
pharmacokinetic
model"
provided
by
the
Agency
to
the
Panel,
in
which
the
ERDEM
model
overestimates
the
peak
total
amount
of
C14
in
blood
by
a
factor
of
2.5.
An
additional
differential
term
describing
the
release
of
carbaryl
from
dermis
to
blood
may
need
to
be
added
to
this
model.
Other
results
from
the
simulation
for
simultaneous
oral
and
dermal
exposure
to
carbaryl
all
show
some
degree
of
differences
between
model
prediction
and
the
observations.
The
Dermal
Thin
Film
technology,
which
is
being
53
of
113
developed
in
ORD,
also
could
be
used
to
investigate
the
extent
of
carbaryl
or
N­
methyl
carbamate
retention
in
dermis
and
the
rate
of
release
to
blood.

The
other
missing
component
in
the
ERDEM
model
is
a
description
of
how
the
baseline
AChE
level
in
brain
tissue
is
incorporated.
The
current
approach
of
using
animal
baseline
AChE
activities
as
the
input
levels
is
inadequate
and
may
be
wrong.
The
fundamental
limitation
is
that
baseline
AChE
brain
level
in
humans
of
all
ages
is
hard
to
obtain.
This
is
why
the
SAP
encourages
EPA
and
the
registrants
to
expand
the
focus
on
building
PBPK/
PD
model
to
include
both
brain
tissue
and
red
blood
cells
(
RBC).
As
long
as
the
RBC
AChE
inhibition
can
be
modeled
appropriately
and
adequately
in
the
PBPK/
PD
model,
the
brain
tissue
AChE
inhibition
can
then
be
predicted
with
a
much
better
accuracy.

Finally,
on
model
changes,
the
model
appears
to
treat
AChE
inhibition
as
a
reversible
binding
mechanism
when
in
fact,
it
is
more
of
a
metabolism
mechanism;
this
needs
to
be
corrected
in
later
versions
of
the
model.

The
runtime
version
of
ERDEM
was
made
available
by
the
Agency
and
it
was
possible
to
execute
this
version
without
problems.
It
produces
a
logfile
where
all
relevant
information
on
the
set
up,
the
performance
and
the
results
of
the
program
can
be
inspected.
This
wealth
of
information
requires
answers
to
several
questions:

 
How
will
the
model
output
be
organized?
 
Which
tables
and
which
graphs
of
the
numerical
results
will
be
stored
in
which
format
in
which
directory
of
the
user's
PC?
Will
these
be
files
that
can
be
easily
edited?
 
How
will
the
model
characteristics
be
made
available
so
that
modeling
can
be
replicated?

The
model
seems
to
be
general
enough
to
cover
the
PBPK/
PD
modeling
of
single
carbamates.
However,
it
is
difficult
for
the
Panel
to
discuss
additions
or
subtractions
to
this
model
at
this
point
because
it
has
not
been
applied
rigorously
and
to
a
broader
class
of
agents.
As
long
as
other
NMCs
behave
like
carbaryl,
this
carbaryl
PBPK/
PD
model
should
be
able
to
be
used
for
other
NMCs
and
mixtures.
Extra
steps
would
need
to
be
added
for
NMCs
which
requires
metabolic
activation
and
for
those
that
have
more
than
one
active
metabolite
(
e.
g.,
aldicarb).
Caution
should
be
used
when
the
components
of
the
PD
model
are
used
for
other
NMCs
without
additional
data.
The
major
concern
is
that
the
reactivation
time,
or
rate
of
AChE
inhibition,
will
vary
by
NMC.
The
other
concern
is
that
AChE­
inhibition
by
NMCs,
rather
than
by
carbaryl
alone,
is
assumed
to
be
dose­
additive.
While
additional
animal
studies
are
being
conducted
to
determine
whether
this
assumption
is
valid,
caution
should
again
be
taken.
The
Panel
encourages
the
Agency
to
evaluate
how
the
model
performs
when
all
seven
carbamates
and
their
metabolites
are
modeled
simultaneously
using
different
exposure
routes.
54
of
113
The
ERDEM
model
excludes
a
large
part
of
the
pharmacodynamics
(
PD)
of
carbamate
toxicity.
Documents
provided
by
the
Agency
indicate
that
other
PD
endpoints
are
currently
under
investigation
(
e.
g.
motor
activity)
which
may
represent
more
advanced
health
endpoints
than
AChE
inhibition.
As
the
Agency
develops
the
ERDEM
model,
they
are
encouraged
to
pursue
these
other
PD
endpoints.

The
model
uses
various
allometric
scaling
factors
for
adjusting
model
parameters
including
the
determination
of
respective
parameters
for
young
children.
Price
et
al.
(
2003)
have
collected
a
series
of
equations
for
estimating
organ
and
tissue
volumes
using
National
Health
and
Nutrition
Examination
Survey
(
NHANES)
anthropometric
data.
It
was
not
clear
from
the
documents
and
the
presentation
whether
those
equations
have
been
used
for
the
model,
and
if
not,
why
they
have
been
excluded.

The
terms
and
techniques
being
used
by
the
Agency
for
predictions
and
data
comparisons
at
the
99.9
percentile
(
e.
g.
Figure
20
of
the
aforementioned
Agency
draft)
need
clarification.
The
model
predictions
do
not
appear
to
be
99.9%
predictions
of
cumulative
mass
but
are
instead,
based
on
a
highly
unlikely
exposure
event.
The
analysis
does
not
appear
to
include
other
uncertainties
and
variabilities
in
the
model
that
might
impact
the
predictions.
Moreover,
comparing
model
predictions
to
data
at
the
tails
of
the
distribution
is
questionable
at
this
time.
The
research
going
into
developing
the
PBPK/
PD
model
is
based
largely
on
general,
aggregated
observations.
Until
the
model
has
been
exercised
in
greater
detail,
the
Panel
suggests
focusing
the
discussion
on
predictions
at
the
median
and
within
the
interquartile
range.

An
additional
comment
was
made
regarding
the
priority
for
developing
the
structure
ERDEM.
One
Panel
member
suggested
that
an
additional
component
be
added
to
ERDEM
for
modeling
fetal
in
utero
exposures,
with
the
ability
to
model
endpoints
at
target
tissues,
such
as
the
fetal
brain.

Issue
2.3:
Statistical
Model
Evaluation
Considerations
Development
of
PBPK/
PD
models
is
an
iterative
process
such
that
the
model
is
improved
and
revised
as
more
data
and
information
become
available.
In
a
regulatory
setting,
it's
not
unusual
that
model
development
begins
before
all
data
sets
have
been
collected.
Currently
when
using
ERDEM,
an
initial
model
structure
is
developed
based
on
the
species
physiology
and
known
chemistry
of
the
chemical
and
metabolites.
The
initial
model
structure
consists
of
the
differential
equations
and
variables
that
correspond
to
the
relevant
compartments
and
metabolic
transformations.
The
initial
parameter
values
are
estimates
made
by
the
researcher,
often
based
on
models
of
related
chemicals.
The
model
is
considered
provisional
until
the
available
data
are
evaluated.
The
parameter
values
and
model
structure
are
then
updated
to
reflect
the
available
data.
As
new
data
are
made
available,
they
are
evaluated
concurrently
with
the
existing
data
against
the
model
simulations,
and
the
model
is
revised
accordingly.
This
iterative
process
has
been
followed
for
the
55
of
113
current
carbaryl
PBPK/
PD
model,
where
model
evaluation
was
based
on
visual
inspection
and
linear
regression
between
the
model
results
and
data
points
(
not
included
in
the
report).

Question
2.3
Please
comment
on
statistical
or
mathematical
analyses
which
could
inform
the
need
for
model
revisions
as
new
data
are
made
available.

Panel
Response
The
Panel
discussion
on
this
question
centered
on
the
following
three
issues:

 
the
nature
and
use
of
data
for
determining
model
complexity
and
parameter
estimation,
 
approaches
to
assessing
goodness­
of­
fit
of
the
model,
and
 
the
need
for
assessing
model
complexity
and
parameterization
sensitivity.

Model
complexity
and
parameter
estimation
The
example
model
presented
to
the
Panel
is
complex,
with
over
80
parameters:
25
physiological
parameters,
24
physiochemical
parameters,
35
biochemical
parameters
and
10
inhibition
and
recovery
rate
parameters.
Norm
settings
for
these
parameters
are
provided
in
Tables
3
to
9
of
the
draft
report
"
Assessment
of
carbaryl
exposure
following
turf
application
using
a
physiologically
based
pharmacokinetic
model."
The
response
data,
dose
metrics,
available
to
assess
the
goodness
of
model
predictions
are
a
small
dataset
provided
in
Table
1.
Figures
7
to
19
demonstrate
that
the
model
when
evaluated
at
the
norm
settings
of
the
80
parameters
tends
to
follow
the
observed
pattern
of
recovery,
sometimes
overestimating
the
amount
of
14C
radiation
or
the
amount
of
carbaryl,
etc
and
other
times
underestimating.
From
a
statistical
point
of
view,
the
model
is
used
to
describe
a
complex
set
of
multivariate
data
and
hence
it
is
not
surprising
that
some
over­
and
underestimation
occurs.

The
model
is
not
parameterized
to
a
specific
individual,
nor
does
it
use
a
global
parameter
estimation
method,
such
as
non­
linear
least
squares,
maximum
likelihood
or
other
statistical
approach.
Some
Panel
members
were
not
surprised
by
this,
suggesting
that
the
typical
model
building
approach
is
to
determine
parameter
values
for
individual
compartments
using
available
data
for
these
compartments.
Other
Panel
members
felt
that
more
formal
model­
building
and
parameter
estimation
should
have
been
displayed.

The
Panel
discussed
the
model
construction
and
parameter
estimation
process
in
some
detail.
Two
approaches
are
available
for
parameter
estimation.
In
the
global
approach,
the
whole
problem
is
formulated
as
a
large,
multidimensional
search
for
the
values
of
the
parameters
that
minimize
the
residual
sums
of
squared
deviations
between
56
of
113
the
model
predictions
and
any
available
data.
This
can
be
done
with
or
without
constraints
on
the
parameter
values.
Unfortunately,
with
highly
parameterized
models
and
little
data,
model
fit
can
be
amazingly
good,
but
the
fitted
model
may
be
relatively
useless
for
prediction
purposes.
This
is
not
unlike
the
phenomenon
of
fitting
higher
order
polynomials
to
limited
data;
at
some
point
the
flexibility
of
the
model
allows
the
fitted
line
to
perfectly
intersect
all
available
data
points.

As
an
alternative
to
a
global
approach,
least
squares
methods
may
be
employed
in
a
piece­
wise
manner.
Each
parameter
or
related
set
of
parameters
is
treated
as
a
separate
analysis.
This
approach
should
make
it
easier
to
incorporate
constraint
information
and
utilize
or
test
intermediate
variable
values.
The
documents
provided
to
the
Panel
discuss
model
fits
for
various
compartments
for
the
parent
compound
as
well
as
the
metabolites
with
various
endpoints.
It
was
pointed
out
that
when
a
model
is
constructed
or
parameterized
by
successively
incorporating
information
from
different
data
sets
with
adjustments
made
to
model
parameter
values
with
each
addition,
the
order
in
which
the
data
are
incorporated
and
the
changes
made
will
influence
the
final
model
form
and
associated
parameter
values.
This
is
not
unlike
what
happens
using
stepwise
regression
in
the
presence
of
mild
co­
linearity.
The
strategy
of
model
building,
model
refinement
and
parameter
estimation
are
all
interrelated;
changes
in
any
one
affects
the
other
and
together
influence
the
final
model
and
its
predictability.
Therefore,
transparent
model
building
should
include
keeping
track
of
what
data
are
used
to
determine
settings
for
each
parameter.
A
chronology
of
model
adaptations
also
should
be
available
to
reviewers.

An
intermediate
to
the
piece­
wise
approach
and
the
global
approach
is
to
fit
the
model
by
using
the
data
to
inform
us
on
only
a
couple
of
parameters
on
any
particular
run.
This
reduces
the
dimensionality
of
the
fitting
process
by
making
the
fit
of
the
parameters
conditional
on
the
other
parameters
being
held
fixed
at
their
current
"
confirmed"
values.
This
is
a
useful
approach
when
using
data
from
many
different
experiments.
In
a
sense
one
calibrates
the
model
for
each
set
of
data
and
examines
the
predictions
from
the
fitted
model
from
each
set.
This
approach
can
illuminate
when
additional
information
will
be
useful
in
improving
model
predictions.

The
Panel
also
discussed
the
more
difficult
issues
of
the
dimensionality
and
complexity
needed
in
the
model.
It
seems
that
the
goal
of
a
PB/
PK
model
is
to
mimic
reality
and
integrate
current
understanding
of
carbaryl­
related
processes.
Compartments
and
parameters
in
the
model
are
essentially
there,
based
on
theoretical
justifications
and
not
necessarily
because
of
their
ability
to
be
directly
observed
and/
or
calibrated
using
achievable
observations.
Some
compartments
and/
or
parameters
may
have
only
indirect
relationships
to
measurable
outcomes
and
as
such
are
only
slightly
correlated
with
these
outcomes.
This
model
allows
us
to
ask
"
what
if"
questions
whose
answers
cannot
be
directly
derived
from
observations.
A
number
of
Panel
members
felt
that
the
PB/
PK
model
should
be
used
as
a
check
against
simpler
models
that
can
be,
and
are,
calibrated
with
existing
data
to
important
outcomes.
It
was
felt
that
there
is
still
the
need
to
collate
the
relevant
empirical
evidence,
conduct
appropriate
parameterizations
that
pass
statistical
57
of
113
muster,
and
express
the
results
in
a
way
that
allows
us
to
characterize
uncertainty
in
the
model
predictions.
Reference
was
made
to
comparing
the
PBPK/
PD
model
results
with
the
results
from
the
simpler
and
empirically­
based
PK
model
discussed
in
a
preceding
session
(
Session
1:
Issues
related
to
cumulative
hazard
assessment).
Comparisons
such
as
these
offer
insights
into
the
amount
of
model
complexity
that
is
necessary
to
describe
relevant
outcomes.

Characterization
of
prediction
uncertainty
for
the
full
PBPK/
PD
model
is
difficult
because
there
are
both
model
form
uncertainties
as
well
as
parameter
uncertainties.
For
example,
there
was
discussion
relating
to
the
situation
where
the
compartment­
specific
data
suggest
one
value
for
a
parameter
(
e.
g.,
dermal
permeability
coefficient),
but
the
parameter
value
had
to
be
dramatically
changed
from
this
data­
derived
value
in
order
to
get
a
better
global
model
fit.
At
least
one
Panel
member
felt
that
this
is
exactly
the
kind
of
situation
that
would
justify
the
need
to
reexamine
not
just
the
particular
assigned
parameter
value
but
also
the
form
of
the
compartment
model.
Contradictions
such
as
these
lead
to
increased
scientific
understanding
and
better
final
models.
This
also
brought
up
the
issue
of
how
new
information
is
used
to
modify
the
model
and/
or
model
parameters.
There
seems
to
be
a
need
for
protocols
to
help
decide
when
new
data
improve
on
existing
data.
At
least
one
Panel
member
suggested
that
human
data
are
preferred
over
animal
data,
in
vivo
data
preferred
over
in
vitro
data
and
that
direct
observations
be
preferred
over
derived
values.
Incoming
new
data
also
could
be
investigated
with
a
series
of
models
of
increasing
complexity
(
again,
in
a
manner
similar
to
what
is
done
in
model
building
in
multiple
regressions).

Some
Panel
members
belief
that
a
combination
of
model
simplification
(
judicial
use
of
Occam's
Razor)
and
formal
parameter
estimation
via
traditional
least
squares
or
Bayesian
methods
(
e.
g.
formal
true
Bayesian
or
hierarchical
Bayesian
updating
techniques,
Gelman
et
al
1996;
Bois,
2000)
will
result
in
a
model
whose
predictions
are
easier
to
evaluate.
Many
would
appreciate
seeing
how
various
model
simplification
approaches
would
affect
the
final
model
fit,
e.
g.
stripping
off
parts
of
a
model,
combining
compartments
or
eliminating
compartments.

For
a
given
model
formulation
and
where
possible,
parameter
estimates
should
be
accompanied
by
estimated
standard
errors
and
model
predictions
presented
with
associated
prediction
intervals.
It
was
noted
that
the
confidence
intervals
presented
in
Table
C2
(
p.
103)
and
illustrated
in
Figure
C1
(
p.
102)
(
draft
document
"
Assessment
of
carbaryl
exposure
following
turf
application
using
a
physiologically
based
pharmacokinetic
model")
are
confusing;
reporting
lower
limits
that
are
negative
and
plotting
only
the
upper
limits.
It
was
not
clear
if
one­
or
two­
sided
intervals
are
used,
nor
was
it
clear
which
are
desired.

Also,
it
was
noticed
that
much
of
the
data
used
for
parameter
estimation
is
censored
in
some
way.
Traditional
least
squares
methodology
is
not
directly
applicable
in
this
case,
and
how
censored
values
are
handled
can
have
a
dramatic
effect
on
estimates
58
of
113
and
overall
model
fit.
The
more
formal
statistical
estimation
approaches
may
require
censored
data
likelihoods
or
weighted
likelihoods
which
are
quite
difficult
to
handle
properly.
The
problem
is
multiplied
when
one
is
attempting
to
handle
multiple
datasets
in
the
process.

Goodness­
of­
fit
Goodness­
of­
fit
of
the
model
is
highly
related
to
the
complexity
of
the
model
and
parameter
estimation.
For
this
reason,
the
discussion
on
goodness­
of­
fit
was
interspersed
throughout
the
Panel
deliberations.

The
applicability
of
statistical
goodness­
of­
fit
methods
to
assess
the
adequacy
of
model
predictions
is
limited
by
the
complexity
of
the
PBPK/
PD
model,
the
type
of
data,
the
limited
amount
of
data
available
to
fit
the
model,
and
the
aims
of
the
modeling.
Several
views
were
expressed
among
the
Panel
members
on
the
degree
to
which
statistical
approaches
can
and
should
be
used.

Some
Panel
members
preferred
a
more
qualitative
assessment
of
model
goodnessof
fit,
offering
that
adequacy
might
be
judged
by
answers
to
the
following
questions.

 
Does
the
model
capture
the
main
points
of
the
biological
processes?
 
Does
the
model
adequately
describe
available
outcomes
data
for
important
dose
levels
and
time
ranges?
 
What
can
be
learned
by
fitting
the
model
to
each
of
the
available
data
sets?
 
Which
changes
to
model
parameters
are
needed
to
improve
the
fit
in
a
specific
key
compartment
and
do
these
changes
result
in
deterioration
of
the
fit
in
other
compartments?

A
number
of
suggestions
were
made
for
more
quantitative
ways
to
assess
model
goodness­
of­
fit.

 
If
least
squares
methodology
is
used
to
estimate
model
parameters,
the
residual
mean
square
can
be
used
to
assess
model
fit.
Modifications
to
this
approach
include
the
Akaike
Information
Criterion
(
AIC)
which
joins
the
estimated
mean
square
error
to
the
number
of
parameters
actually
estimated
in
the
model.
 
Evaluate
model
complexity
by
building
a
set
of
nested
sub­
models
by
systematically
removing
or
simplifying
compartments.
Again,
residual
mean
squares,
AIC
or
other
criterion
can
then
be
compared,
either
informally
or
via
formal
statistical
tests
where
available,
to
assess
the
significance
of
differences
between
full
and
reduced
models.
 
Implement
dynamic
graphics
which
would
allow
visualization
of
the
changes
in
model
fit
produced
when
one
or
a
small
number
of
parameters
change
their
values
continuously
(
movie­
type
visualization
of
model
sensitivity).
59
of
113
 
Utilize
some
form
of
the
Press
statistic
(
(
)
2
Pr
 
=
 

ess
i
i
i
g
y
y
)
to
evaluate
the
goodness
of
model
predictions
to
data
not
used
to
calibrate
the
model.
 
Utilize
other
criteria,
such
as
the
minimum
absolute
distance
between
observed
and
predicted
values
(
MAD,
 
=
 

MAD
i
i
i
g
y
y
),
or
curve
comparison
statistics
like
an
Anderson­
Darling
distance
to
measure
goodness­
of­
fit.
 
Consider
weighting
some
times
more
than
others
in
the
goodness­
of­
fit
criteria
and
evaluate
how
this
changes
model
fits
(
e.
g.,
a
weighted
Press
statistic,
(
)
2
 
=
 

WP
i
i
i
i
g
w
y
y
).

 
If
a
likelihood
function
can
be
created,
then
likelihood
goodness­
of­
fit
statistics
and
associated
tests
become
available.

Establishing
a
quantitative
criterion
for
assessing
level
of
model
fit
is
important
in
determining
at
what
point
one
achieves
acceptable
model
fit.
What
is
needed
to
set
this
value
is
the
sampling
distribution
of
the
goodness­
of­
fit
statistic.
It
is
unlikely
that
this
will
be
analytically
available
and
hence
some
form
of
Monte
Carlo
simulation
study
will
be
needed
to
develop
this
distribution.
Goodness­
of­
fit
statistics
are
available
to
assess
not
just
the
overall
goodness­
of­
fit
of
the
model
to
observed
outcomes,
but
can
be
used
to
identify
where
lack
of
fit
occurs
e.
g.
does
the
model
underestimate
brain
concentrations,
overestimate
blood
levels,
etc.

The
entire
discussion
by
the
Panel
emphasized
the
need
in
this
study
to
identify
the
critical
end
points
for
the
model.
It
was
not
clear
whether
it
was
more
important
for
the
model
to
correctly
predict
peak
concentrations,
times
at
which
peak
concentrations
were
reached,
peak
or
times
for
specific
compartments
(
brain
or
blood)
or
some
other
aggregate
measure
of
prediction
fit
(
as
mentioned
above).
Some
Panel
members
were
not
certain
that
fits
on
the
temporal
scale
were
best.
Other
Panel
members
were
concerned
with
whether
model
fits
should
be
made
on
untransformed
concentration
values
or
log
transformed
concentrations.
Model
fits
to
untransformed
concentrations
tends
to
allow
large
deviations,
typically
at
higher
concentrations
to
dominate
the
goodness­
of­
fit
statistic.
Assuming
multiplicative
errors
and
natural
log
scaling
allows
better
fits
for
low
doses
and
helps
avoid
situations
where
parameter
estimates
are
inappropriately
estimated
as
negative.

There
was
some
discussion
about
choice
of
the
settings
for
the
human
adult
model.
Typically
such
models
are
designed
for
20
year
old
males
assuming
6%
body
fat.
The
Agency
may
wish
to
utilize
NHANES
data
or
the
findings
of
the
Price
et
al
(
2003)
study
to
define
a
more
realistic
typical
individual.

Sensitivity
analysis
The
ERDEM
PBPK/
PD
model
is
a
complex
system
that
involves
many
60
of
113
interdependent
functional
relationships.
Estimates
or
assigned
values
for
many
parameters
of
this
system
have
varying
degrees
of
current
empirical
support
and
all
are
subject
to
a
degree
of
natural
variability
as
well
as
"
reducible"
uncertainty
for
specific
applications.
There
was
general
consensus
among
the
Panel
of
the
need
to
perform
a
sensitivity
analysis
for
the
PBPK/
PD
model.
A
limited
sensitivity
analysis
was
done
but
a
more
extensive
analysis
needs
to
be
performed.
In
general,
it
was
felt
that
there
may
be
only
a
few
parameters
that
correlate
with
model
predictions.
A
common
theme
of
the
discussion
was
that,
in
general,
researchers
work
very
diligently
to
build
the
best
model
they
are
capable
of
doing
and
then
maintain
skepticism
in
the
model
outputs.
A
sensitivity
analysis
is
one
way
this
is
done.

Sensitivity
analyses
typically
include
an
assessment
of
the
effect
of
parameter
uncertainty
on
model
predictions.
This
is
typically
assessed
by
applying
prior
distributions
to
parameters
and
assessing
the
degree
to
which
changes
in
parameter
values
correlate
with
changes
in
model
predictions.
With
highly
parameterized
models
this
is
a
very
difficult
task.
But
it
was
felt
that
a
proper
uncertainty
analysis
allows
users
of
the
model,
and
eventually
regulatory
assessors,
the
ability
to
place
model
predictions
in
their
proper
context.

In
terms
of
propagating
uncertainties
to
endpoint
predictions,
the
EPA
should
put
more
attention
to
errors
caused
by
systematic
model
misspecification
in
the
PBPK/
PD
model.
This
includes
concerns
with
over
parameterization.
The
uncertainty
(
and
variability)
in
model
predictions
depends
heavily
on
the
selected
model.
If
the
model
is
incorrect,
depending
on
the
degree
of
error,
the
uncertainty
in
the
final
model
predictions
also
will
be
incorrect.

In
assessing
uncertainty,
it
seems
reasonable
to
include
terms
and
pathways
in
the
model
that
may
have
small
contributions
to
the
overall
uncertainty,
simply
because
their
inclusion
usually
doesn't
make
much
difference
and
leaving
them
out
begs
the
question
of
what
effect
they
might
have
had.
A
process
to
evaluate
reductions
in
the
dimensionality
and
complexity
of
the
model
is
needed
to
support
calibration
of
the
models.
Some
Panel
members
felt
that
the
dimensionality
of
the
model
should
be
solely
the
responsibility
of
the
model
builder,
but
others
would
like
to
see
an
analysis
of
how
model
form
changes
affect
predictions
and
changes
sensitivity.

The
Panel
also
discussed
the
idea
of
letting
the
model
form
itself
be
uncertain.
For
example,
the
model
developer
could
generate
a
suite
of
models,
all
of
which
are
acceptable
but
which
illustrate
the
range
of
understanding
of
the
form
of
the
underlying
processes
and
expert
opinion
on
how
compartments
might
interact.
Some
models
might
be
highly
defined,
consisting
of
many
compartments
and
parameters,
whereas
others
could
be
very
simple,
based
on
purely
empirical
relations.
It
also
may
be
possible
to
assign
a
weight
to
each
model
proportional
to
the
likelihood
that
the
particular
model
properly
describes
the
PBPK/
PD
system.
As
in
a
Monte
Carlo
sampling
procedure,
one
could
then
generate
model
predictions
of
endpoints
from
each
of
the
models,
and
from
multiple
samples
of
the
61
of
113
parameter
uncertainties
for
the
respective
model.
This
would
result
in
a
collection
of
model
predictions
based
on
uncertainty
in
both
parameters
and
model
specification
and
the
resulting
prediction
distributions
or
prediction
bands
would
better
convey
overall
uncertainty.
This
information
complements
the
measures
of
goodness­
of­
fit
resulting
from
formal
parameter
estimation.

Not
all
Panel
members
wanted
to
go
quite
as
far
as
the
full
parameter
and
model
uncertainty
analysis
discussed
in
the
previous
paragraph.
One
member
preferred
a
wellstructured
statistical
analysis
to
a
post­
hoc
uncertainty
analysis;
allocating
90%
effort
to
parameter
estimation
and
only
10%
to
assessing
uncertainty.
Others
felt
that
it
was
crucially
important
that
the
underlying
uncertainty
be
assessed
in
explicit
probabilistic
simulations.
One
Panel
member
mentioned
that
a
Bayesian
Network
approach
(
Borsuk
et
al.,
2001,
2004)
might
be
useful
but
another
Panel
member
stated
that
this
approach
has
been
examined
and
found
unacceptable
for
use
with
dynamic
system
models
such
as
these
PBPK/
PD
models.

There
was
discussion
that
advances
in
computational
algorithms
and
computer
technologies
have
eliminated
many
of
the
computational
barriers
to
full
uncertainty
analysis.
What
is
required
is
careful
scientific
specification
of
the
model
structure
and
its
implied
conditional,
probabilistic
relationships.
Specification
of
prior
probabilities
for
parameter
values
requires
good
empirical
data
or
carefully
elicited
expert
judgment.
It
does
not
remove
the
need
to
test
how
individual
model
components
and
changes
in
model
specification
will
influence
model
outputs.
Nor
does
it
eliminate
the
need
for
sensitivity
analyses
to
the
model
outputs.

Other
related
comments
made
during
the
Panel
discussions
include:

 
The
models
discussed
do
not
seem
to
explicitly
incorporate
inter­
individual
variability.
 
The
animal
data
typically
used
in
calibration
are
aggregated
and
assumed
homogeneous.
 
Model
fits
might
be
more
useful
if
they
are
targeted
at
the
midpoint
or
center
of
the
response
distribution
rather
than
at
the
99.9th
percentile
as
seemed
to
be
done
in
the
report
to
the
Panel.
 
The
judgment
of
those
model
fits
would
be
substantially
improved
if
individual
response
values
were
included
on
plots
and
not
only
summary
data
(
means).
 
Related
to
the
quality
of
the
mathematical
and
statistical
analysis
is
the
capability
of
the
interface
between
the
model
application
and
stand­
alone
statistical
analysis
packages.
The
amount
of
data
that
can
be
created
during
a
simulation
run
needs
a
robust
statistical
analysis
package
with
modern
tools
of
statistical
diagnostics
and
regression
methods.
 
After
creating
Figures
7­
19,
the
report
talks
about
typical
or
expected
ranges
for
a
subset
of
the
rate
parameters
as
summarized
in
Table
13
(
p.
48).
It
would
have
been
informative
to
overlay
the
most
extreme
predicted
curves
onto
one
or
more
62
of
113
of
Figures
7­
19.
This
would
provide
a
"
bound"
on
the
prediction,
allowing
us
to
see
if
the
observations
at
least
fit
within
high­
low
predicted
patterns.
This
is
not
exactly
a
confidence
band
but
something
more
than
the
"
point
estimate"
presented
in
original
graphs.

Issue
2.4:
Risk
Metric
Historically,
EPA
has
calculated
margins
of
exposure
(
MOE)
in
its
risk
assessments
for
the
N­
methyl
carbamate
pesticides.
These
MOEs
are
calculated
by
dividing
environmental
exposure
concentrations
by
a
point
of
departure
identified
from
toxicity
studies.
These
points
of
departure
are
typically
no­
observed­
adverseeffect
levels
(
NOAELs)
or
benchmark
dose
estimates
(
BMDs).
For
the
N­
methyl
carbamates,
these
NOAELs
or
BMD
estimates
are
generally
based
on
peak
cholinesterase
inhibition.
The
use
of
PPBK/
PD
models
provides
the
opportunity
to
consider
toxicological
endpoints
other
than
peak
cholinesterase
inhibition.
Some
potential
toxicological
endpoints
include
1)
peak
concentration
of
the
pesticide
(
or
key
metabolite)
at
the
site
of
action;
2)
total
pesticide
(
or
key
metabolite)
at
the
site
of
action
over
a
period
of
time
(
e.
g.
area
under
curve);
3)
peak
cholinesterase
inhibition;
4)
inhibition
at
or
above
a
pre­
defined
level
of
inhibition
(
e.
g.
BMD10);
4)
duration
of
time
for
inhibition
at
or
above
a
pre­
defined
level
of
inhibition.
The
current
report
explicitly
provides
the
peak
concentration
of
carbaryl
and
peak
cholinesterase
inhibition;
although
the
other
metrics
are
easily
accessible
from
a
model
developed
in
the
ERDEM
platform.

Question
2.4
Given
the
toxicological
characteristics
of
carbaryl
and
other
N­
methyl
carbamate
pesticides,
please
comment
on
the
degree
to
which
these
toxicological
endpoints
are
appropriate
for
purposes
of
developing
a
risk
assessment.

Panel
Response
PBPK/
PD
modeling
has
the
advantage
over
the
conventional
NOEL
and
MOE
approach
in
incorporating
the
current
understanding
of
mechanism
of
toxicity
in
describing
the
relationship
between
components
of
risk
assessment,
e.
g.,
toxicity,
exposure,
and
risk.
The
unique
challenge
of
modeling
the
relationship
between
toxicity
and
exposure
for
carbaryl
and
other
carbamate
pesticides
is
that
the
pesticide
and
its
active
metabolite(
s)
are
rapidly
released
after
binding
to
cholinesterase.

The
Panel
generally
agreed
that,
as
far
as
enzyme
binding
is
concerned,
the
inhibition
of
cholinesterase
activity
is
the
ultimate
metric
for
expressing
toxicity
and
risk.
This
could
include
both
considerations
of
the
peak
inhibition
level,
the
duration
of
sustained
inhibition
as
expressed
in
the
area
under
the
curve
(
AUC),
and
the
duration
63
of
113
above
a
pre­
defined
inhibition
level.
On
the
other
hand,
the
profiles
of
carbamate
and
its
active
metabolites
at
the
site
of
action
are
pertinent
metrics
of
exposure.
These
metrics
are
important
for
improving
the
understanding
of
both
short­
and
long­
term
exposures
(
i.
e.,
prolonged
duration
or
repeated
exposures)
and
can
provide
a
linkage
to
the
biological
effects
and
the
biomonitoring
data
in
humans.

Thus,
all
of
the
five
metrics
suggested
by
the
Agency
are
of
interest
and
each
of
them
can
potentially
be
a
valid
metric
for
risk
assessment,
although
some
expressions
may
require
further
clarification.
At
this
early
stage
of
developing
the
PBPK/
PD
model
for
carbaryl
and
possibly
extending
its
application
to
other
N­
methyl
carbamates
and
their
cumulative
exposure,
the
Agency
is
encouraged
to
explore
the
merit
of
each
of
these
endpoints.
In
fact,
multiple
expressions
of
some
of
these
metrics
can
enhance
a
logical
and
clear
presentation
of
risk
assessment
and
facilitate
the
understanding
and
communication
between
the
exposure
and
risk
assessors
and
between
risk
assessors
and
risk
managers.

Specific
comments
for
each
metric
are
provided
below
with
respect
to
their
pertinence
in
risk
assessment
and
consideration
of
time
factor
in
exposure
and
toxicity.

1)
peak
concentration
of
the
pesticide
(
or
key
metabolite)
at
the
site
of
action.

The
peak
concentrations
of
pesticide
or
its
key
metabolites
at
the
site
of
action
are
among
the
primary
variables
presented
by
the
ERDEM
model.
The
concentration
of
the
parent
compound
is
the
key
expression
of
exposure
for
evaluating
the
toxicity
of
carbaryl.
This
and
the
next
metric
(
i.
e.,
AUC)
are
the
ultimate
parameters
of
pharmacokinetic
events
after
exposure,
interfacing
with
the
pharmacodynamic
and
toxicity
metrics.
When
properly
modeled,
the
peak
concentration
informs
pathway
saturation
and
the
various
degrees
of
probability
of
states
of
the
carbaryl,
e.
g.,
free­
circulating,
bound
to
cholinesterase,
metabolized
as
released
from
the
enzyme
complex.
The
peak
level
metric
also
supplies
information
on
the
speed
at
which
the
pesticide
is
taken
up,
the
time­
scale
over
which
effects
are
likely
to
occur,
and
the
maximum
level
of
effect
likely
to
occur.

The
peak
level
also
could
serve
as
a
pointer
to
compound­
specific
toxicities
that
lie
outside
the
common
mechanism
and
is
especially
relevant
in
an
acute
setting.
One
reason
is
that
short­
term
adaptations
reduce
or
eliminate
physiological
disturbances
from
minor
to
moderate
levels
of
cholinesterase
inhibition.
When
moderate
levels
of
inhibition
are
sustained,
several
compensatory
mechanisms
come
into
play
almost
immediately,
including
reduction
of
acetylcholine
release
as
well
as
receptor­
desensitization
and,
later
on,
down
regulation.
Synaptic
homeostasis
is
quickly
restored.
Hence,
the
AUC
(
see
discussions
under
the
next
metric),
is
unlikely
to
be
a
good
measure
of
acute
toxicity.

2)
total
pesticide
(
or
key
metabolite)
at
the
site
of
action
over
a
period
of
time
(
e.
g.
area
under
curve)
64
of
113
The
AUC
metric
includes
the
magnitude
of
effect
implicitly,
but
also
duration
of
effect.
The
concentration
of
the
pesticide
or
its
active
metabolite
in
and
of
itself
is
irrelevant
to
the
characterization
of
risk
in
isolation
from
the
extent
of
enzyme
inhibition.
Thus,
comments
regarding
this
metric
are
not
limited
to
the
AUC
of
pesticide
or
its
active
metabolites
but
also
the
metric
of
AUC
of
cholinesterase
inhibition.

An
important
consideration
is
that
the
compensatory
responses
that
might
be
taken
by
the
organism
in
response
to
perturbation
in
acetylcholine
signal
strength
are
not
necessarily
benign.
If
sustained
adjustments
are
made
in
particular
synapses
in
the
postsynaptic
sensitivity
or
the
presynaptic
extent
of
acetylcholine
release,
these
may
well
have
subtle
long
term
consequences
for
the
transmission
of
signals
that
are
important
parts
of
neuron/
neuron
and
neuron/
muscle
communication.
Thus,
the
toxicological
response
to
long
sustained
exposures
could
be
very
different
from
the
acute
exposure
scenarios.
Total
AUC
of
inhibition
or
total
time
above
a
pre­
defined
level
of
inhibition
(
metric
5)
can
be
important.
In
these
cases,
compensatory
mechanisms
may
fail,
or
the
consequences
of
the
compensations
may
themselves
be
adverse.
Therefore,
models
developed
for
the
ERDEM
platform
should
calculate
and
report
these
metrics.

The
different
toxicological
implications
between
the
peak
cholinesterase
inhibition
(
next
metric)
versus
the
AUC
of
inhibition
are
simply
illustrated:
in
a
poisoning
event,
90%
inhibition
over
20
minutes
is
a
more
serious
effect
than
10%
inhibition
over
180
minutes.
While
the
AUC
metric
would
give
the
same
results
for
the
above
two
scenarios,
the
health
outcome
is
likely
to
be
substantially
different.
On
the
other
hand,
for
non­
acute
poisoning
events,
e.
g.,
10%
inhibition,
the
integrated
AUC
approach
would
be
appropriate.
It
may
be,
for
example,
that
long­
term,
low­
level
cholinesterase
inhibition
gives
rise
to
specific
health
outcomes
of
interest.
Further,
repeated
assaults
may
offer
additional
effects.
The
scenarios
outlined
in
the
document
presented
to
the
Panel
­
adult
applicator
and
child
playing
on
treated
turf
offer
examples
that
could
give
rise
to
similar
integrated
effects
with
entirely
different
profiles.

3)
Peak
cholinesterase
inhibition
As
described
above,
peak
inhibition
of
cholinesterase
activity
is
the
most
important
endpoint
for
evaluating
acute
toxicity
with
carbamate
anticholinesterases.
It
is
the
most
direct
measure
of
adverse
effect
mediated
through
the
common
mechanism
of
toxicity
for
carbamates.
Thus,
for
short
term
effects,
peak
inhibition
is
a
good
first
judgment
for
the
dosimeter
that
is
most
likely
to
be
predictive
of
toxicity.

The
expression
of
peak
cholinesterase
inhibition
at
a
given
dose
is
comparable
to
the
current
toxicity
data
that
defines
the
threshold
as
the
no­
observed­
effect
level
(
NOEL)
based
on
the
peak
inhibition
measured
at
the
assumed
peak
time
of
effects.
Thus,
the
PBPK/
PD
model
can
be
used
to
estimate
the
peak
inhibition
after
human
exposure
and
compared
to
the
level
defined
as
adverse
when
exceeded,
i.
e.,
threshold
cholinesterase
inhibition.
Depending
on
the
extent
that
a
PBPK/
PD
model
may
address
the
inter­
species
65
of
113
and
inter­
individual
variation
in
sensitivity
(
e.
g.,
two
individuals
exposed
to
identical
level
of
carbaryl
may
show
different
levels
of
inhibition
due
to
differential
susceptibility),
uncertainty
factor(
s)
may
be
modeled
or
applied
in
this
comparison.

The
similarity
of
this
metric
to
the
conventional
NOEL
or
BMD
has
the
added
advantage
of
facilitating
risk
communication
while
transitioning
into
the
mechanistic
approach.
Modeling
peak
inhibition
also
allows
the
possibility
of
correlating
this
metric
to
other
related
neurological
endpoints
(
e.
g.,
neurobehavioral
and
clinical
signs).

An
apparent
unknown
is
what
dosimeter
is
closest
to
a
causal
determinant
of
the
subsequent
pharmacodynamic
processes
 
from
overt
symptoms
in
adults
to
putative
marginal
strengthening
of
signaling
along
some
pathways
rather
than
others
during
development
to
more
subtle
adaptations
of
the
synaptic
responsiveness
to
later
cholinergic
stimuli.
This
is
where
we
really
need
progress
in
basic
science
to
show
how
cells
set
and
reset
their
set
points
for
responsiveness
in
relation
to
the
degree
and
duration
of
cholinesterase
inhibition.
Thus,
for
current
assessments,
both
peak
inhibition
and
the
AUC
(
the
amount
of
inhibition
times
time)
should
be
used
as
dose
metrics
for
assessing
MOE
and
risks
of
response.

4)
Inhibition
at
or
above
a
pre­
defined
level
of
inhibition
(
e.
g.
BMD10)

One
Panel
member
commented
that
this
metric
requires
some
clarification.
Simplistically,
a
PBPK/
PD
model
can
define
the
dose
that
results
in
a
pre­
defined
peak
level
of
cholinesterase
inhibition,
equivalent
to
the
BMD
(
e.
g.,
10%
inhibition
in
the
brain).
In
a
conventional
approach,
this
metric
may
then
be
used
to
generate
the
relative
potency
factor
for
acute
toxicity
of
multiple
carbamates;
if
it
is
determined
that
peak
inhibition
is
the
valid
metric
for
expressing
acute
toxicity.
The
main
drawback
is
that
this
expression
does
not
implicitly
account
for
the
time
factor
as
the
next
metric
(
i.
e.,
duration
of
time
for
inhibition
at
or
above
a
pre­
defined
level
of
inhibition).
Instead,
this
metric
suggests
a
dichotomous
variable
­
exposure
resulting
in
inhibition
above
a
fixed
level.
Thus,
this
metric
may
be
less
useful
than
knowing
how
long
such
an
inhibition
occurred.
Clearly
there
is
a
difference
in
inhibition
above
10%
that
lasted
for
five
seconds
versus
a
similar
inhibition,
due
to
chronic
exposure,
that
lasts
for
many
days.

5)
Duration
of
time
for
inhibition
at
or
above
a
pre­
defined
level
of
inhibition.

Being
able
to
take
into
account
the
duration/
time
factor
is
one
satisfying
feature
of
this
metric.
Assuming
that
the
pre­
defined
level
is
selected
with
some
intrinsic
meaning
(
e.
g.,
a
health
outcome
of
interest),
the
duration
of
the
insult
experienced
by
the
body
represents
a
significant
measure.
Short­
duration
insults
are
likely
to
produce
less
effect
than
long­
term
sustained
insult
of
the
same
magnitude.
The
selection
of
this
metric
presupposes
that
a
pre­
defined
level
is
below
some
frank
effect
level.
Otherwise,
the
effect
can
be
defined
by
other
means.
Furthermore,
the
duration
should
be
relative
to
the
expected
clearance
rate.
For
example,
it
can
be
asked:
is
this
elevated
level
maintained
for
66
of
113
longer
than
the
lifetime
of
carbaryl
in
the
body
and
thus
may
be
indicative
of
an
ongoing,
albeit
low­
level,
exposure?
67
of
113
REFERENCES
Bois,
F.
(
2000)
Statistical
Analysis
of
Fisher
et
al.
PBPK
Model
of
Trichloroethylene
Kinetics.
Env.
Health
Perspectives
8,
Supplement
2:
275­
282.

Borsuk,
M.
E.,
C.
A.
Stow,
and
K.
H.
Reckhow
(
2004).
A
Bayesian
network
of
eutrophication
models
for
synthesis,
prediction,
and
uncertainty
analysis.
Ecological
Modelling
173:
219­
239.

Borsuk,
M.
E.,
C.
A.
Stow,
D.
Higdon,
and
K.
H.
Reckhow
(
2001).
A
Bayesian
hierarchical
model
to
predict
benthic
oxygen
demand
from
organic
matter
loading
in
estuaries
and
coastal
zones.
Ecological
Modelling
143:
165­
181.

Gelman,
A.,
F.
Bois,
and
J.
Jiang
(
1996).
Physiological
Pharmacokinetic
Analysis
Using
Population
Modeling
and
Informative
Prior
Distributions.
J.
of
Am.
Stat.
Association
91(
436):
1400­
1412.

Ginsberg,
G.
L.,
Pepelko,
W.
E.,
Goble,
R.
L.,
and
Hattis,
D.
B.
(
1996).
Comparison
of
Contact
Site
Cancer
Potency
Across
Dose
Routes:
Case
Study
with
Epichlorohydrin.
Risk
Analysis
16:
667­
681.

Ginsberg,
G.,
Hattis,
D.,
Russ,
A.,
and
Sonawane,
B.
(
2004).
Physiologically­
based
pharmacokinetic
(
PBPK)
modeling
of
caffeine
and
theophylline
in
neonates
and
adults:
Implications
for
assessing
children's
risks
from
environmental
agents.
Journal
of
Toxicology
and
Environmental
Health
67:
297­
329.

Price,
P.
S.,
Connolly,
R.
B.,
Chaisson,
C.
F.,
Gross,
E.
A.,
Young,
J.
S.,
Mathis,
E.
T.,
Tedder,
D.
R.
(
2003).
Modeling
interindividual
variation
in
physiological
factors
used
in
PBPK
models
of
humans.
Critical
Reviews
in
Toxicology
33:
469­
503.

Walker,
K.,
Hattis,
D.,
Russ,
A.,
and
Ginsberg,
G.
(
2004).
Physiologically­
Based
Toxicokinetic
Modeling
for
Acrylamide
 
Risk
Implications
of
Polymorphisms
and
Developmental
Changes
in
Selected
Metabolic
Enzymes.
Report
from
the
George
Perkins
Marsh
Institute,
Clark
University,
and
the
Connecticut
Department
of
Public
Health
to
the
U.
S.
Environmental
Protection
Agency
under
Cooperative
Agreement
#
827195­
0,
December
2004.
68
of
113
SAP
Minutes
No.
2005­
01
A
Set
of
Scientific
Issues
Being
Considered
by
the
Environmental
Protection
Agency
Regarding:

N­
METHYL
CARBAMATE
CUMULATIVE
RISK
ASSESSMENT:
PILOT
CUMULATIVE
ANALYSIS
SESSION
3:
DRINKING
WATER
EXPOSURE
ANALYSIS
February
17,
2005
FIFRA
Scientific
Advisory
Panel
Meeting
held
at
the
Holiday
Inn
National
Airport
Arlington,
VA
Mr.
Joseph
E.
Bailey
Steven
G.
Heeringa,
Ph.
D.
Designated
Federal
Official
FIFRA
SAP
Session
Chair
FIFRA
Scientific
Advisory
Panel
FIFRA
Scientific
Advisory
Panel
Date:
April
15,
2005
Date:
April
15,
2005
69
of
113
Federal
Insecticide,
Fungicide
and
Rodenticide
Act
Scientific
Advisory
Panel
Meeting
February
17,
2005
N­
methyl
Carbamate
Cumulative
Risk
Assessment:
Pilot
Cumulative
Analysis
Session
3:
Drinking
Water
Exposure
Analysis
PARTICIPANTS
FIFRA
SAP
Session
Chair
Steven
G.
Heeringa,
Ph.
D.,
Research
Scientist
&
Director
for
Statistical
Design,
Institute
for
Social
Research,
University
of
Michigan,
Ann
Arbor,
MI
Designated
Federal
Official
Joseph
E.
Bailey,
FIFRA
Scientific
Advisory
Panel
Staff,
Office
of
Science
Coordination
and
Policy,
EPA
FIFRA
Scientific
Advisory
Panel
Members
Stuart
Handwerger,
M.
D.,
Professor
of
Pediatrics,
University
of
Cincinnati
Children's
Hospital
Medical
Center,
Cincinnati,
OH
Gary
E.
Isom,
Ph.
D.,
Professor
of
Toxicology,
School
of
Pharmacy
&
Pharmacal
Sciences,
Purdue
University,
West
Lafayette,
IN
Kenneth
M.
Portier,
Ph.
D.,
Associate
Professor,
Statistics,
Institute
of
Food
and
Agricultural
Sciences,
University
of
Florida,
Gainesville,
FL
FQPA
Science
Review
Board
Members:

John
Adgate,
Ph.
D.,
Assistant
Professor,
Division
of
Environmental
Health
Sciences,
University
of
Minnesota,
School
of
Public
Health,
Minneapolis,
MN
George
B.
Corcoran,
Ph.
D.,
Professor
&
Chairman,
Department
of
Pharmaceutical
Sciences,
Eugene
Applebaum
College
of
Pharmacy
&
Health
Sciences,
Wayne
State
University,
Detroit,
MI
Lutz
Edler,
Ph.
D.,
Head,
Biostatistics
Unit
C060,
German
Cancer
Research
Center,
Heidelberg,
Germany
70
of
113
Bernard
Engel,
Ph.
D.,
Professor,
Agricultural
&
Biological
Engineering,
Purdue
University,
West
Lafayette,
IN
Scott
Ferson,
Ph.
D.,
Senior
Scientist,
Applied
Biomathematics,
Setauket,
NY
Lawrence
J.
Fischer,
Ph.
D.,
Director,
Center
for
Integrative
Toxicology,
Michigan
State
University,
East
Lansing,
MI
Natalie
Freeman,
Ph.
D.,
Associate
Professor,
Department
of
Physiological
Sciences,
College
of
Veterinary
Medicine,
University
of
Florida,
Gainesville,
FL
James
P.
Kehrer,
Ph.
D.,
Director,
Center
for
Molecular
&
Cellular
Toxicology,
College
of
Pharmacy,
The
University
of
Texas
at
Austin,
Austin,
TX
Chensheng
Lu,
Ph.
D.,
Assistant
Professor,
Department
of
Environmental
&
Occupational
Health,
Rollins
School
of
Public
Health,
Emory
University,
Atlanta,
GA
Peter
D.
M.
Macdonald,
D.
Phil.,
Professor
of
Mathematics
&
Statistics,
McMaster
University,
Hamilton,
Ontario,
Canada
David
MacIntosh,
Sc.
D.,
Senior
Associate,
Environmental
Health
&
Engineering,
Inc.,
Newton,
MA
Robert
W.
Malone,
Ph.
D.,
Agricultural
Engineer,
USDA­
ARS,
National
Soil
Tilth
Laboratory,
Ames,
IA
Christopher
J.
Portier,
Ph.
D.,
Director,
Environmental
Toxicology
Program,
National
Institute
of
Environmental
Health
Sciences,
Research
Triangle
Park,
NC
Nu­
may
Ruby
Reed,
Ph.
D.,
D.
A.
B.
T.,
Staff
Toxicologist,
Department
of
Pesticide
Regulation,
California
Environmental
Protection
Agency,
Sacramento,
CA
P.
Barry
Ryan,
Ph.
D.,
Professor,
Environmental
&
Occupational
Health,
Rollins
School
of
Public
Health,
Emory
University,
Atlanta,
GA
Michael
D.
Sohn,
Ph.
D.,
Scientist,
Environmental
Energy
Technologies
Division,
Lawrence
Berkeley
National
Laboratory,
University
of
California,
Berkeley,
CA
Tammo
S.
Steenhuis,
Ph.
D.,
Professor
of
Watershed
Management,
Cornell
University,
Ithaca,
NY
Michael
D.
Wheeler,
Ph.
D.,
Assistant
Professor,
Departments
of
Pharmacology
&
Medicine,
University
of
North
Carolina,
Skipper
Bowles
Center
for
Alcohol
Studies,
Chapel
Hill,
NC
71
of
113
INTRODUCTION
In
Session
3
of
this
meeting,
the
FIFRA
SAP
met
to
consider
and
review
the
N­
methyl
carbamate
pesticide
cumulative
risk
assessment:
pilot
cumulative
analysis,
specifically
issues
related
to
drinking
water
exposure
assessment.
OPP
solicited
comment
from
the
SAP
on
the
use
of
existing
ground­
water
models
to
provide
a
pilot
ground­
water
exposure
assessment
for
the
N­
methyl
carbamate
pesticides.
Session
3
included
presentations
by
Mr.
Nelson
Thurman
and
Dr.
Dirk
Young
(
Environmental
Fate
and
Effects
Division,
Office
of
Pesticide
Programs)
and
Dr.
Tom
Nolan
(
U.
S.
Geological
Survey)
pertaining
to
the
use
of
ground
water
exposure
models
and
transport
models
to
help
predict
drinking
water
exposure
estimates
for
the
N­
methyl
carbamate
pesticides.
Dr.
Steven
Bradbury
(
Director,
Environmental
Fate
and
Effects
Division,
Office
of
Pesticide
Programs)
gave
opening
remarks
for
Session
3.
72
of
113
SUMMARY
OF
PANEL
DISCUSSION
AND
RECOMMENDATIONS
The
Agency
solicited
comments
from
the
Panel
on
assessing
regional
groundwater
vulnerability
and
predicting
pesticide
concentrations
in
groundwater
for
estimating
exposures
to
N­
methyl
carbamate
pesticides.

Data
that
can
be
used
to
assess
nationwide
groundwater
vulnerability
are
quite
limited.
Therefore
an
obvious
approach
for
assessing
groundwater
vulnerability
to
pesticides
does
not
exist.
EPA
proposes
to
use
a
shallow
groundwater
vulnerability
assessment
that
was
developed
for
nitrates.
However,
the
nitrate
assessment
may
not
be
appropriate
for
estimating
shallow
groundwater
vulnerability
to
pesticides.
If
the
proposed
approach
is
to
be
used,
it
should
be
compared
with
observed
groundwater
contamination
data.
Alternatively,
an
approach
utilizing
the
nationwide
State
Soil
Geographic
(
STATSGO)
Data
Base
and
the
pesticide
leaching
models
should
be
considered.

To
predict
concentrations
of
N­
methyl
carbamate
pesticides
in
groundwater,
the
proposed
approach
appears
to
be
conservative
and
health
protective
for
most
reasonable
scenarios.
In
all
but
the
most
unusual
cases,
screens
in
the
wells
will
draw
from
a
large
cross
section
of
the
aquifer,
perhaps
some
depth
below
the
groundwater
table.
This
water
will
be
more
dilute
than
the
concentration
of
leachate
in
the
vadose
zone,
and
additional
degradation
may
occur
before
people
drink
the
groundwater.
Also,
the
level
of
detail
incorporated
in
the
model
and
the
modeling
decisions
is
consistent
with
or
a
bit
more
conservative
than
the
assessments
for
other
routes
of
cumulative
exposure
to
people.

The
Office
of
Pesticide
Programs
(
OPP)
presented
illustrative
applications
of
three
groundwater
models
­­­
Root
Zone
Water
Quality
Model
(
RZWQM),
Pesticide
Root
Zone
Model
(
PRZM),
and
Leaching
Estimation
and
CHemistry­
Pesticides
model
(
LEACHP)
­­­
that
it
proposes
to
use.
The
three
groundwater
models
are
very
similar.
However,
in
response
to
specific
questions
brought
to
the
Panel,
important
considerations
include
the
following:

 
Management
practices
and
pesticide
application.
The
models
must
be
able
to
address
the
management
practices
and
pesticide
application
practices
(
e.
g.,
pesticide
incorporation,
management
practices
that
impact
hydrology)
that
are
important
to
N­
methyl
carbamate
and
other
pesticide
movement
to
shallow
groundwater.
Urban
uses
of
N­
methyl
carbamate
pesticides
could
be
important,
but
it
is
unclear
whether
the
models
proposed
could
address
such
uses.
 
Degradation
product
formation.
The
model(
s)
should
simulate
formation
and
movement
of
degradation
products
if
they
are
found
to
significantly
contribute
to
risk.
73
of
113
Preferential
flow.
Macropore
or
preferential
flow
may
be
essential
to
accurately
predict
transport
of
N­
methyl
carbamate
pesticides
and
other
pesticides
to
shallow
groundwater.
 
Hydrology
and
rainfall
characteristics.
Most
models
predict
the
water
balance
accurately.
Correctly
characterizing
rainfall
amount
and
intensity
may
be
especially
important
when
applying
RZWQM,
because
it
will
not
simulate
macropore
flow
unless
the
rainfall
is
sufficiently
intense.
 
Pesticide
fate.
Some
of
the
more
sophisticated
pesticide
fate
routines
in
models
should
possibly
be
avoided.
Both
the
pesticide
degradation
rate
and
the
adsorption
strength
are
highly
variable
between
fields
and
may
explain
the
poor
performance
of
the
models
during
evaluation
efforts.
 
Tile
drainage.
If
N­
methyl
carbamate
pesticides
are
applied
to
areas
that
are
tile
drained,
the
selected
model
should
simulate
tile
drainage
for
accurate
loading
estimates
to
surface
water.
However,
tile
drainage
consideration
may
not
be
necessary
in
the
model
for
the
purposes
of
the
cumulative
risk
assessment,
especially
if
this
is
a
screening
level
tool.
 
Model
calibration,
sensitivity,
testing,
and
ease
of
use.
Model
calibration
may
be
necessary.
If
so,
the
ability
to
easily
calibrate
the
model
will
be
important.
Other
important
considerations
include
the
transparency
in
the
model's
calculations,
the
data
requirements,
post­
analysis
of
the
model
output,
internal
documentation
of
code,
ability
to
modify
code,
ability
to
interface
the
code
with
third­
party
graphical
user
interfaces
(
GUI),
and
ability
to
link
to
statistical
analysis
packages
and
geographical
information
system
(
GIS)
packages.

Surface
water
and
groundwater
behave
quite
differently.
Generally,
groundwater
travels
slowly
and
thus
spatial
averaging
is
a
poor
assumption.
The
pesticide
concentration
in
each
well
depends
on
its
immediate
surroundings
and
cannot
be
represented
by
an
average
concentration.
Because
wells
have
screens
allowing
water
into
them,
temporal
averaging
should
be
considered.

To
account
for
persistence
of
pesticides
in
groundwater,
the
background
levels
could
be
estimated
with
models,
and
decline
in
residues
over
time
(
estimated
based
on
long­
term
trends
in
monitoring)
should
be
considered.
The
background
document
clearly
identified
that
carbamate
transformation
can
be
pH
specific.
Also,
the
transformations
of
some
metabolites
may
be
affected
by
concentration
and
anaerobic
conditions.
74
of
113
PANEL
DELIBERATIONS
AND
RESPONSE
TO
THE
CHARGE
The
specific
issues
to
be
addressed
by
the
Panel
are
keyed
to
the
Agency's
background
documents,
references
and
charge
questions.

Question
1
Selection
of
regional
vulnerable
ground
water
sites:

In
both
the
organophosphate
(
OP)
cumulative
assessment
and
this
carbamate
cumulative
assessment,
OPP
identified
regional
drinking
water
exposure
sites
for
surface
water
sources
of
drinking
water
that
would
represent
one
of
the
more
vulnerable
surface
watersheds
in
each
region.
The
process,
which
was
deemed
a
valid
approach
by
the
2002
SAP
(
USEPA,
2002),
identified
those
areas
where
areas
of
high
combined
cumulative
pesticide
use
coincided
with
drinking
water
sources
which
were
particularly
vulnerable
to
runoff.
This
served
as
a
regional
screening
assessment
in
that
if
the
regional
cumulative
risk
assessment
finds
that
exposure
in
water
is
not
a
significant
contributor
to
the
overall
exposure
in
that
area,
it
will
not
be
a
significant
contributor
in
other
areas
in
the
region.

For
ground
water
sources
of
drinking
water,
OPP
is
proposing
to
use
a
similar
approach
(
described
in
Section
C.
2.
of
the
Drinking
Water
section
of
the
case
study).

National
coverages
of
the
vulnerability
of
aquifers
to
pesticide
contamination
are
limited
in
availability,
and
generally
refer
to
vulnerability
of
the
overlying
soils
and
surficial
geology
to
leaching.
Does
the
SAP
believe
that
the
sources
on
relative
ground
water
vulnerability
identified
in
the
case
study
provide
an
adequate
screening­
level
assessment
of
the
potential
for
contamination
of
shallow
aquifers
to
pesticide
contamination?
If
these
vulnerability
assumptions
are
inadequate,
what
can
be
done
to
improve
the
approach?

Panel
Response
In
the
oral
presentation,
EPA
staff
indicated
that
the
vulnerability
assessment
conducted
for
groundwater
to
date
was
not
as
systematic
as
that
done
for
surface
water.
The
Panel
agreed
that
a
more
systematic
assessment
would
be
desirable.

As
indicated
by
the
EPA,
the
national
coverages
of
the
vulnerability
of
aquifers
to
pesticide
contamination
and
related
data
are
limited.
The
vulnerability
assessments
that
are
available
are
limited
in
the
types
of
data
considered
in
their
creation
and
in
the
scales
of
the
data
used
(
e.
g.,
the
nationwide
DRASTIC
assessments).
There
are
some
state
and
75
of
113
regional
groundwater
vulnerability
assessments,
such
as
those
conducted
in
Indiana
(
Cooper
et
al.,
1997).
The
results
of
these
regional
assessments
point
out
the
importance
of
the
scale
of
underlying
data
used
in
their
creation
(
Cooper
et
al.,
1997).
The
DRASTIC
groundwater
vulnerability
estimates
for
Indiana
created
with
1:
250,000
scale
GIS
data
resulted
in
significantly
different
estimates
of
vulnerability
than
those
created
using
courser
nationally­
derived
data.
Further,
Cooper
et
al.
(
1997)
demonstrated
the
value
of
testing
the
vulnerability
maps
with
observed
groundwater
quality
data.
Therefore,
there
are
challenges
in
identification
of
groundwater
vulnerability
nationally.

To
assess
groundwater
vulnerability,
EPA
used
the
method
of
Nolan
et
al.
(
2002)
and
three
criteria
to
identify
the
most
vulnerable
regional
drinking
water
exposure
sites.
The
first
two
criteria
were
areas
where
carbamate
pesticides
are
applied
and
areas
where
groundwater
is
used
for
drinking.
The
third
criterion
was
to
identify
the
areas
most
vulnerable
to
groundwater
contamination.

The
vulnerability
assessment
approach
does
not
address
off­
label
and
improper
uses
of
pesticides.
Agricultural
pesticide
applications
might
generally
show
fairly
high
adherence
to
prescribed
uses
and
application
rates,
but
residential
applications
might
show
considerably
less
fidelity
to
label
uses.
One
of
the
pesticides
being
considered
has
residential
applications.
It
is
conceivable
that
EPA's
presumption
that
usage
follows
label
guidance
might
seriously
underestimate
vulnerabilities
in
some
suburban
areas.

The
Nolan
et
al.
(
2002)
method
to
assess
groundwater
vulnerability
was
developed
for
nitrate,
which
may
transport
through
the
vadose
zone
or
an
aquifer
differently
than
pesticides.
For
example,
vulnerable
areas
for
pesticides
may
largely
be
a
function
of
soil
organic
carbon
and
macropore
flow,
which
may
not
be
as
large
a
factor
for
nitrate
transport
to
shallow
groundwater.
The
method
used
by
Nolan
et
al.
(
2002)
also
considers
human
population
which
may
be
unique
to
groundwater
vulnerability
to
nitrate.
The
model
coefficient
associated
with
the
depth
to
groundwater
also
may
have
the
wrong
sign
on
it
in
this
model
if
it
is
used
for
N­
methyl
carbamate
pesticides.
Therefore,
areas
may
be
vulnerable
to
pesticide
leaching
to
shallow
groundwater
that
are
not
identified
by
the
Nolan
et
al.
(
2002)
approach;
the
vulnerability
GIS
coverage
that
has
been
used
may
not
be
the
most
appropriate
approach
for
identifying
vulnerability
of
groundwater
to
carbamates
and
other
pesticides.

If
the
Nolan
et
al.
(
2002)
vulnerability
approach
is
to
be
used
for
identifying
areas
vulnerable
to
pesticides
reaching
shallow
groundwater,
it
should
be
compared
to
existing
data
on
known
contaminated
aquifers,
both
current
and
past,
to
assess
its
quality.
The
vulnerability
model
should
agree
well
with
field
observations.
Where
there
is
disagreement,
the
EPA
should
investigate
to
understand
why
they
differ.

If
one
examines
the
next
step
proposed
by
EPA
in
modeling
groundwater
vulnerability,
there
is
possibly
a
disconnect
between
the
proposed
groundwater
vulnerability
approach
and
the
modeling
approach.
For
example,
the
groundwater
76
of
113
vulnerability
screening
GIS
map
was
created
considering
confined
as
well
as
unconfined
aquifers.
However,
the
modeling
approach
only
considers
very
shallow
unconfined
aquifers.
The
use
of
STATSGO
(
a
nationwide
soil
GIS
dataset)
may
be
a
more
appropriate
mechanism
for
vulnerability
screening.
The
use
of
STATSGO
combined
with
the
models
would
allow
the
most
vulnerable
settings
to
be
identified.
Also,
using
STATSGO
would
be
consistent
with
the
proposed
approach
for
modeling
pesticide
exposure
in
shallow
groundwater,
since
soil
properties
play
an
important
role
in
parameterizing
the
models,
while
aquifer
properties
are
not
considered.

Question
2
Use
of
leaching
models
for
ground­
water
exposure
assessments:

The
Agency
is
basing
its
ground­
water
exposure
assessment
on
private
rural
wells
drawing
its
drinking
water
from
an
unconfined
aquifer.
The
estimated
exposure
in
drinking
water
from
these
wells
is
based
on
the
concentration
estimated
at
the
top
of
this
aquifer.

The
three
models
that
the
Agency
is
considering
for
use
in
the
cumulative
assessment
and
in
refined
ground
water
exposure
in
aggregate
(
individual
chemical)
assessments
 
LEACHP,
PRZM,
and
RZWQM
 
are
leaching
models
that
predict
pesticide
concentrations
in
water
at
some
depth
below
the
surface.
As
a
result,
the
Agency
is
using
estimated
pesticide
concentrations
in
the
vadose­
zone
to
represent
concentrations
in
shallow
ground
water.
As
such,
estimates
would
represent
potential
drinking
water
exposure
from
wells
drawing
from
shallow,
unconfined
aquifers.

Is
this
approach
a
reasonable,
health­
protective
approach
for
use
in
both
cumulative
and
aggregate
drinking
water
exposure
assessments?
If
this
approach
produces
an
exceedance
of
essentially
safe
exposure
levels,
in
what
manner
could
a
better
estimate
of
exposure
to
pesticides
in
water
be
derived
from
existing
data
and
modeling
approaches?

Panel
Response
For
most
reasonable
scenarios,
the
proposed
approach
appears
to
be
conservative
and
health
protective.
In
most
cases,
the
screens
of
wells
are
some
depth
below
the
groundwater
table,
allowing
time
for
additional
degradation
before
people
drink
the
groundwater.
The
level
of
detail
in
the
assessment
and
the
assumptions
are
consistent
or
a
bit
more
conservative
than
the
assumptions
made
for
exposure
from
other
pathways.
However,
there
are
several
cases
where
caution
is
needed
and
the
approach
may
not
be
health
protective.
77
of
113
If
the
pesticide
is
not
entirely
in
the
dissolved
phase
(
as
assumed
in
these
models)
and
greatly
in
excess
of
normal
agricultural
use
rate
for
example
due
to
either
a
large
accidental
spill,
an
unusual
surface
runoff
event
or
cleaning
of
tanks
close
to
the
well,
the
pesticide
might
reach
a
shallow
unconfined
aquifer
in
concentrations
far
in
excess
of
model
predictions.

The
use
of
an
aquifer
model
combined
with
the
estimates
of
pesticides
reaching
shallow
groundwater
without
calibration
for
local
conditions
is
cautioned.
Calibrations
with
experimental
data
would
likely
provide
better
estimates
of
actual
pesticide
levels
reaching
drinking
water.
One
of
the
challenges
will
be
adequately
representing
the
aquifers
because
data
will
be
limited
and
site
specific
conditions
could
be
essential
for
accurately
estimating
pesticide
concentration.

Data
used
for
model
validation
might
not
represent
the
"
real"
concentration
leaching
into
the
aquifer.
Caution
should
be
used
when
estimating
pesticide
transport
to
shallow
groundwater
based
on
suction
lysimeter
and
soil
sample
data.
If
macropores
are
present
in
the
soil,
pesticides
can
move
from
the
soil
surface
to
shallow
groundwater
with
little
evidence
of
this
movement
in
the
unsaturated
zone
(
Malone
et
al.,
2000).
The
best
locations
for
pesticide
samplers
are
in
the
capillary
fringe
of
the
groundwater
or
above
textural
interface
layers.
Wick
pan
samplers
are
preferred
above
gravity
pan
samplers.
Tile
lines
give
the
overall
best
integrated
sample
(
Boll
et
al.,
1992;
Shalit
et
al.,
1995;
Boll
et
al.,
1997).

The
Agency's
tiered
approach
is
supported.
Tiered
approach
is
commonly
used
in
risk
assessment
as
a
good
use
of
limited
resources
by
first
looking
into
the
worst
case
scenario
using
conservative
assumptions.
This
initial
tier
of
analysis
is
less
time
consuming
and
data
intensive.
A
refining
tier
of
analysis
for
a
more
realistic
scenario
is
performed
if
the
worst
case
scenario
is
deemed
unrealistic
and
shows
exceedance
of
safe
exposure
level.

The
Panel
raised
several
questions
during
the
Agency
presentation
concerning
how
the
models
will
be
evaluated
for
the
closeness
of
their
prediction
to
the
realistic
situations.
The
Panel
recommended
a
more
rigorous
review
of
the
chosen
models
for
N­
methyl
carbamate
pesticides,
specifically
for
their
use
in
cumulative
risk
assessment.
When
the
worst
case
scenario
exposure
prediction
exceeds
the
Agency
pre­
determined
safe
level,
and
that
prediction
is
deemed
a
gross
over­
estimation,
the
input
parameters,
including
the
default
concordance
parameters
(
e.
g.,
in
RZWQM),
should
be
reviewed
specific
for
the
cumulative
assessment
group
of
the
N­
methyl
carbamate
chemicals.
Attempts
should
be
made
to
compare
model
results
to
available
monitoring
data.
78
of
113
Question
3
Addressing
the
cumulative
risk
assessment
needs:

The
Agency
has
considered
analyses
of
the
capabilities
of
three
ground
water
models
for
use
in
the
carbamate
cumulative
exposure
assessment
and
in
individual
chemical
(
aggregate)
assessments.
The
major
areas
of
evaluation
 
hydrology,
management
practices,
pesticide
transport
processes,
and
ease
of
use
of
the
model
 
are
described
in
the
background
document
Drinking
Water
Exposure
Assessment:
Ground
Water
Model
Evaluation.
For
the
cumulative
assessment,
OPP
will
compare
results
of
all
three
models
with
each
other
and
with
available
monitoring.
The
ultimate
evaluation
goal
for
the
models
is
how
well
each
meets
the
requirement
of
providing
reasonable,
health­
protective
estimates
of
pesticide
residues
for
use
in
aggregate
and
cumulative
drinking
water
exposure
assessments.

In
the
SAP's
estimation,
how
well
do
the
three
ground
water
models
OPP
proposes
to
use
 
RZWQM,
PRZM,
and
LEACHP
 
compare
in
addressing
hydrology?
Macropore
flow?
Rainfall
characteristics?
Management
practices?
Pesticide
fate
and
transport?
Formation
and
movement
of
transformation
products?
Can
the
panel
recommend
other
criteria
that
should
be
considered
in
evaluating
the
effectiveness
of
these
models
for
estimating
drinking
water
exposure
for
regulatory
purposes?

Panel
Response
The
Panel's
deliberations
on
each
of
the
parameters
are
discussed
in
the
following
sections,
along
with
recommendations
for
additional
parameters
that
should
be
considered
in
the
models.

Management
Practices
and
Pesticide
Application
It
is
necessary
to
identify
which
processes
are
most
important
for
accurately
modeling
N­
methyl
carbamate
pesticides.
For
example,
incorporation
of
applied
pesticides
into
the
soil
should
probably
be
considered
because
this
is
a
recommended
agricultural
practice
in
conjunction
with
the
application
of
some
N­
methyl
carbamate
pesticides.
Another
important
factor
to
consider
is
the
amount
of
pesticide
applied.
The
quantity
used
can
usually
be
estimated
based
on
sales
records
and
on
label
instructions.
The
management
practices
(
such
as
Integrated
Pest
Management)
can
greatly
affect
the
amount
of
pesticide
applied
and
is,
therefore,
an
important
consideration
in
modeling
for
this
purpose.

Management
practices
usually
have
little
effect
on
the
water
balance
with
the
exception
of
those
that
include
a
winter
cover
crop
(
Walter
et
al.,
1979).
In
these
cases,
79
of
113
there
is
less
percolation
during
the
early
spring.
Therefore,
management
practices
that
include
cover
crops
should
be
considered
in
the
simulation
of
the
hydrology.

Another
concern
is
that
agricultural
uses
may
not
be
the
only
source
of
N­
methyl
carbamate
pesticides,
especially
in
suburban
settings,
yet
these
settings
may
still
be
drawing
water
from
surficial
aquifer
wells.
Non­
agricultural
use
may
be
small
in
its
overall
contribution,
but
such
uses
may
be
important
and
could
even
dominate
in
selected
situations.
It
may
be
important
for
the
models
to
allow
for
consideration
of
nonagricultural
sources
.

Degradation
Product
Formation
N­
methyl
carbamate
pesticides
in
the
environment
are
relatively
labile
and
readily
undergo
hydrolysis
giving
rise
to
the
amino
acid
and
the
alcohol
form
of
the
leaving
group,
e.
g,
1­
naphthol.
The
leaving
group
component
may
itself
be
somewhat
toxic.
There
are
likely
to
be
other
products
as
well.
These
should
be
traceable
as
part
of
the
modeling.
Therefore,
the
model
should
simulate
formation
and
movement
of
degradation
products.
For
example,
metabolites
of
aldicarb
such
as
aldicarb
sulfoxide
and
aldicarb
sulfone
may
be
more
important
than
the
parent
compound
(
Kraft
and
Helmke,
1991;
Smelt
et
al.,
1995).
This
component
must
be
thoroughly
tested
because
there
have
been
mass
balance
problems
with
transformation
products
in
model
testing.

Preferential
Flow
Macropore
flow
is
likely
essential
to
accurately
model
carbamate
transport
to
shallow
groundwater.
Even
on
loamy
sand
(
Ritter
et
al.,
1996),
macropore
flow
can
be
an
important
process.
Specifically,
N­
methyl
carbamate
pesticides
such
as
carbofuran,
can
be
transported
to
shallow
groundwater
via
preferential
flow
(
Isensee
et
al.,
1990;
Kladivko
et
al,
1999).
Model
codes
including
a
description
of
preferential
flow
processes
required
less
calibration
efforts
to
meet
the
FOCUS
model
performance
criteria
on
a
sandy
loam
soil
than
those
without
such
description
(
Thorsen
et
al.,
1998).

The
methods
to
simulate
preferential
flow
are
vastly
different
between
the
models.
This
should
not
be
a
surprise
since
the
theory
for
simulating
preferential
flow
is
still
under
development
and
not
understood
well.
Of
interest
here,
most
aquifers
are
overlain
by
sandy
soils.
Also,
most
N­
methyl
carbamate
pesticides
are
applied
on
these
sandy
soils.
On
these
types
of
soils,
unstable
fingered
flow
and
fingered
flow
are
important.
Although
RZWQM
is
the
most
sophisticated
model
in
simulating
preferential
flow,
it
does
not
consider
fingered
flow
(
see
Figure
2).
To
model
fingered
flow,
the
maximum
intensity
of
a
storm
might
determine
the
number
of
fingered
flow
paths
(
Selker
et
al.,
1996).
RZWQM
clearly
stands
out
in
its
sophistication
to
model
macropore
flow
and
is
likely
superior
above
the
other
models,
but
none
of
the
models
address
the
unstable
fingered
flow
phenomena
in
sandy
soils.
At
the
present
time,
preferential
flow
including
funnel
flow
in
sandy
soils
could
be
included
by
model
developers,
since
the
initial
theory
has
been
80
of
113
developed
and
could
be
tested
in
field
situations
(
Steenhuis
et
al.,
2001;
Kim
et
al.,
2005).
EPA
should
be
looking
for
these
types
of
models
for
inclusion
in
the
risk
assessment.

Model
users
should
be
aware
that
it
is
simplistic
to
divide
up
the
medium
into
one
preferential
flow
region
and
a
matrix
component
(
Steenhuis
et
al.,
1990).
In
reality,
there
is
a
continuum
between
the
fastest
flow
path
and
the
slowest
flow
paths.
The
experimental
and
theoretical
work
of
Kung
et
al.
(
2000,
2005)
and
Gish
et
al.
(
2004)
confirmed
this
concept.

In
structured
soil
(
e.
g.,
silty,
loamy,
and
clayey
soil),
macropore
characteristics
for
RZWQM
are
easily
parameterized.
Little
is
known
for
sandy
soils
though
and
work
is
needed
to
parameterize
RZWQM
for
sandy
soils.

Hydrology
and
Rainfall
Characteristics
Most
models
predict
the
water
balance
quite
accurately,
including
the
ones
proposed
for
use
by
the
Agency.
Accurate
determination
of
daily
rainfall
amount
is
essential
and
may
be
one
of
the
most
crucial
issues
concerning
accurate
simulation
of
water
balance
(
Wagenet
and
Hutson,
1996).
The
rainfall
rate
is
also
important,
especially
if
macropore
flow
is
identified
as
an
important
process
to
include
in
the
model.
Skopp
et
al.
(
1981)
varied
how
water
was
added
to
a
column,
which
affected
the
rate
of
preferential
flow,
and
consequently
the
breakthrough
curves
were
completely
different
(
also
see
Skopp
and
Gardner,
1992).
Steenhuis
et
al.
(
1990)
developed
a
preferential
flow
model
whereby
flow
was
directly
related
to
the
rate
of
application.

Correctly
characterizing
rainfall
intensity
may
be
especially
important
when
applying
RZWQM,
because
it
will
not
simulate
macropore
flow
unless
the
rainfall
is
sufficiently
intense.
That
is,
24
hour
or
hourly
rainfall
may
not
be
sufficiently
intense
to
result
in
macropore
flow
with
RZWQM.
Pesticides
may
move
to
shallow
groundwater
by
macropore
flow
in
the
sandy
soils
if
heavy
rainfall
occurs
shortly
after
pesticide
application
(
Ritter
et
al.,
1996).
Subsequent
rainfall
events
may
be
less
important
because
the
first
storm
after
application
can
move
solutes
into
the
soil
matrix,
thereby
reducing
the
potential
for
transport
in
macropores
(
Shipitalo
et
al.,
1990).
Therefore,
efforts
should
be
applied
to
develop
breakpoint
rainfall
input
files
for
the
first
few
storms
after
pesticide
application.

Pesticide
Fate
Some
of
the
more
sophisticated
pesticide
fate
routines
in
models
should
possibly
be
avoided.
For
example,
the
RZWQM
irreversible
binding
routine
did
not
provide
accurate
simulations
of
metribuzin
in
percolate,
suggesting
this
concept
may
need
to
be
explored
further
(
Malone
et
al.,
2004).
Also,
estimating
parameters
for
kinetic
sorption
may
be
difficult.
However,
if
most
pesticide
is
transported
through
percolate
shortly
after
application,
irreversible
binding
and
kinetic
sorption
may
be
less
important
than
short­
term
81
of
113
half­
life
and
equilibrium
sorption.

Both
the
degradation
rate
and
the
adsorption
strength
are
highly
variable
between
fields.
The
variability
of
degradation
rates
between
sites
is
one
of
the
main
problems
for
simulating
the
pesticide
concentration
in
the
aquifers
with
confidence,
and
this
might
be
the
main
reason
that
the
pesticide
concentration
in
the
USGS
model
evaluations
(
included
as
background
documents)
are
so
poorly
simulated.
Pivetz
and
Steenhuis
(
1995)
and
Pivetz
et
al.
(
1996)
concluded
that
the
degradation
rate
may
vary
significantly
in
time.
Reasonable
estimates
can
only
be
obtained
if
prior
experiments
have
been
carried
out
at
the
site.
Because
simple
parameters
such
as
the
linear
adsorption
model
and
the
first
order
degradation
rate
model
are
difficult
to
obtain,
it
will
likely
be
too
difficult
to
apply
more
complicated
formulations
of
these
relationships
nationally.

Enhanced
or
accelerated
degradation
has
been
reported
in
N­
methyl
carbamate
pesticides
such
as
carbofuran
(
Cogger
et
al.,
1998;
Getzin
and
Shanks,
1990)
and
aldicarb
(
Suett
and
Jukes,
1988;
Smelt
et
al.,
1987).
Enhanced
degradation,
defined
as
an
increase
in
pesticide
degradation
with
each
application,
may
lead
to
higher
or
more
frequent
applications
because
insecticidal
efficacy
will
be
reduced.

Tile
Drainage
If
N­
methyl
carbamate
pesticides
are
applied
to
areas
that
are
tile
drained,
then
the
selected
model
should
simulate
tile
drainage
for
accurate
loading
estimates
to
surface
water.
Recent
research
suggests
that
the
pesticide
isoxaflutole
and
its
metabolite
RPA
202248
are
transported
to
the
edge
of
the
field
through
subsurface
drains
at
about
1
m
at
rates
similar
to
runoff
(
personal
communications,
Garey
A.
Fox,
Assistant
Professor,
University
of
Mississippi).
However,
tile
drainage
may
not
be
necessary
in
the
model
for
the
purposes
of
the
cumulative
risk
assessment,
especially
if
this
is
a
screening
level
tool.

Model
Calibration,
Sensitivity,
Testing,
and
Ease
of
Use
The
use
of
any
model
of
pesticide
contamination
from
agricultural
use
is
likely
to
give
only
an
estimation
of
the
contamination
found
in
the
surficial
aquifer.
The
use
of
the
PRZM
system
is
well
established
with
respect
to
EPA
usage.
This
suggests
that
its
use
in
this
context
may
be
supported
above
and
beyond
other
ostensibly
better,
but
perhaps
more
complicated
and
difficult
to
use
models.
Further,
parameterization
of
the
other
models
may
be
difficult
since
data
may
not
be
available
for
scenarios
of
interest.

Experience
with
LeachN,
the
nitrogen
movement
implementation
of
the
Leach
model,
demonstrates
the
difficulty
of
getting
these
models
to
predict
observed
levels
of
materials,
even
with
intensive
measurement
of
soil
characteristics,
initial
conditions
and
control
of
soil
nutrient
additions.
EPA
may
wish
to
create
their
own
tool
for
this
purpose
that
they
understand
and
can
distribute
and
control.
82
of
113
The
sensitivity
of
the
models
to
key
parameters
that
are
likely
to
be
changed
in
application
of
the
models
should
be
well
understood.
Also,
the
sensitivity
of
key
parameters
relative
to
the
group
of
models
being
considered
would
provide
insight
into
the
most
applicable
model
or
models
for
various
situations.
Perhaps
probabilistic
or
nonlinear
analysis
of
the
model
output
might
be
the
best
approach
to
analyzing
the
parameter
interdependencies.

The
ease
of
model
calibration
and
the
ability
to
calibrate
each
model
should
be
considered.
Models
often
must
be
calibrated
to
perform
adequately.
Therefore,
calibrations
will
be
important
considerations
when
using
the
groundwater
models
to
estimate
pesticides
reaching
shallow
groundwater.

The
ability
to
automate
the
use
of
each
model
so
that
it
can
readily
utilize
data
from
databases
and
GIS
to
facilitate
automated
model
runs
will
likely
be
important
for
future
applications
of
the
models.
Models
that
require
the
use
of
a
model­
specific
interface
may
not
work
very
well
or
at
all
for
such
purposes.

Other
criteria
to
take
into
consideration
are:
transparency
in
the
model's
calculations,
the
data
requirements,
post­
analysis
of
the
model
output,
internal
documentation
of
code,
ability
to
modify
code,
ability
to
interface
the
code
with
thirdparty
GUIs,
and
ability
to
link
to
statistical
analysis
packages
and
GIS
packages.
In
analyzing
a
full
cumulative
risk
assessment,
users
and
reviewers
must
have
models
that
are
simple
to
evaluate
during
all
portions
of
the
risk
assessment.

For
model
testing,
as
complete
a
dataset
as
possible
should
be
used.
Some
of
the
discrepancies
between
model
and
data
may
be
resolved
by
averaging
samples
taken
over
a
broader
area.
The
models
have
much
in
common,
so
discrepancies
between
models
must
be
explicable
in
each
case
in
terms
of
known
differences
in
model
assumptions.
The
dataset
should
include
soil
samples,
runoff,
and
percolate,
because
neglecting
one
of
these
dissipation
pathways
may
provide
an
incomplete
picture
of
model
performance
(
Malone
et
al.,
1999).
When
preferential
flow
occurs,
soil
samples
and
suction
lysimeters
may
not
adequately
characterize
chemical
movement
to
shallow
groundwater,
and
percolate
samples
may
be
necessary
(
Malone
et
al.,
2000).

The
effects
of
water
purification
processes
on
carbamate
levels
also
should
potentially
be
considered.
If
such
processes
routinely
eliminate
most
N­
methyl
carbamate
pesticides
from
drinking
water,
this
modeling
would
seem
to
be
superfluous
except
for
those
drinking
untreated
water.
Estimation
of
these
purification
effects
would
thus
seem
critical,
particularly
new
filtration
technologies
and
the
effects
of
chlorination
and
chloramines
disinfectant
technology.
At
the
current
time,
the
potential
for
significant
overestimation
of
carbamate
levels
from
these
models
seems
to
exist.
These
factors,
together
with
the
low
level
of
exposure
to
N­
methyl
carbamate
pesticides
from
drinking
83
of
113
water
relative
to
other
exposure
sources,
suggest
this
modeling
may
be
of
value
for
only
a
subset
of
the
public.

Question
4
Estimating
cumulative
carbamate
exposures
in
ground
water:

Available
monitoring
data,
primarily
from
the
USGS
NAWQA
program,
indicate
that
more
than
one
carbamate
in
the
cumulative
action
group
may
occur
together
in
ground
water
(
see
the
drinking
water
exposure
section
of
the
case
study).
Co­
occurrence
in
ground
water
results
when
more
than
one
carbamate
is
used
at
different
times
on
the
same
crop,
on
different
crops
in
rotation
on
the
same
fields,
or
on
different
crops
grown
on
adjacent
fields.

For
surface
water
sources
of
drinking
water,
OPP
adjusted
estimated
pesticide
concentrations
in
the
modeled
reservoir
by
percent
crop
area
and
percent
crop
treated
in
the
watershed
to
reflect
the
dilution
of
untreated
areas
on
the
total
pesticide
load
reaching
the
reservoir
(
see
the
description
of
the
surface
water
exposure
assessment
in
the
analysis
methods
of
the
Drinking
Water
section
of
the
case
study).

Given
that
less
mixing
of
water
from
different
fields
is
expected
in
shallow
aquifers,
should
OPP
use
similar
adjustments
to
ground
water
concentrations
estimated
from
the
leaching
models?
If
not,
what
recommendations
does
the
Panel
have
to
account
for
the
potential
contributions
from
different
fields
treated
with
carbamates?

Panel
Response
Surface
water
and
groundwater
behave
quite
differently.
Surface
water
travels
relatively
rapidly
from
any
location
in
the
watershed
to
the
outlet,
and
the
Agency's
adjustment
for
surface
water
edge
of
field
concentration
by
percent
crop
area
and
percent
crop
treated
in
the
watershed
is
reasonable.
Thus,
the
assumption
is
made
that
pesticide
concentration
in
surface
drinking
water
can
be
represented
by
the
average
spatial
concentration
of
the
runoff.

The
assumption
of
taking
spatial
averages
in
the
watershed
is
not
valid
for
groundwater.
Generally,
groundwater
travels
slowly,
usually
less
than
100
m/
year.
The
pesticide
concentration
in
each
well
depends
on
its
surroundings
and
cannot
be
represented
by
an
average
concentration.
Thus,
the
people
drinking
water
from
the
wells
are
exposed
to
different
concentrations.
This
is
important
for
the
risk
analysis
where
highest
concentrations
are
often
of
interest.
84
of
113
The
different
concentrations
for
different
wells
are
schematically
depicted
in
Figure
1.
Pesticides
are
applied
to
a
field
as
indicated.
The
pesticides
travel
in
a
distinct
path
to
the
river
through
the
ground
water.
Figure
1
illustrates
that
only
the
shallow
well
downstream
from
the
field
contains
pesticide,
while
the
other
wells
are
pesticide
free.
The
drinking
water
well
in
Figure
1
samples
water
from
a
larger
volume
and
will
be
affected
by
more
than
one
field.

The
screens
of
shallow
wells
are
typically
a
minimum
of
4
feet
long
and
sample
pesticides
over
these
depths.
This
means
that
the
daily
input
at
the
top
of
the
aquifer
is
averaged
over
a
certain
time
period.
To
help
describe
this
better,
Figure
2
depicts
a
close
up
view
of
the
situation
near
the
pesticide
field
from
Figure
1.
Only
the
water
that
fell
on
top
of
the
field
is
shown
in
the
groundwater
depicted.
It
can
be
seen
that
the
oldest
water
is
the
deepest,
while
the
water
which
just
arrived
is
near
the
surface
of
the
aquifer.
While
the
water
is
moving
downward,
it
is,
at
the
same
time,
moving
sideways
toward
the
river.
The
line
in
the
groundwater
in
Figure
2
is
a
streamline.
In
this
case
water
that
falls
at
the
upper
edge
of
the
field
follows
this
path.

Although
spatial
averages
cannot
be
taken
for
wells,
temporal
averaging
can
be
done.
It
is
not
difficult
to
calculate
a
reasonable
time
period
over
which
the
pesticide
concentrations
from
the
vadose
pesticide
model
should
be
averaged.
The
depth
of
"
recharge
layer"
per
year
is
simply
the
recharge
per
year,
r,
divided
by
the
saturated
moisture
content,
 s.
The
depth
of
water
that
represents
water
reaching
the
aquifer
each
year,
D,
can
be
calculated
taking
into
account
the
adsorption
partition
coefficient,
k,
and
the
density
of
the
soil,
 ,
as:

The
time,
T,
over
which
the
pesticides
arriving
to
the
groundwater
should
be
averaged
can
then
be
simply
written
as:

D
S
T
=

where
S
is
the
screen
length.

The
location
of
stream
lines
can
be
found
with
the
method
presented
by
Gelhar
and
Williams
(
1974).
This
method
was
used
in
a
model
called
MOUSE
(
Steenhuis
et
al.,
1987).
The
model
had
many
similarities
to
the
models
currently
used
by
EPA
(
but
without
preferential
flow
paths).
The
details
of
simulating
streamlines
are
given
in
these
publications.
The
method
consists
of
two
simple
equations;
one
that
determines
the
displacement
in
the
horizontal
direction
and
the
other
in
the
vertical
direction.
D
r
k
s
=
+
 
 
85
of
113
The
two
surface
water
modifiers,
percentage
of
crop
area
(
PCA)
and
percentage
of
crop
treated
(
PCT),
can
reasonably
be
used
in
ground
water
prediction
also.
However,
the
persistence
of
N­
methyl
carbamate
pesticides
in
ground
water
would
require
considerations
of
multiple
years
of
pesticide
use.
This
is
different
from
focusing
the
surface
water
scenarios
on
spikes
associated
with
runoff
and
its
relatively
fast
decline
to
low
or
nondetectable
levels
after
runoff
events.
Considerations
also
can
be
given
for
time
of
pesticide
use
and
ground
water
movement.
In
terms
of
PCA,
the
pattern
of
change
in
land
use
over
time
should
be
accounted
for
(
e.
g.,
to
consider
pesticide
use
at
sites
that
were
once
agricultural).
When
available,
multiple
years
of
data
on
PCT
also
may
be
useful
to
modify
the
default
assumption
of
100%
PCT.

Figure
1.
Schematic
of
flow
paths
in
a
valley
with
a
river.
Only
the
shallow
well
downstream
of
the
field
is
affected
by
the
field
in
which
the
pesticides
are
applied.
86
of
113
Figure
2.
Simple
schematic
of
ground
water
movement
under
a
pesticide
movement.

Question
5
Several
carbamates
are
known
to
persist
for
long
periods
of
time
after
they
reach
ground
water,
particularly
slightly
acidic
to
acidic
ground
water
(
see
the
drinking
water
exposure
section
of
the
case
study).
This
differs
from
estimates
of
exposures
in
surface
water,
where
pesticide
concentrations
tend
to
occur
as
spikes
associated
with
runoff
and
decline
to
low
or
nondetectable
levels
after
runoff
events.
Pesticide
concentrations
in
acidic
ground
water
are
slower
to
respond
to
changes
in
use
patterns
and
mitigation
actions
than
would
be
expected
in
surface
water.

What
recommendations
does
the
Panel
have
for
addressing
carbamate
persistence
in
ground
water
in
order
to
provide
a
reasonable,
health­
protective
estimate
of
residue
levels
in
shallow
ground
water
sources
of
drinking
water?
87
of
113
Panel
Response
The
Agency
is
considering
three
possible
approaches:

1)
at
one
extreme,
assume
no
background
residues
(
drinking
water
exposures
would
reflect
only
what
is
estimated
by
modeling),
i.
e.,
all
residues
in
groundwater
are
"
fresh";

2)
at
the
other
extreme,
assume
a
baseline
background
concentration
(
based
on
available
monitoring),
with
model
estimates
as
additions
and
no
decline;

3)
in
between,
include
the
background
levels
with
model
estimates,
but
provide
an
estimate
of
decline
in
residues
over
time
(
estimate
based
on
long­
term
trends
in
monitoring)

The
EPA's
3rd
suggested
approach
from
the
"
background
document"
for
Session
3
appears
sensible.
However,
it
is
unclear
how
EPA
would
obtain
the
"
long­
term
trends
in
monitoring".

The
background
document
clearly
identified
that
carbamate
transformation
can
be
pH
specific.
Also,
the
transformations
of
aldicarb
metabolites
are
concentration
specific
(
Smelt
et
al.,
1995).
That
is,
higher
concentrations
degrade
more
slowly
than
lower
concentrations.
At
least
two
peer­
reviewed
sources
that
studied
aldicarb
transformation
used
initial
concentrations
greater
than
150
ug/
L.
Some
studies
of
aldicarb
degradation
suggest
that
anaerobic
conditions
increase
degradation
(
Smelt
et
al.,
1995;
Kraft
and
Helmke,
1992).

The
question
of
persistence
has
underlying
implications
on
the
travel
distance
of
the
pesticide
through
a
continuous
aquifer.
There
are
a
few
references
on
estimating
characteristic
travel
distance
of
a
persistent
pesticide
based
on
its
chemical
properties
(
Bennet
et
al.,
1999,
1998).
88
of
113
REFERENCES
Bennett,
D.
H.,
McKone,
T.
E.,
Kastenberg,
W.
E.,
Matthies,
M.
(
1998)
General
Formulation
of
Characteristic
Travel
Distance
for
Semi­
Volatile
Organic
Chemicals
in
a
Multi­
media
Environment,
Environ.
Sci.
Technol.
32:
4023­
4030.

Bennett,
D.
H.,
Kastenberg,
W.
E.,
McKone,
T.
E.
(
1999)
General
Formulation
of
Characteristic
Time
for
Persistent
Chemicals
in
a
Multimedia
Environment,
Environ.
Sci.
Technol.
33:
503­
509.

Boll,
J.,
Steenhuis,
T.
S.,
and
Selker,
J.
S.
(
1992)
Fiberglass
Wicks
for
Sampling
Water
and
Solutes
in
the
Vadose
Zone.
Soil
Sci.
Soc.
Am.
J.
56:
701­
707.

Boll,
J.,
Selker,
J.
S.,
Shalit,
G.,
and
Steenhuis,
T.
S.
(
1997)
Frequency
Distribution
of
Water
and
Solute
Transport
Properties
Derived
from
Pan
Sampler
Data.
Water
Resour.
Res.
33:
2655­
2664.

Cogger
CG,
Stark
J.
D.,
Bristow
P.
R.,
Getzin
L.
W.,
Montgomery
M.
(
1998)
Transport
and
Persistence
of
Pesticides
in
Alluvial
Soils:
Ii.
Carbofuran.
J.
Environ.
Qual.
27(
3):
551­
556.

Cooper,
B.,
Navulur,
K.,
Engel,
B.,
and
Hahn,
L.
(
1997)
Groundwater
vulnerability
evaluation
to
pesticide
and
nitrate
pollution
on
a
regional
scale
using
GIS.
http://
pasture.
ecn.
purdue.
edu/~
aggrass/
groundwater
Gelhar
L.
D.
and
Williams,
J.
L.
(
1974)
Ground
water
quality
modeling.
Ground
Water
12:
399­
408.

Getzin
L.
W.
and
Shanks,
C.
H.
(
1990)
Enhanced
Degradation
Of
Carbofuran
In
Pacific­
Northwest
Soils.
J.
Environ.
Sci.
Health
Part
B.
25:
433­
446
Gish
T.
J.,
Kung,
K­
J.
S.,
Perry,
D.
C.,
Posner,
J.,
Bubenzer,
G.,
Helling,
C.
S.,
Kladivko,
E.
J.
and
Steenhuis,
T.
S.
(
2004)
Impact
of
preferential
flow
at
varying
irrigation
rates
by
quantifying
mass
fluxes.
J.
Environ.
Qual.
33(
3):
1033­
1040
Isensee,
A.
R.,
Nash,
R.
G.,
Helling,
C.
S.
(
1990)
Effect
Of
Conventional
Vs
No­
Tillage
On
Pesticide
Leaching
To
Shallow
Groundwater.
J.
Environ.
Qual.
19(
3):
434­
440.

Kim,
Y.­
J.,
Darnault,
C.
J.
G.,
Bailey,
N.
O.,
Parlange,
J.­
Y.
and
Steenhuis,
T.
S.
(
2005)
Equation
for
Describing
Solute
Transport
in
Field
Soils
with
Preferential
Flow
Paths.
Soil
Sci.
Soc.
Am.
J.
69.
<
in
press>
89
of
113
Kladivko,
E.
J.,
Grochulska,
J.,
Turco,
R.
F.,
Van
Scoyoc,
G.
E.,
and
Eigel,
J.
D.
(
1999)
Pesticide
and
Nitrate
Transport
into
Subsurface
Tile
Drains
of
Different
Spacings.
J.
Environ.
Qual.
28(
3):
997­
1004.

Kraft,
G.
J,
and
Helmke,
P.
A.
(
1991)
Aldicarb
Residue
Degradation
Rates
At
3
Depths
of
a
Wisconsin
Central
Sand
Plain
Aquifer.
Pest.
Sci.
33(
1):
47­
55.

Kraft,
G.
J.
and
Helmke,
P.
A.
(
1992)
Dependence
Of
Aldicarb
Residue
Degradation
Rates
On
Groundwater
Chemistry
In
The
Wisconsin
Central
Sands.
J.
Environ.
Qual.
21(
3):
368­
372.

Kung,
K.­
J.
S.,
Kladivko,
E.
J.,
Gish,
T.
J.,
Steenhuis,
T.
S.,
Bubenzer,
G.
and
Helling,
C.
S.
(
2000)
Quantifying
Preferential
Flow
by
Breakthrough
of
Sequentially
Applied
Tracers:
Silt
Loam
Soil.
Soil
Sci.
Soc.
Am.
J.
64:
1296­
1304.

Kung,
K.­
J.
S.,
Hanke,
M.,
Helling,
C.
S.,
Kladivko,
E.
J.,
Gish,
T.
J.,
Steenhuis,
T.
S.
and
Jaynes,
D.
B.
(
2005)
Quantifying
Pore
Size
Spectrum
of
Macropore­
Type
Preferential
Pathways.
Soil
Sci.
Soc.
Am.
J.
(
in
press)

Malone,
R.
W.,
Warner,
R.
C.,
Workman,
S.
R.
and
Byers,
M.
E.
(
1999)
Modeling
Surface
And
Subsurface
Pesticide
Transport
Under
Three
Field
Conditions
Using
PRZM­
3
And
GLEAMS
Transactions
of
the
ASAE.
42(
5):
1275­
1288.

Malone,
R.
W.,
Shipitalo,
M.
J.,
Douglass,
L.
W.,
Owens,
L.
B.,
Nelsen,
T.
C.,
Warner,
R.
C.,
and
Byers,
M.
E.
(
2000)
Assessing
Herbicide
Movement
Using
Soil
Samples
Versus
Percolate
Samples
Transactions
of
the
ASAE.
43(
2):
343­
348.

Malone,
R.
W.,
Ma,
L.
W.,
Wauchope
R.
D.,
Ahuja,
L.
R.,
Rojas,
K.
W.,
Ma,
Q.
L.,
Warner,
R.
and
Byers,
M.
(
2004)
Modeling
hydrology,
metribuzin
degradation
and
metribuzin
transport
in
macroporous
tilled
and
no­
till
silt
loam
soil
using
RZWQM.
Pest
Manage.
Sci.
60(
3):
253­
266.

Nolan,
B.
T.,
Hitt,
K.
J.
and
Ruddy,
B.
C.
(
2002)
Map
of
the
probability
that
nitrate
exceeds
4
mg/
L
in
shallow
ground
waters
of
the
United
States,
based
on
the
LR
model.
Accessed
April
13,
2004
at
URL:
http://
water.
usgs.
gov/
lookup/
getspatial?
gwrisk
Pivetz,
B.
E.
and
Steenhuis,
T.
S.
(
1995)
Soil
Matrix
and
Macropore
Biodegradation
of
2,4­
D.
J.
Env.
Qual.
24:
564­
570.

Pivetz,
B.
E.,
Kelsey,
J.
W.,
Steenhuis,
T.
S.
and
Alexander,
M.
(
1996)
A
Procedure
to
Calculate
Biodegradation
During
Preferential
Flow
Through
Heterogeneous
Soil
Columns.
Soil
Sci.
Soc.
Am.
J.
60:
381­
388.
90
of
113
Ritter,
W.
F.,
Chirnside,
A.
E.
M.
and
Scarborough,
R.
W.
(
1996)
Movement
And
Degradation
Of
Triazines,
Alachlor,
And
Metolachlor
In
Sandy
Soils.
J.
Environ.
Sci.
Health
Part
A.
31(
10):
2699­
2721.

Selker,
J.
S.,
Steenhuis,
T.
S.
and
Parlange,
J.­
Y.
(
1996)
An
Engineering
Approach
to
Fingered
Vadose
Pollutant
Transport.
In:
Fingered
Flow
in
Unsaturated
Soil:
From
Nature
to
Model,
T.
S.
Steenhuis,
C.
J.
Ritsema,
and
L.
W.
Dekker,
Eds.
Special
Issue
of
Geoderma
70:
197­
206.

Shalit,
G.,
Steenhuis,
T.
S.,
Hakvoort,
H.
M.,
Boll,
J.,
Geohring,
J.
D.
and
van
Es,
H.
M.
(
1995)
Subsurface
Drainage
Water
Quality
from
a
Structured
Soil.
ASCE
J.
Irrig.
Drain.
121:
239­
247.

Shipitalo,
M.
J.,
Edwards,
W.
M.,
Dick,
W.
A.
and
Owens,
L.
B.
(
1990).
Initial
Storm
Effects
On
Macropore
Transport
Of
Surface­
Applied
Chemicals
In
No­
Till
Soil.
Soil
Sci.
Soc.
Am.
J.
54(
6):
1530­
1536.

Skopp,
J.,
Gardner,
W.
R.
and
Tyler,
C.
J.
(
1981)
Solute
movement
in
structured
soils
­
2­
region
model
with
small
interaction
Soil
Sci.
Soc.
Am.
J.
45(
5):
837­
842
Skopp,
J.
and
Gardner,
W.
R.
(
1992)
Miscible
displacement
­
an
interacting
flow
region
model.
Soil
Sci.
Soc.
Am.
J.
56
(
6):
1680­
1686
Smelt,
J.
H.,
Crum,
S.
J.
H.,
Teunissen,
W.
and
Leistra,
M.
(
1987)
Accelerated
degradation
of
aldicarb,
oxamyl
and
ethoprophos
after
repeated
soil
treatments.
Crop
Protection
6:
295­
303.

Smelt,
J.
H.,
Vandepeppelgroen,
A.
E.
and
Leistra,
M.
(
1995)
Transformation
Of
Aldicarb
Sulfoxide
And
Aldicarb
Sulfone
In
4
Water­
Saturated
Sandy
Subsoils.
Pest.
Sci.
44
(
4):
323­
334.

Steenhuis,
T.
S.,
Pacenka,
S.
and
Porter,
K.
S.
(
1987)
MOUSE:
A
Management
Model
for
Evaluating
Groundwater
Contamination
from
Diffuse
Surface
Sources
Aided
by
Computer
Graphics.
Appl.
Agric.
Res.
2:
277­
289.

Steenhuis,
T.
S.,
Parlange,
J.­
Y.
and
Andreini,
M.
S.
(
1990)
A
Numerical
Model
for
Preferential
Solute
Movement
in
Structured
Soils.
Geoderma
46:
193­
208.

Steenhuis,
T.
S.,
Kim,
Y.­
J.,
Parlange,
J.­
Y.,
Akhtar,
M.
S.,
Richards,
B.
K.,
Dung,
K.­
J.
S.,
Gish,
T.
J.,
Dekker,
L.
W.,
Ritsema,
C.
J.,
and
Aburime,
S.
O.
(
2001)
An
Equation
for
Describing
Solute
Transport
in
Field
Soils
with
Preferential
Flow
Paths.
Preferential
Flow,
Water
Movements
and
Chemical
Transport
in
the
Environment.
D.
D.
Bosch
and
K.
W.
King,
Editors.
ASAE
Proceedings.
Second
International
Symposiums.
Honolulu,
HI.
137­
140.
91
of
113
Suett,
D.
L.
and
Jukes,
A.
A.
(
1988)
Accelerated
degradation
of
aldicarb
and
its
oxidation
precuts
in
previously
treated
soils.
Crop
Protection
7:
147­
152.

Thorsen,
M.,
Jorgensen,
P.
R.,
Felding,
G.,
Jacobsen,
O.
H.,
Spliid,
N.
H.
and
Refsgaard,
J.
C.
(
1998).
Evaluation
Of
A
Stepwise
Procedure
For
Comparative
Validation
Of
Pesticide
Leaching
Models.
J.
Environ.
Qual..
27(
5):
1183­
1193
Wagenet,
R.
J.
and
Hutson,
J.
L.
(
1996)
Scale­
dependency
of
solute
transport
modeling/
GIS
applications
J.
Environ.
Qual.
25
(
3):
499­
510
Walter,
M.
F.,
Steenhuis,
T.
S.
and
Haith,
D.
A.
(
1979)
Nonpoint
Source
Pollution
Control
by
Soil
and
Water
Conservation
Practices.
Trans.
ASAE
22:
834­
840.
92
of
113
SAP
Minutes
No.
2005­
01
A
Set
of
Scientific
Issues
Being
Considered
by
the
Environmental
Protection
Agency
Regarding:

N­
METHYL
CARBAMATE
CUMULATIVE
RISK
ASSESSMENT:
PILOT
CUMULATIVE
ANALYSIS
SESSION
4:
N­
METHYL
CARBAMATE
EXPOSURE
ASSESSMENT:
A
PILOT
CASE
STUDY
February
18,
2005
FIFRA
Scientific
Advisory
Panel
Meeting
held
at
the
Holiday
Inn
National
Airport
Arlington,
VA
Mr.
Joseph
E.
Bailey
Steven
G.
Heeringa,
Ph.
D.
Designated
Federal
Official
FIFRA
SAP
Session
Chair
FIFRA
Scientific
Advisory
Panel
FIFRA
Scientific
Advisory
Panel
Date:
April
15,
2005
Date:
April
15,
2005
93
of
113
Federal
Insecticide,
Fungicide
and
Rodenticide
Act
Scientific
Advisory
Panel
Meeting
February
18,
2005
N­
methyl
Carbamate
Cumulative
Risk
Assessment:
Pilot
Cumulative
Analysis
Session
4:
N­
methyl
Carbamate
Exposure
Assessment:
A
Pilot
Case
Study
PARTICIPANTS
FIFRA
SAP
Session
Chair
Steven
G.
Heeringa,
Ph.
D.,
Research
Scientist
&
Director
for
Statistical
Design,
Institute
for
Social
Research,
University
of
Michigan,
Ann
Arbor,
MI
Designated
Federal
Official
Joseph
E.
Bailey,
FIFRA
Scientific
Advisory
Panel
Staff,
Office
of
Science
Coordination
and
Policy,
EPA
FIFRA
Scientific
Advisory
Panel
Members
Stuart
Handwerger,
M.
D.,
Professor
of
Pediatrics,
University
of
Cincinnati
Children's
Hospital
Medical
Center,
Cincinnati,
OH
Gary
E.
Isom,
Ph.
D.,
Professor
of
Toxicology,
School
of
Pharmacy
&
Pharmacal
Sciences,
Purdue
University,
West
Lafayette,
IN
Kenneth
M.
Portier,
Ph.
D.,
Associate
Professor,
Statistics,
Institute
of
Food
and
Agricultural
Sciences,
University
of
Florida,
Gainesville,
FL
FQPA
Science
Review
Board
Members:

John
Adgate,
Ph.
D.,
Assistant
Professor,
Division
of
Environmental
Health
Sciences,
University
of
Minnesota,
School
of
Public
Health,
Minneapolis,
MN
William
Brimijoin,
Ph.
D.,
Chair,
Pharmacology,
Mayo
Clinic
and
Medical
School,
Rochester,
MN
George
B.
Corcoran,
Ph.
D.,
Professor
&
Chairman,
Department
of
Pharmaceutical
Sciences,
Eugene
Applebaum
College
of
Pharmacy
&
Health
Sciences,
Wayne
State
University,
Detroit,
MI
94
of
113
Lutz
Edler,
Ph.
D.,
Head,
Biostatistics
Unit
C060,
German
Cancer
Research
Center,
Heidelberg,
Germany
Bernard
Engel,
Ph.
D.,
Professor,
Agricultural
&
Biological
Engineering,
Purdue
University,
West
Lafayette,
IN
Scott
Ferson,
Ph.
D.,
Senior
Scientist,
Applied
Biomathematics,
Setauket,
NY
Lawrence
J.
Fischer,
Ph.
D.,
Director,
Center
for
Integrative
Toxicology,
Michigan
State
University,
East
Lansing,
MI
Natalie
Freeman,
Ph.
D.,
Associate
Professor,
Department
of
Physiological
Sciences,
College
of
Veterinary
Medicine,
University
of
Florida,
Gainesville,
FL
Dale
Hattis,
Ph.
D.,
Research
Professor,
Center
for
Technology,
Environment
&
Development
(
CENTED),
George
Perkins
Marsh
Institute,
Clark
University,
Worcester,
MA
James
P.
Kehrer,
Ph.
D.,
Director,
Center
for
Molecular
&
Cellular
Toxicology,
College
of
Pharmacy,
The
University
of
Texas
at
Austin,
Austin,
TX
Chensheng
Lu,
Ph.
D.,
Assistant
Professor,
Department
of
Environmental
&
Occupational
Health,
Rollins
School
of
Public
Health,
Emory
University,
Atlanta,
GA
Peter
D.
M.
Macdonald,
D.
Phil.,
Professor
of
Mathematics
&
Statistics,
McMaster
University,
Hamilton,
Ontario,
Canada
David
MacIntosh,
Sc.
D.,
Senior
Associate,
Environmental
Health
&
Engineering,
Inc.,
Newton,
MA
Robert
W.
Malone,
Ph.
D.,
Agricultural
Engineer,
USDA­
ARS,
National
Soil
Tilth
Laboratory,
Ames,
IA
Christopher
J.
Portier,
Ph.
D.,
Director,
Environmental
Toxicology
Program,
National
Institute
of
Environmental
Health
Sciences,
Research
Triangle
Park,
NC
Nu­
may
Ruby
Reed,
Ph.
D.,
D.
A.
B.
T.,
Staff
Toxicologist,
Department
of
Pesticide
Regulation,
California
Environmental
Protection
Agency,
Sacramento,
CA
P.
Barry
Ryan,
Ph.
D.,
Professor,
Environmental
&
Occupational
Health,
Rollins
School
of
Public
Health,
Emory
University,
Atlanta,
GA
Tammo
S.
Steenhuis,
Ph.
D.,
Professor
of
Watershed
Management,
Cornell
University,
Ithaca,
NY
95
of
113
Michael
D.
Wheeler,
Ph.
D.,
Assistant
Professor,
Departments
of
Pharmacology
&
Medicine,
University
of
North
Carolina,
Skipper
Bowles
Center
for
Alcohol
Studies,
Chapel
Hill,
NC
96
of
113
INTRODUCTION
In
Session
4
of
this
meeting,
the
FIFRA
SAP
met
to
consider
and
review
the
N­
methyl
carbamate
pesticide
cumulative
risk
assessment:
pilot
cumulative
analysis,
specifically
issues
related
to
the
N­
methyl
carbamate
exposure
assessment:
a
pilot
case
study.
OPP
solicited
comment
from
the
SAP
on
a
case
study
which
uses
the
Relative
Potency
Factor
(
RPF)
approach
to
perform
a
cumulative
exposure
assessment
for
a
group
of
10
N­
methyl
carbamate
pesticides
that
have
been
previously
determined
to
represent
a
common
assessment
group
based
on
the
chemicals'
similar
mechanism
of
toxicity.
For
Session
4,
Dr.
Anna
Lowit
and
Mr.
David
Miller
(
Health
Effects
Division,
Office
of
Pesticide
Programs)
provided
an
introduction
and
background
on
the
pilot
case
study.
Mr.
David
Hrdy,
Mr.
Jeff
Evans,
Dr.
Steve
Nako,
and
Mr.
David
Miller
(
Health
Effects
Division,
Office
of
Pesticide
Programs)
provided
a
detailed
presentation
on
the
pilot
exposure
assessment
case
study,
including
discussions
on
dietary
assessment,
residential
exposure
inputs,
residential
use
data
and
aggregate
exposure,
cumulative
exposure
assessment
and
model
comparison.
Opening
remarks
for
Session
4
were
provided
by
Dr.
Tina
Levine
(
Acting
Director,
Health
Effects
Division,
Office
of
Pesticide
Programs).
97
of
113
SUMMARY
OF
PANEL
DISCUSSION
AND
RECOMMENDATIONS
The
Panel
members
agree
that
the
overall
case
study
is
a
thoughtful
and
useful
analysis.
EPA
has
gained
much
knowledge
from
development
of
the
organophosphate
(
OP)
cumulative
risk
assessment
and
the
Agency
is
to
be
commended
on
the
carefully
developed
exploratory
analysis
for
the
N­
methyl
carbamates,
given
the
varying
quality
of
the
datasets
that
were
available.
The
level
of
detail
is
adequate
for
a
case
study,
but
additional
effort
is
needed
in
the
full
assessment,
especially
with
regard
to
non­
food
scenarios.
It
was
felt
that
uncertainty
analyses
will
help
to
identify
the
datasets
that
should
be
examined
most
closely.

The
default
assumption
of
eating
events
close
together
in
time,
in
the
context
of
the
pilot
case
study
for
the
N­
methyl
carbamate
exposure
assessment,
is
a
reasonable
approach.
Allowing
for
recovery
seems
difficult
given
the
unknown
durations
that
would
have
to
be
assigned
to
defensibly
address
this
issue.
The
extent
to
which
eating
frequency
affects
the
outcome
of
this
analysis
is
a
testable
proposition
that
could
be
explored
as
a
distributed
analysis.
The
Panel
noted
that
a
supplementary
analysis
using
a
stratified
sample
of
high
end
consumers
would
help
justify
the
viability
of
this
approach.

The
Panel
made
several
recommendations
regarding
the
cumulative
N­
methyl
carbamate
pesticide
case
study
in
terms
of
the
data
availability,
reliability,
limitations,
and
evaluations
for
use
in
representing
patterns
of
pesticide
co­
occurence.
In
addition,
recommendations
were
made
about
pesticide
residue
degradation
estimates,
residue
data
availability,
and
the
need
to
use
regional
and
temporal
variability
in
residues
in
the
exposure
scenarios
chosen
to
be
assessed
by
the
Agency.

The
case
study
presented
to
the
Panel
clearly
demonstrated
that
available
data,
although
limited,
could
be
combined
to
conduct
a
cumulative
exposure
assessment
for
the
N­
methyl
carbamate
pesticides.
The
major
concerns
of
the
Panel
were
related
to
the
age
and
quality
of
the
data
available,
and
its
impact
on
pesticide
use
patterns
that
have
been
changing.
It
was
recommended
that
National
Health
and
Nutrition
Examination
Survey
(
NHANES)
biomonitoring
data
be
used
in
combination
with
PBPK
models
to
serve
as
a
bounding
analysis
for
the
cumulative
risk
assessment.
98
of
113
PANEL
DELIBERATIONS
AND
RESPONSE
TO
THE
CHARGE
The
specific
issues
to
be
addressed
by
the
Panel
are
keyed
to
the
Agency's
background
documents,
references
and
charge
questions.

Question
1
There
are
several
key
principles
for
conducting
a
cumulative
risk
assessment.
One
such
principle
concerns
the
time
frame
of
both
the
exposure
(
e.
g.,
When
does
exposure
occur?
What
is
the
exposure
duration?)
and
of
the
toxic
effect
(
e.
g.,
What
are
the
time
to
peak
effects
and
the
time
to
recovery?
How
quickly
is
the
effect
reversed?).
Both
should
be
adequately
considered
when
performing
a
cumulative
risk
assessment
so
that
an
individual's
exposure
is
matched
with
relevant
and
appropriate
toxicological
values
in
terms
of
duration
and
timing.
There
are
several
important
considerations
with
respect
to
the
temporal
characteristics
of
the
exposures
and
of
the
cholinesterase
inhibitory
effects
of
N­
methyl
carbamate
pesticides
in
estimating
their
cumulative
risk.

OPP
used
a
Relative
Potency
Factor
(
RPF)
approach
in
this
case
study
which
is
based
on
cholinesterase
inhibition
data
from
acute
dosing
studies
performed
in
the
rat.
A
similar
approach
was
used
in
the
organophosphorus
pesticide
cumulative
risk
assessment
several
years
ago.
This
RPF
approach
expresses
toxicity
of
each
chemical
in
terms
of
"
index
chemical
equivalents".
In
this
approach,
all
exposure
events
within
a
day
are
adjusted
by
their
RPFs
and
summed.
The
three
exposure
models
(
DEEM/
Calendex,
LifeLine,
and
CARES)
used
in
the
case
study
express
exposure
as
a
distribution
of
1
day
(
24
hour
totals)
exposures
within
a
population.

Since
AChE
inhibition
caused
by
the
N­
methyl
carbamate
pesticides
recovers
rapidly
(
minutes
to
hours),
it
might
be
important
to
consider
the
intra­
day
timing
of
exposure
events.
Specifically,
if
the
exposure
events
within
a
day
are
distributed
sufficiently
far
apart
in
time
so
that
significant
recovery
of
AChE
inhibition
occurs
between
any
two
exposure
events,
then
summing
exposures
over
24
hours
might
overestimate
the
risk
associated
with
AChE
inhibition.
For
example,
if
an
individual
consumed
200
mL
of
apple
juice
in
the
early
morning,
an
additional
200
mL
during
the
afternoon,
and
another
200
mL
late
at
night,
this
could
present
a
very
different
risk
picture
than
if
the
total
600
mL
were
presumed
to
be
consumed
at
one
time.

The
current
Food
Commodity
Intake
Database
(
FCID)
and
the
DEEM/
Calendex,
LifeLine,
and
CARES
models
are
set
up
to
consider
food
consumption
on
a
per
day
(
rather
than
per
eating
occasion)
manner.
Thus,
the
exposures
reported
in
this
case
study
reflect
daily
(
24
hour)
exposures.
To
the
extent
that
a
day's
eating
occasion
events
leading
to
high
(
total)
daily
exposure
are
close
together
in
time,
the
RPF
approach
described
in
the
case
study
provides
reasonable
99
of
113
estimates
of
risk.
To
the
extent
that
eating
exposure
events
leading
to
high
total
daily
exposures
are
widely
separated
in
time
such
that
recovery
of
AChE
inhibition
occurs,
the
risks
under
the
RPF
approach
in
this
case
study
may
be
overstated
and
a
more
sophisticated
approach
which
accounts
for
intra­
day
eating
patterns
would
be
more
appropriate.

OPP
has
investigated
the
degree
to
which
high­
end
exposures
can
be
attributed
to
specific
eating
occasions
(
within
a
day)
that
occur
either
closely
spaced
in
time
or
widely
separated
by
time
by
looking
at
the
actual
individual
consumption
events
as
reported
in
the
Continuing
Survey
of
Food
Intakes
by
Individuals
(
CSFII).
Specifically,
OPP
has
looked
at
the
CSFII
 
based
dietary
records
for
individuals
at
several
locations
in
the
upper
end
of
the
exposure
distribution
to
determine
the
extent
to
which
these
daily
high­
end
exposures
can
be
attributed
to
a
single
eating
event,
several
eating
events
spaced
closely
in
time
(
over
several
hours),
or
eating
events
widely
separated
by
time
(
more
than
several
hours).
As
described
in
Section
IV.
H
of
the
case
study
document,
OPP
found
that
that
a
large
fraction
of
daily
records
contributing
to
the
upper
tail
of
the
food
exposure
distribution
represent
single
eating
occasions.
Assuming
that
subsequent,
more
detailed
and
extensive
analyses
provide
confirmation
of
these
preliminary
observations
and
analyses,
OPP
believes
and
that
it
is
unlikely
that
any
more
sophisticated,
temporal­
based
approach
which
better
accounts
for
temporal
separation
of
eating/
exposure
events
will
result
in
substantial
or
significant
changes
in
OPP's
risk
estimates.

Part
A.
EPA
requests
the
SAP
provide
comments
on
this
exploratory
analysis
with
respect
to
its
adequacy
and
appropriateness.
Please
also
provide
suggestions
for
future,
more
detailed
analyses.

Panel
Response
The
Panel
agrees
that
the
overall
case
study
is
a
thoughtful
and
useful
analysis.
EPA
has
gained
much
knowledge
from
development
of
the
organophosphate
(
OP)
cumulative
risk
assessment
and
the
Agency
is
to
be
commended
on
the
carefully
developed
exploratory
analysis,
given
the
varying
quality
of
the
datasets
that
were
available.
The
level
of
detail
is
adequate
for
a
case
study,
but
more
effort
is
needed
in
the
full
assessment,
especially
the
non­
food
scenarios.
Several
additional
exposure
scenarios
should
be
considered
and
they
are
detailed
below.

The
general
framing
of
Question
1
conveys
concern
about
overestimates
of
risk.
The
Panel
thinks
that
EPA
should
be
focusing
on
identifying
areas
where
overestimates
or
underestimates
might
occur,
and
then
treat
these
specific
concerns
as
points
to
consider
in
the
variability
and
uncertainty
analysis.
Clarity
is
an
important
consideration
as
the
Agency
moves
toward
a
full
case
study.
100
of
113
The
assessment
process
is
complex,
with
different
models
contributing
estimates
to
the
overall
assessment.
It
is
crucial
that
a
stochastic
model
that
implements
a
quantitative
uncertainty
analysis
be
performed.
It
is
an
important
addition
and
a
necessary
adjunct
to
the
overall
cumulative
assessment
that
compares
the
various
pathways
of
exposure.

Overall
data
quality
is
highly
variable
within
the
assessment.
In
general,
the
food
assessment
has
the
highest
quality
and
most
abundant
data,
with
lower
quality
and
more
sparse
datasets
for
water,
incidental
ingestion,
and
dermal
pathways.
Presentation
of
model
calibrations
for
these
other
datasets
(
e.
g.,
measurements
of
tap
water
in
the
Spring
season
in
southeast
North
Carolina,
dermal
exposure
post­
application
studies
and
model
outputs
compared
to
urinary
biomonitoring
levels
observed
in
the
US
population)
are
important
in
order
for
the
output
of
the
assessment
to
be
an
accurate
representation
of
the
real
world.

Another
key
analytical
issue
is
the
choice
of
exposure
distributions
to
be
used.
In
particular,
the
Panel
believes
that
the
exposure
distributions
need
careful
consideration
and
should
use
all
available
data
without
artificial
truncation
of
what
are
typically
lognormal
exposure
distributions.
The
concern
that
this
may
produce
unreasonable
assessments
can
be
tested
empirically
if
these
"
high"
values
are
included
but
subjected
to
sensitivity
and
uncertainty
analyses
in
the
overall
cumulative
assessments.
Many
of
the
input
datasets
for
the
analysis
may
already
artificially
truncate
the
exposure
distributions
by
whom
they
sample
and
how
the
samples
are
taken.
Recruiting
by
telephone,
for
example,
tends
to
find
established
people
and
likely
does
not
reflect
current
diversity
in
race,
ethnicity
and
income
in
the
United
States.
Truncations,
if
they
are
used,
should
be
based
on
physical
limits
of
concentration.

The
Panel
believes
that
EPA
should
focus
on
appropriate
time
scales
and
populations.
The
use
of
a
99.9th
percentile
is
important
and
should
be
presented
such
that
it
is
clear
this
means
approximately
4,000
children
in
the
United
States
may
be
at
risk
given
that
there
are
approximately
4
million
births
in
any
given
year.

Comments
on
Specific
Issues
The
omission
of
institutional
exposure
(
i.
e.,
non­
residential,
non­
occupational)
leaves
a
serious
gap
in
the
calculation
of
aggregate
exposure.
Each
age
group
will
spend
time
in
schools,
offices,
shopping
malls,
recreation
and
entertainment
facilities,
etc.,
and
there
will
be
pesticide
use
in
these
places.
It
is
understood
that
suitable
data
may
not
be
available
for
these
places,
yet
it
is
important
that
EPA
acknowledge
these
data
gaps
in
the
model
development.
This
issue
was
identified
by
the
September
27­
29,
2000
SAP
"
Session
V
­
Aggregate
and
Cumulative
Assessments
Using
Lifeline
 
­
A
Case
Study
Using
Three
Hypothetical
Pesticides"
(
p.
31).
http://
www.
epa.
gov/
scipoly/
sap/
2000/
september/
modelsreport.
pdf.

Recall
surveys
can
be
of
dubious
quality.
This
issue
was
identified
by
the
April
30
101
of
113
­
May
1,
2002
SAP
"
Cumulative
and
Aggregate
Risk
Evaluation
System
(
CARES
 
)
Model
Review"
(
p.
22).
http://
www.
epa.
gov/
scipoly/
sap/
2002/
april/
caresmeetingminutes.
pdf
The
Panel
reminded
the
Agency
that
some
of
the
survey
data
available
for
the
pilot
case
study
exposure
assessment
of
the
N­
methyl
carbamate
pesticides
are
dated
and
may
not
closely
reflect
some
current
exposure
scenarios
or
be
representative
of
the
current
population
of
the
U.
S.

Relative
potency
factors
(
i.
e.,
the
ratio
of
the
toxic
potency
of
a
given
chemical
to
that
of
the
index
chemical)
are
used
to
convert
exposures
of
all
chemicals
in
the
group
into
exposure
equivalents
of
the
index
chemical.
The
current
presentation
of
this
approach
for
the
calculation
of
the
cumulative
residue
seems
more
complicated
than
necessary
(
e.
g.
when
it
is
explained
how
the
amount
of
residue
of
each
chemical
is
adjusted
by
multiplying
by
a
Relative
Potency
Factor
(
RPF)
to
get
the
equivalent
residue
of
an
index
chemical).
Consulting
the
Toxic
Equivalency
Factor­
Toxic
Equivalent
(
TEF­
TEQ)
approach
for
the
dioxins
and
furans,
where
one
determines
the
TEFs,
obtains
the
TEQ
values
of
each
compound
in
the
mixture
exposure,
and
then
does
the
multiplication
and
the
summation,
could
serve
as
a
good
example
to
communicate
this
approach.
However,
this
simple
multiplication
and
calculation
holds
only
if
it
can
be
shown
that
the
carbamates
act
according
to
the
principle
of
simple
similar
additive
action
at
the
chosen
endpoint
for
risk
assessment.
This
means
that
the
isoboles
are
parallel
straight
lines
as
is
assumed
for
the
dioxins.
This
is
a
crucial
assumption
underlying
all
the
analyses
at
this
stage.

Several
of
the
contact
categories
need
to
be
expanded
or
better
justified
(
see
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
N­
methyl
Carbamates,
Table
10.
Specific
Exposure
Routes
and
Pathways/
Scenarios).
Additional
contact
routes
for
children
should
include
dermal/
oral
contact
with
gardens,
ornamentals,
and
trees.
Children
have
been
observed
having
skin
contact
with
vegetation
and
placing
vegetation
in
their
mouths.
Whether
this
is
a
major
source
of
dermal
or
non­
dietary
ingestion
is
unclear;
there
may
not
be
enough
data
to
make
a
determination
as
to
its
impact
on
children's
exposure.

In
certain
scenarios,
EPA
should
provide
better
support
to
justify
the
exclusion
of
specific
exposure
routes
such
as
"
ingestion
of
soil
and
mouthing
of
grass"
(
see
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
Nmethyl
Carbamates,
Section
VI
­
D).
EPA
states
that
the
reason
for
not
including
an
exposure
route
involving
oral
ingestion
of
soil
and
mouthing
of
grass
was
justified
because
of
the
'
little
impact'
such
routes
had
on
the
individual
chemical
risk
assessments.
This
justification
needs
better
support
by
further
exploring
what
difference
the
cumulative
effect
of
several
`
little
impacts'
might
make
in
the
cumulative
risk
assessment.

In
terms
of
clarity
of
the
document,
when
discussing
the
residential
exposure
scenarios,
EPA
states
that
"
The
full
range
of
exposure
values
­­
expressed
as
uniform,
log­
102
of
113
normal,
triangular,
or
cumulative
distributions
­­
are
used,
where
appropriate,
rather
than
relying
on
point
estimates."
(
see
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
N­
methyl
Carbamates,
Section
VI­
A).
While
the
Panel
applauds
this
move
away
from
point
estimates,
there
needs
to
be
clear
descriptions
as
to
why
a
particular
distribution
was
used
rather
than
another
one.

The
food
assessment
is
vulnerable
in
terms
of
extremes
in
food
consumption,
home
grown
vegetables
and
home
pesticide
applications.
Furthermore,
most
of
the
residues
are
a
single
record
of
composites,
and
deconvolution
of
residue
data
should
be
considered.
The
food
dataset
should
not
be
artificially
restricted
(
e.
g.,
exclusion
of
over
tolerance
values
from
the
USDA
Pesticide
Data
Program
(
PDP)
dataset).

Part
B.
Given
the
results
of
the
initial
exploratory
analysis,
EPA
believes
that
a
more
sophisticated
time­
based
intra­
day
model
(
e.
g.,
PBPK
in
which
the
timing
of
intra­
day
eating
events
is
explicitly
incorporated)
for
exposures
through
the
food
pathway
would
not
substantially
change
the
assessment
of
potential
risks
through
this
pathway
compared
to
the
results
produced
using
the
RPF
methodology
used
in
the
case
study
in
which
24
hour
food
consumption
data
is
used.
Please
explain
why
you
agree
or
disagree.

Panel
Response
The
default
assumption
of
eating
events
close
together
in
time
is
a
proper
approach
in
the
context
of
this
pilot
case
study
for
the
N­
methyl
carbamate
exposure
assessment.
Allowing
for
recovery
seems
difficult
given
the
unknown
durations
that
would
have
to
be
assigned
to
defensibly
address
this
issue.
The
extent
to
which
eating
frequency
affects
the
outcome
of
this
case
study
analysis
is
a
testable
proposition
that
could
be
explored
as
a
distributed
analysis.
The
Panel
noted
that
a
supplementary
analysis
using
a
stratified
sample
of
consumers
at
the
high
end
of
pesticide
ingestion
would
help
justify
the
viability
of
this
approach.
However,
a
supplementary
analysis
should
only
be
done
if
the
underlying
data
are
adequate
to
support
the
analysis.
The
Panel
is
skeptical
that
this
proposition
is
testable
with
existing
datasets.
For
example,
the
assessment
presents
comparison
of
records
for
a
number
of
eating
occasions
and
found
few,
if
any,
with
four
or
more
per
day
(
see
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
N­
methyl
Carbamates,
Section
IV­
H).
Food
consumption
patterns,
particularly
for
fruit
juices
in
some
families,
is
on
an
ad
libitum
pattern
rather
than
as
a
specified
number
of
events
per
day.
The
structure
of
the
CSFII
surveys
would
not
adequately
pick
up
this
type
of
exposure.
Therefore,
the
concern
is
that
the
data
do
not
adequately
capture
human
behavior,
and
exposures
may
not
be
seen
because
of
the
quality
of
the
available
data
and
not
because
the
potential
for
exposure
did
not
exist.

Question
2
A
key
concept
that
is
unique
to
cumulative
risk
assessments
is
the
concept
of
103
of
113
co­
occurrence
of
residues
(
and
thus
co­
or
simultaneous
exposure)
to
members
of
the
Common
Assessment
Group
(
CAG).
Specifically,
a
cumulative
assessment
must
appropriately
consider
residues
that
co­
occur
in
time
and
space
since
these
exposures
must
by
combined
and
considered
jointly.
This
is
true
for
exposures
through
the
food
pathway,
the
drinking
water
pathway,
and
the
residential
pathway.
The
USDA's
PDP
data
program
uses
multi­
analyte
methods
and
thus
simultaneously
measures
co­
occurring
residues
in
samples.
The
drinking
water
concentrations
generated
from
the
PRZM­
EXAMS
model
considered
regional
Nmethyl
carbamate
pesticide
use
and
usage
practices
and
thus
implicitly
considered
co­
occuring
residues.

Exposures
through
the
residential
pathways
can
also
co­
occur.
One
of
the
unique
aspects
of
the
N­
methyl
carbamate
pesticide
cumulative
exposure
assessment
is
the
use
of
the
Residential
Exposure
Joint
Venture
(
REJV)
data
that
provides
current
information
on
co­
occurring
use
patterns
for
residential
exposure.
The
US
EPA
National
Home
and
Garden
Pesticide
Use
Survey
(
NHGPUS)
can
also
be
used
to
develop
residential
use
profiles.
The
PDP,
PRZM­
EXAMS,
and
REJV/
NHGPUS
data
were
used,
to
varying
degrees,
in
this
case
study.

Please
comment
on
the
use
of
the
pesticide
use/
usage
data
(
e.
g.,
REJV
and
NHGPUS)
to
account
for
co­
occurring
use
patterns
in
assessing
residential
exposures.

Panel
Response
Data
for
Co­
occurrence
The
Residential
Exposure
Joint
Venture
(
referred
to
hereafter
as
the
Joint
Venture)
and
the
National
Home
and
Garden
Pesticide
Use
Survey
(
referred
to
hereafter
as
the
Home
and
Garden
Survey)
appear
to
be
the
two
main
databases
for
characterizing
the
co­
occurring
pesticide
use
patterns.
However,
their
use
in
assessing
the
co­
occurrence
of
pesticide
use
would
need
to
be
carefully
considered
and
justified
as
it
is
possible
that
neither
database
fully
captures
the
diversity
of
the
US
population
as
it
presently
exists.
The
overall
representativeness
of
both
surveys
should
be
evaluated
to
ensure
they
accurately
reflect
the
racial,
ethnic,
and
income
diversity
that
currently
exists
in
the
U.
S.
population.

The
Panel
appreciates
the
efforts
of
the
Joint
Venture
task
force
that
surveyed
and
collected
more
updated
data
on
pesticide
use
patterns.
However,
without
access
to
this
proprietary
database,
the
Panel
is
unable
to
provide
a
proper
review.
The
Panel
is
also
unable
at
this
time,
to
determine
if
there
might
be
alternative
ways
to
utilize
the
data
(
e.
g.,
the
reports
from
respondents
who
provided
less
than
12
months
diaries)
or
any
new
dimensions
to
exposure
analysis
other
than
what
was
presented
by
the
Agency.
104
of
113
The
Panel
understands
that
the
Agency's
effort
to
evaluate
the
use
of
Joint
Venture
data
is
on­
going
and
encourages
a
careful
review
of
the
validity
and
reliability
of
the
Joint
Venture
data,
in
particular
the
potential
effects
of
response
bias
and
errors
in
reporting
as
indicated
by
product­
application
combinations
that
are
not
plausible.

Additional
Comments
on
Specific
Issues
Regarding
the
Joint
Venture
Database.

As
mentioned
earlier
in
the
Panel's
response
to
Question
1
­
Part
A,
telephone
recruiting
may
not
provide
the
most
representative
sampling
of
the
population.
The
dependence
on
telephone
for
recruiting
may
bias
the
sample
to
a
wealthier
than
average
population
and
may
perhaps
also
bias
against
younger
users
that
rely
solely
on
cell
phones
whose
numbers
are
not
readily
available.
The
Agency
is
encouraged
to
systematically
address
the
overall
issues
of
telephone
recruiting
and
consider
a
sub­
study
of
nonresponders
that
includes
a
telephone
or
personal
interview
as
well
as
a
validity
study
of
responders
that
includes
personal
or
telephone
interviews.

The
demonstration
of
validity
should
also
include
response
rates,
and
the
adjustments
done
to
them
need
to
be
understandable
to
the
Panel
and
others
who
will
review
the
cumulative
risk
assessment.
The
extreme
loss
of
response
rate
after
the
initial
screening
points
to
the
need
for
a
careful
evaluation
of
the
Joint
Venture
database.
The
National
Family
Opinion
(
NFO)
adjusted
survey
results
for
6
demographic
factors
in
an
attempt
to
make
the
data
representative
of
the
overall
U.
S.
population.
The
factors
chosen
to
make
such
adjustments
should
be
related
to
the
use
of
pesticides,
and
specifically,
to
the
N­
methyl
carbamate
pesticides.

Issues
of
statistical
inference
by
region
or
state
are
important,
as
is
the
number
of
homes
represented
by
the
survey
in
demographic
subgroups.
The
limitation
of
sample
size
becomes
a
greater
issue
when
the
database
is
divided
into
more
refined
geographic
locations.

The
analysis
of
the
database
should
include
the
concordance
between
the
inventory
data
and
application
records
(
accounting
for
labeled
application
method
where
appropriate).

Some
of
the
data,
as
presented
by
the
Agency,
stood
out
and
may
indicate
the
need
for
a
more
careful
evaluation.
For
example,
the
69%
derived
proportion
of
applications
to
trees
was
by
a
handwand
sprayer
(
see
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
N­
methyl
Carbamates,
Table
A.
1.2
­
Residential
Use
of
Chemical
A,
REJV/
1).
Also,
the
67%
usage
rate
over
a
year
seems
low
compared
to
other
surveys.
Both
Adgate
et
al.
(
2000)
and
the
Home
and
Garden
Survey
reported
usage
rates
in
the
80­
90%
range.

The
co­
occurrence
information
provided
by
the
Joint
Venture
is
unique
and
105
of
113
valuable,
yet
the
limitations
of
those
data
should
be
fully
explored
and
stated.
For
example,
the
table
of
conditional
probabilities
(
see
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
N­
methyl
Carbamates,
Table
A.
1.8
­
Chemical
A
Scenario
Co­
occurance
Probability
Matrix)
for
co­
occurrence
of
pesticide
applications
to
lawns
must
be
based
on
a
notably
small
number
of
observations
(
e.
g.,
1
or
2
homes).

Given
the
proprietary
nature
of
the
Joint
Venture
dataset,
and
the
need
for
transparency
in
Agency
decision­
making,
EPA
should
carefully
consider
the
pros
and
cons
of
using
this
data
source
and
the
specific
information
it
can
provide.
The
Agency
is
encouraged
to
investigate
the
reliability
of
the
co­
occurrence
data
represented
in
the
Joint
Venture,
to
identify
the
important
limitations,
weaknesses,
and
uncertainties
associated
with
the
amount
of
information
therein,
and
to
incorporate
this
knowledge
into
the
Nmethyl
carbamate
pesticide
cumulative
risk
assessment
(
CRA).

Additional
Comments
on
Specific
Issues
Regarding
Home
and
Garden
Survey
The
main
concern
regarding
the
Home
and
Garden
Survey
database
is
the
lack
of
assurance
that
the
pattern
of
pesticide
use
has
remained
the
same
since
the
survey
was
taken
more
than
15
years
ago.
In
many
respects,
the
household
pesticide
use
practices
have
likely
changed
over
the
years.
For
example,
one
cannot
assume
that
pesticide
use
practices
and
types
of
pesticides
used
for
various
applications
are
the
same
now
as
then.
If
the
amount
of
advertising
that
is
conducted
for
marketing
residential
pesticides
is
effective,
there
should
be
at
least
increased
product
purchasing
if
not
usage.
The
demographic
distribution
of
responders
in
the
study
may
not
reflect
current
demographics
and
therefore
may
either
underestimate
or
overestimate
usage
patterns.
In
addition,
the
choice
of
pesticide
may
also
have
changed
in
the
past
15
years.
It
would
be
desirable
to
have
an
updated
Home
and
Garden
Survey
in
order
to
have
some
assurance
that
the
survey
results
and
extracted
co­
occurrences
are
still
valid.

The
Joint
Venture
can
be
expected
to
reflect
more
closely
the
development
of
better
spraying
equipment,
improvements
in
use
labeling,
and
the
awareness
of
domestic
pesticide
use
currently.
However,
it
is
unclear
how
much
information
from
that
data
source
is/
will
be
used.
Thus,
at
this
stage,
the
Panel
strongly
recommends
that
the
Agency
compare
and
contrast
the
results
from
Home
and
Garden
Survey
and
the
Joint
Venture
data
where
possible.
The
advantages
and
disadvantages
of
using
information
from
one
or
both
surveys
should
be
determined.
The
comparison
also
could
allow
a
better
assessment
of
how
the
Home
and
Garden
Survey
data
can
be
utilized
in
Lifeline
to
imply
current
use
patterns.
An
Agency
effort
to
provide
easy
public
access
to
the
two
parts
of
the
Home
and
Garden
Survey
report
is
also
recommended.

Pesticide
Residue
Data
The
Panel
commented
on
the
approach
used
for
pesticide
degradation
parameters
106
of
113
to
define
foliar
residue
in
contact
exposure
(
see
Estimation
Of
Cumulative
Exposure
From
N­
Methyl
Carbamate
Pesticides:
A
Case
Study
Of
10
N­
Methyl
Carbamates,
Section
VIF
For
example,
setting
the
residue
to
zero
for
Chemical
A
30
days
after
its
application
does
not
appear
to
be
"
conservative"
because,
based
on
its
half
life
of
9
days,
10%
residue
is
expected
to
remain.
As
for
chemical
G,
a
linear
degradation
rate
is
highly
unusual
because
almost
all
environmental
processes
are
first­
order
or
quasi
first­
order.
The
Agency
is
encouraged
to
revisit
the
assumptions
about
chemical
degradation
with
respect
to
realism
versus
conservatism.

The
Panel
recommends
that
the
regional
and
temporal
patterns
of
N­
methyl
carbamate
pesticide
residues
be
explored
across
all
pathways
of
exposure
(
not
just
for
pathways
associated
with
applications
around
the
residence
and
for
drinking
water)
and
reflected
in
the
exposure
analysis
estimates.
For
example,
the
concentrations
of
certain
organophosphates
in
food
display
a
temporal
pattern
with
the
highest
concentrations
occurring
in
the
spring
and
summer
months.
It
seems
reasonable
to
also
expect
that
Nmethyl
carbamate
pesticides
used
in
domestic
agriculture
will
display
a
similar
pattern.
The
Agency
is
encouraged
to
further
explore
the
potential
seasonal
patterns
of
the
PDP
data
that
are
used
in
assessing
dietary
exposures.

Exposure
Scenarios
The
studies
used
to
develop
the
unit
exposures
for
each
application
scenario
should
be
made
available
to
the
Panel
to
evaluate
their
validity
and
reliability.
Many
factors
influence
the
results
of
such
tests
and
the
Panel
is
not
currently
aware
of
the
conditions
during
the
test.
Some
of
the
important
factors
include:
sprayer­
target
geometry;
meteorology;
spray
nozzle
type
and
settings,
if
adjustable;
duration
of
application
events;
sampling
methods
and
locations;
analytical
methods
and
performances;
data
reduction
and
analysis
methods.

Ready­
to­
use
products
have
higher
upper
range
dermal
unit
exposure
values
(
mg
exposure
per
unit
of
active
ingredient
applied)
than
hand
wand,
etc.
Yet,
the
amount
applied
with
ready­
to­
use
products
is
relatively
low,
so
this
application
technique
seems
the
least
important
(
at
least
for
many
scenarios).
However,
questions
remain
regarding
how
the
duration
of
exposure
is
considered;
e.
g.,
if
the
exposure
is
assumed
to
be
instantaneous
or
spread
over
the
duration
of
the
application
event.

The
Panel
commented
on
the
following
specific
exposure
scenarios
that
are
not
addressed
by
the
Agency:

1.
Omitting
the
hose­
end
sprayer
application
method
for
ornamental/
tree
sites
may
not
represent
a
"
conservative"
approach.
Not
explicitly
accounting
for
use
on
trees
may
thus
be
an
important
limitation
of
the
Lifeline
model,
and
could
lead
to
underestimates
of
upper­
end
residential
use
exposures
for
products
applied
to
107
of
113
trees.
The
Agency
is
encouraged
to
explore
the
sensitivity
of
their
results
to
this
limitation.
2.
The
Agency
should
account
for
pesticide
use
by
property
managers
or
commercial
pesticide
applicators.
3.
Hand­
to­
mouth
transfer
of
pesticides
by
applicators
(
adults)
due
to
tobacco
smoking
is
a
likely
exposure
pathway,
but
it
is
not
included
in
the
pilot
exposure
case
study.
Whenever
the
smoking
information
for
adult
applicators
is
available,
this
oral
ingestion
route
should
be
incorporated
in
the
assessment.
Dermal
and
oral
exposures
are
also
likely
to
occur
in
children
age
3
to
5
when
N­
methyl
carbamate
pesticides
are
used
in
home
gardens.
Studies
have
indicated
that
children
of
parents
who
use
pesticides
in
their
home
gardens
(
either
vegetable
or
flower)
have
significantly
higher
urinary
OP
metabolite
concentrations
than
children
whose
parents
reported
no
use
of
pesticides
in
their
garden
(
Lu
et
al.
2001).
These
two
likely
exposure
routes
should
be
included
in
the
assessment.
4.
The
co­
occurrences
considered
by
the
Agency
in
this
current
analysis
are
limited
exclusively
to
pesticide
occurrence
in
media
in
and
around
the
home.
As
a
result,
locations
frequented
by
people
outside
the
home
are
not
considered.
For
children's
exposure,
important
areas
not
yet
considered
might
include
schools
and
day
care
centers
where
a
premium
is
placed
on
demonstrable
hygiene.
The
Agency
should
at
least
discuss
the
possibility
of
exposures
outside
the
home
and
the
associated
risks.
5.
In
the
current
assessment,
exposures
associated
with
household
outdoor
applications
of
N­
methyl
carbamate
pesticides
require
contact
with
the
media
treated
by
the
applicant
(
e.
g.,
lawn,
garden
and
vegetables).
Several
studies
in
the
scientific
literature
have
demonstrated
that
common
residential
use
insecticides
applied
outdoors
migrate
into
the
home
by
being
tracked
inside
(
Lewis
et
al.,
2001;
Stout
and
Leidy,
2000;
Nishioka
et
al.,
2001).
The
Agency
should
consider
the
importance
of
this
potential
exposure
pathway
in
their
continuing
assessment
of
N­
methyl
carbamate
pesticides.
6.
The
present
analysis
correctly
puts
a
large
amount
of
effort
into
distinguishing
between
the
age
groups.
This
is
important
and
needed
because
of
the
possible
higher
risk
of
young
children.
On
the
other
hand,
age
is
just
one
factor
that
can
be
considered.
Others
that
could
be
considered
include
vegetarians
versus
non
 
vegetarians
or
other
groups
of
persons
with
specific
behavioral
differences
which
cause
heterogeneity.
The
simple
PK
model
would
allow
this
distinction
since
it
adjusts
for
the
incorporation
of
a
number
of
experimental
factors.
7.
It
is
not
immediately
clear
how
certain
pathways
were
omitted
from
the
garden
scenarios.
The
Agency
should
consider
the
tendency
of
children
to
mimic
adult
behaviors.

Question
3
The
data
sources
and
methodologies
used
in
the
N­
methyl
carbamate
case
study
are
similar
in
many
respects
to
the
data
sources
and
methodologies
used
in
the
108
of
113
Cumulative
Risk
Assessment
for
the
OP
pesticides.
For
example,
in
both
assessments
the
evaluation
of
exposure
of
the
food
pathway
relied
to
a
great
extent
on
the
USDA's
PDP
data,
the
evaluation
of
exposure
through
the
water
pathway
used
PRZM­
EXAMS
modeling
data,
and
the
evaluation
of
exposure
through
the
residential
pathway
used
standard
SOP
algorithms
along
with
label
information,
professional
judgments,
literature
values,
and
survey
data
(
REJV,
NHGPUS).

Please
comment
on
the
data
sources
used
in
the
cumulative
exposure
assessment
and
on
how
EPA
has
considered
and
incorporated
the
data.
Does
the
SAP
have
any
suggestions
or
recommendations
regarding
additional
available
data
sources
that
EPA
may
wish
to
investigate?

Panel
Response
The
data
sources
used
in
the
cumulative
exposure
assessment
were
those
currently
available.
The
Panel
raised
a
number
of
concerns
about
the
age
and
quality
of
the
datasets
and
urged
EPA
to
test
the
outcome
of
its
assessment
against
the
NHANES
carbamate
metabolite
data
as
a
means
of
reality
testing.
Concerns
were
raised
about
secular
trends
in
pesticide
use,
the
age
of
some
of
the
data,
how
the
data
are
used,
how
the
data
are
interpreted,
and
the
types
of
distributions
developed
from
the
data.

Sources
Used
in
Assessment
The
case
study
presented
to
the
SAP
clearly
demonstrated
that
available
data,
although
limited,
could
be
combined
to
conduct
a
cumulative
exposure
assessment
for
the
N­
methyl
carbamate
pesticides.
All
3
models
were
built
upon
the
same
datasets;
therefore,
the
similar
outputs,
particularly
at
the
high­
end
exposure
profiles
from
all
models,
are
not
surprising.
While
the
improvement
of
these
models
is
on­
going,
the
Agency
should
begin
the
process
of
validating
all
of
the
models
in
terms
of
adequacy
and
appropriateness
by
using
a
different
dataset.
The
immediately
available
data
for
such
an
exercise
are
data
collected
by
Bayer
from
their
carbaryl
turf
application
study.
Assuming
the
study
took
place
in
the
southeast
region
of
North
Carolina
and
therefore
took
into
account
the
component
of
drinking
water
exposure
in
the
model
simulation,
the
outputs
from
these
models
can
then
be
used
as
inputs
to
the
PBPK
model
for
estimating
24­
hour
average
1­
naphthol
levels.
The
modeled
1­
naphthol
levels
can
then
be
compared
to
1­
naphthol
levels
reported
by
Bayer.
From
here,
the
Agency
could
decide
which
model
would
provide
the
most
robust
and
accurate
prediction,
and
if
not,
what
modifications
are
needed
in
the
CRA
models
so
that
the
gap
between
predicted
vs.
observed
levels
can
be
brought
closer.

Comments
Regarding
Specific
Issues
1.
Secular
trends:
Several
Panel
members
raised
the
issue
of
possible
change
in
use
patterns
and
trends
away
from
using
cholinesterase
inhibiting
pesticides.
The
Panelists
question
whether
the
Agency
used
any
factors
to
adjust
for
this
potential
factor.
There
109
of
113
have
been
significant
secular
trends
in
overall
pesticide
use
with
shifts
away
from
acetylcholinesterase
inhibitors
because
of
their
inherent
toxicity.
While
there
has
not
been
an
outright
ban
or
prohibition
of
any
carbamates,
we
are
likely
to
see
far
less
household
use
of
such
pesticides.
At
the
very
least,
the
mix
of
carbamates
used
in
residential
and
agricultural
settings
has
likely
shifted.
It
is
unclear
in
the
background
documents
provided
to
the
Panel
what
correction
factors
were
used.

2.
Age
of
data:
There
was
considerable
concern
among
the
Panel
about
the
age
of
data
collected
with
the
NHGPUS
15
years
ago.
One
cannot
assume
that
pesticide
use
practices
and
types
of
pesticides
used
for
various
applications
are
the
same
now
as
then.
Also,
the
demographic
distribution
of
responders
in
that
study
may
not
reflect
current
demographics
and
therefore
may
either
underestimate
or
overestimate
usage
patterns.
In
addition,
the
choice
of
pesticide
also
may
have
changed
in
the
past
15
years.
Since
NHGPUS
was
published
around
1990,
and
likely
represent
data
collected
around
1988,
it
is
quite
dated.

3.
Data
used:
Some
of
the
data
used
are
regionally
specific
(
water)
while
other
appear
not
to
be
regionally
specific.
The
regional
use
of
data
meant
to
be
representative
of
the
nation
is
inappropriate
and
likely
to
be
error
prone.
Since
the
case
study
is
in
the
Southeast,
it
is
appropriate
to
try
and
use
data
specific
to
that
region.
The
authors
should
be
clear
on
the
potential
hazards
of
using
such
data
on
regions
smaller
than
that
for
which
they
are
truly
representative.

This
assessment
uses
food
residue
data
collected
over
a
nine
year
period,
1994
through
2002.
A
more
recent
and
smaller
window
would
be
more
appropriate.
As
stated
in
the
document,
the
primary
reason
for
this
choice
is
to
maximize
the
number
of
food
commodities.
With
this
design,
there
is
a
trade
off
between
precision
and
bias
that
needs
to
be
addressed.
It
would
be
interesting
to
see
how
robust
the
estimates
are
when
early
years
are
deleted.

Support
for
some
of
the
assumptions
may
be
made
by
correlating
overall
exposure
with
pesticide
sales
figures,
at
least
if
one
collects
data
over
a
long
period
of
time.

It
was
also
felt
that
deconvolution
methods
should
be
used
to
impute
individual
serving
exposure
values
from
composite
samples.
110
of
113
4.
Data
Interpretation:
Observational
data
for
children
contact
with
pets
are
from
a
very
small
dataset
and
do
not
cover
activities
such
as
sleeping
with
pets.
It
also
is
based
on
a
single
period
of
observation
per
child.

5.
Data
robustness:
It
is
important
to
get
away
from
professional
judgments
that
are
more
difficult
to
evaluate
by
outside
reviewers
and
sometimes
less
transparent
in
how
they
were
developed,
and
use
existing
data
bases.
Robustness
of
databases
is
always
a
concern,
particularly
when
they
were
not
developed
specifically
to
meet
the
needs
of
the
Agency
for
use
in
modeling
exercises
such
as
this
one.

6.
Uncertainty
Analysis:
It
is
very
important
to
accompany
the
assessment
with
a
substantial
uncertainty
analysis.
The
model
comparisons
of
the
exposure
results
are
presented
in
Table
15
of
the
background
document,
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
N­
methyl
Carbamates,
with
six
significant
digits.
It
seems
unlikely,
for
a
host
of
reasons,
that
such
precision
really
attends
these
numbers.
Estimating
the
99.9th
percentile,
for
instance,
at
this
precision
would
require
many
more
replications
than
were
likely
deployed
in
the
Monte
Carlo
simulations.
The
model
uncertainty
expressed
by
the
tables
suggests
only
one
or
two
significant
digits.
There
are,
perhaps,
even
fewer
after
the
measurement
uncertainties
and
the
sampling
error
that
arises
from
having
few
data
samples
in
some
of
the
cells
are
accounted
for.

The
Agency
attempted
to
quantify
the
reliability
of
the
simulation
results.
It
should
be
pointed
out,
however,
that
changing
the
numerical
seed
and
re­
running
simulations
is
not
a
sufficient
way
to
conduct
uncertainty
analyses.
At
the
very
best,
this
approach
can
only
assess
the
uncertainty
arising
from
the
use
of
a
particular
sequence
of
random
numbers.
We
might
call
this
the
`
simulation
uncertainty'.
This
is
perhaps
the
easiest
kind
of
uncertainty
to
address,
but
it
is
likely
to
be
the
smallest
kind
affecting
the
overall
results;
or
at
least
can
always
be
made
to
be
the
smallest,
simply
by
increasing
the
number
of
Monte
Carlo
replications.
The
Panel
emphasizes
that
the
consequences
for
the
reliability
of
the
results
attributable
to
model
uncertainty,
sampling
uncertainty,
and
measurement
uncertainty
in
general
will
likely
be
much
greater,
and
will
probably
lead
to
the
results
having
no
significant
figures
at
all.
In
such
a
case,
an
explicit
uncertainty
analysis
is
crucial
to
interpreting
the
import
of
the
calculations.
Measurement
uncertainty
may
be
especially
hard
to
estimate
because
the
various
model
approaches
have
been
based
on
overlapping
and
essentially
similar
datasets.

The
Panel
voiced
concerns
about
truncation
several
times.
It
was
felt
that
the
Agency
should
not
truncate
a
parameterized
distribution
without
a
reasoned
argument
to
justify
this
decision.
The
same
issue
arises
with
empirical
distributions.
Using
an
empirical
distribution
is
perhaps
the
most
egregious
kind
of
unjustified
truncation.
As
several
Panelists
mentioned,
higher
values
are
almost
certain
to
be
obtained
than
the
original
data
contained
if
more
samples
are
collected.
It
is
impossible
to
know
the
impact
of
truncations
unless
an
uncertainty
analysis
is
conducted
that
addresses
the
consequences
111
of
113
of
ignoring
the
tails.
The
Panel
encourages
EPA
to
continue
to
study
the
model
stability
and
sensitivity.

One
Panel
member
suggested
that
it
is
important
to
assess
the
appropriateness,
sufficiency
and
precision
of
the
outcome
of
this
cumulative
risk
assessment
for
the
Nmethyl
carbamate
pesticides.
For
example,
are
N­
methyl
carbamate
pesticides
in
drinking
water
alone
capable
of
elevating
the
exposure
profile
during
a
certain
period
of
time
during
a
year
for
all
ages
of
modeled
individuals
at
the
95th
percentile?
The
other
interesting
scenario
about
the
outcome
of
this
cumulative
exposure
assessment
is
that,
assuming
seasonal
contribution
of
N­
methyl
carbamate
cumulative
exposure
from
drinking
water
is
real,
and
most,
if
not
all,
of
the
N­
methyl
carbamate
pesticide
residues
come
from
the
seasonal
agricultural
uses
in
the
modeled
region,
this
seasonal
effect
is
not
reflected
in
the
food
component.

According
to
data
presented
in
Table
16
of
the
background
document
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
Nmethyl
Carbamates,
approximately
65%
of
total
N­
methyl
carbamate
dietary
exposure
comes
from
a
single
food
type,
which
is
citrus
fruits,
including
orange,
tangerine,
and
grapefruit.
If
the
availability
of
citrus
fruits
to
the
modeled
individuals
is
limited
only
to
a
certain
period
of
time
in
a
year,
say
winter
months,
this
seasonal
effect
obviously
does
not
transmit
to
the
total
exposure
profile.
It
was
felt
that
longitudinal
correlations
across
seasons
are
included
in
the
analyses.
There
was
discussion
of
the
longitudinal
nature
of
the
PDP
studies
and
the
inclusion
of
foreign
food
items
and
items
from
other
areas
of
the
country
(
i.
e.,
Florida
versus
California
oranges).

Diurnal
patterns
of
exposure
are
very
important
for
biomonitoring
studies
of
exposure
and
effect
and
are
not
really
addressed
in
the
Agency
assessments.
If
we
measure
a
parent/
metabolite
at
one
time
of
day,
yet
exposures
vary
across
the
day,
we
are
likely
to
get
substantially
different
inferences
for
exposure
and
risk
depending
on
when
the
exposures
occurred.
A
morning
exposure
may
be
completely
eliminated
by
late
evening.
If
a
urine
sample
is
taken
in
the
late
evening,
the
exposure
inferred
will
be
substantially
smaller
than
what
might
be
inferred
from
a
noon
urine
sample.

Other
Data
Sources
There
are
other
sources
of
data
that
can
be
used
to
challenge
and
evaluate
the
qualitative
accuracy
of
the
data
on
exposure
being
generated
in
this
analysis.
Specifically,
the
Centers
for
Disease
Control
and
Prevention
(
CDC)
has
produced
the
"
Second
National
Report
on
Human
Exposure
to
Environmental
Chemicals"
(
CDC
Publication
02­
0716)
which
determined
levels
of
1­
Naphthol
(
carbaryl
metabolite),
2­
isopropoxyphenol
(
metabolite
of
propoxur)
and
carbofuranphenol
(
metabolite
of
benzofuracarb
and
carbofuran)
in
human
urine.
Through
"
best
estimate"
and
"
conservative"
assumptions
and
use
of
the
PBPK
model
presented
during
this
SAP
meeting,
it
would
be
possible
to
bound
the
amount
ingested
prior
to
the
measurement
and
compare
this
with
the
range
seen
in
the
112
of
113
exposure
assessment.
For
example,
1­
naphthol
is
a
metabolic
product
of
carbaryl,
naphthalene
and
some
components
of
cigarette
smoke.
Assuming
all
of
this
is
from
carbaryl
is
a
"
conservative"
choice.
Calculating
a
likely
urinary
level
from
the
exposure
estimates
and
the
PBPK
model
and
comparing
this
to
what
was
seen
could
provide
the
maximum
percentage
in
urine
that
would
be
due
to
carbaryl
if
the
exposure
estimates
are
correct.
While
this
still
presents
some
challenges
(
how
long
between
urinations,
what
to
do
about
the
large
number
of
observations
below
the
limit­
of­
detection,
other
sources
of
the
metabolite,
etc.),
it
is
a
reality
check
worth
doing.
The
median
urinary
level
of
1­
naphthol
in
children
6­
11
years
of
age
was
1.1
micrograms/
L,
and
the
95%­
tile
5.6
micrograms/
L.
From
the
PBPK
model
and
the
data
used
to
form
that
model,
it
appears
that
1­
naphthol
will
be
a
minor
urinary
metabolite
of
carbaryl,
accounting
for
less
than
5%
(
high
metabolism
but
small
urinary
excretion)
and
suggesting
about
a
20­
fold
ingestion
relative
to
urinary
output;
however,
this
could
easily
be
calculated
more
accurately
by
those
who
are
familiar
with
the
model.
For
2­
isopropoxyphenol,
both
the
median
and
the
95%­
tile
are
below
the
limit­
of­
detection.
For
carbofuranphenol,
the
median
is
below
the
limit­
of­
detection
and
the
95%­
tile
is
0.43
micrograms/
L.
The
sample
sizes
for
6­
11
yearolds
are
between
450
and
483
and
for
the
total
dataset
are
between
1917
and
1998.
They
have
the
advantage
of
being
representative
of
the
national
population
and
can
be
divided
into
broad
regions.

In
terms
of
additional
data,
the
Agency
noted
that
a
study
on
lawns
and
gardens
use
of
Chemical
A
conducted
by
Rhone­
Poulenc
in
1998
showed
similar
results
as
estimates
from
the
NHGPUS
(
see
Estimation
of
Cumulative
Exposure
from
N­
methyl
Carbamate
Pesticides:
A
Case
Study
of
10
N­
methyl
Carbamates,
Appendix1
­
Page
148,
Table
A.
1.6).
It
appears
that
this
study
may
provide
a
more
current
context
to
the
NHGPUS
and
the
Agency
is
encouraged
to
present
this
data
set
in
greater
detail.
113
of
113
REFERENCES
Adgate,
J.
L.,
Kukowski,
A.,
Stroebel,
C.,
Shubat,
P.
J.,
Morrell,
S.,
Quackenboss,
J.
J.,
Whitmore,
R.
W.
and
Sexton,
K.
(
2000)
Pesticide
storage
and
use
patterns
in
Minnesota
households
with
children.
J.
Expos.
Anal.
Environ.
Epi.
10:
159­
167
Centers
for
Disease
Control
and
Prevention.
National
Center
for
Environmental
Health.
2003.
Second
National
Report
on
Human
Exposure
to
Environmental
Chemicals.
(
NCEH
Publication
#
02­
0716)

Lewis,
R.
G.,
Fortune,
C.
R.,
Blanchard,
F.
T.,
and
Camann,
D.
E.
(
2001)
Movement
and
deposition
of
two
organophosphorous
pesticides
within
a
residence
after
interior
and
exterior
applications.
J.
Air
Waste
Manage.
Assoc.
51:
339­
351.

Lu,
C.,
Fisker­
Anderson,
J.,
Knutson,
D.,
Moate,
T.
and
Fenske,
R.
(
2001)
Biological
monitoring
survey
of
organophosphorus
pesticide
exposure
among
pre­
school
children
in
the
Seattle
metropolitan
area.
Environ
Health
Perspect.
109(
3):
299­
303.

Nishioka,
M.
G.,
Lewis,
R.
G.,
Brinkman,
M.
C.,
Burkholder,
H.
M.,
Hines,
C.
E.
and
Menkedick,
J.
R.
(
2001)
Distribution
of
2,4­
D
in
air
and
on
surfaces
inside
residences
after
lawn
applications:
comparing
exposure
estimates
from
various
media
for
young
children.
Environm.
Health
Perspect.
109(
11):
1185­
1191.

Stout,
D.
M.
and
Leidy,
R.
B.
(
2000).
A
preliminary
examination
of
the
translocation
of
microencapsulated
cyfluthrin
following
applications
to
the
perimeter
of
residential
dwellings.
J.
Environ.
Sci.
Health,
Part
B.
35(
4):
477­
489.

U.
S.
Environmental
Protection
Agency.
2000.
Report
of
the
Scientific
Advisory
Panel
Meeting
held
September
27­
29,
2000.
"
Session
V
­
Aggregate
and
Cumulative
Assessments
Using
Lifeline
 
­
A
Case
Study
Using
Three
Hypothetical
Pesticides"
(
p.
31).
FIFRA
Scientific
Advisory
Panel.
Office
of
Science
Coordination
and
Policy.
Report
#
2000­
03.
http://
www.
epa.
gov/
scipoly/
sap/
2000/
september/
modelsreport.
pdf.

US
Environmental
Protection
Agency.
2002.
Report
of
the
Scientific
Advisory
Panel
Meeting
held
April
30
­
May
1,
2002.
"
Cumulative
and
Aggregate
Risk
Evaluation
System
(
CARES
 
)
Model
Review"
(
p.
22).
FIFRA
Scientific
Advisory
Panel.
Office
of
Science
Coordination
and
Policy.
Report
#
2002­
02.
http://
www.
epa.
gov/
scipoly/
sap/
2002/
april/
caresmeetingminutes.
pdf
