COMMENTS
SUMMARY
REPORT
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
From
EPA
Bacteriological
Studies
Contract
No.
68­
C­
02­
091
Versar
Work
Assignment
No.
1­
11
Prepared
for:

U.
S.
Environmental
Protection
Agency
Office
of
Water
Office
of
Science
and
Technology
Health
and
Ecological
Criteria
Division
301
Constitution
Ave,
N.
W.
Washington,
D.
C.
20004
Prepared
by:

Versar,
Inc.
6850
Versar
Center
Springfield,
Virginia
22151
February
2004
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
TABLE
OF
CONTENTS
I.
INTRODUCTION
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Page
1
II.
CHARGE
TO
THE
PEER
REVIEWERS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Page
3
III.
GENERAL
COMMENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Page
4
IV.
RESPONSE
TO
CHARGE
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Page
5
V.
SPECIFIC
COMMENTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Page
9
APPENDIX
A:
Reviewer
Comments:
Joseph
Eisenberg
APPENDIX
B:
Reviewer
Comments:
Charles
McGee
APPENDIX
C:
Reviewer
Comments:
Mark
Sobsey
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
1
of
16
I.
INTRODUCTION
The
United
States
Environmental
Protection
Agency
(
EPA),
Office
of
Water
is
charged
with
protecting
public
health
and
the
environment
from
adverse
exposure
to
chemicals
and
microbials
in
water
media,
such
as
ambient
and
drinking
waters,
waste
water/
sewage
sludge
and
sediments.
In
support
of
this
mission,
the
Office
of
Water/
Office
of
Science
and
Technology
(
OST)
develops
health
standards,
health
criteria,
health
advisories,
and
technical
guidance
documents
for
water
and
waterrelated
media.

In
1986,
EPA
published
Ambient
Water
Quality
Criteria
for
Bacteria
 
1986.
That
document
contained
EPA's
recommended
water
quality
criteria
for
bacteria
to
protect
bathers
from
gastrointestinal
illness
in
recreational
waters.
The
water
quality
criteria
identified
levels
of
indicator
bacteria,
namely
Escherichia
coli
(
E.
coli)
and
enterococci,
that
demonstrate
the
presence
of
fecal
pollution
and
which
should
not
be
exceeded
to
protect
bathers
in
fresh
and
marine
recreational
waters.
Indicator
organisms
such
as
these
have
long
been
used
to
protect
bathers
from
illnesses
that
may
be
contracted
from
recreational
activities
in
surface
waters
contaminated
by
fecal
pollution.
These
organisms
generally
do
not
cause
illness
directly,
but
have
demonstrated
characteristics
that
make
them
good
indicators
of
harmful
pathogens
in
waterbodies.
Prior
to
its
1986
recommendations,
EPA
recommended
the
use
of
fecal
coliforms
as
an
indicator
organism
to
protect
bathers
from
gastrointestinal
illness
in
recreational
waters.
However,
EPA
conducted
epidemiological
studies
and
evaluated
the
use
of
several
organisms
as
indicators,
including
fecal
coliforms,
E.
coli,
and
enterococci,
and
subsequently
recommended
in
1986
the
use
of
E.
coli
for
fresh
recreational
waters
and
enterococci
for
fresh
and
marine
recreational
waters
because
they
were
better
predictors
of
acute
gastrointestinal
illness
than
fecal
coliforms.
Some
states
and
authorized
tribes
have
replaced
their
fecal
coliform
criteria
with
water
quality
criteria
for
E.
coli
and/
or
enterococci;
however,
many
other
states
and
authorized
tribes
have
not
yet
made
this
transition.

In
the
1986
criteria
document,
EPA
recommended
the
use
of
a
risk
level
associated
with
8
illnesses
per
1000
swimmers
in
fresh
waters
and
19
illnesses
per
1000
in
marine
waters.
This
represents
approximately
a
1­
2%
risk
that
recreators
will
suffer
from
gastrointestinal
illness
from
swimming
in
ambient
recreational
waters.
These
risk
levels
were
identified
based
on
the
concentrations
of
E.
coli
and
enterococci
that
roughly
correlated
to
the
previous
fecal
coliform
criterion.
However,
EPA
believes
that
it
is
appropriate
for
states
and
authorized
tribes
to
exercise
their
risk
management
discretion
when
protecting
recreational
waters.
Based
on
a
review
of
the
studies
used
in
the
derivation
of
EPA's
§
304(
a)
criteria
for
bacteria,
EPA
recommends
states
and
authorized
tribes
select
a
risk
level
for
fresh
waters
between
0.8
and
1.0
percent.
However,
some
have
suggested
that
EPA
may
be
recommending
risk
level
bounds
for
freshwater
that
are
too
restrictive
given
the
type
of
data
and
analysis
performed,
particularly
given
the
risk
level
bounds
recommended
for
marine
waters.

Under
this
work
assignment,
Sections
1.5
and
1.6
of
the
"
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
Bacteria"
were
externally
reviewed
by
a
panel
of
three
peer
reviewers.
The
peer
reviewers
were
asked
to
address
major
issues
associated
with
the
approach
used
to
determine
the
appropriate
risk
level
range
for
recreators
in
fresh
waters.
The
peer
review
panel
included
experts
in
statistical
analysis,
particularly
those
associated
with
microbiology
and
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
2
of
16
epidemiological
studies.
The
three
reviewers
were
Joseph
Eisenberg,
Charles
McGee,
and
Mark
Sobsey.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
3
of
16
II.
CHARGE
TO
THE
PEER
REVIEWERS
The
peer
reviewers
were
charged
with
responding
to
the
following
Technical
Charge:

1.
Given
the
constraints
of
the
data
available,
is
the
risk
analysis
in
the
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
Bacteria
appropriate?

2.
Is
it
scientifically
defensible
to
extrapolate
the
relationship
(
in
terms
of
linear
regression
or
other
quantitative
means)
between
bacterial
indicator
density
and
illness
rate
for
fresh
waters
beyond
the
1%
risk
level?

3.
How
much
further
could
one
extrapolate
and
what
would
be
the
rationale
for
extrapolating
further?
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
4
of
16
III.
GENERAL
COMMENTS
Joseph
Eisenberg
I
have
some
questions
on
the
use
of
geometric
means
for
estimating
the
dose
of
exposure.
It
is
the
arithmetic
mean
that
provides
the
appropriate
average
exposure
over
time?
The
geometric
mean,
which
is
a
better
estimate
of
the
median,
will
tend
to
underestimate
the
average
level
of
exposure.

It
would
be
nice
to
see
Figures
1.3
and
1.4
for
enterococci
in
fresh
and
marine
waters.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
5
of
16
IV.
RESPONSE
TO
CHARGE
Joseph
Eisenberg
The
guidelines
are
appropriate
as
written
precisely
because
they
do
not
go
beyond
the
limits
of
the
data.
See
answers
to
2
and
3
for
further
clarification.

Charles
McGee
For
many
years,
this
analysis
has
been
the
subject
of
much
debate.
However,
the
experimental
design,
the
quality
of
the
data
gathered
and
duplication
of
the
results
by
other
researchers
has
made
the
risk
analysis
put
forth
in
the
guidance
defensible.

A
current
challenge
to
the
original
research
upon
which
the
risk
analysis
is
based
is
whether
the
spatial
and
temporal
variability
of
the
beach
water
quality
was
captured
in
the
experimental
design.
In
any
study,
the
strength
of
the
relationships
between
two
variables
is
dependent
on
the
precision
of
the
measurement
of
those
variables.
EPA's
own
EMPACT
study
and
research
on
recreational
water
contamination
carried
out
on
the
west
coast
has
demonstrated
the
significance
of
this
variability.
In
preparing
my
answer
to
this
question,
I
reviewed
some
of
the
original
EPA
publications,
and
I
was
convinced
that
the
study
design
adequately
addressed
this
concern.
A
second
issue
that
should
be
addressed
is
how
measurement
error
was
taken
into
account
in
the
calculation
of
risks
criteria
for
indicator
bacteria
densities
(
see
Table
4
in
EPA440/
5­
84­
002
[
Ambient
Water
Quality
Criteria
for
Bacteria
­
1986]).
Specifically,
the
question
was
whether
water
quality
variation
can
be
better
explained
by
systematic
patterns
or
is
variability
simply
the
result
of
random
error.
Dr.
Joon
Ha
Kim
and
I
had
considered
this
point
in
a
journal
article
evaluating
error
associated
with
California's
marine
water
quality
monitoring
and
public
notification
procedures.
Upon
submission
of
this
article
to
the
journal,
one
of
the
article's
reviewer's
comments
triggered
a
rewrite
of
the
article.
The
concepts
in
that
rewrite
not
only
apply
to
the
expression
of
error
in
the
formula
in
Table
4
but
to
all
three
questions
posed
in
this
peer
review.
Because
of
time
constraints,
I
could
not
contribute
to
the
rewrite
and
had
my
name
removed
as
an
author.
I
have
spoken
with
Dr.
Kim,
and
he
has
agreed
to
allow
you
to
contact
him
for
an
advance
copy
of
his
new
article.
He
would
also
like
to
suggest
some
other
data
analyses
of
the
EPA
and
Santa
Monica
Bay
data.
Dr.
Kim's
address
is:

Joon
Ha
Kim,
Ph.
D.
944A
Engineering
Tower
Chemical
Engineering
&
Materials
Science
University
of
California,
Irvine
Irvine,
California
92697­
2575
Phone
number
949­
824­
7754
[
Note:
articles
provided
by
Dr.
Kim
are
provided
in
Appendix
B,
following
Charles
McGee's
comments.]
1.
Given
the
constraints
of
the
data
available,
is
the
risk
analysis
in
the
"
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
Bacteria"
appropriate?
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
6
of
16
Mark
Sobsey
The
answer
is
"
no".
The
reasons
for
this
answer
will
be
given
below
[
see
comments
under
"
Specific
Comments"
section]
in
more
detail.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
7
of
16
Joseph
Eisenberg
No.
The
linear
regression
model
is
a
data
driven
model;
i.
e.,
there
is
no
mechanistic
rationale
for
the
model
structure.
The
predictions
beyond
the
data,
therefore,
are
not
reliable
or
defensible.

One
potential
way
to
extrapolate
is
to
assume
the
sigmoidal
dose­
response
relationship,
based
on
data
from
dosing
trials.
These
trials
have
been
conducted
for
various
pathogens
and
administer
higher
doses
than
observed
in
epidemiology
studies.
Given
these
trial
data,
one
can
assume
that
the
relationship
is
linear
at
low
doses,
increases
exponentially
at
higher
doses,
and
eventually
saturates
at
yet
higher
doses.
This
is
basically
what
is
said
in
the
document
under
review,
and
is
illustrated
in
Figure
1.2.
The
problem
becomes
where
to
place
the
cut
point.
One
can
be
fairly
confident
that
the
cut
point
is
beyond
the
last
observed
data
point
from
the
epidemiology
studies
(
e.
g.,
beyond
236
/
100ml
and
a
10/
1000
risk
for
E
coli
in
fresh
water.
However,
since
the
epidemiology
data
is
illness
based
and
the
dosing
trial
data
is
pathogen­
specific,
it
is
difficult
to
estimate
this
cut­
off
using
the
higher
doses
from
the
dosing
trials.
One
approach
may
be
to
use
sensitivity
studies
to
looking
at
highly
infectious
organisms
and
less
infectious
organisms.

Charles
McGee
Unless
there
were
measurements
of
water
quality
and
illness
that
would
allow
the
linear
regression
to
be
defined
beyond
the
1%
level,
the
answer
to
this
question
is
no.
However,
Dr.
Kim
and
I
suggest
that
data
could
be
used
from
other
studies
with
a
similar
experimental
design
(
such
as
the
Santa
Monica
Bay)
to
supplement
EPA's
original
regression
calculations.
Dr.
Kim
has
already
examined
the
Santa
Monica
Bay
data
and
verified
that
a
similar
illness
relationship
holds
up
to
the
1%
risk
level.
Having
proven
that,
then
it
would
be
defensible
to
use
data
from
that
study
or
others
to
examine
illness
rates
beyond
the
original
1%.

Mark
Sobsey
It
is
scientifically
defensible
in
principle
to
perform
downward
extrapolations
to
lower
levels
of
risk
on
the
basis
of
data
for
microbial
dose
and
health
effects
response.
This
is
done
often
in
quantitative
microbial
risk
assessment
and
in
other
health
effects
analyses.
However,
the
scientific
validity
of
doing
this
for
the
U.
S.
EPA
data
of
this
study
and
by
the
downward
extrapolation
method
employed
is
not
scientifically
defensible.
There
is
simply
too
much
variability
and
uncertainty
in
the
data
to
justify
this
downward
extrapolation.
Furthermore,
the
simple
log­
linear
regression
model
used
for
this
downward
extrapolation
is
not
adequately
explained
or
justified
and
it
is
not
compared
to
other
more
robust
and
scientifically
valid
downward
extrapolation
models
for
such
data.
Furthermore,
the
analyses
does
not
report
any
sensitivity
analyses
that
would
indicate
to
what
extent
the
output
results
would
changes
due
to
changes
in
microbial
water
quality
or
changes
in
health
effects
outcome
(
illness
rates).
Nearly
all
of
the
potential
sources
of
bias
that
would
be
factors
influencing
the
results
were
not
addressed
in
either
the
collection
or
presentation
of
the
data
or
by
accounting
or
controlling
for
them
in
the
analyses
performed.
2.
Is
it
scientifically
defensible
to
extrapolate
the
relationship
(
in
terms
of
linear
regression
or
other
quantitative
means)
between
bacterial
indicator
density
and
illness
rate
for
fresh
waters
beyond
the
1%
risk
level?
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
8
of
16
Joseph
Eisenberg
Based
on
my
answer
to
Question
2,
I
would
not
recommend
extrapolation.
The
only
reason
to
extrapolate
beyond
the
data
would
be
to
provide
water
quality
guidelines
for
greater
than
a
10/
1000
risk
level
for
freshwater
exposures.
However
the
reason
for
using
risk
levels
greater
than
10/
1000
depends
on
how
the
acceptable
level
of
risk
is
set,
which
is
beyond
the
scope
of
this
document.

That
being
said,
based
on
some
of
the
supporting
documents
sent
to
me,
it
may
be
defensible
to
provide
guidelines
above
10/
1000
for
certain
situations.
For
example,
Figure
7
(
Cabelli
1983,
EPA
600/
1­
80­
031
[
Health
Effects
Criteria
for
Marine
Recreational
Waters],
p.
36)
suggests
that
the
data
would
allow
guidance
for
12
 
13
per
1000
illness
for
enterococci
in
fresh
waters.
Likewise,
Figures
1
and
2
(
Dufour
1984,
EPA
600/
1­
84­
004
[
Health
Effects
Criteria
for
Fresh
Recreational
Waters],
p.
26)
suggest
that
the
data
would
allow
guidance
to
just
below
30
per
1000
illnesses
for
enterococci
in
marine
waters.

Charles
McGee
The
answer
to
this
question
will
be
limited
by
water
quality
measured
in
other
studies.
The
extrapolation
has
not
been
done
yet,
but
Dr.
Kim
would
like
the
opportunity
to
do
so
using
the
Santa
Monica
Bay
data.

Mark
Sobsey
As
indicated
in
the
response
to
Question
2,
it
is
scientifically
defensible
to
perform
downward
extrapolations
to
ranges
of
dose­
response
that
are
well
below
the
levels
of
the
observable
data
range.
These
extrapolations
can
be
by
as
much
as
several
orders
of
magnitude
in
some
quantitative
microbial
dose­
response
and
risk
assessment
analyses.
However,
as
indicated
in
the
response
to
Question
2,
such
downward
extrapolations
are
not
justified
for
these
data.
This
is
because
of
the
limitations
in
the
quantity
and
quality
of
the
data,
the
failure
to
account
for
or
control
for
bias
in
either
the
data
collection
and
data
analyses,
and
the
limitations
of
the
downward
extrapolation
analyses.
Specifically
only
one
simple
log­
linear
regression
model
was
used,
there
was
an
inadequate
effort
to
address
variability
and
uncertain
and
there
is
a
lack
of
any
sensitivity
analyses.
3.
How
much
further
could
one
extrapolate
and
what
would
be
the
rationale
for
extrapolating
further?
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
9
of
16
V.
SPECIFIC
COMMENTS
Mark
Sobsey
Analysis
and
Discussion
Limitations
of
Study
Data
and
Analyses
This
risk
analysis
is
inadequate
because
the
quantity
and
quality
of
the
data
used
are
inadequate
and
because
the
analytical
approach
and
execution
also
are
inadequate.
Overall,
the
U.
S.
EPA
data
and
the
analytic
approach
are
inadequate
to
address
major
sources
of
bias
that
can
influence
the
data
quality
and
the
analytic
approaches
from
which
to
estimate
health
risks
in
relation
to
water
quality.

A
major
limitation
of
the
proposed
"
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
Bacteria"
is
the
overall
quality
of
the
microbiological
and
human
health
effects
data
used
for
the
analysis
that
provides
the
basis
for
the
criteria.
More
specifically,
the
data
come
from
studies
at
only
three
marine
beach
locations
and
only
two
freshwater
beach
locations.
These
studies
did
not
adequately
address
other
and
more
diversified
sources
of
fecal
contamination,
such
as
more
highly
treated
sewage
effluents
in
which
the
ratios
of
fecal
indicator
bacteria
to
pathogens
may
be
different
than
those
at
the
few
beaches
studied.
These
few
studies
and
study
sites
also
do
not
adequately
represent
some
other
sources
of
fecal
contamination
that
can
impact
bathing
water
and
carry
pathogens,
such
as
non­
point
sources
of
human
fecal
contamination
(
e.
g.,
septic
tank­
soil
absorption
systems
and
waste
discharges
from
boats
in
nearby
marinas)
or
non­
human
fecal
contamination
sources,
such
as
waterfowl
and
animal
agricultural
waste.
In
many
beach
locations
these
other
sources
are
the
major
sources
of
fecal
contamination
and
they
may
results
in
different
relationships
among
fecal
indicator
bacteria,
pathogens
and
attendant
human
health
risks.

Additionally,
the
important
advances
in
statistical
methods
that
have
been
made
in
the
last
two
decades
and
are
now
widely
used
in
health
effects
research
for
dose­
response
relationships
and
in
quantitative
microbial
risk
assessment
were
not
applied
in
this
study.
Advanced
regression
and
multivariate
analyses
methods
were
not
applied
in
these
studies
and
they
should
have
been.
Such
advanced
analytical
methods
are
also
much
better
for
addressing
variability
and
uncertainty
and
in
controlling
for
bias.

There
are
many
sources
of
bias
in
these
studies
and
these
sources
of
bias
were
not
adequately
addressed
or
controlled
for
in
the
U.
S.
EPA
studies.
These
sources
of
bias,
most
of
which
apply
to
the
U.
S.
EPA
studies,
are
summarized
in
Table
1.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
10
of
16
TABLE
1.
Types
of
Biases
Potentially
Encountered
in
Recreational
Water
Quality
Health
Effects
Studies
and
Their
Potential
Effects
Type
of
Bias
Description
Use
of
indicator
microbes
to
assess
water
quality
of
exposure
Temporal
and
spatial
indicator
variation
is
substantial
and
difficult
to
relate
to
individual
bathers
(
Fleisher,
1990),
unless
study
design
is
experimental
(
Kay
et
al.,
1994;
Fleisher
et
al.,
1996a).
This
is
a
limitation
of
the
U.
S.
EPA
data.
Limited
precision
of
methods
for
counting
indicator
organisms,
causing
measurement
error
(
Fleisher,
1990;
Fleisher
et
al.,
1993);
bacterial
indicators
may
not
be
representative
of
viruses,
which
may
be
important
etiological
agents
of
swimming
associated
gastrointestinal
illness.
This
is
another
limitation
of
the
U.
S.
EPA
data.

Use
of
seasonal
means
to
assess
water
quality
Some
studies
use
seasonal
or
other
collapsed
or
grouped
means
and
not
daily
measurements
of
indicator
organisms
to
characterize
individual
exposure,
thus
adding
substantial
inaccuracy.
This
is
a
limitation
of
the
U.
S.
EPA
analysis.

Assessment
of
exposure
pathway
Certain
studies
do
not
account
for
the
potential
infection
pathway
to
definite
exposure,
e.
g.,
mainly
head
immersion
or
ingestion
of
water
for
gastrointestinal
symptoms.
Difficulties
in
exposure
recall
further
increase
inaccuracy
of
individual
exposure.
These
were
limitations
in
the
U.
S.
EPA
studies.

Non­
control
for
confounders
Non­
control
for
confounders
(
e.
g.,
food
and
drink
intake,
age,
sex,
history
of
certain
diseases,
drug
use,
personal
contact,
additional
bathing,
sun,
socioeconomic
factors,
etc.),
may
influence
the
observed
association.
These
were
limitations
in
the
U.
S.
EPA
studies.

Selection
of
unrepresentative
study
population
Results
reported
for
certain
study
populations
(
e.
g.,
limited
age
groups
or
from
regions
with
certain
endemicities)
are
a
priori
not
directly
transferable
to
populations
with
other
characteristics.
This
was
a
limitation
of
the
U.
S.
EPA
studies.

Self­
reporting
of
symptoms
Most
observational
studies
relied
on
self­
reporting
of
symptoms
by
study
populations.
Validation
of
symptoms
by
medical
examination
(
Kay
et
al.,
1994;
Fleisher
et
al.,
1996a)
would
reduce
potential
bias.
External
factors,
such
as
media
or
publicity,
may
have
influenced
self­
reporting.
This
was
a
limitation
of
the
U.
S.
EPA
study.

Response
rate
Response
rates
were
>
70%
in
all,
and
>
80%
in
most,
studies.
Differential
reporting,
e.
g.,
higher
response
among
participants
experiencing
symptoms,
would
probably
not
have
major
consequences.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
Type
of
Bias
Description
2/
04
Page
11
of
16
Recruitment
method
Recruitment
methods
were
to
approach
persons
on
beaches
in
almost
all
observational
studies
and
by
advertisement
for
randomized
controlled
studies.

Interviewer
effect
Differences
in
methodology
of
data
collection
among
interviewers
may
influence
the
study
results.
(
Adapted
from
Pruss,
1998;
Stavros
and
Langford,
2002;
WHO,
2001)

Data
Quality
and
Quantity
Used
for
the
Risk
Analysis
The
quality
and
quantity
of
the
data
used
for
the
analysis
are
too
limited
and
inadequate
to
provide
reliable
national
estimates
of
the
relationships
between
human
exposure
to
pathogens
(
dose)
as
estimated
by
measuring
fecal
indicator
bacteria
such
as
E.
coli
and
enterococci
and
human
responses
(
health
effects)
in
bathers.
More
and
better
data
are
available
from
numerous
studies,
some
in
the
USA
and
some
in
other
countries,
and
they
should
have
been
used
for
these
analyses.
In
addition,
even
for
the
single
set
of
data
that
was
analyzed,
additional
analytical
methods
should
have
been
employed
to
provide
potentially
better
estimates
of
the
relationships
between
bacterial
quality
of
water
and
human
health
effects
in
bathers.

Limited
data
for
a
few
geographic
locations
and
beaches
were
used
for
the
analyses.
For
marine
water
beaches
only
three
different
geographic
locations
were
used
as
study
sites,
New
York
City,
Lake
Pontchartrain,
LA,
and
Boston,
MA.
New
York
City
had
two
beaches,
one
relatively
polluted
and
one
relatively
unpolluted.
The
Lake
Pontchartrain
study
had
two
beaches,
both
of
which
were
impacted
by
less
defined
sources
of
fecal
contamination
than
the
point
sources
found
at
other
study
locations.
Fecal
contamination
was
believed
to
be
caused
by
stormwater
discharges
reaching
the
beach
via
canals
and
bayous
which
empty
into
the
Lake
and
elevated
fecal
indicator
bacteria
levels
were
observed
in
association
with
storm
events.
In
Boston
Harbor
two
beaches
were
studied.
The
pollution
sources
impacting
these
beaches
were
not
as
well
defined
as
those
at
the
New
York
City
beaches.
One
of
the
beaches
had
fecal
indicator
bacteria
levels
about
1
order
of
magnitude
higher
than
the
other.
While
the
3
marine
beach
locations
show
some
diversity
and
have
2
or
more
different
beach
sites
for
study,
there
are
considerably
more
marine
beaches
with
greater
diversity
of
data
that
have
been
studied
for
relationships
of
bacteriological
water
quality
and
health
effects
in
swimmers
than
are
represented
here.

For
freshwater
beaches,
only
two
geographic
locations,
Erie,
PA,
and
Keystone
Lake,
near
Tulsa,
OK,
were
used
to
represent
all
freshwater
beaches
of
the
entire
country.
At
each
geographic
location,
only
two
beaches
were
used,
one
of
which
was
closer
to
a
point
source
discharge
of
sewage
effluent
and
the
other
of
which
was
more
remote
from
the
sewage
effluent
source.
The
distances
from
the
sewage
effluent
source
to
the
beaches
were
not
consistent
from
one
geographical
study
site
to
the
other,
the
degree
of
dilution
of
the
sewage
in
the
ambient
water
was
not
reported,
and
the
quality
of
the
sewage
effluent
varied.
In
one
location
it
was
chlorinated
secondary
effluent
with
unreported
chlorine
doses,
unreported
contact
times
and
unreported
residual
concentrations
of
combined
and
free
chlorine
levels
in
the
effluent
when
discharged.
In
the
other
location,
the
effluent
differed
from
one
study
year
to
the
other.
Initially
it
was
undisinfected
effluent
from
two
"
full
retention"
lagoons
of
unreported
type
(
anaerobic,
facultative
or
aerobic),
retention
time
and
operating
conditions
(
e.
g.,
were
the
two
lagoons
operated
in
series
or
in
parallel?).
In
the
second
year
the
effluent
was
treated
in
a
lagoon
of
unreported
type
and
retention
time,
followed
by
aeration
(
of
unreported
duration
and
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
12
of
16
inadequate
process
description)
followed
by
chlorination
with
unreported
chlorine
dose,
unreported
contact
time
and
unreported
residual
levels
of
free
or
combined
chlorine
in
the
discharged
effluent.

The
data
for
the
two
geographic
locations
of
freshwater
beaches
had
considerable
variability
and
uncertainty.
For
example
in
the
1979
study
year
at
Lake
Erie,
the
indicator
densities
were
unexpectedly
low
at
both
beaches,
in
1980
they
were
high
and
occasionally
extremely
high,
and
in
1982
they
were
moderately
high
relative
to
those
observed
in
1979.

At
the
Lake
Erie
beaches,
the
1980
data
do
not
reflect
bacterial
indicator
densities
consistent
with
proximity
of
the
beach
to
the
fecal
waste
source.
In
particular,
the
E.
coli
densities
are
higher
at
the
beach
more
distant
from
the
pollution
source.
The
investigators
suggest
that
these
inconsistent
results
may
have
been
caused
by
heavy
rains
which
occurred
in
the
four
days
before
the
start
of
the
beach
study
trials.
In
a
four­
day
period,
8.15
inches
of
rain
was
measured,
which
caused
the
lake
elevation
to
rise
and
the
turbidity
to
increase.
The
effect
of
these
unusual
events
on
the
swimmer
illness
rates
is
unknown.
These
observations
indicate
a
highly
variable
and
not
necessarily
representative
set
of
conditions
at
these
two
study
locations
that
call
into
question
their
national
representativeness
of
freshwater
beaches.

There
are
also
important
limitation
in
the
quality,
quantity
and
representativeness
of
the
bacteriological
data
for
water
quality.
At
each
geographic
location,
only
a
few
bacteriological
measurements
were
made
on
any
given
day
of
observations.
At
each
marine
beach,
samples
were
taken
at
2­
3
locations
in
chest
deep
water
(
location)
and
a
given
day
of
exposure,
with
only
a
few
(
3­
4)
samples
collected
between
the
hours
of
11
AM
and
5
PM
(
time
of
exposure).
The
actual
numbers
of
water
samples
taken
per
location,
site
and
study
day
were
not
specified
for
the
studies
at
the
freshwater
beaches.
However,
it
is
said
that
the
experimental
design
and
approach
was
similar
to
that
for
the
marine
beaches
studies.

It
is
unlikely
that
estimating
the
bacteriological
quality
of
water
based
on
a
relatively
small
number
of
samples
collected
only
in
chest­
deep
water
is
representative
of
the
exposure
of
all
bathers.
Children
and
many
other
people
never
venture
into
chest
deep
water
and
their
exposures
are
likely
to
be
better
represented
by
water
that
is
ankle­
deep,
knee­
deep
or
waist
deep.
Some
serious
swimmers
are
also
more
likely
to
be
exposed
to
water
beyond
the
chest­
deep
area.
In
some
subsequent
studies
on
bathing
water
quality
and
health
done
by
other
investigators,
water
samples
were
collected
from
a
grid
of
sample
sites
representing
different
depths
and
they
were
collected
are
more
frequent
intervals.
Such
sample
provides
better
estimates
of
the
bacteriological
quality
of
the
water
at
specific
locations
and
times
that
can
be
referenced
or
related
to
the
exposure
of
specific
bathers
who
bathed
at
specific
locations
for
specific
time
periods.

The
best
of
the
studies
that
related
exposure
to
recreational
bathing
water
of
measured
bacteriological
quality
to
human
health
effects
in
those
exposed
by
such
bathing
were
randomized
controlled
trials
in
which
subjected
were
recruited
and
randomly
assigned
to
a
bathing
group
or
a
non­
bathing
group.
Bathers
were
asked
to
spend
specific
times
in
the
water
(
10
minutes)
and
asked
to
immerse
their
heads
at
least
three
times
(
Kay
et
al.,
1994;
2001;
Fleisher
et
al.,
1998).
Exposure
measurements
of
bacteriological
quality
of
the
water
were
made
at
the
time
and
location
of
exposure
and
at
three
different
depths,
specifically
surf,
mid
and
chest.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
13
of
16
Other
Source
of
Relevant
Data
Not
Used
in
this
Analysis
Since
these
few
early
studies
by
the
U.
S.
EPA,
there
have
been
numerous
studies
of
water
quality
at
bathing
beaches
in
relation
to
human
health
effects
in
swimmers.
The
U.
S.
EPA
should
have
used
the
marine
water
and
freshwater
data
from
these
other
studies
in
its
analyses
and
risk
assessment.
In
a
review
article
published
in
1998,
Pruss
reported
on
the
findings
of
4
additional
studies
conducted
at
freshwater
beaches
in
the
UK,
France
and
Canada
and
14
marine
water
beaches
in
countries
in
North
America,
Europe,
the
Middle
East,
Asia,
the
Western
Pacific
(
Australia
and
New
Zealand)
and
South
Africa.
More
recently,
Wade
et
al.
(
2003)
conducted
a
systematic
review
and
meta­
analysis
of
a
total
of
15
marine
water
studies
and
8
freshwater
studies
and
on
the
relationships
of
bacterial
water
quality
and
health,
besides
the
studies
done
by
the
U.
S.
EPA.
These
more
recent
studies
provide
a
more
diversified
and
more
representative
database
than
the
few
studies
used
by
the
U.
S.
EPA.
These
more
recent
analyses
reveal
wider
ranges
of
bacterial
densities
than
studied
by
the
U.
S.
EPA
and
document
different
dose­
response
relationships
for
bacterial
densities
and
human
health
effects
than
those
observed
by
the
U.
S.
EPA
in
its
studies.
In
the
study
by
Wade
et
al.
(
2003)
the
analyses
of
the
data
for
freshwater
beaches
showed
elevated
relative
risks
of
swimming­
associated
illness
both
above
and
below
the
U.
S.
EPA
guideline
value
for
enterococci.
For
E.
coli,
studies
below
the
U.
S.
EPA
guideline
value
were
not
associated
with
increased
illness,
while
exposures
above
the
U.
S.
EPA
guideline
value
were.
These
findings
suggest
that
the
current
U.
S.
EPA
guideline
values
for
the
two
different
indicators
provide
different
and
inconsistent
levels
of
bather
protection
from
swimming­
associated
illness
in
freshwater
beaches.
Therefore,
the
two
different
indicators
and
their
associated
guideline
values
are
not
interchangeable
in
terms
of
their
levels
of
protection.
Based
on
the
U.
S.
EPA's
analysis
of
its
own
data
for
freshwater
studies,
these
two
indicators
are
considered
interchangeable
and
give
equivalent
levels
of
protection.
For
the
analyses
of
the
marine
water
beaches,
Wade
et
al.
(
2003)
found
that
enterococci
were
the
indicator
that
most
strongly
predicted
increased
health
risks.
By
categorical
analysis,
the
relative
health
risks
did
not
continue
to
increase
in
studies
with
bacterial
densities
greater
than
104
cfu/
100
ml.
This
indicates
a
potential
threshold
effect
for
risk
of
GI
illness.
By
weighted
regression
analysis
there
was
an
association
between
enterococci
density
and
the
natural
log
of
the
relative
risk
of
health
effects.
The
relative
risk
for
GI
illness
increased
1.3
times
for
every
log10
increase
in
enterococci
density
in
water.
In
relation
to
the
current
U.
S.
EPA
guideline
for
enterococci
in
marine
waters,
summary
relative
risks
for
GI
illness
below
the
U.
S.
EPA
guideline
value
were
lower
and
not
statistically
significant.
Relative
risks
for
GI
illness
above
the
U.
S.
EPA
guideline
value
were
elevated
and
statistically
significant.

Analytical
Approach
and
Execution
The
approach
to
the
analyses
of
the
microbiological
data
is
limited
and
probably
flawed.
More
and
better
approaches
to
the
types
of
data
gathered
and
analyzed
and
the
type
of
analyses
of
the
data
should
have
been
conducted.
The
U.
S.
EPA
should
have
attempted
to
provide
better
estimates
of
the
concentrations
of
bacteria
in
the
water
to
which
people
were
exposed,
particularly
with
respect
to
the
spatial
and
temporal
relationships
of
the
exposures.
The
bacterial
data
for
each
location
were
analyzed
by
grouping
them
by
location
and
then
season
and
computing
a
geometric
mean
concentration
of
bacteria.
Apparently,
only
a
few
measures
of
bacteria
concentration
were
made
for
a
given
beach,
with
samples
taken
at
2­
3
locations
n
chest
deep
water
(
location)
and
a
given
day
of
exposure,
with
only
a
few
(
3­
4)
samples
collected
between
the
hours
of
11
AM
and
5
PM
(
time
of
exposure).
Geometric
mean
concentrations
were
computed
for
this
exposure
location
and
day
and
then
used
in
a
linear
regression
analysis
to
determine
relationships
between
illness
rates
and
average
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
14
of
16
water
quality.
The
resulting
linear
regression
model
was
then
used
to
derive
standard
deviations
for
the
dose­
response
relationship.

The
approach
used
by
the
U.
S.
EPA
in
which
all
data
were
combined
for
the
development
of
the
distribution
of
bacterial
concentrations
that
was
then
used
for
comparisons
with
human
health
effects
has
serious
drawbacks
and
limitations.
This
is
because
the
standard
deviation
of
the
probability
density
function
or
distribution
affects
the
probability
of
exposure
to
polluted
water
and
thus
the
risk
of
illness.
The
use
of
a
single
distribution
and
resulting
parameter
value
to
define
a
guideline
value
across
all
waters
does
not
adequately
address
local
variability.
Local
variations
in
standard
deviation
will
mean
that
risks
of
illness
will
vary
even
though
the
same
guideline
value
is
in
place.
Simply
stated,
combining
the
data
into
a
single
distribution
does
not
address
local
variability
in
the
distribution
of
bacteria
density
on
a
site­
specific
basis
and
the
relationships
of
these
bacterial
densities
to
local
heath
risks.

In
addition,
it
is
not
clear
the
U.
S.
EPA
used
robust
criteria
to
determine
if
bacterial
concentrations
could
be
legitimately
log10­
transformed
to
create
the
log10
distribution
that
was
used
in
the
doseresponse
analyses.
Statistical
tests
should
have
been
done
to
test
for
normality
or
to
determine
if
the
hypothesis
of
normality
can
be
rejected.
Such
analyses
should
have
been
done
for
the
data
from
individual
beaches
and
study
sites
as
well
as
on
the
combined
data.
The
need
to
test
for
the
normality
of
the
data
on
a
site­
specific
basis
is
because
of
the
need
to
link
bacterial
densities
with
specific
exposures
that
result
in
health
effects.

Apparently,
the
U.
S.
EPA
did
not
consider
alternative
analytical
strategies
for
these
data
that
do
not
rely
on
the
use
of
the
data
taken
directly
from
a
log10­
transformed
distribution.
One
possible
alternative
approach
in
situations
where
normality
(
or
log­
normality)
is
violated
is
to
use
a
Bootstrapping
procedure.
In
this
case
a
Monte
Carlo
style
procedure
is
applied
to
bootstrapped
samples
of
the
actual
empirical
distributions
of
bacterial
concentration,
rather
than
the
parametrically
generated
distribution.
The
bootstrapping
procedure
draws
a
large
number
(
say
1000)
of
"
resamples",
of
size
equal
to
the
original
sample,
from
this
original
sample
randomly
with
replacement.
No
such
alternative
analytical
approach
was
attempted
by
the
U.
S.
EPA.

The
basis
for
defining
the
different
risk
levels
as
upper
percentiles
(
e.
g.,
75th,
82nd,
90th
and
95th)
is
poorly
documented
and
justified
and
the
basis
for
focusing
only
the
human
health
risks
in
the
range
of
0.8
to
1%
also
is
poorly
documented
and
inadequately
explained
or
justified.
These
standard
deviation
estimates
were
used
to
consider
various
upper
percentiles
(
75th,
82d,
90th
and
95%)
for
which
were
calculate
various
upper
percentiles
values
of
allowable
bacterial
density
per
100
ml
corresponding
to
different
health
risk
levels
in
the
range
of
0.8
to
1%.
The
flaws
and
weakness
of
this
approach
are
that
the
data
were
transformed
to
log10
values
and
these
log10­
transformed
values
were
used
in
the
analyses
and
the
analysis
of
the
log­
transformed
data
used
only
a
simple
linear
regression
model
to
examine
the
relationship
between
indicator
density
and
human
health
risk
of
bathers.
No
other
forms
of
the
data
were
used
for
analyses
(
such
as
arithmetic
forms
of
the
data
with
no
transformations)
and
no
other
models
were
applied
to
the
data
(
such
as
Beta­
Poisson,
twopopulation
or
other
dose­
response
models
now
widely
used
for
quantitative
microbial
risk
assessment).
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
15
of
16
In
the
study
of
Kay
et
al.
(
1994)
an
epidemiological
relationship
described
the
excess
risk
of
illness
from
exposure
to
water
containing
fecal
indicator
bacteria.
This
relationship
was
best
as
a
doseresponse
relationship
linking
water
quality
exposure
(
x),
indexed
by
the
fecal
streptococci
density
at
chest
depth
water,
and
the
excess
probability
of
gastroenteritis
(
y)
is
given
by
the
following
(
for
exposures
between
32
and
158
fecal
streptococci
per
100ml):

where,
m
is
the
natural
logarithm
of
the
odds
of
getting
gastroenteritis
from
bathing,
derived
from
the
logistic
regression
equation:

and
the
term
p32
is
the
probability
of
gastroenteritis
where
x
=
32
cfu
per
100
ml
(
p=
0.0866)
and
adjusts
the
relationship
to
reflect
excess
rather
than
absolute
probability
of
illness
relative
to
those
who
do
not
bathe.

In
addition
to
the
consideration
of
the
dose­
response
model
used,
there
are
valid
reasons
to
believe
that
the
form
of
microbiological
data
used,
specifically
the
log10­
transformed
data
of
the
U.
S.
EPA,
is
likely
to
underestimate
exposure
in
dose­
response
analyses
and
that
the
liner
regression
model
applied
to
these
data
is
likely
to
provide
a
downward
extrapolation
that
is
a
less
reliable
portrayal
of
the
actual
dose­
response
relationship.
This
is
not
always
the
case,
as
was
found
in
the
studies
by
Kay
et
al.,
where
the
bacteriological
data
were
best
described
by
a
log­
normal
probability
density
function.
However,
there
are
scientifically
valid
reasons
to
use
arithmetic
data
for
dose­
response
analyses
and
determine
if
this
from
of
the
data
better
describes
the
extremes
of
exposure
and
resulting
health
effects
at
both
the
low
and
high
ends
of
the
distribution
of
bacteria
concentration.
Overall,
there
are
serious
concerns
about
the
extent
to
which
reliable
downward
extrapolations
can
be
made
from
the
U.
S.
EPA
data,
given
all
of
the
potential
sources
of
bias,
the
use
of
log10­
transformed
data
rather
than
arithmetic
data,
the
use
of
only
a
simple
log­
linear
regression
model,
and
the
rather
shallow
slope
of
the
dose­
response
relationship
in
the
human
health
effects
response
range
of
interest
(
corresponding
to
0.8
to
1.0%
risk).

In
addition,
there
was
very
inadequate
treatment
of
variability
and
uncertainty
in
the
analyses
for
either
bacteriological
quality
of
the
water,
the
extent
of
bather
exposure
and
the
temporal
and
spatial
relationships
between
exposure
and
resulting
health
effects
in
the
exposed.
These
deficiencies
make
it
inappropriate
to
attempt
to
do
downward
extrapolations
of
the
water
quality
(
bacterial
concentration)
­
health
effects)
dose­
response)
relationships
in
the
data
range
of
interest
for
freshwaters
or
in
general.
Developing
water
quality
criteria
and
regulatory
guidelines
based
on
such
analyses
can
not
be
supported
or
justified
in
the
opinion
of
this
reviewer.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Page
16
of
16
References
Cited
Fleisher
JM,
Kay
D,
Wyer
MD,
A.
F.
Godfree
Estimates
of
the
severity
of
illnesses
associated
with
bathing
in
marine
recreational
waters
contaminated
with
domestic
sewage.
Int
J
Epidemiol.
1998
Aug;
27(
4):
722­
6.

Kay
D,
Fleisher
JM,
Salmon
RL,
Jones
F,
Wyer
MD,
Godfree
AF,
Zelenauch­
Jacquotte
Z,
R.
Shore
(
1994)
Predicting
likelihood
of
gastroenteritis
from
sea
bathing:
results
from
randomised
exposure.
Lancet.
1994
Oct
1;
344(
8927):
905­
9.

Kay,
D.,
J.
Fleisher,
M.
D.
Wyer,
and
R.
I.
Salmon
(
2001)
Re­
analysis
of
the
Seabathing
Data
from
the
UK
Randomised
Trials.
A
Report
to
DETR.
Aberystwyth,
University
of
Wales,
Centre
for
Research
into
the
Environment
and
health,
17
pages.

Pruss,
A.
(
1998)
Review
of
epidemiological
studies
on
health
effects
from
exposure
to
recreational
water.
International
Journal
of
Epidemiology
27(
1):
1­
9.

Stavros,
G.
and
I.
H.
Langford
(
2002)
Coastal
Bathing
Water
Quality
and
Human
Health
Risks:
A
Review
of
Legislation,
Policy
and
Epidemiology,
with
an
Assessment
of
Current
UK
Water
Quality,
Proposed
Standards,
and
Disease
Burden
in
England
and
Wales.
CSERGE
Working
Paper
ECM
02­
06.
Centre
for
Social
and
Economic
Research
on
the
Global
Environment,
University
of
East
Anglia,
UK.

Wade
TJ,
Pai
N,
Eisenberg
JN,
Colford
JM
Jr.
(
2003)
Do
U.
S.
Environmental
Protection
Agency
water
quality
guidelines
for
recreational
waters
prevent
gastrointestinal
illness?
A
systematic
review
and
meta­
analysis.
Environ
Health
Perspect.
2003
Jun;
111(
8):
1102­
9.

WHO
(
2001)
Bathing
Water
Quality
and
Human
Health:
Faecal
Pollution.
Outcome
of
an
Expert
Consultation,
Farnham,
UK,
April
2001.
Co­
sponsored
by
Department
of
the
Environment,
Transport
and
the
Regions,
United
Kingdom.
WHO/
SDE/
WSH/
01.2.
Geneva:
World
Health
Organization.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
APPENDIX
A
Joseph
Eisenberg
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Peer
review
of
the
EPA
document"
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
Bacteria"

For
Versar
Reviewed
by
Joe
Eisenberg
12/
20/
03
Overview
The
predicted
risk
level
(
risk
of
GI
illness
per
1000
swimmers)
for
a
given
water
quality
(
geometric
mean
density
of
an
indicator
organism)
is
based
on
a
linear
regression
model.
This
model
was
estimated
using
beach
study
data
(
e.
g.,
9
data
points
for
E
coli
in
fresh
waters).
In
addition
to
providing
point
risk
estimates
for
geometric
mean
values
of
indicator
densities,
the
regression
model
also
allows
for
the
calculation
of
75,
82,
90,
and
95%
confidence
limits,
which
were
used
to
fill
in
the
percentile
values
in
Table
1­
1
and
1­
2.

Comments
on
text
I
have
some
questions
on
the
use
of
geometric
means
for
estimating
the
dose
of
exposure.
It
is
the
arithmetic
mean
that
provides
the
appropriate
average
exposure
over
time.
The
geometric
mean,
which
is
a
better
estimate
of
the
median,
will
tend
to
underestimate
the
average
level
of
exposure.

It
would
be
nice
to
see
Figures
1.3
and
1.4
for
enterococci
in
fresh
and
marine
waters.

1.
Given
the
constraints
of
the
data
available,
is
the
risk
analysis
in
the
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
Bacteria
appropriate?

The
guidelines
are
appropriate
as
written
precisely
because
they
do
not
go
beyond
the
limits
of
the
data.
See
answers
to
2
and
3
for
further
clarification.

2.
Is
it
scientifically
defensible
to
extrapolate
the
relationship
(
in
terms
of
linear
regression
or
other
quantitative
means)
between
bacterial
indicator
density
and
illness
rate
for
fresh
waters
beyond
the
1%
risk
level?

No.
The
linear
regression
model
is
a
data
driven
model;
i.
e.,
there
is
no
mechanistic
rationale
for
the
model
structure.
The
predictions
beyond
the
data,
therefore,
are
not
reliable
or
defensible.

One
potential
way
to
extrapolate
is
to
assume
the
sigmoidal
dose­
response
relationship,
based
on
data
from
dosing
trials.
These
trials
have
been
conducted
for
various
pathogens
and
administer
higher
doses
than
observed
in
epidemiology
studies.
Given
these
trial
data,
one
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
can
assume
that
the
relationship
is
linear
at
low
doses,
increases
exponentially
at
higher
doses,
and
eventually
saturates
at
yet
higher
doses.
This
is
basically
what
is
said
in
the
document
under
review,
and
is
illustrated
in
Figure
1.2.
The
problem
becomes
where
to
place
the
cut
point.
One
can
be
fairly
confident
that
the
cut
point
is
beyond
the
last
observed
data
point
from
the
epidemiology
studies
(
e.
g.,
beyond
236
/
100ml
and
a
10/
1000
risk
for
E
coli
in
fresh
water.
However,
since
the
epidemiology
data
is
illness
based
and
the
dosing
trial
data
is
pathogen­
specific,
it
is
difficult
to
estimate
this
cut­
off
using
the
higher
doses
from
the
dosing
trials.
One
approach
may
be
to
use
sensitivity
studies
to
looking
at
highly
infectious
organisms
and
less
infectious
organisms.

3.
How
much
further
could
one
extrapolate
and
what
would
be
the
rationale
for
extrapolating
further?

Based
on
my
answer
to
Question
2,
I
would
not
recommend
extrapolation.
The
only
reason
to
extrapolate
beyond
the
data
would
be
to
provide
water
quality
guidelines
for
greater
than
a
10/
1000
risk
level
for
freshwater
exposures.
However
the
reason
for
using
risk
levels
greater
than
10/
1000
depends
on
how
the
acceptable
level
of
risk
is
set,
which
is
beyond
the
scope
of
this
document.

That
being
said,
based
on
some
of
the
supporting
documents
sent
to
me,
it
may
be
defensible
to
provide
guidelines
above
10/
1000
for
certain
situations.
For
example,
Figure
7
(
Cabelli
1983,
EPA
600/
1­
84­
004,
p36)
suggests
that
the
data
would
allow
guidance
for
12
 
13
per
1000
illness
for
enterococci
in
fresh
waters.
Likewise,
Figure
1
and
2
(
Dufour
1984,
EPA
600/
1­
84­
004,
p26)
suggest
that
the
data
would
allow
guidance
to
just
below
30
per
1000
illnesses
for
enterococci
in
marine
waters.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
APPENDIX
B
Charles
McGee
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Diane
S.
Sinkowski
Environmental
Engineer
Exposure/
Risk
Assessment
Division
Versar,
Inc.
6850
Versar
Center
Springfield,
VA
22151
Dear
Ms.
Sinknwski,

In
regards
to:
Work
Assignment
#
1­
11;
Peer
Review
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
The
following
are
my
responses
to
the
three
questions
raised
in
the
above
work
assignment:

1.
Given
the
constraints
of
the
data
available,
is
the
risk
analysis
in
the
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
Bacteria
appropriate?

Response:
For
many
years,
this
analysis
has
been
the
subject
of
much
debate.
However,
the
experimental
design,
the
quality
of
the
data
gathered
and
duplication
of
the
results
by
other
researchers
has
made
the
risk
analysis
put
forth
in
the
guidance
defensible.

A
current
challenge
to
the
original
research
upon
which
the
risk
analysis
is
based
is
whether
the
spatial
and
temporal
variability
of
the
beach
water
quality
was
captured
in
the
experimental
design.
In
any
study,
the
strength
o
f
t
he
relationships
between
two
variables
is
dependent
on
the
precision
of
the
measurement
of
those
variables.
EPA's
own
EMPACT
study
and
research
on
recreational
water
contamination
carried
out
on
the
west
coast
has
demonstrated
the
significance
of
this
variability.
In
preparing
my
answer
to
this
question,
I
reviewed
some
of
the
original
EPA
publications,
and
I
was
convinced
that
the
study
design
adequately
addressed
this
concern.

A
second
issue
that
should
be
addressed
is
how
measurement
error
was
taken
into
account
in
the
calculation
of
risks
criteria
for
indicator
bacteria
densities
(
see
Table
4
in
EPA440/
5­
84­
002).
Specifically,
the
question
was
whether
water
quality
variation
can
be
better
explained
by
systematic
patterns
or
is
variability
simply
the
result
of
random
error.
Dr.
Joon
Ha
Kim
and
I
had
considered
this
point
in
a
journal
article
evaluating
error
associated
with
California's
marine
water
quality
monitoring
and
public
notification
procedures.
Upon
submission
of
this
article
to
the
journal,
one
of
the
article's
reviewer's
comments
triggered
a
rewrite
of
the
article.
The
concepts
in
that
rewrite
not
only
apply
to
the
expression
of
error
in
the
formula
in
Table
4
but
to
all
three
questions
posed
in
this
peer
review.
Because
of
time
constraints,
I
could
not
contribute
to
the
rewrite
and
had
my
name
removed
as
an
author.
I
have
spoken
with
Dr.
Kim,
and
he
has
agreed
to
allow
you
to
contact
him
for
an
advance
copy
of
his
new
article.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
He
would
also
like
to
suggest
some
other
data
analyses
of
the
EPA
and
Santa
Monica
Bay
data.
Dr.
Kim's
address
is:

Joon
Ha
Kim,
Ph.
D.

944A
Engineering
Tower
Chemical
Engineering
&
Materials
Science
University
of
California,
Irvine
Irvine,
California
92697­
2575
Phone
number
949­
824­
7754
2.
Is
it
scientifically
defensible
to
extrapolate
the
relationship
(
in
terms
of
linear
regression
or
other
quantitative
means)
between
bacterial
indicator
density
and
illness
rate
for
fresh
waters
beyond
the
1%
risk
level?

Response:
Unless
there
were
measurements
of
water
quality
and
illness
that
would
allow
the
linear
regression
to
be
defined
beyond
the
1%
level,
the
answer
to
this
question
is
no.
However,
Dr.
Kim
and
I
suggest
that
data
could
be
used
from
other
studies
with
a
similar
experimental
design
(
such
as
the
Santa
Monica
Bay)
to
supplement
EPA's
original
regression
calculations.
Dr.
Kim
has
already
examined
the
Santa
Monica
Bay
data
and
verified
that
a
similar
illness
relationship
holds
up
to
the
1%
risk
level.
Having
proven
that,
then
it
would
be
defensible
to
use
data
from
that
study
or
others
to
examine
illness
rates
beyond
the
original
1%.

3.
How
much
further
could
one
extrapolate
and
what
would
be
the
rational
for
extrapolating
further?

Response:
The
answer
to
this
question
will
be
limited
by
water
quality
measured
in
other
studies.
The
extrapolation
has
not
been
done
yet,
but
Dr.
Kim
would
like
the
opportunity
to
do
so
using
the
Santa
Monica
Bay
data.

Thank
you
for
the
opportunity
to
comment
on
these
questions,
but
my
best
advice
would
be
to
contact
Dr.
Kim
to
actually
perform
the
statistical
analysis
that
could
further
strengthen
the
underpinnings
of
the
guidance
document.

Best
regards,
Charles
D.
McGee
Laboratory
Supervisor
Orange
County
Sanitation
District
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
APPENDIX
C
Mark
Sobsey
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Peer
Review
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
Mark
D.
Sobsey
University
of
North
Carolina
CB#
7431,
McGavran­
Greenberg
Hall,
Room
4114a
Chapel
Hill,
NC
27599­
7431
Responses
to
the
three
specific
questions
posed
to
the
reviewers:

1.
Given
the
constraints
of
the
data
available,
is
the
risk
analysis
in
the
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
Bacteria"
appropriate?

Response:
The
answer
is
"
no".
The
reasons
for
this
answer
will
be
given
below
in
more
detail.

2.
Is
it
scientifically
defensible
to
extrapolate
the
relationship
(
in
terms
of
linear
regression
or
other
quantitative
means)
between
bacterial
indicator
density
and
illness
rate
for
fresh
waters
beyond
the
1%
risk
level?

Response:
It
is
scientifically
defensible
in
principle
to
perform
downward
extrapolations
to
lower
levels
of
risk
on
the
basis
of
data
for
microbial
dose
and
health
effects
response.
This
is
done
often
in
quantitative
microbial
risk
assessment
and
in
other
health
effects
analyses.
However,
the
scientific
validity
of
doing
this
for
the
US
EPA
data
of
this
study
and
by
the
downward
extrapolation
method
employed
is
not
scientifically
defensible.
There
is
simply
too
much
variability
and
uncertainty
in
the
data
to
justify
this
downward
extrapolation.
Furthermore,
the
simple
log­
linear
regression
model
used
for
this
downward
extrapolation
is
not
adequately
explained
or
justified
and
it
is
not
compared
to
other
more
robust
and
scientifically
valid
downward
extrapolation
models
for
such
data.
Furthermore,
the
analyses
does
not
report
any
sensitivity
analyses
that
would
indicate
to
what
extent
the
output
results
would
changes
due
to
changes
in
microbial
water
quality
or
changes
in
health
effects
outcome
(
illness
rates).
Nearly
all
of
the
potential
sources
of
bias
that
would
be
factors
influencing
the
results
were
not
addressed
in
either
the
collection
or
presentation
of
the
data
or
by
accounting
or
controlling
for
them
in
the
analyses
performed.

3.
How
much
further
could
one
extrapolate
and
what
would
be
the
rationale
for
extrapolation
further.

Response:
As
indicated
in
the
response
to
question
2,
it
is
scientifically
defensible
to
perform
downward
extrapolations
to
ranges
of
dose­
response
that
are
well
below
the
levels
of
the
observable
data
range.
These
extrapolations
can
be
by
as
much
as
several
orders
of
magnitude
in
some
quantitative
microbial
doseresponse
and
risk
assessment
analyses.
However,
as
indicated
in
the
response
to
question
2,
such
downward
extrapolations
are
not
justified
for
these
data.
This
is
because
of
the
limitations
in
the
quantity
and
quality
of
the
data,
the
failure
to
account
for
or
control
for
bias
in
either
the
data
collection
and
data
analyses,
and
the
limitations
of
the
downward
extrapolation
analyses.
Specifically
only
one
simple
log­
linear
regression
model
was
used,
there
was
an
inadequate
effort
to
address
variability
and
uncertain
and
there
is
a
lack
of
any
sensitivity
analyses.

ANALYSIS
AND
DISCUSSION
Limitations
of
Study
Data
and
Analyses
This
risk
analysis
is
inadequate
because
the
quantity
and
quality
of
the
data
used
are
inadequate
and
because
the
analytical
approach
and
execution
also
are
inadequate.
Overall,
the
US
EPA
data
and
the
analytic
approach
are
inadequate
to
address
major
sources
of
bias
that
can
influence
the
data
quality
and
the
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
analytic
approaches
from
which
to
estimate
health
risks
in
relation
to
water
quality.

A
major
limitation
of
the
proposed
"
Implementation
Guidance
for
Ambient
Water
Quality
Criteria
for
bacteria"
is
the
overall
quality
of
the
microbiological
and
human
health
effects
data
used
for
the
analysis
that
provides
the
basis
for
the
criteria.
More
specifically,
the
data
come
from
studies
at
only
three
marine
beach
locations
and
only
two
freshwater
beach
locations.
These
studies
did
not
adequately
address
other
and
more
diversified
sources
of
fecal
contamination,
such
as
more
highly
treated
sewage
effluents
in
which
the
ratios
of
fecal
indicator
bacteria
to
pathogens
may
be
different
than
those
at
the
few
beaches
studied.
These
few
studies
and
study
sites
also
do
not
adequately
represent
some
other
sources
of
fecal
contamination
that
can
impact
bathing
water
and
carry
pathogens,
such
as
non­
point
sources
of
human
fecal
contamination
(
e.
g.,
septic
tank­
soil
absorption
systems
and
waste
discharges
from
boats
in
nearby
marinas)
or
non­
human
fecal
contamination
sources,
such
as
waterfowl
and
animal
agricultural
waste.
In
many
beach
locations
these
other
sources
are
the
major
sources
of
fecal
contamination
and
they
may
results
in
different
relationships
among
fecal
indicator
bacteria,
pathogens
and
attendant
human
health
risks.

Additionally,
the
important
advances
in
statistical
methods
that
have
been
made
in
the
last
two
decades
and
are
now
widely
used
in
health
effects
research
for
dose­
response
relationships
and
in
quantitative
microbial
risk
assessment
were
not
applied
in
this
study.
Advanced
regression
and
multivariate
analyses
methods
were
not
applied
in
these
studies
and
they
should
have
been.
Such
advanced
analytical
methods
are
also
much
better
for
addressing
variability
and
uncertainty
and
in
controlling
for
bias.

There
are
many
sources
of
bias
in
these
studies
and
these
sources
of
bias
were
not
adequately
addressed
or
controlled
for
in
the
EPA
studies.
These
sources
of
bias,
most
of
which
apply
to
the
US
EPA
studies,
are
summarized
in
Table
1
below.

Type
of
Bias
Description
Use
of
indicator
microbes
to
assess
water
quality
of
exposure
Temporal
and
spatial
indicator
variation
is
substantial
and
difficult
to
relate
to
individual
bathers
(
Fleisher,
1990),
unless
study
design
is
experimental
(
Kay
et
al.,
1994;
Fleisher
et
al.,
1996a).
This
is
a
limitation
of
the
US
EPA
data.
Limited
precision
of
methods
for
counting
indicator
organisms,
causing
measurement
error
(
Fleisher,
1990;
Fleisher
et
al.,
1993);
bacterial
indicators
may
not
be
representative
of
viruses,
which
may
be
important
etiological
agents
of
swimming
associated
gastrointestinal
illness.
This
is
another
limitation
of
the
US
EPA
data.

Use
of
seasonal
means
to
assess
water
quality
Some
studies
use
seasonal
or
other
collapsed
or
grouped
means
and
not
daily
measurements
of
indicator
organisms
to
characterize
individual
exposure,
thus
adding
substantial
inaccuracy.
This
is
a
limitation
of
the
US
EPA
analysis.

Assessment
of
exposure
pathway
Certain
studies
do
not
account
for
the
potential
infection
pathway
to
definite
exposure,
e.
g.,
mainly
head
immersion
or
ingestion
of
water
for
gastrointestinal
symptoms.
Difficulties
in
exposure
recall
further
increase
inaccuracy
of
individual
exposure.
These
were
limitations
in
the
US
EPA
studies.

Non­
control
for
confounders
Non­
control
for
confounders
(
e.
g.,
food
and
drink
intake,
age,
sex,
history
of
certain
diseases,
drug
use,
personal
contact,
additional
bathing,
sun,
socio­
economic
factors,
etc.),
may
influence
the
observed
association.
These
were
limitations
in
the
US
EPA
studies.

Selection
of
Results
reported
for
certain
study
populations
(
e.
g.,
limited
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
unrepresentative
study
population
age
groups
or
from
regions
with
certain
endemicities)
are
a
priori
not
directly
transferable
to
populations
with
other
characteristics.
This
was
a
limitation
of
the
US
EPA
studies.

Self­
reporting
of
symptoms
Most
observational
studies
relied
on
self­
reporting
of
symptoms
by
study
populations.
Validation
of
symptoms
by
medical
examination
(
Kay
et
al.,
1994;
Fleisher
et
al.,
1996a)
would
reduce
potential
bias.
External
factors,
such
as
media
or
publicity,
may
have
influenced
self­
reporting.
This
was
a
limitation
of
the
US
EPA
study.

Response
rate
Response
rates
were
>
70%
in
all,
and
>
80%
in
most,
studies.
Differential
reporting,
e.
g.,
higher
response
among
participants
experiencing
symptoms,
would
probably
not
have
major
consequences.

Recruitment
method
Recruitment
methods
were
to
approach
persons
on
beaches
in
almost
all
observational
studies
and
by
advertisement
for
randomized
controlled
studies.

Interviewer
effect
Differences
in
methodology
of
data
collection
among
interviewers
may
influence
the
study
results.
TABLE
1.
Types
of
Biases
Potentially
Encountered
in
Recreational
Water
Quality
Health
Effects
Studies
and
Their
Potential
Effects
(
Adapted
from
Pruss,
1998;
Stavros
and
Langford,
2002;
WHO,
2001)

Data
Quality
and
Quantity
Used
for
the
Risk
Analysis
The
quality
and
quantity
of
the
data
used
for
the
analysis
are
too
limited
and
inadequate
to
provide
reliable
national
estimates
of
the
relationships
between
human
exposure
to
pathogens
(
dose)
as
estimated
by
measuring
fecal
indicator
bacteria
such
as
E.
coli
and
enterococci
and
human
responses
(
health
effects)
in
bathers.
More
and
better
data
are
available
from
numerous
studies,
some
in
the
USA
and
some
in
other
countries,
and
they
should
have
been
used
for
these
analyses.
In
addition,
even
for
the
single
set
of
data
that
was
analyzed,
additional
analytical
methods
should
have
been
employed
to
provide
potentially
better
estimates
of
the
relationships
between
bacterial
quality
of
water
and
human
health
effects
in
bathers.

Limited
data
for
a
few
geographic
locations
and
beaches
were
used
for
the
analyses.
For
marine
water
beaches
only
three
different
geographic
locations
were
used
as
study
sites,
New
York
City,
Lake
Pontchartrain,
LA.,
and
Boston,
MA.
New
York
City
had
two
beaches,
one
relatively
polluted
and
one
relatively
unpolluted.
The
Lake
Pontchartrain
study
had
two
beaches,
both
of
which
were
impacted
by
less
defined
sources
of
fecal
contamination
than
the
point
sources
found
at
other
study
locations.
Fecal
contamination
was
believed
to
be
caused
by
stormwater
discharges
reaching
the
beach
via
canals
and
bayous
which
empty
into
the
Lake
and
elevated
fecal
indicator
bacteria
levels
were
observed
in
association
with
storm
events.
In
Boston
harbor
two
beaches
were
studied.
The
pollution
sources
impacting
these
beaches
were
not
as
well
defined
as
those
at
the
New
York
City
beaches.
One
of
the
beaches
had
fecal
indicator
bacteria
levels
about
1
order
of
magnitude
higher
than
the
other.
While
the
3
marine
beach
locations
show
some
diversity
and
have
2
or
more
different
beach
sites
for
study,
there
are
considerably
more
marine
beaches
with
greater
diversity
of
data
that
have
been
studied
for
relationships
of
bacteriological
water
quality
and
health
effects
in
swimmers
than
are
represented
here.

For
freshwater
beaches,
only
two
geographic
locations,
Erie,
PA
and
Keystone
Lake,
near
Tulsa,
OK,
were
used
to
represent
all
freshwater
beaches
of
the
entire
country
(.
At
each
geographic
location,
only
two
beaches
were
used,
one
of
which
was
closer
to
a
point
source
discharge
of
sewage
effluent
and
the
other
of
which
was
more
remote
from
the
sewage
effluent
source.
The
distances
from
the
sewage
effluent
source
to
the
beaches
were
not
consistent
from
one
geographical
study
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
site
to
the
other,
the
degree
of
dilution
of
the
sewage
in
the
ambient
water
was
not
reported,
and
the
quality
of
the
sewage
effluent
varied.
In
one
location
it
was
chlorinated
secondary
effluent
with
unreported
chlorine
doses,
unreported
contact
times
and
unreported
residual
concentrations
of
combined
and
free
chlorine
levels
in
the
effluent
when
discharged.
In
the
other
location,
the
effluent
differed
from
one
study
year
to
the
other.
Initially
it
was
undisinfected
effluent
from
two
"
full
retention"
lagoons
of
unreported
type
(
anaerobic,
facultative
or
aerobic),
retention
time
and
operating
conditions
(
e.
g.,
were
the
two
lagoons
operated
in
series
or
in
parallel?).
In
the
second
year
the
effluent
was
treated
in
a
lagoon
of
unreported
type
and
retention
time,
followed
by
aeration
(
of
unreported
duration
and
inadequate
process
description)
followed
by
chlorination
with
unreported
chlorine
dose,
unreported
contact
time
and
unreported
residual
levels
of
free
or
combined
chlorine
in
the
discharged
effluent.

The
data
for
the
two
geographic
locations
of
freshwater
beaches
had
considerable
variability
and
uncertainty.
For
example
in
the
1979
study
year
at
Lake
Erie,
the
indicator
densities
were
unexpectedly
low
at
both
beaches,
in
1980
they
were
high
and
occasionally
extremely
high,
and
in
1982
they
were
moderately
high
relative
to
those
observed
in
1979.

At
the
Lake
Erie
beaches,
the
1980
data
do
not
reflect
bacterial
indicator
densities
consistent
with
proximity
of
the
beach
to
the
fecal
waste
source.
In
particular,
the
E.
coli
densities
are
higher
at
the
beach
more
distant
from
the
pollution
source.
The
investigators
suggest
that
these
inconsistent
results
may
have
been
caused
by
heavy
rains
which
occurred
in
the
four
days
before
the
start
of
the
beach
study
trials.
In
a
four­
day
period,
8.15
inches
of
rain
was
measured,
which
caused
the
lake
elevation
to
rise
and
the
turbidity
to
increase
The
effect
of
these
unusual
events
on
the
swimmer
illness
rates
is
unknown.
These
observations
indicate
a
highly
variable
and
not
necessarily
representative
set
of
conditions
at
these
two
study
locations
that
call
into
question
their
national
representativeness
of
freshwater
beaches.

There
are
also
important
limitation
in
the
quality,
quantity
and
representativeness
of
the
bacteriological
data
for
water
quality.
At
each
geographic
location,
only
a
few
bacteriological
measurements
were
made
on
any
given
day
of
observations.
At
each
marine
beach,
samples
were
taken
at
2­
3
locations
in
chest
deep
water
(
location)
and
a
given
day
of
exposure,
with
only
a
few
(
3­
4)
samples
collected
between
the
hours
of
11
AM
and
5
PM
(
time
of
exposure).
The
actual
numbers
of
water
samples
taken
per
location,
site
and
study
day
were
not
specified
for
the
studies
at
the
freshwater
beaches.
However,
it
is
said
that
the
experimental
design
and
approach
was
similar
to
that
for
the
marine
beaches
studies.

It
is
unlikely
that
estimating
the
bacteriological
quality
of
water
based
on
a
relatively
small
number
of
samples
collected
only
in
chest­
deep
water
is
representative
of
the
exposure
of
all
bathers.
Children
and
many
other
people
never
venture
into
chest
deep
water
and
their
exposures
are
likely
to
be
better
represented
by
water
that
is
ankle­
deep,
knee­
deep
or
waist
deep.
Some
serious
swimmers
are
also
more
likely
to
be
exposed
to
water
beyond
the
chest­
deep
area.
In
some
subsequent
studies
on
bathing
water
quality
and
health
done
by
other
investigators,
water
samples
were
collected
from
a
grid
of
sample
sites
representing
different
depths
and
they
were
collected
are
more
frequent
intervals.
Such
sample
provides
better
estimates
of
the
bacteriological
quality
of
the
water
at
specific
locations
and
times
that
can
be
referenced
or
related
to
the
exposure
of
specific
bathers
who
bathed
at
specific
locations
for
specific
time
periods.

The
best
of
the
studies
that
related
exposure
to
recreational
bathing
water
of
measured
bacteriological
quality
to
human
health
effects
in
those
exposed
by
such
bathing
were
randomized
controlled
trials
in
which
subjected
were
recruited
and
randomly
assigned
to
a
bathing
group
or
a
non­
bathing
group.
Bathers
were
asked
to
spend
specific
times
in
the
water
(
10
minutes)
and
asked
to
immerse
their
heads
at
least
three
times
(
Kay
et
al.,
1994;
2001;
Fleisher
et
al.,
1998).
Exposure
measurements
of
bacteriological
quality
of
the
water
were
made
at
the
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
time
and
location
of
exposure
and
at
three
different
depths,
specifically
surf,
mid
and
chest.

Other
Source
of
Relevant
Data
Not
Used
in
this
Analysis
Since
these
few
early
studies
by
the
US
EPA,
there
have
been
numerous
studies
of
water
quality
at
bathing
beaches
in
relation
to
human
health
effects
in
swimmers.
The
US
EPA
should
have
used
the
marine
water
and
freshwater
data
from
these
other
studies
in
its
analyses
and
risk
assessment.
In
a
review
article
published
in
1998,
Preuss
reported
on
the
findings
of
4
additional
studies
conducted
at
freshwater
beaches
in
the
UK,
France
and
Canada
and
14
marine
water
beaches
in
countries
in
North
America,
Europe,
the
Middle
East,
Asia,
the
Western
Pacific
(
Australia
and
New
Zealand)
and
South
Africa.
More
recently,
Wade
et
al.
(
2003)
conducted
a
systematic
review
and
meta­
analysis
of
a
total
of
15
marine
water
studies
and
8
freshwater
studies
and
on
the
relationships
of
bacterial
water
quality
and
health,
besides
the
studies
done
by
the
US
EPA.
These
more
recent
studies
provide
a
more
diversified
and
more
representative
database
than
the
few
studies
used
by
the
US
EPA.
These
more
recent
analyses
reveal
wider
ranges
of
bacterial
densities
than
studied
by
the
US
EPA
and
document
different
doseresponse
relationships
for
bacterial
densities
and
human
health
effects
than
those
observed
by
the
US
EPA
in
its
studies.
In
the
study
by
Wade
et
al.
(
2003)
the
analyses
of
the
data
for
freshwater
beaches
showed
elevated
relative
risks
of
swimming­
associated
illness
both
above
and
below
the
US
EPA
guideline
value
for
enterococci.
For
E.
coli,
studies
below
the
US
EPA
guideline
value
were
not
associated
with
increased
illness,
while
exposures
above
the
US
EPA
guideline
value
were.
These
findings
suggest
that
the
current
US
EPA
guideline
values
for
the
two
different
indicators
provide
different
and
inconsistent
levels
of
bather
protection
from
swimming­
associated
illness
in
freshwater
beaches.
Therefore,
the
two
different
indicators
and
their
associated
guideline
values
are
not
interchangeable
in
terms
of
their
levels
of
protection.
Based
on
the
EPA's
analysis
of
its
own
data
for
freshwater
studies,
these
two
indicators
are
considered
interchangeable
and
give
equivalent
levels
of
protection.
For
the
analyses
of
the
marine
water
beaches,
Wade
et
al.
(
2003)
found
that
enterococci
were
the
indicator
that
most
strongly
predicted
increased
health
risks.
By
categorical
analysis,
the
relative
health
risks
did
not
continue
to
increase
in
studies
with
bacterial
densities
greater
than
104
cfu/
100
ml.
This
indicates
a
potential
threshold
effect
for
risk
of
GI
illness.
By
weighted
regression
analysis
there
was
an
association
between
enterococci
density
and
the
natural
log
of
the
relative
risk
of
health
effects.
The
relative
risk
for
GI
illness
increased
1.3
times
for
every
log10
increase
in
enterococci
density
in
water.
In
relation
to
the
current
EPA
guideline
for
enterococci
in
marine
waters,
summary
relative
risks
for
GI
illness
below
the
EPA
guideline
value
were
lower
and
not
statistically
significant.
Relative
risks
for
GI
illness
above
the
EPA
guideline
value
were
elevated
and
statistically
significant.

Analytical
Approach
and
Execution
The
approach
to
the
analyses
of
the
microbiological
data
is
limited
and
probably
flawed.
More
and
better
approaches
to
the
types
of
data
gathered
and
analyzed
and
the
type
of
analyses
of
the
data
should
have
been
conducted.
The
EPA
should
have
attempted
to
provide
better
estimates
of
the
concentrations
of
bacteria
in
the
water
to
which
people
were
exposed,
particularly
with
respect
to
the
spatial
and
temporal
relationships
of
the
exposures.
The
bacterial
data
for
each
location
were
analyzed
by
grouping
them
by
location
and
then
season
and
computing
a
geometric
mean
concentration
of
bacteria.
Apparently,
only
a
few
measures
of
bacteria
concentration
were
made
for
a
given
beach,
with
samples
taken
at
2­
3
locations
n
chest
deep
water
(
location)
and
a
given
day
of
exposure,
with
only
a
few
(
3­
4)
samples
collected
between
the
hours
of
11
AM
and
5
PM
(
time
of
exposure).
Geometric
mean
concentrations
were
computed
for
this
exposure
location
and
day
and
then
used
in
a
linear
regression
analysis
to
determine
relationships
between
illness
rates
and
average
water
quality.
The
resulting
linear
regression
model
was
then
used
to
derive
standard
deviations
for
the
doseresponse
relationship.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
The
approach
used
by
the
US
EPA
in
which
all
data
were
combined
for
the
development
of
the
distribution
of
bacterial
concentrations
that
was
then
used
for
comparisons
with
human
health
effects
has
serious
drawbacks
and
limitations.
This
is
because
the
standard
deviation
of
the
probability
density
function
or
distribution
affects
the
probability
of
exposure
to
polluted
water
and
thus
the
risk
of
illness.
The
use
of
a
single
distribution
and
resulting
parameter
value
to
define
a
guideline
value
across
all
waters
does
not
adequately
address
local
variability.
Local
variations
in
standard
deviation
will
mean
that
risks
of
illness
will
vary
even
though
the
same
guideline
value
is
in
place.
Simply
stated,
combining
the
data
into
a
single
distribution
does
not
address
local
variability
in
the
distribution
of
bacteria
density
on
a
site­
specific
basis
and
the
relationships
of
these
bacterial
densities
to
local
heath
risks
In
addition,
it
is
not
clear
the
US
EPA
used
robust
criteria
to
determine
if
bacterial
concentrations
could
be
legitimately
log10
transformed
to
create
the
log10
distribution
that
was
used
in
the
dose­
response
analyses.
Statistical
tests
should
have
been
done
to
test
for
normality
or
to
determine
if
the
hypothesis
of
normality
can
be
rejected.
Such
analyses
should
have
been
done
for
the
data
from
individual
beaches
and
study
sites
as
well
as
on
the
combined
data.
The
need
to
test
for
the
normality
of
the
data
on
a
site­
specific
basis
is
because
of
the
need
to
link
bacterial
densities
with
specific
exposures
that
result
in
health
effects.

Apparently,
the
US
EPA
did
not
consider
alternative
analytical
strategies
for
these
data
that
do
not
rely
on
the
use
of
the
data
taken
directly
from
a
log10­
transformed
distribution.
One
possible
alternative
approach
in
situations
where
normality
(
or
log­
normality)
is
violated
is
to
use
a
Bootstrapping
procedure.
In
this
case
a
Monte
Carlo
style
procedure
is
applied
to
bootstrapped
samples
of
the
actual
empirical
distributions
of
bacterial
concentration,
rather
than
the
parametrically
generated
distribution.
The
bootstrapping
procedure
draws
a
large
number
(
say
1000)
of
"
resamples",
of
size
equal
to
the
original
sample,
from
this
original
sample
randomly
with
replacement.
No
such
alternative
analytical
approach
was
attempted
by
the
use
EPA.

The
basis
for
defining
the
di
fferent
risk
levels
as
upper
percentiles
(
e.
g.,
75
th,
82
nd,
90
th
and
95th)
is
poorly
documented
and
justified
and
the
basis
for
focusing
only
the
human
health
risks
in
the
range
of
0.8
to
1%
also
is
poorly
documented
and
inadequately
explained
or
justified.
These
standard
deviation
estimates
were
used
to
consider
various
upper
percentiles
(
75
th,
82
nd,
90
th
and
95%)
for
which
were
calculate
various
upper
percentiles
values
of
allowable
bacterial
density
per
100
ml
corresponding
to
different
health
risk
levels
in
the
range
of
0.8
to
1%.
The
flaws
and
weakness
of
this
approach
are
that
the
data
were
transformed
to
log10
values
and
these
log10
transformed
values
were
used
in
the
analyses
and
the
analysis
of
the
logtransformed
data
used
only
a
simple
linear
regression
model
to
examine
the
relationship
between
indicator
density
and
human
health
risk
of
bathers.
No
other
forms
of
the
data
were
used
for
analyses
(
such
as
arithmetic
forms
of
the
data
with
no
transformations)
and
no
other
models
were
applied
to
the
data
(
such
as
Beta­
Poisson,
two­
population
or
other
dose­
response
models
now
widely
used
for
quantitative
microbial
risk
assessment).

In
the
study
of
Kay
et
al.
(
1994)
an
epidemiological
relationship
described
the
excess
risk
of
illness
from
exposure
to
water
containing
fecal
indicator
bacteria.
This
relationship
was
best
as
a
dose­
response
relationship
linking
water
quality
exposure
(
x),
indexed
by
the
fecal
streptococci
density
at
chest
depth
water,
and
the
excess
probability
of
gastroenteritis
(
y)
is
given
by
the
following
(
for
exposures
between
32
and
158
fecal
streptococci
per
100ml):
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
where,
m
is
the
natural
logarithm
of
the
odds
of
getting
gastroenteritis
from
bathing,
derived
from
the
logistic
regression
equation:

and
the
term
p32
is
the
probability
of
gastroenteritis
where
x
=
32
cfu
per
100ml
(
p=
0.0866)
and
adjusts
the
relationship
to
reflect
excess
rather
than
absolute
probability
of
illness
relative
to
those
who
do
not
bathe.

In
addition
to
the
consideration
of
the
dose­
response
model
used,
there
are
valid
reasons
to
believe
that
the
form
of
microbiological
data
used,
specifically
the
log10Ctransformed
data
of
the
US
EPA,
is
likely
to
underestimate
exposure
in
dose­
response
analyses
and
that
the
liner
regression
model
applied
to
these
data
is
likely
to
provide
a
downward
extrapolation
that
is
a
less
reliable
portrayal
of
the
actual
dose­
response
relationship.
This
is
not
always
the
case,
as
was
found
in
the
studies
by
Kay
et
al.,
where
the
bacteriological
data
were
best
described
by
a
log­
normal
probability
density
function.
However,
there
are
scientifically
valid
reasons
to
use
arithmetic
data
for
dose­
response
analyses
and
determine
if
this
from
of
the
data
better
describes
the
extremes
of
exposure
and
resulting
health
effects
at
both
the
low
and
high
ends
of
the
distribution
of
bacteria
concentration.

Overall,
there
are
serious
concerns
about
the
extent
to
which
reliable
downward
extrapolations
can
be
made
from
the
US
EPA
data,
given
all
of
the
potential
sources
of
bias,
the
use
of
loh10­
transformed
data
rather
than
arithmetic
data,
the
use
of
only
a
simple
log­
linear
regression
model,
and
the
rather
shallow
slope
of
the
dose­
response
relationship
in
the
human
health
effects
response
range
of
interest
(
corresponding
to
0.8
to
1.0%
risk).

In
addition,
there
was
very
inadequate
treatment
of
variability
and
uncertainty
in
the
analyses
for
either
bacteriological
quality
of
the
water,
the
extent
of
bather
exposure
and
the
temporal
and
spatial
relationships
between
exposure
and
resulting
health
effects
in
the
exposed.
These
deficiencies
make
it
inappropriate
to
attempt
to
do
downward
extrapolations
of
the
water
quality
(
bacterial
concentration)­
health
effects
)
dose­
response)
relationships
in
the
data
range
of
interest
for
freshwaters
or
in
general.
Developing
water
quality
criteria
and
regulatory
guidelines
based
on
such
analyses
can
not
be
supported
or
justified
in
the
opinion
of
this
reviewer.

References
Cited
Fleisher
JM,
Kay
D,
Wyer
MD,
A.
F.
Godfree
Estimates
of
the
severity
of
illnesses
associated
with
bathing
in
marine
recreational
waters
contaminated
with
domestic
sewage.
Int
J
Epidemiol.
1998
Aug;
27(
4):
722­
6.
Kay
D,
Fleisher
JM,
Salmon
RL,
Jones
F,
Wyer
MD,
Godfree
AF,
Zelenauch­
Jacquotte
Z,
R.
Shore
(
1994)
Predicting
likelihood
of
gastroenteritis
from
sea
bathing:
results
from
randomised
exposure.
Lancet.
1994
Oct
1;
344(
8927):
905­
9.
Kay,
D.,
J.
Fleisher,
M.
D.
Wyer,
and
R.
I.
Salmon
(
2001)
Re­
analysis
of
the
Seabathing
Data
from
the
UK
Randomised
Trials.
A
Report
to
DETR.
Aberystwyth,
University
of
Wales,
Centre
for
Research
into
the
Environment
and
health,
17
pages.
Pruss,
A.
(
1998)
Review
of
epidemiological
studies
on
health
effects
from
exposure
to
recreational
water.
International
Journal
of
Epidemiology
27(
1):
1­
9.
Stavros,
G.
and
I.
H.
Langford
(
2002)
Coastal
Bathing
Water
Quality
and
Human
Health
Risks:
A
Review
of
Legislation,
Policy
and
Epidemiology,
with
an
Assessment
of
Current
UK
Water
Quality,
Proposed
Standards,
and
Disease
Burden
in
England
and
Wales.
CSERGE
Working
Paper
ECM
02­
06.
Centre
for
Social
and
Economic
Research
on
the
Global
Environment,
University
of
East
Anglia,
UK.
External
Peer
Review
of
EPA
Analysis
of
Epidemiological
Data
from
EPA
Bacteriological
Studies
2/
04
Wade
TJ,
Pai
N,
Eisenberg
JN,
Colford
JM
Jr.
(
2003)
Do
U.
S.
Environmental
Protection
Agency
water
quality
guidelines
for
recreational
waters
prevent
gastrointestinal
illness?
A
systematic
review
and
meta­
analysis.
Environ
Health
Perspect.
2003
Jun;
111(
8):
1102­
9.
WHO
(
2001)
Bathing
Water
Quality
and
Human
Health:
Faecal
Pollution.
Outcome
of
an
Expert
Consultation,
Farnham,
UK,
April
2001.
Co­
sponsored
by
Department
of
the
Environment,
Transport
and
the
Regions,
United
Kingdom.
WHO/
SDE/
WSH/
01.2.
Geneva:
World
Health
Organization.
