ESTIMATING
POTENCY
SCORES:
AN
EXERCISE
Joyce
Morrissey
Donohue
Ph.
D.,
R.
D.
US
EPA,
Office
of
Water,
Office
of
Science
and
Technology,
Health
and
Ecological
Criteria
Division
8/
26/
04
i
TABLE
OF
CONTENTS
ABSTRACT
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
BACKGROUND
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
1
INTRODUCTION
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
2
PROCEDURE
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
3
Figure
1.
Decile
Distribution
of
RfD
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4
Figure
2.
Logarithmic
Distribution
of
RfD
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
Table
1.
Scoring
Equations
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
6
Table
2.
Potency
Scores
for
Chemicals
in
the
Learning
Set
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7
Table
3.
Potency
Scores
for
Chemicals
Not
in
the
Learning
Set
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8
RESULTS
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8
APPENDIX
I
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A1
Figure
3.
Logarithmic
Distribution
of
NOAEL
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A2
Figure
4.
Logarithmic
Distribution
of
LOAEL
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A3
Figure
5.
Logarithmic
Distribution
of
LD50
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A4
Figure
6.
Logarithmic
Distribution
of
Cancer
Potency
Values
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A5
1
ESTIMATING
POTENCY
SCORES:
AN
EXERCISE.

ABSTRACT
The
Office
of
Water
(
OW)
has
developed
and
tested
a
potential
approach
to
score
or
rank
the
cancer
and
non­
cancer
potencies
of
potential
drinking
water
contaminants.
This
approach
uses
measured
or
modeled
values
of
potency
[
Reference
Dose
(
RfD),
No­
Observed­
Adverse­
Effect­
Level
(
NOAEL),
Lowest­
Observed­
Adverse­
Effect­
Level
(
LOAEL),
LD
50
and
10
­
4
(
one
in
ten
thousand)
estimate
of
cancer
risk].
The
scoring
for
each
of
these
parameters
was
calibrated
based
on
data
from
a
set
of
200
known
drinking
water
contaminants
and
chemicals
generally
recognized
as
safe.
The
distribution
of
the
RfD,
NOAEL,
LOAEL,
LD
50,
and
10
­
4­
risk
values
for
this
set
of
chemicals
was
similar
to
a
normal
distribution
curve
when
log
transformed
(
i.
e.,
arranged
by
powers
of
10).
The
distribution
was
used
to
develop
a
scoring
equation
for
each
potency
parameter
by
equating
a
value
of
five
to
the
"
most
frequent"
power
of
ten
in
the
distribution.
Scores
were
fairly
consistent
across
related
parameters
for
individual
chemicals.
Agreement
across
scores
was
weakest
for
the
weakly
toxic
chemicals.
[
The
opinions
expressed
represent
those
of
the
author
and
do
not
necessarily
reflect
the
opinions
of
the
U.
S.
Environmental
Protection
Agency.]

1.0
BACKGROUND
The
1996
Safe
Drinking
Water
Act
requires
EPA
to
develop
a
list
of
contaminants
with
the
potential
for
regulation
on
a
five­
year
cycle.
This
list,
known
as
the
Contaminant
Candidate
List
(
CCL),
serves
as
the
primary
source
of
the
priority
contaminants
for
the
EPA's
drinking
water
program.
The
CCL
also
identifies
the
group
of
contaminants
from
which
the
Agency
will
make
regulatory
determinations
three
and
one­
half
years
after
publication
of
the
list.

After
publication
of
the
first
CCL,
EPA
requested
assistance
from
the
National
Research
Council
(
NRC)
of
the
National
Academy
of
Sciences
in
providing
suggestions
for
a
comprehensive,
transparent
process
to
prioritize
and
select
contaminants
for
inclusion
on
the
CCL.
The
NRC
recommendations
were
published
in
2001.
They
provide
a
framework
for
the
process
in
which
potential
contaminants
are
initially
selected
from
a
large
universe
of
contaminants
of
concern
with
a
preliminary
screen
based
on
occurrence
and
health
effects
information.
The
CCL
is
chosen
from
these
screened
contaminants
considering
scored
attributes
for
health
effects
(
potency
and
severity)
and
occurrence
(
prevalence,
magnitude,
persistence,
and
mobility).
The
data
requirements
for
the
selection
of
the
CCL
provide
a
higher
degree
of
scrutiny
than
those
used
for
the
preliminary
screen.
This
exercise
uses
the
NRC
recommended
framework
to
evaluate
and
score
the
relative
potency
of
contaminants
based
on
toxicity
parameters.

2.0
INTRODUCTION
1
An
RfD
is
an
estimate
(
with
uncertainty
spanning
perhaps
one
order
of
magnitude)
of
a
daily
exposure
to
humans
(
including
sensitive
individuals)
that
is
likely
to
be
without
appreciable
risk
of
deleterious
effects
during
a
lifetime.
It
is
expressed
in
mg/
kg/
day.
ATSDR
lifetime
MRLs,
WHO
TDIs,
WHO
and
FDA
ADIs,
and
nutrient
Tolerable
Upper
Limit
Values
(
ULs)
are
roughly
equivalent
to
RfDs.

2
For
this
exercise
cancer
potency
was
evaluated
as
the
concentration
in
drinking
water
equivalent
to
an
excess
cancer
risk
of
one
case
in
10,000
(
10­
4).
This
value
is
given
in
the
OW
Drinking
Water
Standards
and
Health
Advisory
Tables
and
also
is
included
in
all
IRIS
Summary
documents.
When
the
10­
4
risk
value
is
not
available,
it
can
be
calculated
from
a
cancer
slope
factor.

2
Potency
is
a
value
that
indicates
the
power
of
a
contaminant
to
cause
adverse
health
effects.
In
the
case
of
chemicals,
that
power
is
apparent
in
the
dose
required
to
cause
the
most
sensitive
manifestation
of
an
adverse
health
effect,
or
to
generate
a
particular
excess
cancer
risk.
Potency
for
chemicals
is
reflected
in
several
standard
toxicological
parameters
which
are
discussed
below.

There
are
a
number
of
approaches
that
have
the
potential
to
be
useful
in
scoring
potency.
However,
no
matter
which
approach
is
selected,
the
methods
will
require
calibrating
the
scores
to
normalize
the
scale.
There
must
be
some
agreement
about
the
types
of
data
that
will
be
used
in
the
scoring
algorithm
and
the
magnitude
of
the
potency
value
that
represents
a
given
number
on
the
potency
scale.
With
this
in
mind,
an
initial
"
learning
set"
of
about
two
hundred
chemicals
was
developed
in
order
to
make
a
preliminary
assessment
of
one
calibration
approach.
The
chemicals
considered
included:

°
Regulated
chemicals,
°
Unregulated
chemicals
with
lifetime
health
advisories,
and
°
A
small
group
of
chemicals
(
nutrients/
food
additives)
which
are
generally
considered
as
relatively
non­
toxic
and
which
have
toxicity
values
similar
to
lifetime
health
advisories.

The
following
toxicity
parameters
were
collected
for
the
learning
set
chemicals,
and
their
numeric
distribution
across
the
range
of
values
were
examined
(
see
the
footnotes
below
for
definitions
of
the
terms).

°
RfD
1
or
equivalent
°
Cancer
potency2
(
10­
4
risk
concentration
in
water)
3
NOAEL
is
a
No
Observed
Adverse
Effect
Level.
It
is
the
highest
dose
in
a
toxicological
study
or
a
group
of
studies
that
has
no
observed
adverse
effect.

4
LOAEL
is
a
Lowest
Observed
Adverse
Effect
Level.
It
is
the
lowest
dose
in
a
toxicological
study
or
a
group
of
studies
that
causes
an
adverse
health
effect.

5
A
oral
LD
50
is
an
estimate
of
the
oral
dose
that
will
cause
the
death
of
50%
of
the
exposed
animals.
LD
50
data
are
based
on
acute
exposures
with
limited
post­
exposure
observations
of
the
animals
for
mortality,
clinical
signs,
and
gross
pathology.

6
The
2002
Edition
of
the
Drinking
Water
Regulations
and
Health
Advisories
was
used
for
the
RfD
and
10­
4
risk
values.

3
°
NOAEL3
and/
or
LOAEL4
associated
with
the
RfD
°
Rat
oral
LD
50
5
values.

Several
approaches
to
characterize
the
distribution
of
values
for
the
different
toxicity
parameters
were
employed
in
this
exercise.
The
approaches
are
described
in
the
Procedure
section
below.

The
toxicity
parameters
listed
above
were
selected
because
they
were
readily
available
on
the
Integrated
Risk
Information
System
(
IRIS),
in
OW
health
advisory
documents,
or
in
the
Institute
of
Medicine
(
IOM)
Dietary
Reference
Intake
(
DRI)
documents
for
nutrients.
They
also
were
selected
because
they
are
the
values
that
are
most
likely
to
be
found
in
readily
accessible
sources
such
as
IRIS,
or
to
be
amenable
to
evaluation
using
Quantitative
Structure
Activity
Relationship
(
QSAR)
programs.

3.0
PROCEDURE
The
data
for
the
learning
set
were
obtained
from
the
following
sources:

°
Integrated
Risk
Information
System
(
IRIS)
°
Office
of
Water
Health
Advisory
Documents6
°
Registry
of
Toxic
Effects
of
Chemical
Substances
(
Mostly
LD
50
values)
°
Tolerable
Upper
Levels
from
the
Institute
of
Medicine
Dietary
Reference
Intakes.

Once
the
data
had
been
collected,
they
were
graphically
distributed
across
the
range
in
two
ways.
For
the
initial
evaluation,
the
range
was
divided
into
approximately
ten
equal
units
(
deciles).
This
distribution
was
found
to
be
highly
skewed
with
a
large
majority
of
the
values
falling
in
the
decile
of
highest
toxicity
(
see
Figure
1
for
an
example).
Two
factors
influenced
this
situation.
The
first
factor
is
the
fact
that
the
range
of
values
was
up
to
twelve
orders
of
magnitude
for
the
parameters
evaluated.
The
second
factor
is
the
fact
that
the
set
of
contaminants
contained
both
toxic
chemicals
as
well
as
those
generally
regarded
as
safe.
In
general,
there
are
far
more
toxicological
data
available
in
the
literature
on
chemicals
considered
to
be
toxic
than
for
4
0
20
40
60
80
100
120
140
160
0
<=
0.1
>
0.1­
0.2
>
0.2­
0.3
>
0.3­
0.4
>
0.4­
0.5
>
0.5­
0.6
>
0.6­
0.7
>
0.7­
0.8
>
0.8­
0.9
>
0.9
RfD
Frequency
those,
like
the
nutrients,
that
are
only
weakly
toxic.
It
is
difficult
to
identify
standard
toxicity
values
for
most
chemicals
that
are
generally
regarded
as
safe.
As
a
result,
the
second
distribution
evaluated
was
based
on
logarithms
(
base
10)
of
the
toxicity
parameters
rounded
to
the
nearest
integer.
This
provided
a
far
better
distribution
of
toxicity
values
across
the
range
(
see
Figure
2
as
an
example)
and
was
used
for
subsequent
calibration
of
the
data
for
scoring.

Figure
1.
Decile
Distribution
of
RfD
Values
5
0
10
20
30
40
50
60
70
80
<=
­
6
­
5
­
4
­
3
­
2
­
1
0
1
2
3
4
More
Round(
Log10(
RfD))
Frequency
Figure
2.
Logarithmic
Distribution
of
RfD
Values
6
The
distribution
that
provided
the
best
spread
of
the
scores
(
the
log­
based
distribution)
was
used
to
establish
a
scoring
equation
for
potency
for
each
measure
of
toxicity.
This
was
accomplished
by
assigning
the
most
frequent
(
modal)
value
in
the
distribution
a
score
of
5
on
a
10
point
scale
and
solving
the
equation
for
the
variable
that
would
make
that
distributional
value
equal
a
score
of
5.
For
example,
in
Figure
2,
the
most
frequent
value
is
a
rounded
logarithm
of
­
2
(
0.01).
The
scoring
equation
for
the
RfD
values
was
the
developed
as
follows:

5
=
10
­
(
most
frequent
rounded
log
+
X)
5
=
10
­
(­
2
+
X)
5
=
10
+
2
­
X
5
=
12
­
X
5
­
12
=
­
X
­
7
=
­
X
7
=
X
Accordingly
the
equation
for
scoring
the
RfD
values
is
Score
=
10
­
(
rounded
log
of
RfD
+
7)

The
scoring
equations
for
the
other
measures
of
toxicity
were
derived
from
the
modal
rounded
logarithm
values
of
their
distributions
in
a
similar
fashion.
The
resultant
equations
are
summarized
in
Table
1.
The
distributions
of
the
rounded
logarithms
are
attached
to
this
report
as
an
Appendix.

Table
1.
Scoring
Equations
RfD
Score
=
10
­
(
Log10
of
RfD
+
7)

NOAEL
=
10
­
(
Log10
of
NOAEL
+
4)

LOAEL
=
10
­
(
Log10
of
LOAEL
+
4)

LD50
=
10
­
(
Log10
of
LD50
+
2)

10­
4
cancer
risk
1
=
10
­
(
Log
10
of
the
10­
4
cancer
risk
+
6)

1.
The
10­
4
cancer
risk
in
water
was
selected
as
the
measure
of
potency
for
carcinogens
because
this
is
the
value
given
in
the
Drinking
Water
Regulations
Health
Advisory
Tables
prepared
by
OW
and
also
is
provided
in
IRIS
Summaries.

Scores
were
restricted
to
whole
number
values
with
a
maximum
of
10
and
a
minimum
of
1
7
All
of
the
chemicals
in
the
learning
set
were
scored
for
each
toxicity
parameter
to
examine
the
consistency
across
scores
for
the
non­
cancer
measures
of
potency.
Some
examples
of
this
evaluation
are
provided
in
Table
2.
Since
the
mechanisms
that
lead
to
the
development
of
cancer
involve
some
biological
responses
that
are
unique
to
tumors,
the
10­
4
cancer
risk
values
were
not
included
in
this
comparison.
The
scores
for
individual
chemicals
were
compared
across
the
toxicity
values,
and
the
agreement
between
scores
was
evaluated.

Table
2.
Potency
Scores
for
Chemicals
in
the
Learning
Set
Chemical
RfD
NOAEL
LOAEL
LD50
Arsenic
(
Sodium
Arsenite
for
LD
50)
7
9
8
6
Calcium
(
Calcium
chloride
for
LD
50)
1
ND
4
5
Cyanazine
6
6
6
6
Dioxin
(
2,3,7,8­
TCDD)
10
ND
10
4
Heptachlor
8
ND
8
7
Hexazinone
4
5
4
5
Iodine
(
Sodium
iodide
for
LD
50)
5
8
8
4
Methyl
Ethyl
Ketone
3
3
3
5
MethylParathion
7
8
7
7
Naphthalene
5
4
4
5
Paraquat
5
6
6
6
Phenol
4
4
4
5
Phosphorous
(
Sodium
Hydrogen
Phosphate
for
LD
50)
1
4
ND
4
Vitamin
D
6
9
9
ND
ND
=
No
data
In
addition,
the
scoring
equations
were
applied
to
chemicals
that
were
not
in
the
learning
set
using
data
available
in
Agency
of
Toxic
Substance
and
Disease
Registry
(
ATSDR)
Toxicological
Profiles.
Those
results
are
summarized
in
Table
3.
Scores
were
evaluated
for
consistency
across
parameters.
8
Table
3.
Potency
Scores
for
Chemicals
Not
in
the
Learning
Set
Chemical/
Potency
Scores
RfD­
equivalent
(
mg/
kg/
day)
NOAEL
(
mg/
kg/
day)
LOAEL
(
mg/
kg/
day)
LD50
(
mg/
kg)

Acrylonitrile
0.04
4.2
14
93
Scores
4
5
5
6
Ethion
0.0004
0.06
0.71
27
Scored
6
7
6
6
Malathion
0.02
2
29
925
Scores
5
6
5
5
Endofulfan
0.002
0.18
ND
1740
Scores
6
7
ND
5
ND
=
No
Data
4.0
RESULTS
As
mentioned
in
the
Procedure
section
above,
the
decile
distribution
was
found
to
be
undesirable
in
developing
an
equation
for
scoring
potency
because
almost
all
of
the
chemicals
are
clustered
at
one
end
of
the
distribution.
This
does
not
provide
good
distributions
of
scores.
With
the
decile
distribution,
almost
all
of
the
chemicals
in
the
learning
set
would
have
a
high
potency
score
of
10.
Very
few
chemicals
would
have
lower
scores.
The
distribution
based
on
the
rounded
Log10
of
the
toxicity
parameter
was
found
to
be
far
superior
to
the
decile
distribution
because
toxicity
parameters
for
different
chemicals
are
spread
out
across
the
range
and
the
most
frequent
Log10
is
approximately
in
the
middle
of
the
range
making
the
curve
roughly
normal.
It
was
for
this
reason
that
the
Log10
distribution
was
selected
for
development
of
the
scoring
equation.
The
distribution
of
toxicity
values
is
still
somewhat
skewed
toward
higher
toxicity
scores;
however,
this
is
a
product
of
a
limited
number
of
weakly
toxic
chemicals
that
were
included
in
the
learning
set.

Some
distributions
for
toxicity
parameters
span
a
range
greater
than
ten
orders
of
magnitude
(
see
Figure
2).
It
was
decided
that
in
cases
where
the
range
of
toxicity
parameters
was
greater
then
ten,
a
calculated
score
less
than
1
would
be
given
a
score
of
1
and
anything
with
a
calculated
score
greater
than
10
would
be
given
a
score
of
10.
This
is
because
these
chemicals
are
at
the
tails
of
the
distributions.
Conversely,
for
the
distributions
that
covered
less
than
10
orders
of
magnitude
(
see
the
LD50
distribution
in
the
Appendix),
no
attempt
was
made
to
normalize
the
scores
across
a
range
of
ten
because
the
learning
set
is
limited
and
could
have
been
expanded
by
searching
for
chemicals
that
are
more
toxic
than
dioxin
(
the
most
toxic
substance
in
9
the
learning
set
with
an
RfD
of
1
x
10­
9
mg/
kg/
day)
and
less
toxic
than
phosphorous
(
the
least
toxic
chemical
in
the
learning
set
with
an
RfD­
equivalent
of
57
mg/
kg/
day
derived
from
the
IOM
UL).

The
distribution
for
cancer
is
the
most
skewed
of
those
examined
because
there
are
a
greater
number
of
chemicals
that
are
more
potent
carcinogens
than
those
in
the
modal
grouping
than
there
are
those
that
are
less
potent.
This
in
not
unusual
when
one
takes
into
account
the
fact
that
cancer
bioassays
are
very
costly
and
there
is
a
greater
incentive
to
invest
resources
in
studying
chemicals
that
have
a
high
likelihood
of
being
potent
carcinogens
than
those
that
are
not.
No
attempt
was
made
to
normalize
the
cancer
scores
across
a
range
of
10.
For
the
chemicals
in
the
learning
set,
the
lowest
cancer
potency
score
is
a
3.

When
the
agreement
of
non­
cancer
scores
across
the
RfD,
NOAEL,
LOAEL
and
LD
50
inputs
was
evaluated,
the
scores
for
any
given
compound
were
found
to
be
generally
consistent
across
parameters.
There
were
216
chemicals
in
the
learning
set;
13.5%
of
those
with
mutiple
noncancer
scores
had
identical
scores
across
all
parameters
(
see
cyanazine
in
Table
2).
For
54.6%,
the
scores
deviated
by
1
integer
(
see
hexazinone
in
Table
2);
20.5%
deviated
by
2
integers
(
see
methyl
ethyl
ketone
in
Table
2).
There
was
a
3­
integer
deviation
for
9.7%,
and
the
majority
of
those
were
inorganic
compounds
(
see
Arsenic
[
sodium
arsenate]
and
iodine
[
sodium
iodide]
in
Table
2).
Only
1.6%
deviated
by
more
than
3
integers
(
see
dioxin
in
Table
2).
Scores
deviated
by
two
integers
or
less
for
88.6
%
of
the
chemicals.
The
differences
between
scores
for
a
given
compound
was
greatest
for
the
relatively
non­
toxic
chemicals
(
see
phosphorous/
sodium
hydrogen
phosphate
in
Table
2).
In
almost
all
cases
the
NOAEL
and
LOAEL
scores
were
higher
than
the
RfD
score,
effectively
negating
the
claim
that
the
inclusion
of
uncertainty
factors
in
the
calculation
of
the
RfD
would
inflate
the
potency
score.
For
those
chemicals
with
low
uncertainty
factors
the
NOAEL
or
LOAEL
scores
were
often
3
or
more
integers
higher
than
the
RfD
scores
(
see
calcium
chloride
and
vitamin
D
in
Table
2).

Since
most
chemicals
with
RfD
values
also
are
likely
to
have
NOAEL,
LOAEL,
and/
or
LD
50
values,
a
policy
decision
is
needed
with
regard
to
how
one
should
select
the
parameter
used
to
score
for
a
noncancer
endpoint.
There
are
a
number
of
options.
One
could
select
the
lowest
or
highest
of
a
suite
of
scores,
or
develop
a
hierarchy
for
choosing
between
RfD,
NOAEL,
and
LOAEL.
It
was
been
suggested
by
those
who
worked
on
this
project
that
the
LD
50
only
be
used
when
NOAEL
and/
or
LOAEL
values
are
not
available
since
it
is
a
measure
of
acute
rather
than
chronic
toxicity.
When
comparing
cancer
and
noncancer
scores,
the
work
group
suggested
that
whichever
one
provides
the
highest
measure
of
potency
be
selected.

The
scoring
equations
were
easily
applied
to
chemicals
not
in
the
learning
set.
The
calculated
scores
were
reasonable
based
on
the
toxicity
of
the
chemical
(
Table
3)
and
provide
support
for
the
proposed
approach
to
scoring
potency.
However,
the
purpose
for
this
exercise
was
merely
to
explore
a
method
for
scoring
potency
by
utilizing
several
standard
toxicological
parameters
that
are
utilized
for
quantitative
risk
assessment
of
chemicals.
Possibilities
exist
for
further
expanding
the
exercise
by:
10
°
Expanding
the
learning
set
°
Examining
other
approaches
to
using
distributions
of
the
learning
set
data
to
calibrate
scoring
and
comparing
the
results.
°
Doing
additional
testing
of
the
calibrated
scoring
methodology
with
chemicals
that
are
not
in
the
learning
set.

This
exercise
did
not
consider
the
severity
of
the
cancer
or
noncancer
effect
identified
by
the
RfD,
LOAEL,
or
cancer
risk.
OW
is
examining
possible
methods
for
scoring
severity
as
part
of
a
separate
exercise.

ACKNOWLEDGMENT:
Greg
Blumenthal
and
Katherine
Sullivan
of
ICF
Consulting
Inc,
Fairfax
VA
provided
technical
support
for
this
project
in
collecting
and
processing
the
data.
APPENDIX
I
DISTRIBUTIONS
OF
THE
ROUNDED
LOGARITHMS
A2
0
5
10
15
20
25
30
35
40
45
­
5
­
4
­
3
­
2
­
1
0
1
2
3
4
5
More
Round(
Log10(
NOAEL))

Frequency
Figure
3.
Logarithmic
Distribution
of
NOAEL
Values
A3
0
5
10
15
20
25
30
35
40
45
<=
­
5
­
4
­
3
­
2
­
1
0
1
2
3
4
5
More
Round(
Log10(
LOAEL))

Frequency
Figure
4.
Logarithmic
Distribution
of
LOAEL
Values
A4
Figure
5.
Logarithmic
Distribution
of
LD50
Values
A5
0
10
20
30
40
50
60
70
80
­
2
­
1
0
1
2
3
4
5
6
7
8
More
Round(
Log10(
LD50))

Frequency
A6
0
5
10
15
20
25
<=­
6
­
5
­
4
­
3
­
2
­
1
0
1
2
3
4
More
Round(
Log10(
E­
4))

Frequency
Figure
6.
Logarithmic
Distribution
of
Cancer
Potency
Values
