\ I/
United States
Environmental Protection
Agency
Office of Health and
Environmental Assessment
Washington DC 20460
Research and Development
EPA/600/S6-87/007 Mar. 1988
&EPA Project Summary
Investigation of Cancer Risk
Assessment Methods
Bruce C. Allen, Annette M. Shipp, Kenny S. Crump, Bryan Kilian,
Mary Lee Hogg, Joe Tudor, and Barbara Keller
The major focus of this study is
making quantitative comparisons of
carcinogenic potency in animals and
humans for 23 chemicals for which
suitable animal and human data exist.
These comparisons are based upon
estimates of risk-related doses (RRDs)
obtained from both animal and human
data. An RRD represents the average
daily dose per body weight of a chem-
ical that would result in an extra cancer
risk of 0.25. Animal data on these and
21 other chemicals of interest to the
U.S. Environmental Protection Agency
(EPA) and the Department of Defense
(DOD) are coded into an animal data
base that permits evaluation using
different risk assessment approaches.
The full report is the result of a two-
year study to examine the assumptions,
other than those involving low-dose
extrapolation, used in quantitative
cancer risk assessment. The study was
funded by the DOD (through an inter-
agency transfer of funds to the EPA),
the EPA, the Electric Power Research
Institute and, in its latter stages, by the
Risk Science Institute.
This Project Summary was devel-
oped by EPA's Office of Health and
Environmental Assessment, Washing-
ton, DC, to announce key findings of
the research project that is fully doc-
umented in four separate volumes of
the same title (see Project Report
ordering information at back).
Introduction
The full report is the result of a two-
year study to examine the assumptions,
other than those" involving low-dose
extrapolation, used in quantitative
cancer risk assessment. The objectives
of the study are:
1. To identify and express quantita-
tively uncertainties that are
involved in the process of risk
estimation, excluding the uncer-
tainties in the low-dose extrapola-
tion model;
2. To examine the impact of the
different assumptions that are
made in risk estimation;
3. To compare results calculated from
human and animal data, including
the identification of the assump-
tions that produce the best corre-
lation of risk estimates between
humans and animals; and
4. To develop guidelines for present-
ing a range of risk estimates based
on different but scientifically
acceptable assumptions or as-
sumptions that have considerable
backing in the scientific
community.
These objectives are pursued using
empirical methods in which carcinogen-
icity data for 44 chemicals are analyzed
systematically in a variety of ways.
Particular attention is placed on those 23
chemicals for which there exist data from
both animal and human studies suitable
for making quantitative comparisons.
Table 1 lists components of a quan-
titative risk assessment based upon
animal data. Each component requires
a decision on the part of the risk assessor
for which there is no unique "correct"
choice. Also listed in Table 1 are various
-------
possible approaches to each component.
The choices that a risk assessor makes
for these components affect the resulting
estimates of risk. The choices for these
components therefore are related to the
uncertainty in assessment of risk from
animal data.
Objective 2 is pursued by making
different risk estimates for the 44
chemicals in the study by systematically
varying the approaches to the compo-
nents listed in Table 1. Examination of
the distributions of the changes in the
estimates associated with different
approaches to the various components
permits the examination of the impact
of the various approaches (assumptions).
These distributions also relate to the
uncertainties in the process of risk
estimation, so this work also applies to
Objective 1.
A major part of the study involves
making comparisons between risk esti-
mates derived from animal data and
those derived from human data for those
23 chemicals for which suitable data
exist for both animals and humans. This
work addresses the question of whether
correlations exist between animal and
human data, and therefore is of funda-
mental importance to the scientific
validity of quantitative risk assessment.
The practice of making quantitative
estimates of human riskfrom animal data
is based upon the hypothesis (heretofore
essentially untested) that such correla-
tions do in fact exist. If quantitative
correlations can be shown to exist, then
these correlations can provide a stronger
scientific basis for risk assessment.
Further, evaluation of the correlations
and determination of these approaches
to the components listed in Table 1 that
produce the best correlations can sug-
gest better risk assessment methods and
assist in evaluating and presenting the
uncertainty in risk estimates derived
using those methods, in accordance with
Objectives 3 and 4.
Conclusions
The major focus of this study is making
quantitative comparisons of carcinogenic
potency in animals and humans for 23
chemicals for which suitable animal and
human data exists. These comparisons
are based upon estimates of "RRDs"
obtained from both animal and human
data. An RRD represents the average
daily dose per body weight of a chemical
that would result in an extra cancer risk
of 0.25. Animal data on these and 21
other chemicals of interest to the EPA
and the DOD are coded into an animal
data base that permits evaluation using
different risk assessment approaches.
The major findings of this study are
as follows:
1. Animal and human RRDs are
strongly correlated. The knowledge
that this correlation exists betwee
animal and human carcinogenicit
data should strengthen the scien
tific basis for cancer risk assess
ment and cause increased confi
dence to be placed in estimates c
human cancer risk made fron
animal data.
Table 1. Approaches to Risk Assessment Components
1 Length of experiment
a^ Use data from any experiment but correct for short observation periods.
b. Use data from experiments which last no less than 90% of the standard expenmen
length of the test animal.
2. Length of dosing
a^_ Use data from any experiment, regardless of exposures duration
b. Use data from experiments that expose animals to the test chemical no less than 80°A
of the standard experiment length.
3. Route of exposure
a. Use data from experiments for which route of exposrue is most similar to that encounterei
by humans.
b. Use data from any experiment, regardless of route of exposure.
c^ Use data from experiments that exposed animals by gavage, inhalation, any oral route.
or by the route most similar to that encountered by humans.
4. Units of dose assumed to give human-animal equivalence
a. mg/kg body wt/day.
b. ppm in diet.
c. ppm in air.
d. mg/kg body wt/lifetime.
e^ mg/m2 surface area/day.
5. Calculations of average dose
a^ Doses expressed as average dose up to termination of experiment.
b. Doses expressed as average dose over the first 80% of the experiment.
6. Animals to use in analysis
a^ Use all animals examined for the particular tumor type
b. Use animals surviving just prior to discovery of the first tumor of the type chosen.
7. Malignancy status to consider
a. Consider malignant tumors only.
b._ Consider both benign and malignant tumors.
8. Tumor type to use
a Use combination of tumor types with significant dose-response.
b. Use total tumor-bearing animals.
c. Use response that occurs in humans.
d^ Use any individual response.
9. Combining data from males and females
§._ Use data from each sex within a study separately.
b. Average the results of different sexes within a study.
10. Combining data from different studies
a^ Consider every study within a species separately
b. Average the results of different studies within a species
11. Combining data from different species.
a. A verage results from all available species.
b. Average results from mice and rats.
c. Use data from a single, preselected species.
d^ Use all species separately.
NOTE: Underline indicates approach used in base analysis (Analysis O).
-------
2. Inthe majority of cases considered,
analysis methods for bioassay data
that utilize lower statistical confi-
dence limits as predictors yield
better predictions of human results
than do the same methods using
maximum likelihood estimates.
3. Analysis methods for animal data
that utilize median lower bound
RRDs detei mined from the ensem-
ble of data for a chemical generally
yield better predictions of human
results than analyses that utilize
minimum RRDs calculated from all
the studies available.
4. Use of the "mg intake/kg body
weight/day" (body weight) method
for animal-to-human extrapolation
generally causes RRDs estimated
from animal and human data to
correspond more closely than the
other methods evaluated, including
the "mg intake/m2 surface area/
day" (surface area) method.
5. The risk assessment approach for
animal data that was intended to
mimic that used by the EPA under-
estimates the RRDs (equivalent to
overestimating human risk)
obtained from the human data in
this study by about an order of
magnitude, on average. However,
it should be understood that the
risk assessment approaches imple-
mented in this study are computer-
automated and do not always
utilize the same data or provide the
same result as the EPA approach.
6. Reasonable risk analysis methods
can be defined for the chemicals
in this study that reduce the resid-
ual loss (roughly the average mul-
tiplicative factor by which the RRD
predictors obtained from the
animal data are inconsistent with
the range of human RRDs consist-
ent with the human data) to 1.7.
This is not the same as saying that
the predictors are accurate to
within a factor of 1.7, because the
estimated ranges of human RRDs
that are consistent with the human
data cover an order of magnitude
or more for most chemicals.
7. It has been possible to identify a
set of analysis methods using the
median lower bound estimates that
are most appropriate for extrapo-
lating risk from animals to humans.
given the current state of knowl-
edge and data analysis. It is pos-
sible to use the information and
results presented in this investiga-
tion to calculate ranges of risk
estimates that are consistent with
the data and also incorporate many
uncertainties associated with the
extrapolation procedure.
8. The many components of risk
assessment are interrelated and
evaluation of risk assessment
methods should focus on the com-
plete risk assessment process
rather than on individual
components.
9. The data base and method used in
this study can provide a useful
basis for evaluating various risk
assessment methods.
This study only compared human and
animal results for a relatively high risk
level. It did not examine the uncertainty
inherent in the low-dose extrapolation
process.
The animal data base and the methods
used in this study provide a useful basis
for evaluating quantitative risk assess-
ment. Their use in the present context
has demonstrated the strong positive
correlation between the animal and
human risk estimates and, hence rele-
vance of animal carcinogenicity exper-
iments to human risk estimation.
Moreover, it has been possible to identify
methods of analysis of the bioassay data,
including the choice of the median lower
bound predictor, that satisfactorily pre-
dict risk-related doses in humans. Appli-
cation of these methods has led to
suggested guidelines concerning the
prediction of human risks and the
presentation of ranges of estimates
incorporating the relevant uncertainties.
There are, however, certain features
of this investigation that should be borne
in mind when evaluating the results of
this study. These are summarized below.
• A risk level of 0.25 is used throughout.
• The bioassay data is rather crude in
several respects. The data deficien-
cies and their impact on the ability to
perform some analyses are discussed
in the document.
• The epidemiological data is of variable
quality. Some degree of subjectivity is
inherent in the estimates of uncer-
tainty associated with the epidemio-
logical RRDs.
• Different forms (complexes) of some
chemicals were grouped together.
• Other approaches to the components
could be defined and investigated.
• The three loss functions employed in
the prediction analysis lack an under-
lying statistical development and so
have been used merely to rank the
analysis methods.
• Many other analysis methods could be
investigated.
Bruce C. Allen. Annette M. Shipp, Kenny S. Crump, Bryan Kilian, Mary Lee
Hogg, Joe Tudor, and Barbara Keller are with Clement Associates, Inc.,
Ruston, LA 71270.
Chao Chen is the EPA Project Officer (see below).
The complete report consists of four volumes entitled "Investigation of Cancer
Risk Assessment Methods:" (Set Order No. PB 88-127 O97/AS; Cost: $80.OO)
"Summary" (Order No. PB88-127 105/AS; Cost: $14.95)
"Volume 1. Introduction and Epidemiology" (Order No. PB 88-127113/AS;
Cost $32.9 5)
"Volume2. Bioassay Data Base" (Order No. PB 88-127121/AS; Cost $25.95)
"Volume 3. Analyses" (Order No. PB 88-127139/AS; Cost $19.95).
The above reports will be available only from: (costs subject to change)
National Technical Information Service
5285 Port Royal Road
Springfield, VA 22161
Telephone: 703-487-4650
The EPA Project Officer can be contacted at: .
Office of Health and Environmental Assessment
U.S. Environmental Protection Agency
Washington, DC 20460
-------
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
).;:^e
' ' - '?a! c fl
* ! „. U
Official Business
Penalty for Private Use $300
EPA/600/S6-87/007
^Cl-"1C>/ -1:>rMo:IVJRl
OOOOS29 PS
IL
------- |