 |
 |
 |
 |
 |
 |
|
|||
 |
 |
 |
|||||||
| Copyright © 2012. National Academy of Sciences. All rights reserved. Terms of Use and Privacy Statement |
Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 239
A Scientific Limitations to
Extrapolating Data on
Cancer Risk from
Animals to Humans
As noted in Chapter 4, CAG is directed by Agency guidelines (U.S. EPA
1976) to determine the carcinogenicity of chemicals and to provide
numerical estimates of excess cancers in the human population that
would result from current use of the compounds under scrutiny. The
question of how or even whether to quantify human cancer risks has
been the object of considerable controversy within both the scientific and
federal regulatory communities for several years (Carter 1979~. If
relevant data were available, numerical estimates would contain less
error, and less controversy would surround the issue. Few would doubt
the scientific validity and precision of estimates of the human risk of
cancer if the estimates were based on sound epidemiologic evidence, but
such evidence is rarely available. Most commonly, only carcinogenicity
test data derived from studies conducted with experimental animals are
available and it is from such data that CAG generally determines whether
and to what extent a compound appears to be a potential carcinogen to
humans.
There is general agreement in the scientific community about a
reasonable basis for qualitatively determining that a substance is a
potential human carcinogen GREG 1979~. The I~G report, currently
under review as federal guidelines to cancer risk assessment, provides a
detailed consideration of this type of qualitative determination. The
reader is referred to this source for additional information. The opinion
is widely held that if a substance is demonstrated to be a carcinogen for
any mammalian species in an appropriately designed and performed
239
OCR for page 240
240
Appendix A
carcinogenesis bioassay, then the substance is likely to pose a potential
cancer risk to humans (Upton 1979~.
However, as we have noted, the qualitative evaluation is only half the
charge given to CAG. They are also expected to estimate the impact of the
carcinogenic compound in terms of quantitative tumor response in the
human population by extrapolating from observed responses in animal
test systems. It is this issue that is the source of so much debate. It is the
opinion of the Committee that the current state of scientific knowledge
does not permit meaningful and safe quantification of cancer risks in
humans, and for that reason EPA'S current practice should be abandoned
or greatly modified. The error in EPA'S risk estimates could be as much as
5 or 6 orders of magnitude, while benefit estimates can be trusted within
1 order of magnitude. (See Chapter 4.) The scientific considerations that
lead us to this opinion are discussed briefly in this appendix.
SOURCES OF ERROR
Potential sources of error in making both qualitative and quantitative
evaluations of carcinogenesis bioassay data are numerous. The determi-
nation that a chemical is carcinogenic rests upon demonstrating a
statistically significant excess of tumors in experimental groups as
compared to control groups. Inherent in this determination is an
assessment of how adequately a study was performed, including the
adequacy of the evaluation of the pathology of the tumors. Since the
number of excess tumors ascertained will be used for determining
quantitative risk, the ascertained and any error inherent in the evaluation
propagated to yield the final error. Furthermore, in order to determine a
meaningful excess incidence of tumors, statistical evaluation of tumor
results must consider all the experimental and control animals, including
premature deaths with or without tumors.
Next, the quantitative data from the bioassay must be extrapolated to
conditions that apply to the induction of tumors in humans. There are
several important differences to consider in comparing the conditions of
experimental studies in animals and those of human exposure to
presumed carcinogens. First, experimental studies are conducted in a
species other than humans, most commonly rodents. Second, differences
often exist between the route of administration of the carcinogenic
compound to experimental animals and the typical route of exposure
observed in human populations. Finally, practical considerations posed
by the limited life spans of the experimental animals used in the studies
and the limited sizes of experimental groups dictated by costs generally
require that large doses of the compound be administered to the
OCR for page 241
Appendix A
241
experimental animals. In contrast, estimated levels of exposure in the
human population may be considerably smaller, often by several orders
of magnitude.
Several other factors must also be considered, and any errors inherent
in these processes quantified and included in the risk estimates. Such
factors include the additive or synergistic ejects of interactions between
other carcinogens and the test compound in the human population, the
biological and genetic diversity of the human population as compared to
the experimental animals studied, and the effects of intercurrent disease
in the human population. These factors would have to be combined Lath
appropriate quantitative data concerning exposure to the compound in
question within the human population.
PERFORMANCE OF CARCINOGENESIS BIOASSAYS
Before the recent effort to describe and adopt standard protocols,
carcinogenesis bioassays varied widely in the manner in which they were
performed. Factors such as the choice of experimental animal, the
number of animals per experimental group, the dose of the test
compound employed, the schedule for administering the compound, the
conditions of housing and maintenance of the animals, and the duration
of the experiment and procedure for terminating it (e.g., serial or
terminal sacrifices, or lifetime holding) were all variables determined by
the investigator. They frequently differed between individual investiga-
tors and even between individual studies by the same investigator. In the
reports stemming from these studies, details of experimental technique
are frequently omitted with the result that specific techniques are not
definable. Frequently the chemical tested is not thoroughly evaluated in
terms of purity and composition. Similarly, diets that were obtained from
commercial sources may have changed in unknown respects between the
interval in which the study was performed and the present. In cases
where several chemicals were evaluated for carcinogenicity at the sane
time, it is rarely if ever evident whether animals exposed to more than
one compound were held in the same room and in proximity to one
another. It is also often unclear whether animals treated with known
strong carcinogens as positive controls were housed together with the
experimental animals.
Many of these uncertainties have been corrected or clarified in more
recent studies, but the results of earlier investigations remain in the
literature often without information vital to their thorough evaluation.
When a compound is being considered for regulation, the early studies
must be part of the evaluation. In its risk assessments, CAG iS responsible
OCR for page 242
242
Appendix A
for evaluating the results of every study available for a given compound
and deciding whether and how to use them. CAG'S assessment is subject
to comment by all interested parties, any one of whom may challenge or
attempt to revise CAG'S evaluation of the adequacy of the relevant
studies.
EVALUATION OF TUMOR PATHOLOGY
The basic data in a carcinogenesis bioassay are tumors observed in the
experimental and control groups. The evaluation of the pathologic
lesions in experimental and control animals of a bioassay is basically a
subjective process. Morphologic lesions, both gross and microscopic,
that may fall anywhere on a continuum of biologic diversity are assigned
to discontinuous categories. The adequacy of the categorization depends
on the insight into the morphologic manifestations of the natural history
of the disease and the thoroughness with which individual categories are
characterized and distinguished. Lesions may fall between clearly
defined categories, and more than one type of lesion can occur
concurrently in a given tissue. Furthermore, the natural histories of some
disease processes in experimental animals are less well characterized
than comparable lesions in humans. In such cases, it is more difficult to
morphologically characterize and define lesions, and pathologists have
less insight into the biological significance of the lesions. These factors
influence the precision of categorization of pathology.
Diagnostic precision is also influenced by personal insights, skill, and
experience. Although pathologic evaluations are admittedly subjective,
rarely, if ever, is an attempt made to place a measure on the precision or
accuracy of these diagnoses, that is, to determine precisely how the
categorization or description of lesions characterize the pathology in that
organism, or how well individual pathologists rate in their assignment of
given lesions to appropriate categories. It is unclear in most cases
whether this error is 20 percent, 10 percent, 5 percent, or 1 percent, and
so on. As a consequence, the diagnoses are generally used as numerical
data without error tolerances. Thus, a major potential source of error in
the risk quantification never enters into a determination of error
tolerances in the risk estimate.
The problem is compounded in older studies in which diagnostic
material may be unavailable for subsequent reevaluation. Furthermore,
diagnostic criteria and pathologic categorization may have changed
during the interval between initial pathologic review and subsequent
publication of a paper. For more recent studies, particularly those
sponsored by the federal government, external review of pathologic
OCR for page 243
Appendix A
243
evaluations has been instituted. This policy should presumably reduce
error, but the extent of imprecision remains unclear.
EVALUATION OF RESULTS
Bioassay results may be evaluated according to incidence of tumors of
all types within animal groups, incidences of tumors in specific locations,
multiplicity of tumors, incidences of tumors of grouped organ sites, and
so on. Methods for evaluating results vary both with the result observed
and the procedures used for performing the bioassay. For example,
different procedures may be used to evaluate studies in which whole
animals survived to a terminal sacrifice as compared to studies in which
excess early mortality occurred or animals were held until they died
spontaneously. The availability of dose-response data provides addition-
al bases for evaluating results. Lee difficulty involved in assessing excess
tumors, therefore, varies with the results of a study.
In cases in which there is a large excess incidence of tumors in the test
group, no comparable tumors in the control group, and both control and
test groups are large, simple comparisons of the difference in tumor
incidence may suffice. If the groups are not large, comparisons of tumor
incidence must be supplemented with estimates of the imprecision of the
experiment. Preferably, such error estimates should measure inherent
inconsistencies or variances between studies in the ejects of a given dose
of a compound. Generally, however, there is only one test group in a
study, or certainly only one group at a given dose level, making such
estimates impossible. Thus, error estimates are generally based upon the
size of the group used to make the comparison between test and control
animals. When tumor responses are small or when there is a significant
incidence of tumors in the control animals, the issue of experimental
error becomes more critical, particularly when experimental groups are
small and reliable estimates of biologic variability and response to the
test compound are not available. For example, if, for a given site, tumors
occur in the control animals, but a greater number of tumors is detected
at this site in the one tested group, is this a real property of the test
compound? Without knowing the variation of tumors in control animals
at the given site, an excess of tumors in the test over the control group
may only lie within the range of biologic variability of the test animals.
For studies in which animals are allowed to live until they die
spontaneously, or studies that involve a terminal sacrifice but in which a
large proportion die before termination, alternative methods for evaluat-
ing tumor responses are necessary. The general approach is an actuarial
or life-table analysis. Specification of a defined end point in the study is
OCR for page 244
244
Appendix A
critical to this approach. For example, the presence of a tumor that is
uniformly lethal to the host is a defined end point. Similarly, tumors that
are uniformly nonlethal and are found incidentally at death also present
a defined end point. In contrast, tumors that may cause the death of an
animal but do not necessarily do so, and may be found as incidental
microscopic lesions, are less clearly defined as end points and pose
difficulties in statistical analysis. Quantitative estimates of differences
between control and test groups are difficult to make, and lack of
precision here poses an even greater problem for quantification.
To summarize, the degree of difficulty in estimating excess tumor
incidence relates in part to the magnitude of tumor excess in the test
animals over the controls and in part to the number of premature deaths
occurring in the test group as compared to the controls.
The preceding considerations apply to the qualitative determination of
the carcinogenicity of a compound. All these factors also apply to the
quantitative determination of the magnitude of carcinogenic response to
a given mode of treatment with a test compound. In addition, as noted,
several other factors must be considered in achieving a quantitative
extrapolation to human cancer risks, and these factors are considered
briefly below. Methods for determining human exposure have been
considered in Chapter 4 and will not be reiterated here. It should be
evident, however, that the error and imprecision inherent in estimates of
human exposure must be propagated to yield the final estimate of tumor
response in the human population and the error in the estimate offered.
EXTRAPOLATION TO LOW DOSES
Typically, constraints of time and money require that carcinogenesis
bioassays be performed in rodents. Because of expense, control and
experimental groups are limited generally to fewer than 100 animals and,
frequently, to even fewer than 20. Consequently, to maximize the
probability of detecting a postive tumor response, very high doses of test
compounds are used. The doses are generally based upon the maximum
tolerated dose that yields no excess subacute toxicity in the test group.
Such doses are generally much higher than the typical dose to which
humans are exposed frequently by several orders of magnitude (on a
milligram per kilogram body weight per lifetime basis).
The choice of these high doses is a pragmatic one, but it poses the
problem of extrapolating from ejects at high doses to tumor responses
anticipated at the extremely low doses typical of human exposure. Since
we do not have a comprehensive, detailed theory of carcinogenesis, we
do not have a method for calculating the real number of tumors that will
OCR for page 245
Appendix A
245
develop in experimental animals at a projected lower dose based upon
observations at higher doses. However, numerous models have been
proposed that make a variety of assumptions about the nature of
carcinogenesis and over a wide range of estimated tumor responses in a
low-dose range. As has recently been reported for saccharin, depending
upon which assumptions are made and consequently which extrapola-
tion procedures are used, the errors in the predicted tumor incidence at
low doses may range over 6 orders of magnitude in this case and could
conceivably be higher in other circumstances (NRC/TOM 1978~. Again, the
estimates are based in science but rest on unproven assumptions, and the
precision of the resulting estimates is unclear.
Because of the uncertainties inherent in extrapolation models, CAG
generally uses the extrapolation that gives the highest reasonable
estimate of cancer incidence within the dose range of human exposure.
The linear nonthreshold model, although not likely to be close to reality
for most compounds, is generally presumed to represent an upper bound
on risk extrapolation in most cases. If, however, only one dose level is
tested, and if the dose is in a saturation plateau of carcinogenic elect,
then a linear nonthreshold extrapolation may underestimate the carcino-
genic potential of the compound for a portion of the dose-response
curve. Thus, under certain circumstances, even this rather "conservative"
extrapolation procedure may provide an underestimate of erect. None-
theless, the estimates are crude, and the extent of propagation of error in
resulting human risk estimates is, again, unclear.
CORRECTION FOR DIFFERENT ROUTES OF ADMINISTRATION
Carcinogenesis bioassays are most typically performed by feeding the
test compound to experimental animals. Less frequently, compounds are
tested by application to the skin, by inhalation, or by subcutaneous or
other routes of injection into the experimental animal. Although people
are frequently exposed via ingestion, other routes of exposure may be
important. Consequently, corrections must be made when the experi-
mental route of exposure diners from the human. These corrections are
generally not based on theory that is as well formulated as that on which
the extrapolation from high to low doses is based.
The technique that CAG generally uses employs an analogy between
the compound in question and a carcinogen that is chemically similar to
the test compound and has been tested by a variety of routes. Short-term
metabolic studies can indicate similarities or differences between the
distribution of the comparison compound and the test compound and
thus provide, by analogy, more insight into the validity of the
OCR for page 246
246
Appendix A
comparison. The efficacy of tumor induction at various sites depending
upon the route of administration can be determined for the comparison
compound, and in this manner an analogy can be constructed.
Unfortunately, there is a long history of peculiarities of individual
chemicals; even chemicals that are structurally quite similar may
manifest very different properties of distribution in the body and
activation to carcinogenic forms. Thus, the precision of the comparisons
is largely unmeasured.
Again, in practice, comparisons are generally made so as to maximize
the estimated eject. Nonetheless, the estimates are crude and the extent
of propagation of error in the resulting human risk estimate is unclear.
EXTRAPOLATION BETWEEN SPECIES
In making extrapolations between ejects noted in rodents and those
anticipated in humans, the extrapolator generally chooses the test species
in which the largest tumor response per unit dose is observed. This is
then considered the most sensitive test species, and it is generally
assumed that the human population will be less sensitive. This assump-
tion is largely based on the evaluation of six compounds for which
quantitative exposure-tumor response data are available for both
experimental animals and human populations. In these six cases, a
reasonable comparability was determined between the extrapolated
human tumor incidences and the animal dose-response data (NRC 1975~.
In each case where the most sensitive animal strain was selected, the
anticipated or calculated human cancer risk was greater than that
observed epidemiologically in the human population. However, such
evidence for only six compounds does not prove the validity of the
assumption, and there may indeed be compounds for which the
extrapolation is not appropriate. For example, epidemiological studies
indicated that benzene and arsenic are carcinogenic in humans, but
experimental studies with these compounds have yet to prove conclusive-
ly that they are carcinogenic in animal bioassays. It is conceivable that in
these two cases, the discrepancy arises from the fact that the human is
the more sensitive species. Thus, in attempting to extrapolate between
species, to make the assumption that the human is less sensitive than the
most sensitive of the test species is not necessarily correct.
OTHER CONSIDERATIONS
Without a well-formulated, comprehensive theory and explanation for
all or most facets of the development of human cancer, several critical
OCR for page 247
Appendix A
247
determinants of risk in the human population may be omitted from the
extrapolation procedures as currently performed. Factors not now
considered include individual variability in the human population in
response to exposure to carcinogens, erects of intercurrent diseases, and
various forms of interactions between carcinogens, co-carcinogens,
promoters, and other factors.
CAG PRACTICES
In determining whether or not a compound is a potential carcinogen,
CAG iS directed by the EPA guidelines to use a "weight-of-evidence"
approach. The guidelines indicate the relative weight to be given to
epidemiological studies, carcinogenesis bioassays, and short-term tests,
but, as noted in Chapter 4, neither the guidelines nor CAG provide
written criteria for following the approach. In fact there is uncertainty
about how the approach is to be applied, particularly in cases where
studies of the same type come to different conclusions regarding
carcinogenicity or where the quality of studies compared diners
substantially. It is unclear whether CAG adheres to a neutral or objective
"weight-of-evidence" approach or whether it places greater weight on
data suggesting carcinogenicity in an effort to avoid underestimating the
potential for human cancer risk.
In making quantitative estimates, CAG'S philosophy is to maximize
each of the individual components employed in the extrapolation to
estimate excess human cancer deaths, so that the real risk to which the
human population may be subject will always be less than the estimated
risk. This is a practical attempt to deal with the problem of limitations of
current scientific knowledge. Since CAG recognizes that actual human
risk is difficult or impossible to determine precisely, it attempts to
estimate an upper bound of probable human risk. For each of the three
extrapolations discussed above high to low dose, test animal to human,
and route of administration CAG uses those assumptions and estimates
that tend to maximize risI;. Even human exposure estimates, not CAG'S
responsibility, are "upper-limit" estimates. The end result is that CAG
propagates through the calculations those error tolerances that can be
estimated, and the final estimate of excess human cancer incidence is
reported as a range. The principal error arises in calculating the excess
proportion of tumors occurring in the test group as compared to those
arising in the control group and in correcting for the size of the
experimental group, a factor that relates to the precision of the result.
CAG results thus show an estimated number of excess cancer incidences
OCR for page 248
248
Appendix A
with a range of confidence although, as noted above, all sources of error
are not considered.
CAG informally urges caution in interpretation of its estimates. For
example, if each of two estimates-say, 410 and 380 incidences falls
within the error tolerance of the other (the usual case), the two figures are
not to be interpreted as significantly different. Nor are they to be
interpreted as an actual excess of human cancer incidence of this precise
magnitude. Instead, CAG'S estimate is to be interpreted as an upper
bound, and, presumably, the actual excess of human cancer incidence
will be less than this number or less than the upper limit of the variability
of this value. What CAG appears to believe is most important, however, is
an extrapolated risk estimate (or better yet, its upper confidence limit)
that falls below an incidence of one excess cancer; CAG tends to view this
type of result as an indication that the compound in question is not a
significant human cancer risk. But CAG'S estimates would be more
valuable if they were accompanied by these informal interpretations.
Currently, the estimates stand as values with tolerance ranges that,
contrary to the "warning-signal" stature recommended in the Agency
guidelines, appear as actual numerical estimates of excess human cancer
incidence attributable to use of the compound in question.
If CAG'S risk estimates are to be used as intended two conditions must
be met: (1) the estimates must be interpreted correctly by the Adminis-
trator, who is required to judge the balance between risks and benefits;
and (2) the estimates must in fact be an upper bound on the real excess
of human cancer incidence attributable to the compound. By presenting
estimates of excess risk as numerical values, even with error tolerances,
CAG provides values that appear to have tangibility and scientific
validity. Although CAG members have attempted to provide Administra-
tors with insight into the usefulness and limitations of the CAG estimates,
it is difficult to judge how well they have succeeded. The estimates
become a matter of record subject to evaluation and interpretation by
individuals who do not have the benefit of CAG'S informal interpretation
of its own results. Thus, the presentation of the estimates without verbal
explanation of how they might best be used exposes the figures to
misinterpretation and misuse. In fact, the values are often erroneously
accepted quite literally as sound scientific estimates with well-defined
error limits.
The second point mentioned above concerns the adequacy of CAG
estimates as an upper bound on real risk. As we have already seen, CAG
has attempted to validate its extrapolation procedures on the basis of an
NRC (1975) report that compares the results of animal studies and human
epidemiologic investigations for six compounds-benzidine, chlornapha
OCR for page 249
Appendix A
249
zinc, DES, aflatoxin B., vinyl chloride, and cigarette smoking for which
dose-response data exist for both human exposure and animal experi-
mentation. The NRC report concluded that, in at least four of the six
cases, there is substantial agreement between the epidemiologically
estimated excess of cancer deaths and the estimates extrapolated from
animal studies. The translation of this limited conclusion, however, into
a working hypothesis that human lifetime cancer incidence in general
can be approximated by extrapolating from the lifetime incidence
induced by similar exposure in laboratory animals is open to question.
For example, CAG'S estimates from human and animal data of the
incremental risk per unit of inspired benzo~a~pyrene varied over 21/2
orders of magnitude. One of the values determined from the animal
studies was 2 orders of magnitude lower than that determined from
epidemiological estimates, and another value was one twentieth of that
estimated from the human studies (CAG 1978~.
In addition to orders of magnitude disagreement between estimates
from animal and human data, epidemiological estimates themselves are
subject to error. These errors arise not only from imprecisions in the
determination of excess cancers, but from difficulty in estimating actual
human exposures. Epidemiological studies of vinyl chloride exposures
and the related cancer risks are cited as a source of reliable dose-
response data in the human population (NRC 19754. Yet, in this example,
where efforts have been made to quantify exposure of the working
population, the estimates are derived and not the product of precise
measurement of doses. Exposures were estimated retrospectively on the
basis of duration of employment and the specific job of individuals
exposed during that period. The majority of the exposure estimates were
derived from measurements of current levels of exposure to vinyl
chloride in specific jobs, using current equipment and reagent stock. The
extent to which these conditions apply to earlier periods of exposure is
unknown. Furthermore, unlike the corresponding animal studies, the
vinyl chloride workers were probably exposed to a combination of other
organic materials that may influence the elects of the vinyl chloride (see
below) (Nicholson et al. 1975~. Consequently, even the case of vinyl
chloride, considered by some a source of sound epidemiological dose-
response data for excess cancer risk, is open to serious question
concerning; the precision of its exposure estimates. In most other
epidemiological studies exposure estimates are less sound than in the
case of vinyl chloride. Even in the case of human exposure to low doses
of irradiation, where because of better dosimetry one might expect more
precise estimates to be available than for chemical substances, expert
opinions on estimated risks vary by 1-2 orders of magnitude (NRC 1979~.
OCR for page 250
250
Appendix A
Furthermore, the six cases with "good" epidem~olog~cal data present-
ed in the NRC report (1975) involved strong carcinogens. The majority of
compounds that CAG will be required to assess will be weaker
carcinogens than those noted above, and epidem~olog~cal evidence will
generally not be available for validation of the extrapolated values.
Therefore, even if one were to accept the validity of the epidemiolog~-
cal/expenmental comparisons of the six tested compounds, one could
not be sure that the relationship would still be valid for the more typical
compounds CAG iS asked to evaluate. Perhaps most distressing is that for
some calculated incremental nsks, benzo~a~pyrene for example, an
estimate extrapolated from animal studies was not ~ fact an upper
bound on the epidem~olog~cally determined excess risk. One cannot be
certain that the 2-order-of-magn~tude discrepancy (underestimate) be-
tween one of the animal studies and the human epidem~olog~cal
estimates will not be exceeded. This calls into question the fundamental
premise that CAG estimates represent upper bounds on human cancer
risk attributable to the use of a compound.
Thus, not only are there uncertainties in the evaluation of bioassay
data, limitations in extrapolation methods, and omissions of factors
estimating risk because of lack of scientific knowledge, but CAG'S
assumption that their estimates are upper bounds can also be questioned.
Within CAG these problems may be understood and recognized as part of
the estimation process. Of greater concern, however, is the fact that once
out of the hands of CAG, the estimates themselves are subject to
misinterpretation.
CONCLUSIONS
The goal of quantifying assessments of human risk of cancer is attractive
in theory. It could provide a comprehensible, quantitative measure
against which to balance benefits and thereby make administrative
decision making easier. It might also lead to some consistency among
regulatory decisions, if the current attempt by federal agencies to agree
on a uniform method of quantifying cancer risks is successful. Despite
these advantages, the Committee concurs with this recent statement by
Arthur Upton (1979), Director of NCI:
Although an attractive idea, quantitative risk assessment involving extrapolation
from animal data is not yet sufficiently developed to be used as a primary basis
for regulating human exposure to carcinogens. Although we are correct in
concluding qualitatively that animal carcinogens are potential human carcino
OCR for page 251
Appendix A
251
gene, quantitative extrapolations involve potentially large errors, some of which
could underestimate the actual human risk from exposure. Scientific knowledge
is currently insufficient to lend precision to this process.
The attempt to precisely estimate quantitative cancer risks raises
controversy because it requires scientific judgments and extrapolations
which transcend the limits to our scientific knowledge. Important
considerations have been omitted from estimates; this indicates they
were either unrecognized (unlikely) or that no method was available for
incorporating them. Furthermore, some of the factors integral to current
quantification methods may have unquantified errors and thus be
potential sources of further errors whose tolerances may be orders of
magnitude in scale.
Substantial additional research is needed to add to current scientific
knowledge before sound quantitative risk estimates can be achieved.
Such research should focus on: mathematical modeling of carcinogenesis
to learn more about dose extrapolations, synergistic and additive effects,
and quantification of the precision and accuracy ranges of pathological
evaluations. Support is also needed for development of sources of data
and references for pathology, critical reviews of old carcinogenesis data,
and development of a bank of well-characterized reference carcinogens
with dose ejects, pharmacodynamics, species differences, and other
information.
The practical value of quantitative risk assessment alluded to above
makes the pursuit of valid estimation a worthy goal. However, current
methods need to be critically tested and scrutinized before they can
become accepted procedure. Clear distinctions should be made between
scientifically supportable components and those that are only best-guess
extrapolations. The possibility of gross error, particularly underesti-
mates, must be indicated. Overestimates involve the monetary costs of
overregulating a compound; but underestimates are detected years later
and are paid for in human deaths from cancer.
In closing, we repeat the theme of this appendix: until the scientific
limitations to extrapolating numerical estimates of human cancer
incidences from animal data are reduced, the Committee recommends
that the practice be abandoned. EPA currently uses such estimates as a
primary basis for regulating human exposure to carcinogens. Although
Upton suggests that "regulatory decisions must be based on an
evaluation of all the relevant information including the quantitative
estimates of risk" (Upton and Nelson 1979), the Committee feels that
until quantitative estimates are more sound, cessation of the quantitation
of human cancer risk estimates appears to be the most certain method to
OCR for page 252
252
Appendix A
prevent the misuse of such estimates. An alternative method for
estimating risk is offered in Chapter 4 and applied in Chapter 7.
REFERENCES
Carcinogen Assessment Group (1978) Preliminary Report on POM Exposures. External
Review Draft. U.S. Environmental Protection Agency, Washington, D.C. 20460.
(Unpublished)
Carter, L.J. (1979) How to assess cancer risks. Science 204:811~13.
Interagency Regulatory Liaison Group (1979) Scientific Bases for Identification of
Potential Carcinogens and Estimation of Risks. Report of the BEG, Work Group on
Risk Assessment. Journal of the National Cancer Institute 63:241-268.
National Research Council (1975) Pest Control: An Assessment of Present and Alternative
Technologies. Volume I, Contemporary Pest Control Practices and Prospects: The
Report of the Executive Committee, Study on Problems of Pest Control, Environmental
Studies Board, Commission on Natural Resources. Washington, D.C.: National
Academy of Sciences.
National Research Council/Institute of Medicine (1978) Saccharin: Technical Assessment
of Risks and Benefits. Committee for Study on Saccharin and Food Safety Policy.
Washington, D.C.: National Academy of Sciences.
National Research Council (1979) The Effects on Populations of Exposure to Low Levels
of Ionizing Radiations. Committee on the Biological Effects of Ionizing Radiation,
Division of Medical Sciences, Assembly of Life Sciences. 1974582-412:45. Washington,
D.C.: U.S. Government Printing Office.
Nicholson, W.J., E.C. Hammond, H. Seidman, and I.J. Selikoff(1975) Mortality experience
of a cohort of vinyl chloride-polyvinyl chloride workers. Annals of the New York
Academy of Sciences 246:225-230.
Upton, A.C. (1979) Quantitative Risk Assessment. Memorandum to Commissioner, U.S.
Food and Drug Administration, April 5, 1979, from the Director, National Cancer
Institute, National Institutes of Health. (Unpublished)
Upton, A.C. and N. Nelson (1979) Cancer Risk Assessment. Joint statement by Upton,
Director, NC! and Nelson, ~rector, Institute of Environmental Medicine, New York
University Medical Center, Sept. 21. (Unpublished)
U.S. Environmental Protection Agency (1976) Health Risk and Economic Impact
Assessments of Suspected Carcinogens: Interim Procedures and Guidelines. 41 Federal
Register (102)21402-21405.
Representative terms from entire chapter:
human population
