EPA’s Use of Data from Intentional Human Dosing Studies in Risk Assessment
As described in Chapter 1, the committee was asked by the Environmental Protection Agency (EPA) to evaluate the use of data from intentional human dosing studies in the agency’s risk-assessment process. The committee examined the relevant questions within the context of EPA’s general framework for risk assessment but was not asked to evaluate that framework. Rather, it was asked to determine whether and in what way data from intentional dosing studies in humans could be appropriately incorporated into EPA’s established approach to risk assessment.
This chapter focuses on how EPA might use the data obtained through research that meets the scientific, ethical, and procedural standards outlined in the preceding chapters. It provides a description of the risk-assessment framework, and its basis, as the starting point for this evaluation. Following that description, the chapter examines the questions of when and how data from intentional dosing studies should be incorporated into EPA’s risk-assessment process. The descriptions that follow will show that there is substantial precedent at EPA for using such data in risk assessment; however, although it is important to recognize this historical use, the committee looked beyond precedent to examine anew the questions at hand.
EPA’S RISK ASSESSMENT FRAMEWORK
Toxicity information from animal studies is routinely used by EPA
and other regulatory and public health agencies to establish levels of lifetime daily intake of chemical substances that are likely to be without adverse effects in the general human population. EPA describes such levels (or doses) as risk Reference Doses (RfDs) for oral exposure and as risk Reference Concentrations (RfCs) for inhalation exposures (Faustmann and Omenn, 2001). EPA and other agencies also may establish levels to protect humans against short-term exposures or to protect specific subpopulations, such as individuals in occupational settings. In addition, EPA develops risk values for carcinogens, often based on animal data.
The use of animal data rests on substantial evidence that a relatively high degree of concordance exists between experimental animal findings and expected outcomes in humans. Their use also is necessitated by the fact that it is not possible to evaluate most forms of toxicity through intentional dosing studies in humans. The predictive value of animal studies for humans is, however, far from exact. Therefore, decisions to use results from animal studies rest, at least in part, on a “science policy” choice in which results from animal studies are generally assumed to hold for humans unless there is highly convincing evidence to the contrary (NRC, 1983; NRC, 1994).Taking into account the extensive batteries of animal studies conducted at high doses that are normally required for human risk assessment and the safety factors and conservative assumptions built into the risk-assessment process, this science policy choice generally reflects a cautious stance; but the possibility that animal studies, no matter how complete, may sometimes fail to reveal adverse health effects that are significant for humans cannot be ignored.
Data from some types of human studies have played a significant role in the establishment of RfDs, RfCs, and other measures of protection (Dourson et al., 1996; EPA, 1999). Data from both intentional dosing studies in humans and epidemiological studies have been used, with the former generally limited to effects resulting from single or very short-term exposures. In a significant number of important cases, EPA has elected to derive risk values for carcinogens from epidemiological data (e.g., benzene, arsenic, chromium [VI], and several others), given the strength of the databases for these compounds. EPA guidelines for risk assessment express a preference for human over animal data, although they clearly note the difficulties in developing human data adequate for such quantitative assessments (EPA, 2003).
Generally, knowledge of the quantitative differences between doses causing adverse effects in animals and those causing adverse effects in humans is not precise. Beginning in the 1950s, scientists in regulatory agencies began applying “safety factors” to data from animal studies to establish “acceptable daily intakes” (Lehmann and Fitzhugh, 1954). Those factors (100-fold when the animal data were derived from chronic stud-
ies) were intended to account for the possibility that humans were, on average, more sensitive to a chemical’s effects than were laboratory animals, and that some humans were more sensitive than the average. This safety factor approach found broad application over the several decades following its introduction, and its use was promoted throughout the 1970s and 1980s by many committees of the National Research Council (NRC, 1983; NRC, 1994).
During the 1980s, EPA dropped the use of the term “safety factor” and substituted the concept of “uncertainty factors” (UFs). Moreover, the agency defined more completely the use of these factors in each discrete step of the process of deriving RfDs and RfCs from animal data (EPA, 2002). Several types of default UFs are used, for example, in deriving RfDs from animal data:
UFA: A factor of up to 10 is used to extrapolate from animals to humans, to account for the possibility that humans are, on average, more susceptible to the effects of chemical exposures than are experimental animals. (The application of UFA to animal data [a NOAEL, or no observed adverse effect level] yields a dose that should be protective for “humans of average sensitivity.”)
UFH: A factor of up to 10 is used to account for variability in response among humans, such that some members of the human population are more susceptible than the average.
If animal data are available and reflect the effects of chronic exposure, then the traditional lifetime RfD is derived by (1) selecting the results from the study or studies of adequate quality that show effects at the lowest dose and (2) identifying the maximum dose from that study at which no adverse effects (in relation to control animals) were observed (a NOAEL).1 The RfD is derived as follows:
If EPA has available data only from subchronic studies, then another factor (UFs) having a value up to 10 may be applied, and if the available toxicity studies do not include a NOAEL, but only an adverse effect dose (a LOAEL, or lowest observed adverse effect level), another factor (UFL) of up to 10 may be included (Dourson et al., 1996). Thus, the derivation of an RfD from a subchronic toxicity study that reveals a LOAEL, but not a NOAEL, may proceed as follows:
Such a derivation would, at a maximum, include a total UF of 10,000, although an agency technical panel has recently proposed a limit on UFs that would put a maximum total value at 3,000 (EPA, 2002). EPA sometimes includes additional factors in situations in which data deficiencies of various types exist (Faustmann and Omenn, 2001).
In connection with requirements of the Food Quality Protection Act (FQPA), EPA may include an additional safety factor to account for concerns regarding children’s exposure to pesticides and susceptibility to pesticide toxicity (see Chapter 2). It should be noted that, in the case of pesticides, EPA generally does not need to include a UFL, a UFS, or a factor for data deficiencies, because pesticide registration requirements ensure that the agency has a complete database available. Thus, pesticide RfDs derived from animal data generally involve UFA, UFH, and a safety factor to protect children (see Chapter 1).
The magnitudes of the various UFS generally range from 1 to 10. In the case of UFA and UFH, deviations from 10 (the typical default value) usually require substantial evidence to demonstrate that a smaller value is adequate (Dourson et al., 1996). There are many examples of regulated chemicals in which values less than 10 have been used by EPA, although 10 is typically the default in pesticide regulation. UF values less than 10 for UFL and UFS are common and depend on case-by-case judgment.
Historically, when human data are used as the basis for RfD derivation, the UFA factor is replaced with actual data (e.g., a NOAEL derived from adequate human studies might be used without the need for a UFA). The value of UFH used in such situations depends on a judgment regarding the quality of the human data and the degree to which the studied populations represent the average or the more sensitive end of the spectrum of human sensitivities (see, for example, EPA’s approach to RfD derivation in the case of methylmercury).
The scientific bases for the default values of 10 for UFA and UFH are limited, although several empirical studies, based on cases in which com-
parative data are available for animals and humans, generally have shown these values to be adequately protective. This conclusion is, however, far from fully substantiated (Rodricks et al., 2001).
IMPLICATIONS OF THE USE OF INTENTIONAL HUMAN DOSING STUDIES IN THE RISK ASSESSMENT PROCESS
Assuming that data from a given study meet the criteria for scientific validity and are found to be of adequate quality as demonstrated in Chapter 3, and assuming that they satisfy the ethical requirements described in Chapters 4, 5, and 6, they can be considered for use in risk assessment. Direct use of such human data would eliminate the need for introducing the uncertainty factor ordinarily used to extrapolate from animal data to humans (UFA).
It must be emphasized that, even if UFA were to be replaced with data from intentional dosing studies in humans, use of such data would have no effect on the other UFs typically used in deriving RfDs or other criteria for health protection. Specifically, the safety factor introduced under FQPA to protect children would not be affected by the replacement of UFA with actual data.
Additional issues arise in considering the appropriateness of eliminating UFA. Depending on the endpoints studied, the data from an intentional dosing study in humans could yield a NOELHU (no observed effect level), a NOAELHU, a LOELHU (lowest observed effect level), and perhaps a LOAELHU (lowest observed adverse effect level) (see Box 7.1 for a summary of the committee’s use of risk terminology for data derived from intentional human dosing studies). It is possible that an intentional human dosing study may yield a LOELHU but not a NOELHU or, conversely, may yield a NOELHU but not a LOELHU. If the study yields a LOELHU but not a NOELHU, because lower doses were not studied, a judgment would need to be made regarding the UF necessary to estimate a NOELHU from the observed LOELHU. At least for data from intentional dosing studies in humans submitted to EPA by third parties, it would seem that there would be little basis for making such a judgment, and a repeat of the study using lower doses would be necessary to identify a NOELHU. This assumes that EPA will choose to conduct risk assessment using human data based on a NOELHU.
In those cases in which a NOELHU is identified, but a LOELHU is not (i.e., there is no identified effect level, adverse or not), different issues come into play. In the case of animal studies, several dose levels are used, with the expectation that a NOEL, a NOAEL, a LOAEL, and levels revealing serious toxicity will be identified (the latter levels could not be used in an intentional dosing study in humans). Ordinarily, EPA and other regulatory agencies do not use data from “NOEL-only” studies unless no other
No observed effect level (NOELHU)
A NOELHU is the highest dose or concentration at which no changes of any kind are seen relative to controls. Depending on the number of doses studied and the ability to detect the LOELHU, the NOELHU could underestimate the actual dose that could be given without a response.
Lowest observed effect level (LOELHU)/ No observed adverse effect level (NOAELHU)
A LOELHU is the lowest dose or concentration at which a biological effect that is not adverse is seen. An example of such an effect would be cholinesterase inhibition by pesticides. A small amount of cholinesterase activity has not been demonstrated to have any adverse health effects. If lower doses are not studied the LOELHU could overestimate the dose that could actually elicit a response. What the committee terms a LOELHU is often referred to by EPA as a no observed adverse effect level (NOAELHU). The committee is careful in its use of the term “NOAELHU” because it is most appropriately used in situations in which a clear LOAELHU has been identified. A NOAELHU is the highest dose or concentration at which no adverse effect is seen relative to controls.
Lowest observed adverse effect level (LOAELHU)
A LOAELHU is the lowest dose or concentration at which an adverse effect is seen. In terms of the committee’s discussion, for intentional human dosing studies there should be high confidence that any anticipated adverse effect is not serious and is reversible.
data are available and decisions must be made (typically in emergency situations; EPA, 2000a). One problem with NOEL-only studies is that they offer no information on the quantitative relationship between the measured NOEL and the unmeasured NOAEL or LOAEL, so that it is not possible to determine whether the minimum effect or toxic level is a small-or large-multiple of the measured NOEL. If the true but unidentified NOAEL or LOAEL is a large multiple of the measured NOEL, then use of the latter in deriving an RfD or similar protective value will lead to unnec-
essarily restrictive limits (i.e., the measured NOEL is almost certainly smaller than the “true NOEL,” which should be close to the NOAEL).2
But a more important issue arises in connection with “NOEL-only” studies in humans (NOELHU-only), and that concerns the possibility that the study participants may be somehow “nonresponsive” (we are using NOELHU to mean a dose producing no response of any type significantly different from that observed in control groups), or that there were problems with the assay employed in that study. If an intentional dosing study in humans shows no effects significantly different from the control (adverse or not), then the possibility that the volunteers chosen are somehow insensitive to the exposure or there is some other study defect that cannot be excluded, and the study should be repeated.
In general, therefore, any useful human study must investigate a range of doses, including at least one dose with an effect and one without. How many doses are studied, and how far apart they are, will determine the precision of estimates of NOELHU and LOELHU (or NOAELHU and LOAELHU). In addition, the finding of an effect confirms the “assay sensitivity” of the study. A study showing no effect, and therefore providing a potentially conservative (i.e., falsely low) estimate of NOELHU that might seem acceptable would be uninformative because of the lack of evidence of assay sensitivity. Thus, if an intentional dosing study in humans reveals no effects significantly different from the control (adverse or not), then the possibility that the volunteers chosen are somehow insensitive to the exposure cannot be excluded, and the study should be repeated. In general, NOELHU-only studies should not be used for formal risk assessments, unless no other data are available and there is a critical need to develop a tentative risk value.
ELIMINATING THE UFA
A significant issue arising in connection with the substitution of a NOELHU or LOELHU for a NOAEL derived from an animal experiment concerns the matter of the uncertainty factor to be used. Obviously, the use of the traditional UFA is not appropriate in the presence of relevant and reliable human data, because it was introduced as a default to account for the possibility that humans are, on average, more sensitive than are experimental animals. The human data replace that default assump-
tion. However, a decision to reduce or eliminate the UFA does not automatically eliminate the uncertainty associated with using a NOELHU or a LOELHU.
This uncertainty arises because of a certain vagueness in EPA’s risk-assessment methodology—namely, the absence of a completely clear understanding of the sensitivity of the segment of the human population that is the intended target when the UFA is applied to a NOAEL from an animal study. Thus, the use of a UFA intended for extrapolation from animals to humans gives rise to the question: what humans? If these humans are thought to be humans of average sensitivity, then what is meant by that term? And how is it possible to know that the research participants in the intentional dosing study that is the basis for the NOELHU and LOELHU are truly representative of the average humans that are the intended target of the application of a UFA? If this cannot be known, does this suggest the need for a UFH that is somewhat larger than the usual default value of 10?
As a general matter, the risk-assessment methodology assumes that humans of average sensitivity are healthy adults, and that healthy adults are usually the participants in intentional dosing studies. But, because of the uncertainty described above, the committee considered the following possibilities:
Should research sponsors be encouraged to conduct two independent intentional dosing studies in humans, using different study populations and testing facilities? Replication of study results provides added confidence regarding the sensitivities of the studied populations and the degree to which those populations can be said to represent individuals of average sensitivity, while failure to replicate findings provides a measure of the variability in responses among healthy adult volunteers and a basis for assuring that risks will not be underestimated (because the data from the more sensitive study population would be used for risk assessment).
If research sponsors chose to conduct only a single intentional dosing study in humans, should EPA consider applying a UFH to the NOELHU that is somewhat larger than the usual factor of 10? This larger UFH would account for the possibility that the participants in the study may be somewhat less sensitive than the hypothetical average to which the traditional UFA is meant to apply.
The committee did not turn these two possibilities into formal recommendations because they would likely alter EPA’s usual approach to risk assessment, but it concluded that they deserve study and further consideration by the agency. In some cases, volunteers are selected for intentional dosing studies specifically because they are known to be somewhat
more sensitive than average. This situation occurs, for example, in some of the short-term air pollution studies conducted by or for EPA. The need for replication or additional UFs in such circumstances is significantly less compelling than those in which healthy volunteers, not known to have special sensitivities, are the participants in an intentional dosing study.
THE CASE OF CHOLINESTERASE INHIBITION
EPA’s Office of Pesticide Programs has received a body of data from intentional dosing studies in humans sponsored by third parties involving measurements of cholinesterase inhibition induced by certain pesticides (see Chapter 1). Generally, such inhibition is taken to be a very sensitive marker of exposure to this class of pesticides: When RfDs are derived on the basis of NOAELs for this effect obtained from animal studies, they are generally lower than RfDs derived from studies of other adverse effects of the pesticide (including studies of chronic duration), so they are chosen as the basis for regulatory standards. (The committee heard detailed discussions and support for this position from EPA scientists and officials during its open meetings.)
There is a long history of use in pesticide regulation of NOELHU or LOELHU from intentional dosing studies in humans of cholinesterase inhibition, but the appropriateness of using data from such studies has come under question, and those questions gave rise to the work of this committee (see Chapter 1). The committee examined a subset of the third-party intentional dosing studies in humans submitted to EPA, and found that although these studies were not developed using the criteria for scientific validity the committee presents in this report, it appears that some of the studies may meet most of those criteria.
The full evaluation of the quality of the submitted data requires highly intense and detailed work that is beyond the scope of this committee’s work and falls within EPA’s regulatory responsibility. EPA is responsible for determining whether, upon close scrutiny, some or all of the submitted intentional dosing studies in humans on cholinesterase inhibition substantially meet the criteria for validity that the committee has elucidated and yield data of sufficient quality. In doing so, the agency must make clear that the particular response measured in all of these cholinesterase inhibition studies is the critical effect on which RfDs are to be based for each of the pesticides considered and that possible RfDs derived for the same pesticide—based on other findings in animals that are more relevant to chronic effects in humans—are not lower in value. If they are, they should be used rather than the RfD based on cholinesterase inhibition. Such a determination is a critical component of the criteria for scientific validity.
OTHER USES OF DATA
Data from intentional dosing studies in humans often can affect and improve risk assessments in an indirect way. For example, data from pharmacokinetic (PK) or comparative metabolism studies may be used to improve the basis for interspecies extrapolation (e.g., through development of physiologically based PK models)(Andersen, 1995). Data from such intentional dosing studies in humans are generally not used directly in risk assessments—they are not used to replace animal NOAELs, for example—but rather may improve the basis for extrapolation to humans of toxicology data obtained in animals. The increasing use of such studies is often encouraged by EPA (EPA, 2003), and their conduct does not ordinarily present the same ethical issues raised by studies in which potential adverse effects are studied (Chapters 3 and 4). Nonetheless, EPA should ensure that the scientific validity and data quality criteria described by this committee are satisfied before using this type of information in its risk-assessment process.
CONCLUSIONS AND RECOMMENDATIONS
Data from intentional dosing studies in humans that have been developed using the criteria for scientific and ethical validity elucidated in this report—and that are shown upon review to be of adequate quality—can be used within the framework for risk assessment developed by EPA. Use of such data will allow the elimination of the uncertainty factor (UFA) ordinarily used to extrapolate from animals to humans of average sensitivity. Other uncertainty factors and the safety factor called for under FQPA to protect children are in no way affected by the use of data from the intentional human dosing studies conducted to date. It is possible that some types of metabolism and pharmacokinetics studies, together with studies of effects on the critical biomarkers, could be pertinent to the UFH if a sufficiently broad population were studied.
To review data submitted from intentional dosing studies for regulatory decision-making purposes (e.g., setting standards), EPA should ensure the availability of sufficient and appropriate in-house expertise of at least the same level that exists for review of animal studies. The results of scientific review of data for regulatory purposes and their use in standard setting should be communicated to the Human Studies Review Board, recommended in Chapter 6. It is the committee’s view that the Human Studies Review Board is advisory only and would not serve as a replacement for the scientific review EPA must perform in making regulatory decisions.
Recommendation 7-1: Review of Scientific Data
EPA’s use of data from third-party intentional human dosing studies involving cholinesterase inhibition is advisable only if the agency undertakes a thorough review of the data (of the type typically undertaken for submitted animal studies and informed by external peer review) and finds that the studies substantially meet the scientific and ethical standards elucidated in this report. If the studies are found to be scientifically and ethically satisfactory, EPA should use the data to establish RfDs.
For those cholinesterase inhibitors that have been thoroughly investigated in high-quality animal studies (including studies of developmental neurotoxicity), and for which it is clear that cholinesterase inhibition is the most sensitive indicator of toxicity, data from intentional human dosing studies may be considered for use in risk assessment. It should be recognized that these circumstances—in which the most sensitive indicators of toxicity are the acute biological effects of chemicals and in which such effects are readily measurable in ethically acceptable human studies—are likely to be highly unusual. The committee’s recommendations regarding the cholinesterase inhibition studies are thus not expected to suggest many other cases in which dosing studies in humans to establish a NOAEL will be of value and justifiable. The committee’s recommendations regarding study justification, in which proponents of intentional dosing studies in humans must document that the endpoints to be measured are the critical determinants of risk, represent a substantial hurdle.
Recommendation 7-2: Use of Existing Cholinesterase Inhibition Studies
The cholinesterase inhibition studies that already have been submitted to EPA, if determined to be scientifically valid and justified for EPA’s regulatory purposes, may be considered for use in risk assessment and standard setting if they were not unethically conducted (see Recommendation 5-7).
As discussed in Recommendation 3-1 (Chapter 3), under stringent conditions data from intentional dosing studies in humans can be used within EPA’s risk-assessment framework. Use of such data may eliminate or modify the 10-fold default uncertainty factor (UFA), ordinarily used to extrapolate from animals to humans of average sensitivity. The safety factor called for under FQPA to protect children will not be affected by the use of data from intentional dosing studies in humans that address the interspecies uncertainty factor.
Recommendation 7-3: Eliminating or Replacing the Interspecies Uncertainty Factor
In considering the use of data from the cholinesterase inhibition studies already submitted to EPA, the agency should clearly communicate to all stakeholders that information used to eliminate the interspecies uncertainty factor (UFA) will have no influence on the use of other uncertainty factors or on the use of the safety factor protecting children as required by FQPA.
Several critical questions remain regarding the use of data from intentional dosing studies in humans. Studies that reveal “no-effects” of any type at any doses used (so-called NOEL-only studies) may provide some data regarding safety, but they are inadequate for use in deriving RfDs or any other formal measure of human protection because they provide no assurance that the study was capable of detecting the effect of interest. Such data should be used only if there are no other data available and there is a compelling public health need to derive a tentative measure of public health protection. Moreover, the relationship between the presumed sensitivity of the study population and the presumed sensitivity of average humans is somewhat ambiguous and needs clarification. Thus, it is not completely clear that the individuals that are the subjects of intentional dosing studies are always “individuals of average sensitivity” and that they are not less sensitive than the “average” individual. Uncertainties regarding these relationships may be dealt with by a requirement for study replication in a different setting or by the use of an uncertainty factor for intraspecies extrapolation (UFH) that is somewhat greater than the usual factor of 10.
Recommendation 7-4: Data from NOEL-Only Studies and the Sensitivity of Study Populations
EPA should reject data from NOEL-only studies for risk assessments if the NOEL is defined as the absence of any biological response, because such studies do not show levels that give rise to an effect (the LOEL [lowest observed effect level]). Such studies provide no assurance that they were adequate to detect the effect of interest. The agency also should consider whether the uncertainty factor used for intraspecies variability (UFH) should be increased to deal with the possibility that study participants may be of less than average sensitivity. A request for study replication also should be considered as a way to address this last issue.
This chapter provided a description of the risk-assessment framework, and its basis, as the starting point for the committee’s evaluation. The committee then examined the questions of when and how data from intentional dosing studies should be incorporated into EPA’s risk-assessment process. There is substantial precedent at EPA for using such data in risk assessment. Direct use of such human data would eliminate the need for introducing the uncertainty factor ordinarily used to extrapolate from animal data to humans (UFA). However, even if the UFA were to be replaced with data from intentional dosing studies in human, the use of this data would have no effect on the other UFs typically used in deriving RfDs or other criteria for health protection. More specifically, the safety factor introduced under FQPA to protect children would in no way be affected by the replacement of UFA with actual data.
To review data submitted from intentional dosing studies for regulatory decision-making purposes (e.g., setting standards), EPA should ensure the availability of sufficient and appropriate in-house expertise, at least at the level that exists for review of animal studies. The results of scientific review of data for regulatory purposes and its use in setting standards should be communicated to the Human Studies Review Board, recommended in Chapter 6.
For those cholinesterase inhibitors that have been thoroughly investigated in high-quality animal studies (including studies of developmental neurotoxicity), and for which it is clear that cholinesterase inhibition is the most sensitive indicator of toxicity, data from intentional human dosing studies may be considered for use in risk assessment. It should be recognized that these circumstances—in which the most sensitive indicators of toxicity are the acute biological effects of chemicals and in which such effects are readily measurable in human studies involving minimal risk—are likely to be highly unusual. In considering the use of data from the cholinesterase inhibition studies already submitted to EPA, the agency should clearly communicate to all stakeholders that information used to eliminate the interspecies uncertainty factor (UFA) or to replace it with a different factor will have no influence on the use of other uncertainty factors or on the use of the safety factor protecting children as required by FQPA.
Several critical questions remain regarding the use of data from intentional dosing studies in humans. Uncertainties regarding these relationships may be reduced by a requirement for study replication in a different setting or by the introduction of a UFH somewhat larger than the usual value of 10.
Andersen, M. E. 1995. Development of physiologically based pharmacokinetic and pharmacodynamic models for applications in toxicology and risk assessment. Toxicology Letter 79:35-44.
Dourson, M. L., S. P. Feltar, and D. Robinson. 1996. Evolution of science-based uncertainty factors. Regulatory Toxicology and Pharmacology 24:108-120.
Environmental Protection Agency (EPA). 1999. Policy and Procedures on Protection of Human Research Subjects in EPA Conducted or Supported Research, EPA Order.
EPA. 2000a. Integrated Risk Information System. Office of Research and Development. National Center for Environmental Assessment. Available at www.epa.gov/iris.index.html.
EPA. 2000b. Benchmark Dose Technical Guidance Document (External Review Draft). Risk Assessment Forum. EPA/630/R-00/001.
EPA. 2002. A Review of the Reference Dose and Reference Concentration Processes: Final Report. EPA/630/P-02/002F.
EPA. Risk Assessment Forum. 2003. Draft Final Guidelines for Carcinogen Risk Assessment (External Review Draft). Risk Assessment Forum. EPA/630/P-03/001A, NCEA-F-0644A.
Faustman, E. M., and G. S. Omenn. 2001. Risk Assessment. In: Klaassen, C.D. ed. Casarett and Doull’s Toxicology: The Basic Science of Poisons. 6th ed. New York: McGraw Hill.
Lehmann, A. J., and O. G. Fitzhugh. 1954. 100-fold margin of safety. Quarterly Bulletin of the Association of Food and Drug Officials of the U.S. 18:33-35.
National Research Council (NRC). 1983. Risk Assessment in the Federal Government: Managing the Process. Washington, D.C.: National Academy Press.
NRC. 1994. Science and Judgment in Risk Assessment. Washington, D.C.: National Academy Press.
Rodricks, J. V., D. W. Gaylor, and D. Turnbull. 2001. Quantitative extrapolations in toxicology. In: Hayes, A. W., ed. Principles and Methods of Toxicology, 365-386. London: Taylor and Francis.