Risk Assessment Methods

Risk assessment is a means to estimate the probability and possible magnitude of a health response associated with a given exposure. For carcinogens, methods of using available data to estimate risk have been relatively well delineated (EPA, 1984, 1986a,b; OSTP, 1984), although validation of the estimates (and of the methods generally) is still seriously incomplete. Animal bioassays are undertaken at rather high doses, where the response (if any) is assumed to be more frequent than at low doses and hence more likely to be observed in a small group of animals. It is then necessary to adopt some mathematical model of the dose-response curve and to extrapolate the response observed at high doses to an estimated response at exposures of interest. The linearized multistage model is widely used to estimate cancer risks associated with environmental exposures (EPA, 1987) and is said to provide an upper-limit estimate of low-dose response. To some degree, the model's wide use reflects its mathematical flexibility. However, biologic support for the assumption of linearity at low doses remains largely inferential and probably wrong in a high proportion of cases (Bailar et al., 1988).

The NRC report on complex mixtures (1988) concluded that, at doses for which the relative risk is less than 1.01 and under the assumptions of the multistage model, the excess risk associated with exposure to several carcinogens can be estimated by adding the excess risks associated with each carcinogen. That report defined a low dose as one associated with an excess risk no more than 1% and with a small relative risk of cancer in the exposed group (no more than 1.01). That result was also inferred to hold for several additional dose-response models used to estimate cancer risks, such as the one-hit, Moolgavkar-Knudson, linear, and multiplicative models. These inferences are not, however, validated by direct evidence.

Under most circumstances, one would like to use the dose at the target organs as the input to dose-response models. That would increase the precision and could decrease the bias of any risk estimate (Krewski et al., 1987; Whittemore et al., 1986); however, the lack of validated pharmacokinetic models to represent mixtures impedes this effort. Further development of pharmacokinetic models should make possible their application to the individual components of a mixture. For materials activated, say, in the liver or detoxified in the kidney, dose at the liver or the kidney might be most important.

The NRC report (1988) also described risk assessment methods for using data from epidemiologic studies, but noted that the problems presented by human heterogeneity, the potential for bias, the lack of a uniform study design, and the variability in data quality lead to a flood of risk assessment methods, as well as uncertainty about results. Studies based on occupational exposures, which are rarely of known magnitude but can be significantly higher than those of the general population, also present some extrapolation problems. The dose-response models used with these data are often among those for which response additivity is presumed to hold at low doses.

This chapter is related to EPA's existing guidelines for carcinogen risk assessment (EPA, 1986a), inasmuch as the methods suggested here could be applied by EPA in conjuction with those guidelines. Understanding of the biology underlying carcinogenic mechanisms is rapidly evolving and already raises some questions about the need to improve models and assumptions or substitute alternative ones. The adoption of alternative models would necessarily require reexamination of the conclusions given here.

The MCLs for the VOCs and other carcinogens in drinking water are generally below those associated with a 10^-3 excess risk (see Tables 6-1 and 4-3)—thus they satisfy the first criterion for defining low dose. For doses associated with excess risks greater than 10^-3 if the NRC approach is correct, synergism might be important; in such cases, however, the dose of a single carcinogen itself would also be of concern.

It is more complicated to demonstrate that the second criterion for low dose—namely. a low relative risk—has been met. For many of the VOCs that have been studied, evidence of carcinogenicity is largely from animal studies, vinyl chloride and benzene being exceptions. Except for vinyl chloride, excess tumors are generally seen for the hematopoietic or lymphatic systems. For these lymphomas the background cancer risks for humans are about 0.018 over a lifetime (Schneiderman et al., 1979). Hence, for the relative risks to be less than 1.01, the incremental risks associated with exposure to these VOCs must be less than 1.8 × 10^-4 to satisfy the low-dose criterion; exposure at the MCLs satisfies the criterion. An argument can be made that there is not necessarily a strict correspondence across sites of cancer for rodents and humans; hence, total cancer rate should be considered. Alternatively, one could consider other individual sites or groups of sites in humans that might be related to sites of observed tumors in rodents. In such an approach, the low-dose criterion is satisfied, because environmental concentrations of the individual VOCs need only be such that the risk of cancer associated with individual toxic compounds is less than 0.003 (0.01 × 0.33, which is the probability of developing cancer during one's lifetime for all races and both sexes) (Schneiderman et al., 1979).

Vinyl chloride exposures in humans are associated with an excess of liver angiosarcoma, a relatively rare cancer. Background incidence rates are not available for that cancer, but for total liver cancers, the lifetime incidence for both sexes is 0.0054 (J. Horme, National Cancer Institute, personal communication, 1988). If total liver cancers are considered, satisfaction of the low-dose criterion requires that excess risks associated with the chemical in drinking water be less than 5 × 10^-5. If total cancers are considered, however, the second criterion is much more easily satisfied for vinyl chloride. The apparent paradox is due to the smallness of the background incidence of liver angiosarcoma, compared with the total cancer incidence.

The subcommittee recognizes that the assumption of response additivity at low doses does not have an extensive empirical foundation. Rather, it rests on theoretical considerations and observations in limited epidemiologic studies. Regarding higher doses, responses greater than additive occur after human exposure to some mixtures of agents, such as cigarette smoke and asbestos, at concentrations that produce a high incidence of effects separately (NRC, 1988).

Conclusions and Recommendations

The acceptance of an underlying dose-response model allows estimates of lifetime cancer risk to be made for individual carcinogens in drinking water. The kind of models also indicate that the risks associated with simultaneous exposure to two or more carcinogens can be added when their concentrations are low, as they are when drinking water standards are satisfied. Hence, in the models advocated in the current EPA guidelines (EPA, 1986a), any potential for more than additivity is ignored in risk assessments of carcinogens present at low doses. Additional research should be conducted to provide a firmer empirical base for those models, so that the techniques for the risk assessment of mixtures can be improved and refined. The present effort is an early step; periodic reevaluation will be necessary.

Any summation of cancer risk estimates must be made with care, because different sets of assumptions can lead to different results. Different models, for example, have been applied to epidemiologic data and animal toxicologic data. In addition, the exposures in epidemiologic studies, although rarely known with any accuracy, are a fortiori in the range of past human exposures and often much closer to environmental exposures of current concern than are those of animal studies.

If upper-bound risk estimates are used, even more caveats need to be applied in the summation exercise. It is true that the sum of the upper bounds is an upper bound of the sum, but the sum of the upper 95% confidence limits is not such a limit for the sum. Moreover, the calculations of these upper bounds for the multistage model (EPA, 1986a) are heavily influenced by the linearity of the underlying dose-response data. Relatively linear data give rise to relatively tight confidence limits; nonlinear data generally do not, because data at the higher doses become less and less informative about lower doses as curvature over the intervening range of doses increases.

For carcinogens, the most important research needs are those associated with the development of better estimates of dose-response relationships and risks for individual carcinogens in drinking water. Additional research also needs to be directed toward a better understanding of the mechanisms of carcinogenesis and the development of improved risk assessment models that better reflect the underlying biology. The replacement of existing models with improved models could alter the conclusions about the assumption of response additivity at low doses.

References

• Bailar. J. C., E. C. C. Crouch. S. Rashid, and D. Spiegelman. 1988. One-hit models of carcinogenesis: Conservative or not? Risk Anal. 8(4):485-497. [PubMed: 3244857]
• EPA (Environmental Protection Agency). 1980. Chloromethane and chlorinated benzenes proposed test role; amendment to proposed health effects standards. Fed Regist. 45(140):48524-48566.
• EPA (U.S. Environmental Protection Agency). 1984. Proposed guidelines for carcinogen risk assessment. Fed. Regist. 49(227):46294-46301.
• EPA (U.S. Environmental Protection Agency). 1985. Proposed guidelines for the health risk assessment of chemical mixtures. Fed. Regist. 50(6):1170-1176.
• EPA (U.S. Environmental Protection Agency). 1986. a. Guidelines for carcinogen risk assessment. Fed. Regist. 51(185):33992.
• EPA (U.S. Environmental Protection Agency). 1986. b. Guidelines for the health risk assessment of chemical mixtures. Fed. Regist. 51(185):34014-34025.
• EPA (U.S. Environmental Protection Agency). 1987. National primary drinking water regulations—Synthetic organic chemicals: monitoring for unregulated contaminants: final rule. Fed. Regist. 52(130):25690-25717.
• Krewski, D., D. J. Murdoch, and J. R. Withey. 1987. The application of pharmacokinetic data in carcinogenic risk assessment. Pp. 441-468 in Drinking Water and Health, Vol. 8. Pharmacokinetics in Risk Assessment. Washington, D.C.: National Academy Press.
• NRC (National Research Council). 1988. Complex Mixtures: Methods for In Vivo Toxicity Testing. Washington, D.C.: National Academy Press. 227 pp. [PubMed: 25032282]
• OSTP (Office of Science and Technology Policy). 1984. Chemical carcinogens: Notice of review of the science and its associated principles. Fed. Regist. 49(100):21594-21661.
• Schneiderman, M. A., P. Decoufle, and C. C. Brown. 1979. Thresholds for environmental cancer: Biologic and statistical considerations. Ann. N.Y. Acad. Sci. 329:92-130.
• Whittemore, A. S., S. C. Grosser, and A. Silvers. 1986. Pharmacokinetics in low dose extrapolation using animal cancer data. Fundam. Appl. Toxicol. 7:183-190. [PubMed: 3758535]