Upper-Bound Estimates of Carcinogenic Risk: More Policy Than Science

Abstract

A major obstacle to the development of reliable quantitative risk estimation models which describe the potential human carcinogenicity of chemicals is the problem of extrapolation between species. One approach has been to define a standardized parameter, carcinogenic potency, which could be used to compare the risks associated with exposure to different chemicals. The Environmental Protection Agency uses a conservative estimate of carcinogenic potency, the upper-bound, rather than the corresponding maximum likelihood estimate. The data presently used to define both the maximum likelihood and upper-bound potency estimates are obtained primarily from high dose, chronic rodent bioassays. These tests provide little data regarding mechanism of action, and the dose range per chemical is usually narrow. Others have demonstrated that a strong correlation exists between maximum likelihood potency estimates for rats and mice which developed tumors during chronic exposure to the same chemical. This has been used to justify the extrapolation from these data to human risk. The high correlation, however, is an artifact caused by two features of the experimental design: (i) the small number of animals per dose group in each study and (ii) the high correlation between the maximum doses tested in rodents for each chemical. We have illustrated this by showing that the correlation between the maximum likelihood estimates of carcinogenic potency for rats and mice which did not develop tumors is as large as the correlation for those that did (r=0.79, n=82 and r=0.83, n=83, respectively). The correlation is also strong between rats developing tumors and mice not developing tumors (r=0.73, n=76), and mice developing tumors and rats not developing tumors (r=0.76, n=93). Correlations of upper-bound carcinogenic potency estimates for positive and negative experiments are of similar magnitudes. These high correlations are inevitable because the maximum likelihood and upper-bound risk estimates are inversely related to the maximum dose tested. Rather than being plausible estimates of carcinogenic risk, these indices are merely convoluted safety factors. The use of chronic bioassay data to determine such carcinogenic potency indices should be replaced with risk estimation models that can utilize mechanistic information.