The Importance of Data on Mechanism of Carcinogenesis in Efforts to Predict Low‐Dose Human Risk 1

The literature reviewed by Krewski et al. identifies validity problems associated with using the limited data from rodent bioassays in efforts to assess carcinogenic risk to humans. Krewski et al. discuss the debate as to whether the good correlation of carcinogenic potencies found between rats and mice should be interpreted as a justification for quantitative extrapolation from rodents to humans. In 1985, Bernstein et aZ.(l) showed that the observed correlation is largely artifactual, as follows: For chemicals that test positive in rodent bioassays, potency estimates based on the one-hit model are constrained to a narrow range surrounding the high dose tested, the maximum tolerated dose (MTD) (unless all dosed animals develop tumors, which rarely occurs). Over large numbers of chemicals, the MTDs for rats and mice are highly correlated and span many orders of magnitude. Hence, the potency correlation between rats and mice follows statistically. A debate followed in several papers. In their review, Krewski et al. report that potency and MTD are highly correlated, regardless of whether the potency estimate uses the one-stage, multistage, or Weibull model. Freedman, Gold, and Stone (2 ) recently examined how much of the observed correlation in potencies between species is artifactual. Our analysis involved two statistical models where the impacts of various assumptions could be calculated. In effect, the first model assumes that interspecies correlation of potencies is purely artifactual; it ignores the correlation between rats and mice of (potency x MTD), which is a rough measure of tumor yield. The second