Uncertainty estimates for low-dose-rate extrapolations of animal carcinogenicity data.

We explore the statistical uncertainty in low-dose-rate extrapolations of animal carcinogenicity data and state some general conclusions about its magnitude. Our conclusions are based on analyses of both actual data sets from animal experiments and a large number of data sets generated by computer simulation. Our results, which should be applicable to continuous exposure to any “direct-acting” carcinogen, indicate that at low dose rates upper statistical confidence limits on added risk are almost certain to be approximately linear in the dose rate, even when the data trend in the observable range is not linear. When the test data show a definite linear trend, both upper and lower confidence limits on the added risk should be approximately linear and should differ by no more than a factor of about 10 for experiments involving several dozen tumorbearing animals. When the trend does not have a definite linear component, the upper and lower confidence limits on added risk can be expected to differ much more widely; only the upper confidence limits will then be stable and meaningful. The upper confidence limits produced by several different mathematical models are all likely to be much the same, although the lower confidence limits may differ widely. These conclusions should apply as long as the dose-response curve is not forced to have a zero slope at zero dose. In other words, if linearity of the dose-response curve in the low dose range is not ruled out by assumption, it is most unlikely that it can be ruled out empirically on the basis of high-dose-rate animal data.