Statistical aspects of the estimation of human risks.

The preceding paper has discussed the general problem of high dose to low dose extrapolation within a single animal species. The purpose of this extrapolation is to estimate the effects of low level exposure to toxic agents known to be associated with undesired effects at high dose levels. Mathematical models of dose response are necessary for this extrapolation process since the low dose effects, expected to be on the order of response rates of 10(-6, are too small to be accurately measured with limited study sample sizes. A number of mathematical dose-response models have been proposed for extrapolation purposes; we have shown how similar they can appear to one another in the range of observable response rates, yet how different they become at lower, unobservable response rates, the region of primary interest. This is the single, most important limitation of this extrapolation methodology. An estimate of risk at a particular low dose, or an estimate of the dose leading to a particular level of risk is highly dependent upon the mathematical form of the presumed dose response and differences of 3 - 4 orders of magnitude are not uncommon. Therefore, all these sources of uncertainty, dose-response model, pharmacokinetic behavior of the toxic agent, thresholds, heterogeneity, and patterns of exposure, lead to substantial uncertainties in estimates of risk based on high to low dose extrapolations.