STATISTICAL EXTRAPOLATION METHODS FOR ESTIMATING RISKS FROM ANIMAL DATA

The statistical methods of predicting human risk based upon laboratory carcinogenesis data typically involve the estimation of a low dose response in the experimental animal. This estimate is then extrapolated from the laboratory animal to man. Clearly, this process is extremely tenuous and depends upon many assumptions, some of which may be quite unjustified. However, the hope is that these necessary risk estimates can be made within an order of magnitude or so. That phase of risk estimation which has received the greatest attention has been the estimation of low dose response from experimental data generated at much higher dose levels. The reason for this attention has been that, compared to species extrapolation, considerably more experimental data has been available; and the problem, by no means simple, seems on the surface to be easier. The problem of low dose estimation comes down to the appropriate mathematical descriptions of a dose response curve and statistical problems associated with its estimation. Further, studies on the effects of the resulting low dose estimates by the model choice are important. In this discussion I will briefly review various types of dose-response models and indicate some current problems that are being investigated.