Design of animal carcinogenicity studies for goodness-of-fit of multistage models.

The effect of bioassay design changes on the variability of risk estimates in the experimental dose region is investigated. Three-dose designs with a control group and other "usual" designs utilizing 200 animals are studied in detail. Constraints on the minimum power of the linear trend test in proportions are used to eliminate designs with unacceptable power. An acceptable design region is compared to designs aimed at improving the low dose extrapolation. A group of designs acceptable for a range of objectives is proposed.