Statistical modeling of animal bioassay data with variable dosing regimens: example--vinyl chloride.

Abstract

We consider animal bioassay experiments with variable dosing regimens in which groups of animals are dosed beginning at different ages and for varying durations. Two response models are discussed and then applied to data from an experiment on vinyl chloride exposure of F-344 rats, B6C3F1 and Swiss CD-1 mice, and Syrian Golden hamsters. The multistage model of Armitage and Doll, as extended by Whittemore, Day and Brown, and Crump and Howe, is used to estimate the dose effect on the ordered stages of tumor development. The data for all endpoints and species/strains examined consistently indicate a predominant effect on the first stage, suggesting that vinyl chloride is primarily a tumor initiator. This is consistent with evidence from two-stage experiments on this chemical. The second response model, new to this article, adjusts for survival nonparametrically. It is used to test for an age difference in susceptibility, to evaluate alternative exposure durations, and to compare the effectiveness of alternative dosing regimens for detecting carcinogenicity.