On use of the multistage dose–response model for assessing laboratory animal carcinogenicity

We explore how well a statistical multistage model describes dose–response patterns in laboratory animal carcinogenicity experiments from a large database of quantal response data. The data are collected from the US EPA’s publicly available IRIS data warehouse and examined statistically to determine how often higher-order values in the multistage predictor yield significant improvements in explanatory power over lower-order values. Our results suggest that the addition of a second-order parameter to the model only improves the fit about 20% of the time, while adding even higher-order terms apparently does not contribute to the fit at all, at least with the study designs we captured in the IRIS database. Also included is an examination of statistical tests for assessing significance of higher-order terms in a multistage dose–response model. It is noted that bootstrap testing methodology appears to offer greater stability for performing the hypothesis tests than a more-common, but possibly unstable, “Wald” test.