Analysis of a random sample of 2-year carcinogen bioassays from the NCI data base

We take as our primary goals the introduction of novel, methods of analysis that have the potential to control error rates and reconcile results of multiple studies of the same chemical. We develop a selection rule to pick, for each study, a tumor site/type combination for analysis, and then apply random effects models to these study-specific summaries. These models combine evidence over studies; producing a chemical-specific assessment of carcinogenicity. We declare 14 of the 25 chemicals "carcinogens," and find that female mice are the most sensitive sex/species combination. We discuss the benefits and drawbacks of our method and outline developments necessary for their incorporation into the risk assessment process.