Predictions of rodent carcinogenicity testing results: interpretation in light of the Lave-Omenn value-of-information model.

The recent National Institute of Environmental Health Sciences/National Toxicology Program Carcinogen Prediction Challenge elicited a valuable array of predictions of the carcinogenicity of chemicals tested in the lifetime rodent bioassay. The data warrant additional analyses of the similarities and differences of the predictive methods. We provide here analyses of the sensitivity, specificity, and false-positive/false-negative tendencies of the different sets of predictions. Our value-of-information model provides guidance to testing agencies and regulatory agencies in determining the social value of additional information and setting up the framework for assessing the social consequences of different test strategies and nontest predictive methods. These considerations deserve attention in the second round of the Carcinogen Prediction Challenge.