Abstract

Abstract The cause-of-death test (Peto 1974; Peto et al. 1980) provides a test for carcinogenicity without requiring extreme lethality assumptions. The approach has been criticized, however, because cause-of-death determinations require subjective decisions by pathologists and may be unreliable. Using a missing data formulation, we derive a simple, intuitive modification of the test that allows for misclassification of cause of death. The modified statistic is a function of the misclassification probabilities. A sensitivity analysis may be performed using a plausible range of values for these probabilities or, in certain situations, the misclassification probabilities can be estimated from the data. Both approaches are applied to data from the ED01 experiment.