Estimating tumor incidence rates in animal carcinogenicity experiments

Tumor incidence is the primary measure of carcinogenesis. This article focuses on estimating time-dependent incidence rates in animal experiments with few sacrifices. When the context of observation is known for none or all of the animals dying with the tumor of interest, previous results are obtained under relaxed assumptions. The link with existing semiparametric and nonparametric procedures based on latent failure times is exploited by using these methods to compute maximum likelihood estimates of the incidence rates without introducing latent random variables. Nonparametric estimators that are appropriate when all contexts of observation are known are generalized to the case in which the contexts of observation are unknown for a subset of the tumor-bearing animals.