Maximum likelihood estimation of the prevalence of nonlethal neoplasms--an application to radon-daughter inhalation studies

A nonparametric maximum likelihood method for estimating prevalence is described that is applicable to the analysis of nonlethal tumors which are discovered incidentally in either sacrificed animals or animals dead from other causes. The method corrects for competing risks and does not require an analytical model for the prevalence as a function of dose and time. It is applied to the results of an experiment in which numerous groups of Sprague-Dawley rats were exposed to different doses of radon daughters at diffeent dose rates. The dependence of the prevalence on dose and time after exposure is derived, and three basic models are considered that correspond to a dose-dependent shift in time, to an acceleration in time, and finally to the proportional hazards model. Mortality-corrected risk estimates are derived from the estimated prevalences. At doses down to 65 WLM (working level months) the results are consistent with linearity in dose, or possibly with sublinearity (dose exponent less than 1); they exclude, in this dose range, a threshold or a proportionality to a higher power of dose