Statistical methodology used in analyses of data from DOE experimental animal studies

This document describes many of the statistical approaches that are being used to analyze data from life-span animal studies conducted under the Department of Energy experimental radiobiology program. The methods, which are intended to be as informative as possible for assessing human health risks, account for time-related factors and competing risks, and are reasonably comparable to methods used for analyzing data from human epidemiologic studies of persons exposed to radiation. The methods described in this report model the hazard, or age-specific risk, as a function of dose and other factors such as dose rate, age at risk, and time since exposure. Both models in which the radiation risk is expressed relative to the baseline risk and models in which this risk is expressed in absolute terms are formulated. Both parametric and non-parametric models for baseline risks are considered, and several dose-response functions are suggested. Tumors in animals are not always the cause of death but instead may be found incidentally to death from other causes. This report gives detailed attention to the context of observation of tumors, and emphasizes an approach that makes use of information provided by the pathologist on whether tumors are fatal or incidental. Special cases are those in which all tumors are observed in a fatal context or in which all tumors are observed in an incidental context. Maximum likelihood theory provides the basis for fitting the suggested models and for making statistical inferences regarding parameters of these models. Approaches in which observations are grouped by intervals of time and possibly other factors are emphasized. This approach is based on iteratively reweighted least squares and uses Poisson weights for tumors considered to be fatal and binomial weights for tumors considered to be incidental