STATISTICAL METHODS AND APPLICATIONS FOR LABORATORY ANIMAL STUDIES

Environmental Health Sciences

Investigators

Linked publications & trials

Abstract

This project focuses on developing new statistical methods, and applying new and existing statistical techniques, to analyze data from laboratory animal studies. The primary endpoint of interest is the tumor incidence rate, which is the age-specific rate of onset for new tumors. We developed a flexible estimator for occult tumors of unspecified lethality. Age-specific estimates require some sort of time adjustment, but standard survival methods for noninformatively censored data apply only when tumors have no effect on longevity or when they cause death immediately. In general, the censoring of an onset time by a death with the tumor is informative, as the lag between onset and death tends to be shorter than that of a strictly nonlethal tumor and longer than that of an instantly lethal tumor. We express the incidence rate in terms of three component functions that are straightforward to estimate. These simple estimates are combined to give an estimate of the tumor incidence rate. This approach can adjust for explanatory variables, requires only one sacrifice time, and does not make stringent parametric assumptions. We also developed a Bayesian analysis that, in addition to explanatory variables from the current experiment, can incorporate information from prior studies. This approach allows data from historical controls to be formally combined with concurrent data. The relative weight given to the historical data depends on the consistency among the control groups in the past studies. Our use of a Gibbs sampler permits realistic interpretation of the results even when the sample sizes are small or the data are sparse, unlike traditional maximum likelihood procedures. We are also extending this method to simultaneously analyze the incidence rates of tumors at more than one site, in contrast to the customary approach of analyzing each tumor site separately. In this general case, we observe which of several occult tumor types are present when an animal dies. Our new analysis will adjust for survival and lethality, as before, but it will also permit simultaneous inference about the incidence of tumors at multiple sites and will account for within-animal correlations among the onset times.