Using historical controls to adjust for covariates in trend tests for binary data

Historical data often play an important role in helping interpret the resul ts of a current study. This article is motivated primarily by one specific application: the analysis of data from rodent carcinogenicity studies. By p roposing a suitable informative prior distribution on the relationship betw een control outcome data and covariates, we derive modified trend test stat istics that incorporate historical control information to adjust for covari ate effects. Frequentist and fully Bayesian methods are presented, and nove l computational techniques are developed to compute the test statistics. Se veral attractive theoretical and computational properties of the proposed p riors are derived. In addition, a semiautomatic elicitation scheme for the priors is developed. Our approach is used to modify a widely used prevalenc e test for carcinogenicity studies. The proposed methodology is applied to data from a National Toxicology Program carcinogenicity experiment and is s hown to provide helpful insight on the results of the analysis.