The use of historical control information in modelling dose response relationships in carcinogenesis

An empirical Bayes approach is proposed as a means of utilizing historical control information in modelling the relationship between the level of exposure to the test agent and the rate of tumour occurrence in carcinogenicity bioassays. Using a simple two-parameter linear logistic model, it is shown that when the historical control response rates are tightly clustered around the response rate in the concurrent control group, the empirical Bayes estimator can be considerably more precise than the usual maximum likelihood estimator ignoring the historical data. Efficiency gains were also noted with respect to estimates of quantiles of the dose response curve such as the ED10. With a three-parameter logistic model in which spontaneously occurring lesions are assumed to be stochastically independent of those caused by the test agent, only modest gains in efficiency using historical control data were observed.