Time Trend in Historical Controls for Tumour Incidences in Long‐Term Animal Studies

In the analysis of neoplastic lesions in long‐term animal studies it is often desirable to incorporate historical evidence provided by the control groups. A model by Dempster and co‐workers is used for this and is extended to include a time trend term. A full Bayesian analysis is performed, based on multidimensional numerical integration. A data set (liver tumours in male mice) is then presented and analysed which shows a pronounced increase in tumour incidence over time. Although Dempster's model includes an error term to deal with the between‐study variation, it does not include a term which handles possible time trends (shifts) in the incidence of lesions. Historical control data collected over several years at Ciba–Geigy do, however, show such shifts for liver tumours in mice. Dempster's model can easily be extended to handle such time trends. However, rather than to rely on the approximate normality of the posterior density, it is believed to be a safer way to use multidimensional numerical integration although it is more expensive in terms of computational effort. When the integrations are arranged in a clever way, an analysis of data by this technique can be performed in a reasonable time by currently available computers.