Use of historical marker data for assessing treatment effects in phase I/II trials when subject selection is determined by baseline marker level.

Although the primary focus of Phase I clinical trials is to assess clinical pharmacology and possible toxicities, any information on the potential effect of treatment would be useful in helping to determine priorities between treatments for further study. We consider the scenario where data are routinely collected on a marker of disease progression on all patients attending a clinic, but the trial is restricted to patients who have a marker level within a defined range at study baseline. Using a two-step approach to estimation, the marker histories are used to give predictions of marker values during trial follow up, assuming no treatment effect and adjusted for the regression to the mean effect, for those subjects selected for the trial. Comparison between the observed responses and the predicted values then forms the basis for moment-based estimation and hypothesis testing for the treatment effect. The method is easily extended to compare summary measures for repeated measurements during follow up in the trial to predicted summary measures. An example using CD4 cell counts in an AIDS study is given.