Combining different phases in the development of medical treatments within a single trial.

In the development of medical treatments, identification of promising therapies and inference on selected treatments are usually performed in subsequent separate trials. An adaptive two-stage design is proposed for the situation of multiple treatments to be compared with a control, allowing integration of both steps within a single confirmatory trial controlling the multiple level alpha. After the interim analysis, the trial may be terminated early or is continued with a second stage, where the set of treatments may be reduced due to lack of efficacy or to safety problems. The procedure is highly flexible with respect to the distributional assumptions, stopping rules and selection criteria and allows a completely free recalculation of the sample size for the second stage. Simulations show that the method may be substantially more powerful than classical one-stage multiple treatment designs with the same total sample size. As in conventional strategies with a series of separate experiments, a reasonable selection strategy has to be applied in order to prevent proceeding with non-optimal treatments. Copyright 1999 John Wiley & Sons, Ltd.