Evaluating treatments when a gender by treatment interaction may exist.

We propose a two-stage procedure for investigating whether males and females respond differently to treatment. The size of the first stage is based on the assumption of homogeneity of treatment effects across genders. Using stage I, we test for a gender by treatment interaction. If non-significant, we compute an overall average treatment effect and terminate the study. If we find an apparent interaction at the end of the first stage, we consider each gender separately. Because we now need to estimate treatment effects separately for each gender, we may have a need to collect additional information in a second stage. We consider the performance of our procedure for a normally distributed endpoint as well as for a survival model.