|
Abstract:
|
In clinical trials, the treatment effect may vary with subgroup, e.g., demographic factor or subject baseline diagnostic category. It is recommended by ICH E9 and FDA guidance for industry (e.g., Diabetes Mellitus) to examine such interaction. The commonly used omnibus interaction test looks into all possible contrasts of treatment effect, thus leading to inefficient utilization of Type I error on contrasts of little clinical interest. We propose an interaction test with clinically more meaningful contrasts based on average treatment effect. The Type I error is spent on focused contrasts, so the test is more powerful to detect interactions that facilitate further understanding of the drug under development. The proposed approach can be extended to the case of investigating the interaction between treatment and certain covariates. Examples are given to illustrate our proposed approach.
|
