Abstract:
|
Inference for the overall treatment effect resulted from a randomized clinical trial has the best credibility. In a traditional paradigm, where a clinical trial is primarily designed to answer a single question regarding the average (overall) treatment effect, randomization effectively prevents systematic confounding and bias when we assess the overall treatment effect. In an increasingly important new paradigm, inference of treatment heterogeneity is key to comparative effectiveness. In this new paradigm, randomization does not prevent systematic bias when we assess treatment by covariate interaction. Through a case study, this paper explains how this phenomenon occurs and examines whether and how this problem could be solved.
|