Simpson's paradox and clinical trials: what you find is not necessarily what you prove.

Abstract

Expensive clinical trials have become the gold standard for evaluating the efficacy of promising new therapeutic agents. Full exploration of the collected data is routine to maximize the yield of the information available. However, potential methodologic flaws in these extensive analyses may not be appreciated by investigators or readers. One such problem with subgroup analyses is discussed, using hypothetical examples and data from a recently completed clinical trial of brain resuscitation as illustrations.