Subgroup analysis
*Janet Wittes, Statistics Collaborative 


Subgroups have long been the bane of many statisticians in clinical trials. We cringe when we hear clinicians say, "I treat patients, not means." We, like they, know that different patients respond differently to the same drug, but we do not know how to predict reliably who will, or won't, respond. So we demand strong evidence of differential subgroup effects to conclude that a subgroup is more (or less) responsive than other subgroups. We warn about the landscape of clinical trials that chased, unproductively, a promising subgroup. But in the light of global trials and targeted therapies, is it time to reassess our skepticism? Or, more narrowly, are there situations where our traditional stance is not useful? This talk provides an overview of methods that aim to identify more reliably subgroups with responses to therapy that differ materially from each other.