Subgroup hypotheses: clinician's friend or researcher's fiction?

Researchers and clinicians may be tempted to consider apparent differences in effects of interventions in subgroups of study populations for the purposes of individualizing care or directing subsequent focused research efforts. Such "subgroup hypotheses" have a high propensity of being artifacts of chance and are rarely supported when stringent criteria are applied. Subgroup hypotheses emerging from single trials should rarely if ever be considered for application. This article describes published criteria for evaluating the believability of a subgroup hypothesis and presents a conceptual framework for understanding the principles underlying those criteria. Examples of both failed and validated subgroup hypotheses drawn from emergency medicine research are given.