Observational etiologic research

Abstract

S ome conceptual features of case-control studies are described elsewhere and shown to be essential for answering the etiologic questions these studies address. Most importantly, the exposure history in a series of case patients is to be compared to the exposure history that would have been expected if exposure did not cause any of these cases. This expected value is estimated in control patients, fairly selected from the source population of the cases. Here we will extend the discussion to confounding in case-control studies and proper adjustment for confounding, which is essential for addressing etiologic questions. Proper appreciation of the purpose of a control group in case-control studies will elucidate some of the fundamental misconceptions surrounding matching. In casecontrol studies matching cannot prevent confounding but instead introduces a selection bias that behaves as a special form of confounding. This is the opposite of what happens in cohort studies, where matching does remove confounding. The difference arises from the fundamental difference between making exposed and unexposed subjects more comparable (i.e., matching in a cohort study) and making case patients and control subjects more comparable (i.e., matching in a case-control study). The former is desirable because it facilitates a comparison of differences induced solely by the difference in exposure, while the latter is undesirable because it hampers the objective of finding the factor that caused case patients to be different from control subjects (i.e., the etiologic factor that caused the disease).