Propensity Score Matching: The ‘Devil is in the Details’ Where More May Be Hidden than You Know

Propensity score matching has been used with increasing frequency in the analyses of non-prespecified subgroups of randomized clinical trials, and in retrospective analyses of clinical trial data sets, registries, observational studies, electronic medical record analyses, and more. The method attempts to adjust post hoc for recognized unbalanced factors at baseline such that the data once analyzed will hopefully approximate or indicate what a prospective randomized data set-the "gold standard" for comparing two or more therapies-would have shown. However, for practical limitations, propensity score matching cannot assess and balance all the factors that come into play in the clinical management of patients and that may be present in the circumstances of the study. Thus, propensity score matching analyses may omit, due to nonrecognition, the effects of several clinically important but not considered factors that can affect the outcomes of the analyses being reported, causing them to possibly be misleading, or hypothesis-generating at best. This review discusses this issue, using several specific examples, and is targeted at clinicians to make them aware of the limitations of such analyses when they apply their results to patients in their care.