Matching on provider is risky

Objectives: To illustrate that matching on provider may exacerbate, not remove, bias. Study Design and Setting: The degree of confounding bias depends in part on the proportions of treatment variation that can be ascribed to confounders and to instruments, respectively. This commentary raises the specific example of bias by matching on hospital induced in a study of coronary artery bypass graft surgery patients and illustrates the effect of matching on provider in a constructed example. Results: Matching on provider removes a "benign" source of treatment variability, leaving unmeasured confounders as potentially the most important determinants of treatment. Conclusions: Researchers need to articulate the presumed source of pseudorandom variation in observational studies and need to take care not to reduce their effect through unnecessary control. (C) 2013 Published by Elsevier Inc.