January 26, 2009January 26, 2009

Editor's CorrespondenceCOMMENTS AND OPINIONS

Arch Intern Med. 2009;169(2):199-205. doi:10.1001/archinternmed.2008.568

We wanted to raise a concern about the article published in the Archives titled “Physician Performance and Racial Disparities in Diabetes Mellitus Care.”1 One of the main conclusions that the authors draw is that very little of the observed disparities in intermediate outcomes among black and white patients with diabetes is mediated by physician-level factors. Although that conclusion may ultimately be correct, we believe that the analytical methods applied by this study reflect an inappropriate application of random effect models. When representing physician-level influences as a random effect, one must assume that the random effects are statistically independent of the regressors in the model. However, if black patients are systematically under the care of lower-performing physicians, the black regressor and the unobserved physician effect will automatically be correlated. This hypothetical situation would be entirely masked by the random effects approach, and a potentially important source of between-physician variance in the outcome would be lost in the analysis.