January 26, 2009January 26, 2009

Editor's CorrespondenceCOMMENTS AND OPINIONS

Arch Intern Med. 2009;169(2):199-205. doi:10.1001/archinternmed.2008.567

Sequist et al1 present interesting findings on racial variations in intermediate outcomes for diabetes among patients seen in one integrated group practice. The authors are to be lauded for assessing measures of control rather than the usual suspects of tests performed. However, even though the authors point out that partitioning variation into between-physician and within-physician components “may not capture the full spectrum of explanatory factors related to differential outcomes within a physician's panel . . . ” and “ . . . should not be used to assign sole responsibility to an individual physician . . . ,”1(p1146) I believe that the authors proceed too far in that direction in their conclusions, perhaps too easy to do when discussing the rubric of “within-physician variation.” Statistically, what the authors refer to as “within-physician variation” is all the difference between black and white patients that is not explained by measured patient characteristics (with sociodemographic measures obtained from administrative data being rather blunt instruments for trying to understand such factors as personal behavior and possible genetic effects) and nonrandom, between-physician differences; ie, “within-physician” means everything not otherwise explained. This does not mean either that this portion of the racial disparities is explained by the physicians or that physicians are necessarily the best target for programs to ameliorate these examples. In fact, the observation that between-physician differences accounted for very little of the observed differences suggests that, unless one hypothesizes that all of the studied physicians were consciously or unconsciously discriminating against their black patients, the disparities likely have their origins in factors not related to their physicians. As an analogy, differences in diabetes outcomes between black and white patients that are consistent across states and not explained by measured individual factors could be called “within-state variation,” but it would likely be erroneous to conclude that the states were substantially responsible for the differences or the best targets for interventions.