Profiling care provided by different groups of physicians: effects of patient case-mix (bias) and physician-level clustering on quality assessment results.

BACKGROUND: Patient characteristics (case-mix bias) and physician-level variation (clustering) are often overlooked in profiling the quality of care provided by different groups of physicians, such as specialties.
OBJECTIVE: To examine the effect of case-mix bias and physician-level clustering on differences in quality of diabetes care between specialty groups participating in the American Diabetes Association's Provider Recognition Program.
DESIGN: Retrospective record review of both process and outcome measures over 1 year and a cross-sectional patient survey. The sample included 29 solo and group practice sites in diverse regions of the United States. Of the 29 sites, 15 were endocrinology sites and 14 were primary care sites. PATIENTS: 1750 adults with diabetes. MEASUREMENTS: Process measures included frequency of hemoglobin A(1c), lipid, and urine protein testing; blood pressure measurement; and foot and eye examinations. Outcome measures included A(1c)level, blood pressure, lipid levels, and patient satisfaction. Patient case-mix variables included age, sex, health status, level of education, ethnic minority status, and duration of diabetes.
RESULTS: Unadjusted differences between endocrinologists and generalists were statistically significant for most process and outcome measures. Inclusion of patient case-mix variables reduced the statistical significance of specialty differences for some quality measures. After accounting for the substantial physician-level clustering, observed differences between specialties were no longer statistically significant for any of the quality measures except patient satisfaction.
CONCLUSIONS: The findings underscore the importance of designing physician profiling studies with sufficient power to account for physician-level variation (clustering) as well as patient case-mix. Studies that are not designed with both sufficient numbers of physicians and patients per physician may distort differences in quality of care between physician groups.