Physician-level variation in practice patterns in the VA healthcare system

The objective of this study was to determine whether there is enough variation in utilization of medical services which can be reliably attributed to primary care providers to justify practice profiling. A total of 221 primary care providers caring for 81,775 patients with a variety of medical conditions in eight Veterans Affairs medical centers and their associated clinics during calendar year 1999 were used for the study. After controlling for case-mix variation, the physician-level variation in utilization among providers who had at least 75 patients was evaluated. Adjusted Clinical Groups (ACGs) were used for case-mix adjustment. Physician effect was measured by the intraclass correlation coefficient (ICC). Patient-level utilization outcomes were total hospital bed days of care, laboratory costs, radiology costs, and pharmacy costs for each patient during calendar year 1999. The physician effect (ICC) ranged from .04 for bed days to .12 for laboratory costs. The reliabilities of the physician effect estimates computed using the Spearman-Brown prediction formula ranged from .75 for bed days to .91 for laboratory costs for providers with 75 or more patients. The profiling methodology reliably detected practice differences among providers in all utilization outcomes studied. However, the value of profiling as a means of improving quality of care or efficiency is yet untested.