Physician services mix and physician specialty: an application of fuzzy clustering

Underlying proposals to alter the specialty distribution of the U.S. physician workforce is the tacit assumption that physicians can be unambiguously categorized as either primary care or nonprimary care providers. The ways in which physicians provide care, however, are far more complicated than what a simple dichotomy suggests. This study employs a fuzzy c-means clustering methodology to assign physicians into multiple specialties based on the actual services provided to their patients. The results of this study suggest that there is a reasonable structure to the data. The PC and S validity functions yield somewhat conflicting best solutions, however. The S function indicates the 2-partition as the best solution, while the PC criterion indicates the 3-partition. There is also some evidence that the 8-partition captures the substructure present in the data. Further examination of the partitions for c=2, c=3, and c=8 is necessary to determine which of these solutions yield the most meaningful result.