Abstract:
|
This analysis illustrates the use of fixed effects (FE) and random effects (RE) regression models for medical provider profiling using two examples: physician choices between: 1) laparoscopic and open cholecystectomy; and 2) vaginal and cesarean delivery. Data sources include insurance claims, and enrollment and provider data from a large insurer in Rochester, New York. Both RE and FE models were used with logistic regression. The binary dependent variable indicated the use of one of the two possible procedures in the binary choice set. Independent variables included patient and provider characteristics. Upon estimation, FE models accounted for significant amount of the variation in treatment choice. In addition, the FE models were able to produce a case-mix adjusted index of the physician's individual propensity to treat. In contrast, the RE models showed medical training and experience systematically affected choice, while accounting for physician-level heterogeneity in treatment choices. Used in combination, however, these methods offer researchers and healthcare managers a valuable surveillance tool to prioritize provider education and disease management activities.
|