Bayesian inference for a two-part hierarchical model: An application to profiling providers in managed health care

Profiling is currently an important, and hotly debated, topic in health care and other industries looking for ways to control costs, increase profitability, and increase service quality. Managed care in particular has seen a proliferation in the use of statistical profiling methodology, particularly with regard to monitoring expenditure data. This article focuses on the specific problem of developing statistical methods appropriate for profiling physician contributions to patient pharmacy expenditures incurred in a managed care setting. The two-part hierarchical model with a correlated random-effects structure considered here accounts for both the skewed, zero-inflated nature of pharmacy expenditure data and the fact that patient pharmacy expenditures are correlated within physicians. The random-effects structure has an attractive interpretation in terms of a conceptual model for physician prescribing patterns. Using this model, we propose to rank physicians based on an appropriately constructed provider-level performance measure. This information is subsequently used to develop a novel financial incentive scheme. Inference is conducted in a Bayesian framework using Markov chain Monte Carlo.