Keywords: Reversible jump Markov chain Monte Carlo, linear mixed models, model-based clustering
The economics of healthcare are difficult to analyze, yet there is an increasing need to better understand the variation in costs of health care to more efficiently serve patients. Previous authors have identified geographical region and surgeon volume, a surrogate for surgeon experience defined by the number of surgeries performed in a year, as key drivers of the variation in the costs of health care. Various other studies have found decreased mortality rates in hospitals with high volumes of a particular surgery, while others found reduced lengths of stay in addition to the improvement in clinical outcomes. We assess the combined impact of geography and surgeon volume, while allowing for shrinkage across hospitals with the introduction of hospital-specific random effects, on the total charges associated with four different neurological surgeries: pituitary gland surgery (PGS), deep brain stimulation (DBS), vestibular schwannoma (VS), and caratoid endarterectomy (CEA). Using Bayesian models for the total charges, we simultaneously explore the effect of geography and surgeon experience while clustering the random effects for hospitals.