A Hierarchical Multivariate Two-Part Model for Profiling Providers' Effects on Health Care Charges

• Activity Number: 123
• Type: Topic Contributed
• Date/Time: Monday, August 8, 2005 : 10:30 AM to 12:20 PM
• Sponsor: Section on Health Policy Statistics
• Author(s): John Robinson*+ and Scott Zeger and Christopher Forrest
• Companies: Johns Hopkins University and Johns Hopkins University and Johns Hopkins University
• Keywords: Hierarchical model ; Two-part model ; Multivariate analysis ; Gibbs sampler ; Data augmentation ; Provider profiling

Abstract

When profiling providers' effects on charges for health care services, fitting a hierarchical, multivariate model can capture associations among charges for different services at both the provider and patient levels. For estimating covariate effects on charges for a single service when some values are zeros, a two-part model is commonly employed. Here, we extend the two-part model for use with multiple services, using a log-linear model and a general multivariate lognormal model to form the likelihood level of a hierarchical model. The log-linear likelihood is reparameterized so provider effects on any use of each service are marginal with respect to use of other services. The general multivariate lognormal likelihood is structured so variances of log of positive charges are provider-specific. The Gibbs sampler used for model fitting includes a data augmentation step to accommodate for the fact that values of log of positive charges are undefined for unused services. Using this hierarchical, multivariate model, we analyze effects of primary care physicians on their patients' annual charges for two services, primary care and specialty care, and obtain interesting results.