Online Program

A Bayesian Two-Part Latent Class Model for Longitudinal Medical Expenditure Data
*Brian Neelon, Duke University 
Sharon-Lise T. Normand, Harvard Medical School 
James O'Malley, Harvard Medical School 

Keywords: Bayesian analysis, latent class model, longitudinal data analysis, medical expenditure data, two-part model.

We develop a Bayesian latent class model for semi-continuous data characterized by a non-negative distribution with a point mass at zero. Within each class, we fit a two-part random effects model to separately model the probability of a nonzero response and the mean positive response. The regression coefficients and random effect covariances are allowed to vary across classes, thus permitting class-varying correlation structures between the two components of the model. For posterior computation, we propose an efficient MCMC algorithm that combines full-conditional Gibbs and Metropolis steps. Generalizations to zero-inflated count data and multiple outcomes are discussed, and the approach is applied to a study of mental health expenditures among federal employees.