Regency B
Bayesian Analysis of Sparse Multivariate Matched Proportions Data (303968)
Katherine Hobbs Knutson, Partners Healthcare
*Mark J Meyer, Georgetown University
Keywords: Multivariate Matched Proportions, Systems of Care, Sparse Data, Bayesian Statistics, Marginal Probability Models, Prior Specification
Multivariate matched proportions (MMP) data arises when multiple sets of paired binary measurements are taken on the same subject and shows up in a variety of contexts including the surveillance of adverse events in drug regimens, disease classification, and agreement between care providers. While some recent work proposes frequentist methods to model MMP data, the issue of sparsity, where no or very few responses are recorded for one or more sets, is unaddressed. However, sparsity is a common problem. In this talk, we propose several Bayesian methods that model sparse MMP data via the prior specification including multinomial-Dirichlet models and marginal probability models with robust t-priors. Our work is motivated by sparse data from a study of care coordination within a System of Care framework. The study compares initiated contacts with components in the framework by primary care physicians to contacts by specialty care. We will discuss the existing methods for MMP data and our Bayesian models, comparing all models in a multivariate analysis of the System of Care data. Finally, we also develop and present examples of user-friendly R code which we make freely available.