Mixture Model for Multilevel Data

• Activity Number: 251
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Haiqun Lin*+ and Shu-Xia Li and Xiao Xu and Harlan Krumholz
• Companies: Yale University and Yale University and Yale University and Yale University
• Keywords: Multilevel ; latent class

Abstract

We develop a multilevel mixture model for characterizing hospital care. The goal is to identify distinct hospital patterns of practice change over time. We propose a latent class model with each categorical latent class representing a unique hospital practice pattern over time. Because patient composition varies over time even within the same hospital, growth mixture models cannot be directly fitted to the multilevel data involving hospitals. The currently available analytic approach is a two-stage approach where the patient-level data is aggregated within each hospital at a given time in the 1st stage and growth mixture model is fitted to the aggregated data in the 2nd stage. In this paper, we propose a simultaneous approach which directly incorporates patient-level data in the evaluation of hospital performance over time and hence appropriately handle the variability among patients within a hospital. We applied our approach to identify distinct hospital patterns of 30-day mortality rates following acute myocardial infarction during 2005-2010. Simulation studies are performed to illustrate the advantage of our simultaneous model over the current two-stage approach.