Clustering of Longitudinal Data Using Mixture of Extended Linear Mixed-Effect Models

• Activity Number: 27
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: SSC
• Author(s): ChangJiang Xu*+ and Antonio Ciampi and Celia M.T. Greenwood and Vicky Tagalakis
• Companies: and McGill University and Jewish General Hospital/McGill University and Jewish General Hospital/McGill University
• Keywords: clustering ; mixture distribution ; mixed effects ; longitudinal data ; EM algorithm

Abstract

We model longitudinal data from a heterogeneous population as samples from a mixture of extended linear mixed-effects models, and develop an expectation-maximization (EM) algorithm based on Monte Carlo (MC) sampling, called EMMC, to estimate the model parameters. The algorithm EMMC is implemented separately using linear mixed-effect model for each cluster, and thus converges much faster than the standard EM algorithm. We present an evaluation of the approach through simulations. We also applied our algorithm to clustering of International Normalized Ratio (INR) trajectories following warfarin initiation. Four clusters of the INR trajectories were determined using model selection criterion AIC. In contrast, when using BIC for model selection, the cluster consisting of the worst cases, where INR cannot be controlled, was missed.