Mixed-effects State-space Models for Longitudinal Data Analysis

• Activity Number: 2
• Type: Invited
• Date/Time: Sunday, August 8, 2004 : 2:00 PM to 3:50 PM
• Sponsor: WNAR
• Author(s): Hulin Wu*+ and Dacheng Liu and Xu-Feng Niu
• Companies: University of Rochester Medical Center and University of Rochester Medical Center and Florida State University
• Keywords: state-space model ; longitudinal data ; mixed-effects model ; differential equations ; Kalman filter ; HIV/AIDS

Abstract

Recent development of nonparametric smoothing/regression methods for longitudinal data allows us to robustly analyze the longitudinal data with minimum assumptions on the model. However, another important direction has not been explored, which is how to efficiently analyze the longitudinal data when the mechanisms generating the data are known and can be described by complicated differential or difference equations. We propose a novel class of mixed-effects state-space models for longitudinal data. If the dynamics of individual observations can be formulated in a system of differential or difference equations, then state-space modeling is straightforward and appealing. Three methods are developed for estimating unknown parameters, i.e. the global two-stage (GTS) method, the EM-based maximum likelihood method (MLE), and the Bayesian approach. Simulation results indicate that all the three methods perform well. Finally, we apply the mixed-effects state-space model to a dataset from an AIDS clinical trial to illustrate the proposed methodology.