Joint Modeling of Multivariate Ordinal Longitudinal Outcomes with Missing Data

• Activity Number: 656
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Zhen Jiang*+ and Adbus S. Wahed
• Companies: University of Pittsburgh and University of Pittsburgh
• Keywords: Multivariate ordered longitudinal data ; Joint model ; Missing data ; Generalized estimating equations ; Inverse probability weight

Abstract

Multivariate ordinal longitudinal outcomes are often observed in clinical research. In many cases, patient's status cannot be fully characterized by a single outcome, and the correlation between multiple outcomes may be of interest. In such cases, joint modeling of multiple outcomes is a natural choice. In this paper, we propose a joint model which assumes that the ordinal outcomes arose from a partitioned latent multivariate normal process. We show how to construct unbiased parameter estimators of this model in the presence of drop out. Specifically, we propose a weighted estimating equation that provides unbiased estimator when data are missing at random (MAR). We present simulations demonstrating how the weighted estimating equation corrects the bias that occurs when using unweighted estimating equation in the presence of MAR.