A Flexible Correlation Structure for Joint Modeling of Multivariate Ordinal Medication Adherence Data

• Activity Number: 368
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Abdus Wahed*+ and Zhen Jiang
• Companies: University of Pittsburgh and Food and Drug Adminstration
• Keywords: Medication Adherence ; Joint Modeling ; Ordinal Data ; Generalized Estimating Equations ; GEE ; Latent Variable Model

Abstract

Adherence to medication is critical in achieving effectiveness of any treatment. Determining factors that influence adherence has been the subject of many clinical studies. Analyzing adherence data is complicated because adherence is often measured on multiple drugs on multiple occasions, resulting in multivariate longitudinal data. This work is motivated by the Virahep-C study, where adherence is measured on two drugs, one resulting in a binary outcome and the other in a three-category ordinal outcome. We propose a joint model which assumes that the ordered outcomes arise from a latent multivariate normal process. This joint model provides a framework for analyzing multivariate ordered longitudinal data with a general correlation structure, covering both within-outcome and within-individual correlation. We propose to draw inference GEE. The methods are demonstrated by analyzing Virahep-C medication adherence data. Simulation studies show that the estimators are unbiased and more efficient than those obtained through fitting separate models for each outcome. The method also yields unbiased estimators for correlation parameters when correlation structure is specified correctly.