Cross-Sectional and Longitudinal Joint Modeling of Repeated Measures of Quasi-Continuous Patient-Reported Outcome and Binary Response Data

• Activity Number: 187
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Kelly H. Zou*+ and Jeno P. Martin and Richard J. Willke
• Companies: Pfizer Inc. and Pfizer Inc. and Pfizer Inc.
• Keywords: Randomized Controlled Trial ; Patient-Reported Outcome ; Receiver-Operating Characteristic Curve ; Trajectory Analysis ; Mixed-Effects Model ; Probit Regression

Abstract

It is challenging to compare the treatment effects based on quasi-continuous patient reported outcome (PRO) scores and a transient binary response indicator on the same subjects in randomized control trial with repeated-measurements. We propose several methods to jointly model the PRO domain score and binary response. Cross-sectionally, two-sample tests are performed on the PRO domain scores between responses within each treatment. A receiver operating characteristic curve analysis, with the response as a gold standard, is used to estimate an optimal threshold along the PRO measurement scale using the Youden index. Longitudinally, cumulative distribution functions of actual and percent change from baseline of a PRO are visual displays of the entire responses. Trajectory analysis of the time-course of the PRO is performed for each treatment. A mixed-effects model of PRO domain scores is conducted to adjust for baseline covariates. Finally, bivariate probit regression and generalized linear mixed modeling of the PRO and response assesses the effect of treatment and additional baseline covariates. These methods are illustrated on a PRO instrument and a time-varying binary response.