Choice of Models for Joint Distribution of Multidimensional Latent Quality of life and Survival.

• Activity Number: 351
• Type: Topic Contributed
• Date/Time: Wednesday, August 14, 2002 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Mounir Mesbah*+
• Affiliation(s): Université de Bretagne-Sud
• Keywords: Quality of Life ; Longitudinal Data ; Latent Variable ; Rasch Model ; Semi Parametric Modelling ; Complexe Models

Abstract

A major problem that frequently arises in longitudinal clinical trials, where Quality of Life (QoL) questionnaires are repeatedly assessed by each individual under study, is that of dropout. Nonignorable dropout may depend on missing components of the longitudinal process.

In this paper, we first present a recent joint model for a longitudinal normal QoL process with nonignorable dropout and time to dropout (J.F. Dupuy and M. Mesbah, 2002). This model allows dropout to occur at any point in time. Consistency, identifiability, and asymptotic normality of its parameter estimators is achieved (J.F. Dupuy, I. Grama, and M. Mesbah, 2002).

We investigate how to extend this model, when the longitudinal QoL covariate is a latent normal variable assessed by a questionnaire giving multivariate dichotomous or polychotomous responses.

Different strategies and models, using either latent QoL trait estimated by a Rasch Model or more generally an Item Response Model, or a simple QoL score, are presented and compared. All these strategies will use available software. A real data set on QoL Cancer clinical trial is analyzed.