A Multistate Markov Chain Model for Longitudinal Quality-of-Life Data Subject to Non-Ignorable Missingness

• Activity Number: 379
• Type: Topic Contributed
• Date/Time: Thursday, August 15, 2002 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Bernard Cole*+ and Marco Bonetti and Alan Zaslavsky and Richard Gelber
• Affiliation(s): Dartmouth Medical School and Harvard University and Harvard University and Harvard University

Abstract

Quality of life is an important outcome in clinical research, especially in cancer clinical trials. Typically, quality-of-life data are collected longitudinally from patients at pre-specified time points both during treatment and subsequent follow-up. Missing data are common in assessing longitudinal quality of life. Furthermore, the probability a patient misses an assessment may be related to the patient's quality of life at the scheduled assessment time. We propose a Markov chain model for the analysis of ordinal categorical outcomes derived from quality-of-life measures. Our model assumes that transitions between quality-of-life states are described by a Markov chain with transition probabilities that may depend on covariates (possibly time-varying), using either generalized logit models or proportional odds models. Logistic regression models are used to model the conditional probabilities of observing a measurement, given the actual value. Estimation is by maximum likelihood, summing over all possible values of the missing measurements. We illustrate the model using data from a breast cancer clinical trial in which quality-of-life data were collected longitudinally.