Informative Drop-Out Models for Longitudinal Binary Data

This dissertation, "Informative Drop-out Models for Longitudinal Binary Data" by Ka-ki, Chau, 周嘉琪, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of the thesis entitled Informative Drop-out Models for Longitudinal Binary Data submitted by Chau, Ka Ki for the degree of Master of Philosophy at The University of Hong Kong in December 2003 Attrition or drop-out is a common phenomenon in longitudinal studies in which repeated observations are made on the same subject over time. Subjects always drop out prematurely, especially when the measurement process is lengthy. The problem of drop-out results in incomplete and unbalanced data which in turn results in loss of eciency and also bias in the analysed results. Regarding this problem, many modeling approaches that deal with missing data have been pro- posed. This variety of possible approaches di(R)ers for di(R)erent types of drop-out processes and also di(R)erent kinds of longitudinal data. In this thesis, we aim to develop new modeling strategies for longitudinal binary data with informative drop-out. Three di(R)erent conditional AR1 models are proposed for the response and a logistic regression model for the drop-out process. In these models, both the probabilities of a positive response and the drop-out indicator of a patient in that occasion are assumed to be logit linear in some covariates and outcomes. To account for the problem of over-dispersion and accommodate population het- ierogeneity, we incorporate random intercepts to one of the proposed models. We implement the models via likelihood and Bayesian frameworks. Since the inclu- sion of random e(R)ects complicates the calculation considerably, we also attempt to investigate the use of Gibbs output within the Bayesian framework to carry out the Monte Carlo Approximation of the complicated likelihood function in- volving random e(R)ects by a classical likelihood approach. We then demonstrate these models on a methadone clinic data. Moreover, we also investigate the sen- sitivity of the assumption of the dropout process on the parameter estimates for the three proposed models through simulation experiments. Results show that the incorporation of the informative drop-out model helps us to understand and interpret the drop-out process across patients better. (290words) ii DOI: 10.5353/th_b2962714 Subjects: Longitudinal method Monte Carlo method Bayesian statistical decision theory Medicine - Research - Statistical methods