Modelling and inference for complex survical data under interval censoring.
Date of Issue2012
School of Physical and Mathematical Sciences
Analysis of interval-censored survival data has been attracting much research interest as such data commonly arise in many clinical and epidemiological studies due to periodic visits of cases. When data are collected from several clinical sites or geographical regions, clustered interval-censored data are then encountered. Because of advances in medical technology, a number of subjects are not susceptible to the event of interest. It leads us to study clustered interval-censored data with the presence of a cured subgroup assumed. In this thesis, we propose a mixture cure modeling procedure to analyze such complex survival data under interval censoring. To reflect the within-cluster correlation of clustered data, random effects are introduced in manner of the GLMM method. We develop the REML estimation for regression parameters and variance component of random effects, and propose an EM algorithm for its implementation in conjunction with a self-consistent iterative algorithm for estimating the nonparametric component. We also provide a score test to adjust the presence of cured subjects in clustered interval-censored data. Under a general class of semiparametric mixture cure transformation models with random effects, we investigate the model identifiability and establish asymptotic properties, including the consistency and asymptotic normality of the parameter estimators, and the consistency with a cube-root-n convergence rate of the estimator for the nonparametric component. We conduct intensive simulations and analyze the smoking cessation data to evaluate the performance of the proposed estimators and methodology.