Marginal Regression Analysis of Clustered and Incomplete Event History Data

Abstract

Event history data, including competing risks and more general multistate process data, are commonly encountered in biomedical studies. In practice, such event history data are often subject to intra-cluster correlation in multicenter studies and are complicated due to informative cluster size, a situation where the outcomes under study are associated with the size of the cluster. In addition, outcomes or covariates are frequently incompletely observed in real-world settings. Ignoring these statistical issues will lead to invalid inferences. In this dissertation, I develop a series of marginal regression methods to address these statistical issues with competing risks and more general multistate process data. The motivation for this research comes from a large multicenter HIV study and a multicenter randomized oncology trial. First, I propose a marginal regression method for clustered competing risks data with missing cause of failure. I consider the semiparametric proportional cause-specific hazards model and propose a maximum partial pseudolikelihood estimator under a plausible missing at random assumption. Second, I consider more general clustered multistate process data and propose a marginal regression framework for the transient state occupation probabilities. The proposed method is based on a weighted functional generalized estimating equation approach. A nonparametric hypothesis test for the covariate effect is also provided. Third, I extend the proposed framework in the second part of the dissertation to account for missing covariates, via a weighted functional pseudo-expected estimating equation approach. I conduct extensive simulation studies to evaluate the finite sample performance of the proposed methods. The proposed methods are applied to the motivating multicenter HIV study and oncology trial datasets.