|Institution:||University of Pittsburgh|
|Full text PDF:||http://d-scholarship.pitt.edu/23995/1/Dissertation_RenYi_4_2015.pdf|
In survival analysis, the failure time of an event is interval-censored when the event is only known to occur between two observation times. Most existing methods for interval-censored data only account for a single cause of failure. However, in many situations a subject may fail due to more than one type of event. Such data scenarios are called competing risks data. Competing events may preclude the occurrence of the event of interest. In the analysis of competing risks, the conventional methods should be used with caution and may lead to nonsensical interpretation. With covariates, the proportional subdistribution hazards model is widely used to model the cumulative incidence function (also known as the subdistribution) of a particular event. This semiparametric regression model has a straightforward interpretation for estimators as it is akin to the Cox proportional hazards model. For interval-censored competing risks data, however, estimation procedures based on the proportional subdistribution hazards model has not been investigated. In this dissertation, we propose estimation and inference procedures that account for both interval censoring and competing risks by adopting the modeling framework of the proportional subdistribution hazards model. The objective is to examine the effects of covariates on the subdistribution of event of interest. The proposed estimating equations effectively utilize the ordering of event time pairs. The technique of inverse probability weighting is used to account for the missing mechanism. Simulation studies show that the proposed methods perform well under realistic scenarios. A lymphoma data set is used to illustrate the performance of the proposed method in comparison to the proportional subdistribution hazards model using the data imputed by midpoint of the observed time interval. Public health significance: Interval-censored competing risks data are often encountered in biomedical research. The method we proposed serves a useful tool for exploring the covariate effects on the event of interest under this challenging censoring mechanism. The information on the effects of covariates has implications for proper clinical management of the different cohorts of patients. It quantifies the relationship between public health strategies and measurement of health status, and determines the efficacy information for possible improvement of interventions.