Zhu, Yayuan2016-01-202016-01-202016-01-202016-01-13http://hdl.handle.net/10012/10171Event history studies based on disease clinic data often face several complications. Specifically, patients visit the clinic irregularly, and the intermittent inspection times depend on the history of disease-related variables; this can cause event or failure times to be dependently interval-censored. Furthermore, failure times could be truncated, treatment assignment is non-randomized and can be confounded, and there are competing risks of the failure time outcomes under study. I propose a class of inverse probability weights applied to estimating functions so that the informative inspection scheme and confounded treatment are appropriately dealt with. As a result, the distribution of failure time outcomes can be consistently estimated. I consider parametric, non- and semi-parametric estimation. Monotone smoothing techniques are employed in a two-stage estimation procedure for the non- or semi-parametric estimation. Simulations for a variety of failure time models are conducted for examining the finite sample performances of proposed estimators. This research is initially motivated by the Psoriatic Arthritis (PsA) Toronto Cohort Study at the Toronto Western Hospital and the proposed methodologies are applied to this cohort study as an illustration.enintermittent observationoutcome-dependent follow-updependent interval censoringinverse-intensity-of-visit weighted estimating functionssurvival analysisDoctoral Thesis