First Advisor

Shafie, Khalil

Document Type


Date Created


Embargo Date



The benefits of longitudinal data in clinical research are immense, owing to the potential to detect phenomenon changes and trends over time. An avalanche of traditional methods exist for the analysis of this data kind, most of which are predicated on the assumption of fixed time design for study individuals. This assumption however, is not always tenable. Consider occurrences that can alter the time or visit profile of subjects in a clinical trial, such as adverse events. These may result in not just irregular visit times, but also unbalanced data on subject outcomes. Hence, one can consider visit times to be informative, because subsequent subject visits can be altered based on current visits. In this dissertation, we developed Bayesian joint models for analyzing such scenarios. More broadly, the models jointly analyzed longitudinal outcomes resulting from the exponential class of distributions (specifically, Poisson, Bernoulli and Gamma longitudinal outcomes) and informative visit times drawn from the exponential distribution. A simulation approach was employed to investigate and validate the influence of controlled variations in visit patterns, prior and sample size schemes on model performance. As an application example, the Bayesian Bernoulli-Exponential joint model was applied to a bladder cancer data to study the effect of previous tumor occurrences on the likelihood of subsequent recurrence in subjects, while also studying the effect of other prognostic factors.


246 pages

Local Identifiers



Spring 2022 Graduate Dean's Citation for Outstanding Thesis, Dissertation, and Scholarly Project

Rights Statement

Copyright is held by the author.