Date of Award

Summer 7-25-2024

Embargo Period

8-14-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Public Health Sciences

College

College of Graduate Studies

First Advisor

Brian Neelon

Abstract

Overdispersed binomial data arise in many clinical and public health research studies. Examples include timeline followback data (TLFB), which are used in addiction research to monitor recent substance use. This dissertation proposes novel beta-binomial (BB) mixed models to address the challenges associated with overdispersed binomial data, including zero-inflation, heterogenous effects, and spatiotemporal correlation. Motivated by a 12-week randomized controlled trial evaluating the efficacy of varenicline tartrate for smoking cessation among adolescents, Aim 1 develops a Bayesian zero-inflated beta-binomial (ZIBB) model to analyze longitudinal and bounded TLFB data — namely, the number of abstinent days in the past week. Because treatment effects appear to level off during the study, we introduce random changepoints for each study group to reflect group-specific changes in treatment efficacy over time. Using the model, we accurately estimate the mean trend for each group and identify critical windows of treatment efficacy, revealing a short-term positive effect of varenicline that tapers off after week 9. Aim 2 investigates treatment heterogeneity among latent classes of participants in the same trial. We propose a BB growth mixture model designed to cluster study participants based on their longitudinal trajectories to investigate treatment effect heterogeneity across latent classes. Within each latent class, we fit a piecewise linear BB mixed model with random changepoints for each study group to detect critical windows of treatment efficacy. The model simultaneously clusters participants who share similar characteristics, estimates the class-specific mean abstinence trends for each study group, and quantifies the treatment effect over time within each class. This analysis identifies two classes: high-abstinent individuals (young adults and light smokers) benefiting from varenicline, and low abstinent individuals for whom varenicline shows no effect. Aim 3 is motivated by a study examining spatiotemporal trends in five cardiovascular risk factors (CRFs) among pregnant women in South Carolina during the COVID-19 pandemic. These CRFs include pre-pregnancy obesity (BMI 30 kg/m2), diabetes (including pre-pregnancy and gestational diabetes), hypertensive disorders of pregnancy, pre-pregnancy hypertension, and kidney failure. Thus, the total number of CRFs ranges from 0 to 5 and exhibits overdispersion due to the correlation among these factors. We extend the ZIBB model to a spatiotemporal setting by developing a spatially varying coefficients model to investigate the complex relationship between CRFs and geographic and temporal disparities among non-Hispanic White (NHW) and non-Hispanic Black (NHB) women. In simulation studies, the model accurately captures trends between the two racial groups and predicts mean CRF scores in each region over time. Our analysis of CRFs in South Carolina reveals that specific counties, such as Chesterfield and Marlboro, exhibit a disproportionately wide gap in racial health disparities and may be ideal candidates for community-level interventions designed to mitigate health disparities. Collectively, this dissertation aims to advance the analytical toolkit for clinical and public health research, offering robust methods to analyze overdispersed binomial outcomes in various settings.

Rights

Copyright is held by the author. All rights reserved.

Download

Available for download on Friday, August 14, 2026

Share

COinS