Graduation Date

Spring 5-10-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Programs

Biostatistics

First Advisor

Cheng Zheng

Second Advisor

Ying Zhang

Third Advisor

Ran Dai

Fourth Advisor

Jianghu Dong

Abstract

With the increasing emphasis on longitudinal studies and patient monitoring, recurrent event data have become more frequently collected, particularly among patients with chronic diseases. Understanding the mediating role of recurrent events in the causal pathway between exposures and time-to-event outcomes has garnered increasing interest not only in clinical research but also in various other fields, including epidemiology, public health, and social sciences. For instance, Opportunistic Infections (OIs), a form of recurrent events, frequently occur in Human Immunodeficiency Virus (HIV)-infected patients and can substantially impact their health and survival, particularly for those diagnosed with Acquired Immunodeficiency Syndrome (AIDS). Investigating how OIs mediate the effects of treatments or baseline conditions on survival outcomes is of great interest. However, existing methods for causal inference in joint analyses of recurrent and terminal events remain limited, particularly in scenarios involving multiple mediators—such as different types of OIs or other time-dependent biomarkers—and in accounting for unmeasured confounding, which is particularly challenging in survival analysis. To address these challenges, we develop novel joint modeling approaches that simultaneously investigate recurrent event mediators and survival outcomes. Our framework is further extended to handle multiple mediators, whether causally or non-causally related, while estimating direct and indirect effects. In Chapter 2, we propose a joint modeling approach that incorporates recurrent events as mediators and survival end- points as outcomes. By relaxing the “sequential ignorability” assumption through shared random effects, our method accounts for time-independent unmeasured confounding between mediators and outcome. Simulation studies demonstrate the robustness of this approach, and an application to an AIDS study shows that the number of OIs mediates the effect of baseline CD4 count on survival outcomes. In Chapter 3, we extend this framework by modeling recurrent events using gap times rather than the total time scale, allowing for varying distribution and covariate effects for each gap time. This approach provides greater flexibility in understanding the differential impact of exposure on early and late OIs. In Chapter 4, we further advance the mediation framework to quantify causal mechanisms when multiple mediators exist between exposure and terminal events through joint modeling. This approach relaxes the assumption of confounding between mediators and outcomes by utilizing the latent shared random effect. Using this method, we analyze the distinct mediation effects of multiple types of OIs, providing a more nuanced understanding of their role in disease progression. In Chapter 5, we extended the model further by integrating repeated biomarker measurements with recurrent events, explicitly addressing their causal interdependence. Applied to the CPCRA study, our method quantifies how recurrent OIs and repeated CD4 counts jointly mediate the effect of treatment or prior AIDS-defining conditions on mortality. The results highlight the opposing mediation effects of these pathways, emphasizing the importance of controlling OIs while improving immune function. Our research contributes to the methodological development of mediation analysis in survival studies, offering a robust framework for investigating multiple mediators, whether causally or non-causally related. These findings provide deeper insights into treatment mechanisms and inform clinical strategies for managing HIV progression.

Comments

2025 Copyright, the authors

Download

Available for download on Saturday, May 01, 2027

Share

COinS