Bayesian Latent Models and Causal Inference for Biological and Health Experiments

Gravelle, Patrick

Full Metadata

Overview

Year:: 2024
Contributor:: Gravelle, Patrick (creator); Gutman, Roee (Advisor); Gatsonis, Constantine (Reader); Mor, Vincent (Reader); Brown University. Department of Biostatistics (sponsor)
Genre:: theses
Subject:: causal inference; Bayesian Statistics; Randomized controlled trials
Extent:: , None p.

Files

Description

Abstract:: In practice, it is rare that the true generating mechanism of a data source is known. This uncertainty presents a challenge to statisticians, who are commonly tasked with specifying and identifying relationships in a given dataset. However, this challenge is not unique, nor equal, for any particular dataset. Investigators may encounter these challenges when modelling data in a variety of settings, including controlled trial or experiments. However, even under controlled settings, there may be latent structures determining the data generation that are not easily discovered or known. Throughout this dissertation, we are interested in methodological developments and comparisons for latent data generating mechanisms across a variety of controlled trial or experiment settings. In particular, we focus on three settings: 1) Stratified Cluster Randomized Trials (SCRTs), 2) Adaptive Randomized Controlled Trials (RCTs), and 3) Longitudinal Controlled Experiments. In the first aim, we examine the operating characteristics of different specifications of generalized linear multilevel models and generalized estimating equations in SCRTs through extensive simulations. The simulations include binary and continuous outcome variables, different outcome error configurations, linear and non-linear relationships between the covariates and outcomes, as well as different number of clusters and individuals within clusters. In the second aim, we propose an adaptive design for an RCT to address noncompliance with multi-component interventions. We frame the design within the counterfactual causal inference framework, and describe both design and analysis procedures to implement this design. Finally, in the third aim, we propose a hierarchical zero-inflated generalized Dirichlet multinomial regression model with cyclic splines to model the mouse behavior time for nine behaviors across all circadian hours, recorded during a longitudinal controlled biological experiment comparing a cohort of mice with a human-equivalent genetic mutation that characterizes the amyotrophic lateral sclerosis (ALS) disease spectrum against a cohort of control mice.
Notes:: Thesis (Ph. D.)--Brown University, 2024

Content

Citation

Gravelle, Patrick, "Bayesian Latent Models and Causal Inference for Biological and Health Experiments" (2024). Biostatistics Theses and Dissertations. Brown Digital Repository. Brown University Library. https://repository.library.brown.edu/studio/item/bdr:nrz55ddt/

Relations

Collection:

Biostatistics Theses and Dissertations

Theses and Dissertations for the Biostatistics department.
...