Skip to main content
eScholarship
Open Access Publications from the University of California

UC San Diego

UC San Diego Electronic Theses and Dissertations bannerUC San Diego

A Pre-Post, Repeated Measures, Cluster-Randomized Clinical Trial Analysis and Exploration of Treatment Effect Heterogeneity

Abstract

The Open and Ask study is a cluster-randomized controlled trial comparing the effectiveness of three behavioral interventions to improve patient-provider communication. The unit of measurement is clinicians, who are evaluated by patients seen at baseline and at one post-intervention follow-up visit. With different patients at each visit, this introduces a pre-post design with repeated measures. The allocation of measures between baseline and follow-up impacts the power of detecting treatment effects and is an important design consideration. Post hoc heterogeneity analyses supplement evaluation of the main treatment effect by identifying subgroups for whom the effect differs in magnitude or direction. Traditional methods for heterogeneity analyses require manual specification of potential effect modifiers which is often unfeasible in high-dimensional settings.

This dissertation comprises four chapters addressing these topics. Chapter 1 consists of the Open and Ask clinical trial analyses, including descriptive, primary, and several secondary analyses. Chapter 2 discusses the optimal distribution of measures in pre-post, repeated measures, randomized clinical trials for 5 analysis methods. Treatment effect and variance formulas for each method are derived and solutions to the optimal distribution of measures for both known and unknown correlation parameters are provided. Chapter 3 provides an overview of modern machine learning methods for treatment effect heterogeneity analyses in high-dimensional settings. The advantages over traditional methods, limitations, and supplemental analysis techniques required for heterogeneous subgroup identification are discussed. Chapter 4 proposes an alternative approach to complement these methods that automates heterogeneous subgroup identification through the use of conventional estimation and inference techniques.