Lulu Guo

Title: Evaluating Treatment Efficacy in Randomized Controlled Trials with Treatment Noncompliance and Multivariate Outcomes
Date: Wednesday, May 15th, 2024
Time: 10:00am
Location: LIB 2020 & Zoom
Supervised by: Hui Xie & X. Joan Hu

Abstract: In many real-world randomized controlled trials (RCTs), noncompliance behaviour often occurs and can greatly complicate assessing the intervention efficacy. Furthermore, multiple outcomes are usually employed to measure underlying complex traits when evaluating the performance of multifaceted behaviour interventions for chronic disease (e.g., arthritis). Statistical procedures ignoring treatment noncompliance and the correlations among multiple outcomes can lead to biased estimates of treatment efficacy and a significant loss of power to detect treatment efficacy. This dissertation aims at developing novel statistical methodologies to evaluate the efficacy of multifaceted behaviour interventions while addressing noncompliance issue and correlated multiple outcomes simultaneously. To deal with noncompliance issues, a principal stratification approach is employed to estimate complier average causal effects. To address the correlated multiple trial outcomes, this dissertation proposed novel methodologies based on mixed-effects regression models and the latent-factor approach. The first work proposes a multivariate longitudinal potential outcome model based on a hierarchical random-effects approach stratified on latent compliance types under all-or-none compliance. The second work proposes a latent-factor multivariate complier average causal effects (MCACE) model for multidimensional longitudinal outcomes with principal strata of compliance types. Under the model, high dimensional outcomes are reduced to low dimensional latent factors, leading to a parsimonious and efficient test of overall CACEs on multiple endpoints, mitigating the multiple testing issues associated with multidimensional endpoints. The third work considers partial compliance and extends to multivariate CACE estimation under the framework of partial compliance. Comprehensive simulation studies demonstrates the validity of the proposal methods and large gains in the estimation efficiency (several-fold increase in statistical power to detect CACEs compared to existing methods). The application of these proposed methodologies to assess the efficacy of a multifaceted behaviour intervention (Arthritis Health Journal) in a longitudinal trial conducted at Arthritis Research Canada yields novel findings not discovered previously.