Innovative Bayesian and Frequentist Adaptive Designs in Late-Phase Oncology Clinical Trials

In this dissertation, we present two methodological works in the areas of Bayesian and frequentist adaptive statistical designs in late-phase oncology clinical trials. Late-phase oncology clinical trials may incorporate various adaptive features and the focus of this dissertation is on Bayesian trial monitoring and sample size re-estimation methods. In the first part of the dissertation, we focus on therapeutic late-phase oncology clinical trials with time-to-event endpoints. Overall survival (OS) serves as the most reliable and preferred clinical endpoint, while endpoints such as progression-free survival (PFS) are used to make early futility or efficacy decisions. Progression-free survival and overall survival are strongly connected endpoints but may not always reflect strong associations in numerical measures, and their relationship is rarely taken into consideration in the design of a clinical trial. We propose a clinical trial design based on an early endpoint of PFS and a final endpoint of OS via the semi-competing risks framework, with trial monitoring conducted utilizing stochastic ordering of posterior predictive probabilities. In the second part of this dissertation, we shift our focus to parallel-arm pragmatic late-phase oncology clinical trials, specifically in cancer care delivery. Parallel-arm cluster-randomized trials (CRT) are increasingly utilized to understand cancer care in real-world settings. Cluster-randomized trials possess a hierarchical data structure in which subjects are clustered within units such as clinics or providers. In CRTs, sample size and power calculations rely on the intraclass correlation coefficient (ICC), which quantifies the degree of between-cluster variability. Underestimation of the ICC may lead to significantly reduced study power to detect the treatment effect. We propose extensions of several popular adaptive sample size re-estimation methods, including the internal pilot study approach and the promising zone-design based on conditional power.