Bayesian Methods for snSMART Designs with External Controls and Dynamic Prediction of Landmark Survival Time in Cancer Clinical Trials

Abstract

In Duchenne muscular dystrophy (DMD) and other rare diseases, recruiting patients into clinical trials is challenging. Additionally, assigning patients to long-term, multi-year placebo arms raises ethical and trial retention concerns. This poses a significant challenge to the traditional sequential drug development paradigm. In this dissertation, we present small sample, sequential, multiple assignment, randomized trial (snSMART) designs and methods that formally incorporate external control data under both the non-longitudinal and longitudinal settings.

After introducing the integration of snSMART with external control data in Chapter 1, Chapter 2 proposes an snSMART design that integrates dose selection and confirmatory assessment into a single trial. This multi-stage design evaluates the effects of multiple doses of a promising drug, rerandomizing patients to appropriate dose levels based on their stage 1 dose response. Our approach enhances the efficiency of treatment effect estimates by: (i) enriching the placebo arm with external control data, and (ii) utilizing data from all stages. We combine data from external controls and different stages using a robust meta-analytic combined (MAC) approach, accounting for various sources of heterogeneity and potential selection bias. Upon reanalyzing data from a DMD trial with our proposed method, MAC-snSMART, we observe that MAC-snSMART estimators offer improved efficiency over the original trial results. The robust MAC-snSMART method frequently provides more accurate estimators than traditional analytical methods. Overall, our proposed methodology provides a promising candidate for efficient drug development in DMD and other rare diseases.

In Chapter 3, we present Bayesian longitudinal piecewise meta-analytic combined (BLPM), a notable advancement on the robust MAC-snSMART method from Chapter 2. This enhancement introduces significant improvements to snSMART research by: (1) enabling longitudinal data analysis, (2) incorporating patient baseline characteristics, (3) utilizing multiple imputation for missing data, (4) reducing heterogeneity with propensity score (PS), and (5) managing stage-wise treatment effect non-exchangeability. These developments significantly increase the snSMART design’s utility and efficiency in rare disease drug development. BLPM applies PS trimming, inverse probability treatment weighting (IPTW), and theMAC framework to navigate heterogeneity and cross-stage treatment effects. Our evaluations, through simulation studies and the reanalysis of a DMD trial, show that BLPM methods consistently achieve the lowest rMSE across tested scenarios, underscoring its potential to enhance rare disease drug development.

In Chapter 4, we propose a multivariate, joint modeling approach to assess the underlying dynamics of progression-free survival (PFS) components to forecast the death times of trial participants. Through Bayesian model averaging (BMA), our proposed method improves the accuracy of the overall survival (OS) forecast by combining joint models developed from each granular component of PFS. A case study of a renal cell carcinoma trial is conducted, and our method provides the most accurate predictions across all tested scenarios. The reliability of our proposed method is verified through extensive simulation studies, which include a scenario where OS is completely independent of PFS. Overall, the proposed methodology emerges as a promising candidate for reliable OS prediction in solid tumor oncology studies.