ENHANCEMENTS IN NOVEL TRIAL DESIGNS IN THE ERA OF DIGITAL HEALTH

Abstract

The evolution towards ubiquitous digitization today has led to increasing interest in developing personalized interventions that can adapt over time according to individual-level information. A dynamic treatment regime (DTR) is a sequence of decision rules to dictate how interventions should be adapted given the individual's intermediate information. An experimental approach to developing or assessing such DTR is to use the sequential multiple assignment randomized trial (SMART) designs. The SMART is a design that involves multiple stages of randomization, usually at important clinical timepoints. However, due to the design's novelty and inherent complexity, gaps exist between theory and implementation, deterring its actual usage by scientists. The research of my dissertation aims to narrow the gap between the SMART theoretical design and practical implementation issues. First, I present the value of SMART in comparison to a conventional trial via a simulation study. This will highlight the efficiency and characteristics of the SMART. Second, I developed a precision-based sample size calculation method for pilot SMART, which has the benefit of keeping the sample size relatively small while gaining additional information on the estimate. Third, I conducted a novel three-stage SMART study, to illustrate a real-case example. The SMART study has multiple novelties to its structure, which may serve as exemplar study for other researchers. Finally, I present new approaches to evaluate SMART data with ordinal and survival outcome types.