2021 Issue

Faculty Spotlight: Ken Cheung

My research lives in the union of adaptive trial methodology, implementation science, and machine learning techniques, with an emphasis on using innovative designs associated with mobile health, or mHealth, applications to understand and support health behaviors. My mHealth-related projects roughly fall into two areas. The first area focuses on the evaluation, validation, and optimization of behavioral intervention technologies via mobile health applications. In a recent project, we examined the efficacy of a mental health app recommendation system (IntelliCare) in terms of maximizing user engagement (Cheung et al., 2018; Mohr et al., 2015) and studied the optimal personal schedule in terms of elicited response to push notifications (Hu et al., 2021). The second area involves leveraging wearable devices and mobile technology in the conduct of adaptive trials to improve statistical efficiency. In an ongoing National Institutes of Health–funded collaboration with Keith Diaz at the Center for Behavioral Cardiovascular Health at Columbia University, we deploy an innovative adaptive design to identify the minimum effective combinations of sedentary break conditions. Briefly, this study introduces sedentary breaks to participants with a goal of reducing their glucose and/or blood pressure, measured throughout the visit days. While each sedentary break regimen is defined by two elements—break frequency and break duration—the study aims to identify minimum combinations of frequency and duration that shift these cardiometabolic parameters.

We developed an adaptive design based on the Bayesian decision-theoretic framework with respect to combinationwise weighted posterior gains and introduce the concepts of adaptive randomization and epsilon-tapering in this application. Using ensemble simulation, we demonstrate that a properly calibrated adaptive design achieves a true positive rate of about 90%, compared to 68% for balanced randomization, while keeping the false discovery rate comparable and sample size fixed. In addition, the adaptive design on average increases the number of participants treated at effective combinations compared to balanced randomization, thus improving the ethical foundation of the study. While the study is currently in the planning stage, the methodology manuscript is to appear in the Annals of Applied Statistics. The abovementioned work is accomplished in collaboration with the ROADMAP (Research On Adaptive Design for Mobile Application Platforms) working group, which is a multi-institution working group involving numerous faculty, postdoctoral researchers, and students in the department.

References

Ken Cheung, PhD
Professor of Biostatistics and Vice Dean of Education