Joint Modeling of Multi-System Wearable Data

• Activity Number: 43 - Discovering Homology in Multi-View Data: New Statistical Methods for Data Integration
• Type: Invited
• Date/Time: Sunday, July 29, 2018 : 4:00 PM to 5:50 PM
• Sponsor: ENAR
• Author(s): Vadim Zipunnikov* and Junrui Di
• Companies: Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health
• Keywords: sleep; physical activity; circadian rhythmicity; matrix decomposition; tensor decompositions

Abstract

Physical activity and sleep trackers as well as heart rate monitors are now extensively used to track quality of sleep, levels of physical activity, and disruptions in circadian rhythms in many clinical studies, for example, to reliably monitor and assess post-intervention changes in patient's status. Thus, mobile technologies has provided an unprecedented opportunity to obtain objective simultaneous assessment of multiple physiological systems in real-time, over weeks or months. We developed a computationally efficient data analytical frameworkte that parsimoniously models multiple biomarkers nested within multiple systems, accounts for interactions of biomarkers within and between physiological systems, and decomposes within-subject and between-subject variability. We illustrate our approach using 7-day of Actiheart data from 106 participants of Baltimore Longitudinal Study of Aging.