Leveraging Wearables Data to Understand Behavioral Patterns and Predict Influenza-like Illnesses

Abstract

Wearable devices have revolutionized digital health by enabling real-time, continuous monitoring of physiological and behavioral data. However, challenges remain in ensuring participant adherence in digital health studies and effectively leveraging wearable data for intelligent diagnostic applications. This dissertation explores two interconnected areas: (1) improving adherence and retention in longitudinal digital health studies, and (2) developing intelligent testing allocation (ITA) models that optimize diagnostic efficiency for infectious diseases using wearable-derived biomarkers.In the first part of this work, we investigate adherence patterns in large-scale digital health studies conducted at Duke University and Evidation Health. By employing unsupervised clustering, survival analysis, and recurrent event modeling, we identify key demographic and behavioral factors that predict participant retention. Our findings reveal that younger participants demonstrate lower adherence, and we propose targeted engagement strategies to enhance study retention. These insights are critical for ensuring high-quality, representative datasets that support machine learning applications in health monitoring. The second part of this dissertation focuses on Intelligent Testing Allocation (ITA), a machine learning-based framework that utilizes wearable sensor data to prioritize diagnostic testing. We demonstrate that physiological features such as resting heart rate and step count can effectively predict infection risk, leading to a 6.5-fold improvement in diagnostic efficiency compared to standard random testing approaches. Through collaboration with the Biomedical Advanced Research and Development Authority (BARDA), we expand ITA into ITA+, incorporating electronic health records (EHR), survey responses, and additional wearable data from 10,000+ patients at the Duke University Health System (DUHS). Our real-world implementation study validates ITA+ as a scalable, deployable tool for infectious disease surveillance, with implications for future pandemics and personalized healthcare applications. By bridging the gap between digital health study design, data-driven diagnostics, and real-world implementation, this dissertation contributes to the advancement of scalable, data-driven healthcare solutions. The findings underscore the potential of wearable devices to transform preventive health monitoring and personalized medicine, setting the stage for future innovations in digital health research.