Applications of Wearable Device Data Analysis to Personal Health Monitoring and Healthy Aging Promotion

Abstract

Demographic aging creates an imbalance between care-seekers and caretakers that necessitates the development of novel strategies to reduce the per-capita use of clinical resources while simultaneously improving health outcomes. Digital health technologies such as smart watches could be used for autonomously monitoring the health of older adults and administering personalized interventions. However, proof-of-concept studies are lacking that demonstrate use-cases for wearable-based personal health monitoring among older adults. Furthermore, in epidemiological research, wearable device data remains underutilized for understanding healthy aging despite its inherent advantages over survey data regarding objectivity and granularity.

In this thesis, we conducted two wearable-based epidemiological studies and two wearable-based personal health monitoring studies. First, we showed that wearables could be used to monitor cognitive function by developing wearable-based gradient boosting models that accurately distinguish between older adults with good and poor cognition. Second, we showed that caffeine may be a viable intervention strategy for reducing sedentary behavior among older adults by uncovering dose-response associations between weight-adjusted caffeine consumption and device-measured physical activity. Third, by leveraging a Hidden Markov Model-based sleep-wake algorithm, accelerometer data from over 36,000 older adults, and nine years of follow-up death registry data, we showed that sleep efficiency and sleep efficiency regularity are more strongly associated with all-cause, cardiovascular, and cancer mortality than sleep duration and four other measures of sleep health. Fourth, we showed that sleep efficiency can be forecasted 4 to 8 hours before sleep onset using gradient boosting and deep learning algorithms trained exclusively on accelerometer data, a crucial first step towards proactive sleep interventions. We further used functional data analytic and non-linear modelling approaches to identify physical activity targets for preventing poor sleep efficiency. This thesis significantly improves upon prior survey-based epidemiological research and identified cognitive function and sleep quality as two domains where wearable-based personal health monitoring systems could be developed.

Date of Award	5 Jul 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Xinyue LI (Supervisor)