UC San Diego

Healthcare has traditionally aimed to optimize treatments on the population-level, yet meaningful differences between individuals limit the utility of one-size-fits-all approaches. Continuous monitoring of physiological signals from wearable devices can support wellness beyond routine clinical visits by enabling detection of subtle changes in physiology or behavior that may prompt earlier diagnoses, guide intervention choices, and encourage individual agency over one’s health. In Chapter One, I demonstrate how repeated measurements within individuals from wearable devices substantially improve statistical power when detecting physiological change. Using weekend changes in heart rate as a naturally recurring event over 46,000 wearable device users, I compare traditional cross-sectional analyses with within-individual sampling methods. Within-individual sampling not only reduced the sample sizes needed to achieve statistical significance but also uncovered inter-individual heterogeneity in responses that aggregate analyses obscure. These findings underscore the importance of individualized baselines for interpreting changes in physiological data. Building on this principle, Chapter Two investigates distal skin temperature as an age-related signal associated with baselines sleep patterns. The rationale is that thermoregulation and sleep are tightly coupled. Thus, given that daytime sleepiness, sleep-wake patterns, and sleep architecture change across the lifespan, we reason that thermoregulation (assessed via distal skin temperature) may reveal sleep-related changes associated with aging. Using wearable device data from over 20,000 individuals, I find age-related differences that align with known changes in sleep and circadian physiology in distal skin temperature patterns around naps and habitual nap times. Extending from a single physiological modality to a multimodal perspective, in Chapter Three, I develop a wearable device-based biological age model that integrates a wide range of features to predict chronological age. Using wearable device data from over 13,000 individuals, I identify which features are most informative for age prediction and validate the model by showing that deviations between predicted and actual age (i.e., age acceleration) correlate with disease incidence. This Dissertation supports the use of explainable models run on wearable device data to improve precision in detecting physiological changes associated with age-related conditions.