Title: Digital Phenoptyping and Early Disease Detection using Consumer Wearable Devices

Abstract: Consumer wearable devices, including smartwatches, rings, and activity bands, continuously capture physiological and behavioral data from millions of individuals. These data streams enable estimation of individual baselines, detection of deviations, and risk stratification—the central promise of precision medicine. However, the field currently lacks standardized methods to rigorously analyze these high-dimensional, longitudinal datasets. In my defense, I will provide a practical blueprint for how high-dimensional wearable data can be rigorously analyzed and translated into clinical insights, by using interpretable, unsupervised, and state-aware modeling approaches. To demonstrate the richness of wearable data, I first introduce a retrospective clinical case report that integrates five years of multi-wearable data with clinical records of a 76-year-old patient, who had structural heart disease and experienced sudden cardiac death immediately after intense workout. This case report highlights the potential role of integrating continuous digital physiology with traditional cardiology for timely evaluation and intervention. Next, I present an interpretable framework for cohort-level phenotyping that leverages timestamped wearable metrics, clinically relevant risk strata, and unsupervised machine learning methods. Applied across two independent real-world cohorts, this method uncovers latent cohort structure, clinically meaningful outliers, and distributions of physiological risk that are otherwise obscured by standard reporting methods. Finally, by applying an ensemble of dimensionality reduction and unsupervised learning techniques on a longitudinal wearable and self-reported stress dataset, I reveal two reproducible modes of stress-physiology association that reflects either sympathetic ("fight-or-flight") or parasympathetic ("rest-and-digest") dominance that individuals transition between over time. My contributions establish a scalable, reproducible, and personalized inference from real-world wearable measurements, supporting future applications in precision medicine.

Zoom: Please contact Leyre Caracuel for the zoom link.