Sensori AI Model Predicts Health from Wrist Movement Data.

Human health and daily function are deeply intertwined with movement patterns, much of which occurs outside clinical settings. Wrist-worn accelerometers continuously capture these movements, but their rich data is often oversimplified into basic behavioral summaries. This research introduces Sensori, a self-supervised foundation model designed to extract general-purpose health representations directly from 24 hours of raw tri-axial wrist movement data. Sensori was developed and validated using extensive datasets from four population-based cohorts across the UK, China, and the US, encompassing 122,640 participants and over 683,000 person-days of free-living recordings. The model successfully condensed daily movement into representations that captured diverse movement behaviors, demographic traits, various health axes, and physical function. Crucially, these representations demonstrated strong generalization capabilities across different populations and measurement environments without requiring retraining. When integrated with standard clinical covariates, Sensori significantly enhanced the classification of 52 prevalent diseases and the risk prediction for 26 incident diseases, showing the most substantial improvements for neurological and psychiatric conditions. These findings highlight the immense potential of continuous wrist movement data as a scalable and rich source of health information for passive monitoring and population-level disease prediction.