Google Research has presented AI-PPG-ACC-HTN, a system designed to passively detect signs of hypertension from standard consumer smartwatch sensors without calibration from an arm cuff. Rather than estimating exact blood pressure at a single moment, it looks for a persistent hypertensive state by combining physiological signals collected over several days.
The system uses photoplethysmography, which measures changes in blood volume through reflected light, and accelerometer data that can capture motion and subtle ballistocardiogram vibrations produced by the heart. A neural network was pretrained on 492,347 hours of unlabeled sensor data from 2,109 consenting participants. The resulting waveform features are averaged over time, combined with the user’s height, and classified into a hypertension probability score.
Researchers evaluated the approach in a prospective study involving 196 participants across five US sites, using 24-hour ambulatory blood pressure monitoring as the clinical reference. For initial screening, the smartwatch system recorded 65.8% sensitivity, 90.0% specificity, and an 80.6% positive predictive value. In a 48-person confirmatory subgroup, it reached 78.4% sensitivity, 90.9% specificity, and a 96.7% positive predictive value.
The researchers describe the work as early feasibility research. The system is not intended to replace clinical cuffs or provide a standalone diagnosis, but could support opportunistic screening by prompting at-risk users to seek formal assessment.