On November 5, Google published a study in Nature: using only the optical heart-rate and accelerometer sensors of an ordinary smartwatch, AI can screen for signs of hypertension without any cuff calibration. The system, called AI-PPG-ACC-HTN, was tested in a prospective study of 196 people against 24-hour ambulatory blood pressure monitoring, the clinical gold standard, and its initial screening beat in-office blood pressure measurement. Hypertension is often called the "silent killer" — more than a billion people worldwide have it, and over half do not know — and smartwatch screening targets exactly that gap.
How a Watch "Sees" Blood Pressure
The system does not directly estimate systolic and diastolic values at a given moment; instead, it judges whether a person has a persistent, long-term hypertensive state. The approach works in two stages: first, a one-dimensional convolutional neural network is pre-trained with self-supervision, consuming 492,347 hours of sensor data from 2,109 volunteers to learn features from 15-second segments of photoplethysmography (PPG) and accelerometry waveforms; then, each person's features are averaged across multiple days to wash out transient noise from movement and temperature, combined with height (a proxy for arterial path length), and fed to a logistic regression classifier that outputs a hypertension probability score. The accelerometer is not just counting steps here — it captures the subtle wrist vibrations as the heart pumps blood into the aorta, the ballistocardiogram signal.
Tested in 196 People: Both Scores Beat In-Office Measurement
Validation took place at five sites in the US, with 196 participants who had no known cardiovascular disease and took no blood pressure medication. In the initial screening scenario, the smartwatch system reached 65.8% sensitivity, 90.0% specificity, and 80.6% positive predictive value, while in-office screening in the same comparison managed only 55.3% sensitivity and 77.8% positive predictive value. In the confirmatory scenario (48 participants with elevated office blood pressure), the watch reached 78.4% sensitivity, 90.9% specificity, and 96.7% positive predictive value, versus 67.6%, 63.6%, and 86.2% for repeat office measurement, and 89.2%, 81.8%, and 94.3% for multi-day home monitoring. In other words, the watch misses fewer cases than the clinic at initial screening and is more accurate at confirmation, reaching overall accuracy close to traditional cuff devices within the existing clinical workflow.
Still a Few Steps from a "Watch That Measures Blood Pressure"
But the authors stress this is early feasibility research, not a standalone diagnosis and not a feature of any product on sale; its role is opportunistic screening — quietly flagging high-risk people during days they already wear a watch, so they go to a clinic for a formal check. The findings already inform the blood pressure trends feature, Health Guardian, in the Google Health app. Google is not the only one on this path: Apple Watch hypertension alerts are likewise alerts, not measurements. What to watch next is validation in larger, more diverse populations, and whether regulators treat it as a screening tool for approval.
