← Back to Home

Detecting Hypertension from Wrist Signals: Smartwatch AI Completes Prospective Validation in 196 People

Without prior calibration using a blood pressure cuff, the algorithm can look for signs of hypertension in everyday wear data. But initial screening still misses about one-third of cases, leaving some way to go before it can serve as a standalone diagnostic tool.

By SURL BioNews

If an everyday watch could remind people to have their blood pressure checked, it could offer another entry point for health monitoring. On November 5, Google Research introduced a study published in Nature that attempts to identify sustained hypertension from wrist sensor signals. Its aim is to identify who may need further testing, rather than directly report blood pressure at that moment.

The system, called AI-PPG-ACC-HTN, combines photoplethysmography and accelerometer data. The former uses reflected light to capture changes in blood volume caused by the pulse, while the latter provides signals from movement and subtle vibrations. The model aggregates several days of recordings and incorporates height to calculate the probability of hypertension, without requiring prior calibration using blood pressure cuff readings.

According to data released by Google, the prospective clinical study enrolled 196 people at five sites in the United States, using 24-hour ambulatory blood pressure monitoring as the reference standard. Participants had no known cardiovascular disease and were not taking blood pressure medications, so the results cannot yet be directly generalized to patients receiving treatment.

In an initial screening setting, the system had a sensitivity of 65.8% and a specificity of 90.0%. In more intuitive terms, based on its performance in this study, about 66 out of every 100 people who actually have hypertension would be identified, while about 90 out of every 100 people without hypertension would receive a negative result. This also means that the absence of a watch alert is still insufficient to rule out hypertension.

The study also analyzed 48 people whose blood pressure was elevated at their first clinic visit to assess the system’s ability to identify sustained hypertension. Sensitivity and specificity were 78.4% and 90.9%, respectively. However, this subgroup was small, and the higher figures alone do not establish that the system is ready to replace existing confirmatory testing.

The significance of this work lies in testing whether everyday wearable data can provide a prompt before formal assessment. Google describes it as an exploratory feasibility study and explicitly states that it cannot provide a standalone diagnosis or replace a standard upper-arm blood pressure monitor. Moving from signal detection to clinical use will require further clarification of its reliability across different populations and whether alerts actually lead to follow-up testing and care.

References

  1. Google Research