Smartwatch AI flags high BP about as well as an office check

Clinical takeaway: A patient who reports a hypertension alert from a smartwatch still needs cuff confirmation, ideally with home or ambulatory monitoring. But treat alerts from systems like this one as a potential signal. 

More than half of US adults with uncontrolled hypertension don’t know they have it. The US Preventive Services Task Force recommends screening all adults, annually for those 40 and older or at higher risk. Limited access to routine care and uneven screening habits still leave cases undetected. 

Wearables seem like a natural solution, but cuffless blood pressure monitoring has been challenging to implement. Most devices typically require calibration against a cuff and track changes over time rather than absolute pressure. Studies of calibration-free detection have typically benchmarked accuracy against home readings, which can blur true performance. This prospective validation tests a smartwatch AI against standard clinic screening, with gold-standard ambulatory monitoring as the reference. 

The watch detected 65.8% of hypertension cases confirmed by 24-hour ambulatory monitoring, versus 55.3% detected by the initial office reading. The difference was not statistically significant. Both held specificity at 90.0%, meaning about one in 10 adults without hypertension still screened positive. 

Among participants with hypertension that the watch missed, 65.4% also had a negative office reading. Among those the watch wrongly screened positive, 75% met some hypertension criteria on ambulatory readings, particularly overnight. Performance rose with wear time and began to level off at about five days with at least one hour of data collected daily. 

Researchers at Google followed 196 US adults across five sites with no known cardiovascular disease and no antihypertensive medication, 68.4% of them from a risk-enriched group. Each wore a Pixel Watch 3 at home for an average of about seven days alongside 24-hour ambulatory monitoring, the reference. Hypertension was defined as a 24-hour average blood pressure of at least 130/80 mmHg, whereas office screening used a threshold of 140/90 mmHg. The system is investigational, and the 90.0% specificity was a threshold the authors set during development. Classifier errors showed no statistical link to skin tone, which was dark in 35.2% of participants. 

The authors say the next tests are randomized trials of whether the system finds patients who would otherwise go undiagnosed and whether that improves outcomes, plus validation in larger, more globally representative groups. This cohort excluded adults with cardiovascular disease or on blood pressure medication, so use in people already treated, or for tracking response to treatment, remains untested. 

Source: Di Achille P, et al. (2026 Nov 5) Nature. Opportunistically Detecting Signs of Hypertension on a Consumer Smartwatch