← Back to Home

Can Smartwatches Detect Signs of Hypertension? Nature Study Validates Seven-Day AI Screening

A prospective study of 196 people tested wrist sensor signals against 24-hour ambulatory blood pressure monitoring. The results support the potential for everyday wearable devices to help identify risk, but have not demonstrated superiority over office measurements and cannot replace cuff-based measurements for diagnosis.

By SURL BioNews

One challenge with hypertension is that many people have no obvious symptoms and do not realize they need to check their blood pressure. If a watch worn every day could identify risk from routine signals, it could increase the likelihood of people undergoing formal testing. A study published in Nature on November 5 provided prospective validation of this approach: AI analysis of seven days of smartwatch data could identify some cases of hypertension, but still missed about one-third of people with the condition.

The system estimates a person’s probability of having hypertension; it does not directly display their current systolic or diastolic blood pressure. It combines changes in blood volume beneath the skin recorded by a photoplethysmography sensor with wrist signals captured by an accelerometer, then incorporates multiple days of data and the person’s height into the model, without requiring prior calibration using a blood pressure cuff. By accumulating signals, the research team hopes to reduce interference from brief movements and environmental changes.

According to Google Research’s description of the research, the model was pretrained using nearly 492,000 hours of sensor data provided by 2,109 consenting participants. Its clinical detection performance was tested in a prospective study involving 196 people at five sites in the United States. These participants had no known cardiovascular disease and were not taking antihypertensive medication. The team used 24-hour ambulatory blood pressure monitoring as the reference and compared smartwatch screening with the initial office blood pressure measurement.

After seven days of monitoring, the smartwatch system had a sensitivity of 65.8%, with a 95% confidence interval of 54.0% to 76.3%, compared with 55.3% for office measurements. Both had a specificity of 90.0%, meaning that about nine in ten people whom the reference test did not classify as having hypertension were correctly identified. Although the smartwatch’s sensitivity was numerically higher, the difference was not statistically significant, so these results cannot support a claim that it identifies people with hypertension better than office measurements.

The positive predictive value of smartwatch screening was 80.6%, meaning that, among these participants, about eight in ten people who received a positive result were confirmed to have hypertension by the reference test. This proportion also varies with the prevalence of hypertension in the population being tested. The team also conducted an exploratory analysis of 48 people whose initial office blood pressure was elevated, finding a sensitivity of 78.4% and a positive predictive value of 96.7%. However, this was a small, selected subgroup, and its results cannot be directly applied to general users.

The study’s value lies in comparing everyday wearable data with a more rigorous clinical reference; the limits of its applicability must be equally clear. These validation data are not yet sufficient to determine how the system performs in people already receiving antihypertensive treatment or those with existing cardiovascular disease. Nor has the study demonstrated that this screening can reduce clinical events such as stroke. Google Research describes it as an early feasibility study: a positive signal could prompt users to seek clinical evaluation, but diagnosis still requires cuff-based measurements, and a negative result cannot rule out hypertension.

References

  1. Nature Portfolio
  2. Nature
  3. Google Research