A smartwatch model that spots likely hypertension without a cuff

Google researchers trained a model on watch pulse and motion data to flag probable high blood pressure, catching more cases than office readings in a small study.

Google Research has built an AI system that flags likely high blood pressure using only a standard smartwatch. It needs no cuff for calibration. The work was published in Nature and described in a Google blog post.

The model uses two sensors already in modern watches, a pulse sensor and a motion sensor. A neural network was trained on almost 500,000 hours of sensor data. The training was self-supervised. The model compared short data segments and learned which ones came from the same person, with no diagnosis labels. The system then averages the learned pulse and heartbeat patterns over several days and outputs a probability of hypertension.

In a validation study of 196 participants at five US research sites, the watch method identified 65.8 percent of people with high blood pressure. Standard in-office measurement identified 55.3 percent. Google says the findings are built into the blood pressure trends feature of the Google Health app. It also says more than half of people with high blood pressure don't know they have it.

This matters for abundance because screening is the bottleneck. If a device people already own can prompt a check, early detection stops depending on a clinic visit or a separate cuff. That would make a basic piece of preventive care cheaper and more widely available.

The caveats are large. It is still a research project and does not replace a diagnosis. The stated aim is to prompt people to get checked. The sample of 196 is small, and the unawareness figure comes from Google. Watch for larger and more varied trials.

Source: The Decoder