AI news ·
Google study finds smartwatch sensors can passively detect signs of hypertension
Google's AI system uses smartwatch PPG and accelerometer sensors to detect hypertension without cuff calibration, hitting 90.9% specificity and 96.7% positive predictive value in confirmatory testing.
A new AI system developed by Google Research uses standard consumer smartwatch sensors to passively detect signs of hypertension without requiring any cuff calibration, according to a study published November 5, 2026 in Nature. The research addresses a critical gap - more than half of the one billion adults worldwide with high blood pressure are unaware they have it, and missing diagnoses in younger adults aged 18 to 44 can delay interventions that prevent stroke or heart failure.
Two sensors, one predictive model
The system, called AI-PPG-ACC-HTN, reframes blood pressure assessment by identifying a person's persistent hypertensive state over multiple days rather than attempting to estimate exact systolic and diastolic values at a single moment. It draws on two sensors already active in modern smartwatches: photoplethysmography (PPG), which tracks blood volume changes at the wrist through reflected light, and accelerometry (ACC), which detects motion and captures the ballistocardiogram - the subtle wrist vibrations caused by blood ejecting from the heart into the aorta.
The machine learning pipeline operates in two stages. First, a 1-D deep convolutional neural network was trained using self-supervised contrastive learning on 492,347 hours of sensor data from 2,109 consenting participants. This waveform encoder learned to map overlapping 15-second PPG and ACC waveforms from the same individual close together in a latent space, discovering features of vascular and mechanical cardiac function without manual labels. Second, these embeddings are mean-aggregated across multiple days of wear and combined with the user's height - a reliable proxy for arterial path length - then fed into a logistic regression classifier that produces a continuous hypertension probability score.
Clinical validation against the gold standard
The team conducted a prospective, multi-center study across five US sites with 196 participants who were free from known cardiovascular disease and not taking blood pressure medications. Performance was benchmarked against 24-hour ambulatory blood pressure monitoring (ABPM), the clinical gold standard for hypertension diagnosis.
In an initial screening scenario, the smartwatch system achieved 65.8% sensitivity and 90.0% specificity, yielding a positive predictive value (PPV) of 80.6%. Office-based screening showed 55.3% sensitivity and 77.8% PPV against the same standard. For confirmatory testing among participants who screened positive in a clinic, the smartwatch reached 78.4% sensitivity, 90.9% specificity, and 96.7% PPV - outperforming return office measurements, which posted 67.6% sensitivity and 63.6% specificity. The smartwatch results also proved competitive with multi-day home monitoring (89.2% sensitivity, 81.8% specificity, 94.3% PPV).
What the model sees
Researchers examined average pulse waveforms from users at the extremes of predicted hypertension probability. Three patterns emerged: a reduced pulse width suggesting stiffer arteries less able to expand, a slower systolic upstroke with a lower d/a ratio indicating altered pulse wave reflection, and distinct modulations in the ballistocardiogram signal - deeper K waves and wider IJK complexes in those with higher predicted risk.
"Smartwatch-based hypertension screening is not intended to replace standard clinical arm cuffs or provide a definitive standalone medical diagnosis," the researchers said. Instead, the work points toward a passive, scalable tool for opportunistic screening that could prompt at-risk individuals to seek formal clinical assessment.
Why this matters for healthcare and science professionals
The study demonstrates that consumer wearables can detect physiological signatures of hypertension without periodic cuff calibration, a limitation of existing commercial solutions that only track relative changes. For clinical researchers and health system designers, the passive, multi-day aggregation approach offers a different screening model - one that reduces reliance on episodic office measurements, which showed lower sensitivity in the same study. Professionals working in AI for Healthcare Courses or population health will recognize the implications: a tool that nudges undiagnosed individuals into care pathways could shift the detection timeline for a condition that currently escapes notice in half of those affected.