Google Research AI detects hypertension signs from smartwatch PPG and motion
Nature study / Google blog: multimodal model vs 24-hour ABPM in 196 adults; initial screening 65.8% sensitivity, 90.0% specificity, 80.6% PPV; research-only, not a commercial diagnosis.
Why it matters
Instead of estimating cuff-like point blood pressures, the system aggregates multi-day PPG and accelerometer embeddings to classify a persistent hypertensive state without cuff calibration. Performance was competitive with office workflows against ABPM gold standard, including confirmatory testing after elevated office readings. Google says findings informed Health Guardian trends research and stresses the method is exploratory, not a device claim. Strengthens the medicine beat beside Evident IVUS.
Key facts
- Published Nature; Google Research blog dated 2026-11-06T11:00:00Z
- 196 adults, five US sites; no known CVD; not on BP meds
- Initial screening: 65.8% sens, 90.0% spec, 80.6% PPV vs ABPM
- Confirmatory subset N=48: 78.4% sens, 90.9% spec, 96.7% PPV
- Explicitly research-only; not a commercial product claim