Smartwatches spot heart rhythm problems with 97% accuracy

Heart rhythm abnormalities carry significant morbidity, mortality and financial cost, yet they are notoriously hard to catch. Many arrhythmias are episodic, flaring and vanishing between appointments, so conventional investigations often come back clean even when something is wrong. A sensor strapped to the wrist all day and night could, in principle, catch what a snapshot test misses. With smartwatches now mainstream, the question is no longer whether people will use them for heart monitoring but whether the readings deserve trust, and this review set out to measure exactly that.

The authors systematically searched Embase, MEDLINE and the Cochrane Library following PRISMA guidelines for studies reporting smartwatch detection of cardiac arrhythmia. Eighteen studies qualified, together measuring diagnostic accuracy in 424,371 people. The arrhythmias examined included atrial fibrillation, bradyarrhythmias and tachyarrhythmias, and premature contractions, with the watch signals based on photoplethysmography, the light-based pulse sensing built into consumer wearables. Summary estimates of sensitivity, specificity and area under the curve were produced with a bivariate model, and study quality was assessed with the QUADAS-2 tool.

The pooled sensitivity for detecting cardiac arrhythmias was 100%, meaning the watches essentially did not miss true cases across the included studies. Specificity came in at 95% and overall accuracy at 97%. The negative predictive value was 100%, so a clean reading was highly reassuring. The notable weak spot was the positive predictive value of 85%: among readings flagged as abnormal, a meaningful share were false positives, which matters enormously when a technology is worn by millions of mostly healthy people.

This meta-analysis puts firm pooled numbers behind what individual studies had suggested: consumer wrist sensors have become genuinely good at spotting rhythm abnormalities, particularly at ruling them out. The authors frame it as evidence of an evolving field of digital disease detection rather than a finished case for screening. They are explicit that accurate evidence accrual and regulatory standards ready to accept these devices are still needed. The open questions are about what happens after an alert, how false positives play out at population scale, and whether detection translates into better outcomes.

If you wear a smartwatch, these numbers are a reason to treat a rhythm notification as information worth following up rather than noise, and equally a reason not to panic, since a flag is not a diagnosis and some alerts will be false alarms. The near-perfect negative predictive value cuts the other way too: the absence of alerts from a device you wear daily carries some genuine reassurance. What this evidence does not yet support is the idea of the watch as a formal screening programme; the regulatory and evidential groundwork for that is still being laid.