Smartwatches spot atrial fibrillation accurately. The catch is who is wearing them

Consumer smartwatches detect atrial fibrillation with an accuracy that looks, on paper, close to clinical — and not only on Apple's hardware. A diagnostic meta-analysis published in JACC: Advances in November 2025 pooled 26 studies covering 17,349 patients and put the overall sensitivity of smartwatch AF detection at 95% (95% CI 92%–97%) and specificity at 97% (95% CI 94%–98%), with a pooled area under the curve of 0.97 [s1]. The important qualifier is not in the accuracy figures but in who generated them, and what the number means when a healthy person wears the watch instead.

Atrial fibrillation is the most common sustained arrhythmia and a major, treatable risk factor for stroke, which is why a wrist device that can flag it has genuine appeal [s1]. This is squarely diagnostic territory, and the evidence deserves to be read with the same care as any screening test.

Non-Apple devices held up

Because Apple dominates the coverage of wearable heart features, it is worth stating what the pooled data showed for the rest of the market. Samsung devices yielded 97% sensitivity (95% CI 92%–99%) and 96% specificity (95% CI 91%–98%); the Withings Scan Watch showed 89% sensitivity (95% CI 75%–96%) and 95% specificity (95% CI 88%–99%) [s1]. Apple Watch, for comparison, reached 94% sensitivity (95% CI 89%–96%) and 97% specificity [s1]. The review also found accuracy was comparable between devices using photoplethysmography — the optical pulse sensor — and those using a single-lead electrocardiogram [s1].

The wide confidence interval on the Withings sensitivity, and the very high statistical heterogeneity the authors report across studies, are a signal that these headline numbers are averages over quite different devices, populations and testing conditions [s1].

That photoplethysmography performed comparably to electrocardiogram-based detection is worth pausing on [s1]. The optical method — the same green-light pulse sensor used for heart rate — screens continuously and passively, while the single-lead ECG requires the user to touch the watch to take a reading. A device can therefore use the always-on optical sensor to notice an irregular rhythm, then prompt the user to capture a confirmatory ECG strip. The meta-analysis suggests both steps are accurate; it does not resolve how often the passive stage fires when nothing is wrong.

The number that changes off the wrist

A test's sensitivity and specificity are properties of the test; how useful a positive result is depends on how common the disease is in the person tested. The studies in this meta-analysis were largely done in patients already suspected of, or known to have, arrhythmia — a group in which AF is common [s1]. In that setting, a positive reading is usually a true positive.

Move the same watch to a young, healthy population where AF is rare, and the arithmetic shifts: even a specific test throws off false positives that can outnumber the true cases, because there are so few true cases to find. The pooled AUC of 0.97 is a property of the test in the studied groups, not a guarantee of how it behaves on an unselected wrist [s1]. Nor is the burden symmetric across devices: the Withings sensitivity of 89%, the lowest in the analysis, means roughly one in nine true cases went undetected in those studies — a reminder that a normal reading is not a clean bill of rhythm, and that "validated" is a claim about a specific device, not the category [s1]. A review guiding non-experts on consumer wearables makes the general version of this point — that validity and accuracy figures have to be read alongside the population and the practical use, not lifted from a specification sheet [s2]. The consequence is real-world: a false AF alert leads to anxiety, an unnecessary clinic visit, sometimes a monitor or an anticoagulation conversation that was never warranted.

What is and is not settled

What the evidence supports is narrow but real. In people with symptoms or known risk, a smartwatch AF alert from a validated device is worth acting on, and the strongest independent syntheses now put heart-rate measurement and AF detection among the better-supported wearable biometrics [s1][s2]. What the evidence does not support is treating an alert from a healthy wrist as a diagnosis. An irregular-rhythm notification is a prompt to seek a proper electrocardiogram, not a substitute for one — the confirmation still comes from the clinic, not the watch [s1][s2].

This distinction between a validated measurement and a clinical claim is exactly the line the FDA drew when it let wearables estimate physiological signals while restricting what they may say about them, which we covered in the agency's general-wellness guidance. For AF, the watch has earned the right to raise a hand. It has not earned the right to make the call.