Your watch says it has detected an irregular rhythm. Has it diagnosed atrial fibrillation?

Not necessarily. The alert may be useful, but it sits at the start of a clinical question rather than the end. Most watches monitor the pulse optically. Atrial fibrillation, or AF, is an electrical rhythm diagnosis. That distinction matters because the next decisions can include further monitoring, assessment of stroke risk and, for some people with confirmed AF, an anticoagulant medicine that reduces clotting.

Consumer technology has made rhythm screening available on millions of wrists. The evidence asks two separate questions: can a device identify a signal worth checking, and does screening people without symptoms prevent strokes? The first answer is encouraging. The second is much less certain.

What the watch is measuring

Many smartwatches use photoplethysmography, or PPG. Light emitted into the skin changes as blood volume rises and falls with each pulse. An algorithm examines the intervals between those pulse waves and may flag a pattern compatible with AF.

PPG does not record the heart’s electrical activity. Movement, poor skin contact, premature beats and other rhythms can also produce an irregular pulse. Some watches can separately record a short, single-lead electrocardiogram (ECG) when the wearer activates that function. That electrical trace can be diagnostically useful if it is clear enough and reviewed by someone qualified to interpret it.

Australian guidance recommends opportunistic AF screening from age 65 and prefers devices that provide a medical-quality ECG trace, because an ECG is required to confirm the diagnosis. The 2024 European guideline makes the same boundary explicit: a PPG alert alone is not sufficient to start AF risk assessment and management.

A useful alert is still not a diagnosis

The Apple Heart Study shows both the reach and the limitation of an optical alert. It enrolled 419,297 US watch owners who reported no previous AF. Over a median 117 days, 2,161 people—0.52%—received an irregular-pulse notification. Those participants were offered a telemedicine review and an ECG patch to wear for up to seven days.

Only 450 returned a patch with analysable data, applied an average of 13 days after the notification. AF appeared on the later patch in 34%, with a 97.5% confidence interval from 29% to 39%. Among the smaller group who received another alert while actually wearing the patch, 72 of 86 had AF at the same time. That produced a positive predictive value of 84%, with a 95% confidence interval from 76% to 92%.

Those percentages answer different questions. The 84% figure describes a notification occurring while an ECG patch was recording. The 34% figure describes AF found days later among the minority who returned a usable patch. AF can come and go, so a normal later recording does not prove the original alert was false. Conversely, a watch notification without an interpretable ECG does not prove AF occurred.

The study had no control group and did not test whether alerts prevented stroke. Its participants owned compatible devices and chose to enrol; they do not represent the whole Australian population. Apple funded the work. No serious app-related adverse events were reported, but 57% of notified survey respondents contacted a health professional outside the study, illustrating how an alert creates downstream care as well as information.

Finding more AF is not the same as preventing more strokes

AF can allow blood to pool in the atrium and form a clot that travels to the brain. In people with confirmed clinical AF, anticoagulants can substantially reduce stroke risk when the likely benefit outweighs bleeding risk. The harder screening question is whether brief, silent episodes found only through intensive monitoring carry the same risk and respond to the same treatment strategy.

GUARD-AF randomised 11,905 US primary-care patients aged 70 or older to a 14-day continuous ECG patch or usual care. After a median 15.3 months, AF had been diagnosed in 5.0% of the screening group and 3.3% of the usual-care group. Anticoagulants were started in 4.2% and 2.8%, respectively.

The extra diagnoses did not translate into fewer stroke admissions during the trial: 0.7% with screening and 0.6% with usual care, a hazard ratio of 1.10 with a wide 95% confidence interval from 0.69 to 1.75. Bleeding admissions were 1.0% and 1.1%. COVID-19 stopped recruitment at fewer than a quarter of the planned 52,000 participants, and stroke events were uncommon. The result is therefore imprecise; it is not proof that screening has no benefit.

The possible benefit is small and population-specific

A 2025 systematic review pooled seven randomised screening trials involving 76,458 people. Their weighted average age was 75.7 years, and they generally had stroke risk factors. Over an average 5.1 years, screening was associated with a 7% relative reduction in stroke or systemic embolism: risk ratio 0.932, with a 95% confidence interval from 0.873 to 0.996.

The estimate only just crossed the conventional threshold for statistical significance. Assuming a 10% stroke risk over 10 years, the authors calculated an absolute reduction of about 0.7 percentage points and an estimated 148 people screened to prevent one event. The confidence interval around that number ranged from 79 to 2,501. When the analysis was restricted to non-invasive screening methods, the stroke result was no longer statistically significant. Major bleeding and overall mortality did not differ.

None of those trials tested ordinary smartwatch self-screening across healthy adults of all ages. They tested organised programmes, mainly in older people at higher baseline risk, using defined ECG protocols and clinical follow-up. A younger person with a consumer alert and a 75-year-old invited into a structured screening trial are not interchangeable populations.

What an alert can reasonably do

A watch alert can preserve a time-stamped clue and prompt an appropriate ECG. Symptoms, the quality of any recorded trace, how long and how often an episode occurs, age, prior stroke, blood pressure, diabetes, heart disease and bleeding risk all change what follows. The presence of AF is one part of an anticoagulation decision, not the whole decision.

A useful question for a clinical conversation is: “Did this device record an interpretable ECG, or did it only detect an irregular pulse—and what confirmation would change management?” That keeps the technology in its proper role. It can help capture an intermittent signal. It cannot, by itself, supply the diagnosis, the stroke-risk estimate or the treatment plan.

Sources

Evidence note: This Insight is general education, not individual medical advice or a diagnosis. An irregular-rhythm alert, symptoms and decisions about monitoring or anticoagulation should be discussed with a qualified clinician.