Seven Trials: AFib Screening Led to Only 7% Fewer Strokes

Slightly, and only just. A meta-analysis of seven randomized trials covering 76,458 people found screening cut stroke or systemic embolism by 7% (risk ratio 0.932, 95% CI 0.873 to 0.996). Major bleeding and all-cause mortality did not change. Excluding the one trial that used implantable loop recorders, the reduction was not significant (0.942, 0.880 to 1.008). The smartwatch study in this evidence measured how often notifications were confirmed, not stroke outcomes.

Does AFib screening reduce stroke in randomized trials?

Slightly, by one pooled estimate. A 2025 meta-analysis combined seven randomized trials in people with no known AFib. The trials covered 76,458 participants. One used implantable loop recorders. The others used non-invasive screening.

Screening was linked to a 7% relative reduction in all-cause stroke or systemic embolism. The risk ratio was 0.932 (95% CI 0.873 to 0.996, P = 0.037). Major bleeding (0.996) and all-cause mortality (0.987) did not change.

The interval ends close to 1. When only non-invasive methods were counted, the reduction was 0.942 (0.880 to 1.008, P = 0.083), not significant. The authors estimate a number needed to screen of 148 to prevent one event over 10 years in a population of 75-year-olds.

The same paper opens by noting that none of the individual trials had shown a significant stroke reduction.

The Cochrane review of systematic screening found a single cluster-randomized trial in people over 65, in which both invitation-to-ECG screening and pulse checks during GP visits found more atrial fibrillation than routine practice, with the gain seen in men and not in women. It measured detection, not strokes (The Cochrane Database of Systematic Reviews, 2016).

What did the large individual trials find?

STROKESTOP invited every 75- to 76-year-old in two Swedish regions to screening, 28,768 people in all. Participants without known AFib registered intermittent ECGs for 14 days. Follow-up was a median of 6.9 years. The primary outcome was a composite of stroke, systemic embolism, bleeding needing hospital care and death. It occurred in 31.9% of the invited group and 33.0% of controls (hazard ratio 0.96, 95% CI 0.92 to 1.00, P = 0.045). The authors called it a small net benefit.

LOOP implanted loop recorders in adults aged 70 to 90 with stroke risk factors. AFib detection tripled. Stroke or systemic embolism was 4.5% against 5.6%, a hazard ratio of 0.80 (0.61 to 1.05). It was not significant.

GUARD-AF screened people aged 70 and older in US primary care with a 14-day patch. Among 11,905 participants, stroke hospitalization was 0.7% against 0.6% (hazard ratio 1.10, 0.69 to 1.75). Enrollment ended early and event rates were low.

A genetic subgroup analysis of LOOP is covered in whether genetic risk decides who benefits.

In SCREEN-AF, 856 people aged 75 and over with high blood pressure were randomized. Two two-week rounds of a continuous ECG patch found atrial fibrillation in 5.3 percent against 0.5 percent with usual care, a tenfold increase, and most of those found started anticoagulation. The trial measured detection, and its authors call for larger trials with hard clinical outcomes (JAMA Cardiology, 2022).

What did the smartwatch study measure?

Not stroke. The Apple Heart Study recruited 419,297 participants without self-reported AFib. Over a median of 117 days, 2,161 people (0.52%) received an irregular pulse notification.

Of 450 notified participants who returned analyzable ECG patches, 34% had AFib on the patch. Of 1,376 notified participants who returned a 90-day survey, 57% contacted health care providers outside the study. The positive predictive value of a notification confirmed by a simultaneous ECG was 0.84 (95% CI 0.76 to 0.92).

That answers how often a notification is right. A second measure, agreement with an irregular tachogram, gave 0.71 (0.69 to 0.74). Apple funded the study. It does not show that notified people had fewer strokes. Detection accuracy is covered in how accurately a wearable ECG detects AFib. Device inconclusive rates are in wearable ECG comparison.

What does this not settle?

The pooled effect is small and sensitive to which trials are included. Its confidence interval nearly touches 1. Removing the one implantable-recorder trial made it non-significant.

Most trials enrolled adults aged 70 or older. Devices ranged from intermittent ECG to patches to implants. That differs from a consumer smartwatch worn by a younger person with no risk factors. Whether the pathway from watch notification to confirmation to treatment lowers stroke risk has not been tested in a randomized trial in this evidence.

Common questions