Anticoagulation for Atrial Fibrillation with Intermediate Stroke Risk (SINGLE-AF)

Authors, Journal, Affiliations, Type, DOI

• Authors: Daehoon Kim, Young Soo Lee, Jaemin Shim, Hee Tae Yu (all four contributed equally), Junbeom Park, Jin-Kyu Park, Il-Young Oh, Ki-Woon Kang, Eue-Keun Choi, Kyoung-Min Park, Hyoung-Seob Park, Dae-Hyeok Kim, Hyung Wook Park, Jun Kim, Jum-Suk Ko, Dongmin Kim, Jong-Youn Kim, Jin-Bae Kim, Jung-Hoon Sung, Tae-Hoon Kim, Jae-Sun Uhm, Hui-Nam Pak, and Boyoung Joung (corresponding author), for the SINGLE-AF Investigators
• Journal: N Engl J Med 2026;395(12):1155–65. Published online 28 August 2026; print issue 24 September 2026. A Quick Take and CME are available at NEJM.org
• Affiliations: 18 centres in South Korea. Lead site Division of Cardiology, Severance Hospital, Yonsei University College of Medicine, Seoul; other sites include Daegu Catholic, Korea University, Ewha Womans, Hanyang, Seoul National University Bundang, Chung-Ang, Samsung Medical Center, Keimyung, Inha, Chonnam National, Asan Medical Center, Wonkwang, Dankook, Gangnam Severance, Kyung Hee, and CHA Bundang
• Type: Investigator-initiated, multicentre, open-label, adjudicator-masked, randomized superiority trial with blinded event adjudication by an independent clinical-events committee. Registered NCT04437654. Rationale and design published separately (Kim D, et al. Heart Rhythm O2 2025;7:545–51)
• DOI: 10.1056/NEJMoa2607978

Overview

SINGLE-AF is the first randomized trial to test directly what guidelines have always answered by extrapolation: whether oral anticoagulation benefits patients with atrial fibrillation at intermediate stroke risk — CHA2DS2-VASc 1 in men or 2 in women, which is equivalent to CHA2DS2-VA = 1 in both sexes — a group for which both the 2023 AHA/ACC and 2024 ESC guidelines give only a Class IIa "consider" recommendation while conceding that randomized evidence is lacking. In 1,803 patients randomized 1:1 to a DOAC or no anticoagulation across 18 South Korean centres, DOAC therapy reduced the 24-month composite of stroke, systemic embolism, major bleeding, or cardiovascular death from 1.5% to 0.5% (difference −1.0 percentage points, 95% CI −2.0 to −0.1; P=0.03; hazard ratio 0.31, 95% CI 0.10–0.94), driven by fewer strokes (1.1% vs 0.3%), with major bleeding essentially the same in both groups and no cardiovascular deaths in either arm. The trial is notable for its restraint in interpretation: the authors state that the benefit estimate is based on very few events and "may overestimate the true magnitude," that confirmatory trials or pooled analyses are needed, and that the untreated group's annual stroke incidence (~0.6%) was lower than older nationwide cohorts from Denmark (1.55%) and Taiwan (2.75%) and matched a contemporary European study (0.7%).

Keywords

Not provided — NEJM research articles do not carry an author keyword list. The document's own indexing terms would be: atrial fibrillation; intermediate stroke risk; CHA2DS2-VASc; CHA2DS2-VA; direct oral anticoagulant; apixaban; rivaroxaban; stroke prevention; major bleeding; randomized controlled trial.

Key Takeaways

Background

• AF increases stroke risk. At higher CHA2DS2-VASc scores (range 0–9), the expected reduction in thromboembolism with oral anticoagulation generally outweighs bleeding risk
• Current US and European guidelines strongly recommend oral anticoagulants for men with a score ≥2 and women with a score ≥3 (Class I)
• For intermediate risk — CHA2DS2-VASc 1 in men and 2 in women — guidelines give only a Class IIa recommendation to consider oral anticoagulants, and the trial authors state plainly that "evidence from randomized trials supporting this approach is limited"
• Observational studies had provided conflicting evidence on net benefit in this group
• DOACs offer a better risk–benefit balance than VKAs in AF, mainly through lower intracranial haemorrhage. Some phase 3 DOAC trials included low-risk patients, but none directly compared anticoagulation with no treatment in intermediate-risk patients — the gap SINGLE-AF was designed to fill

Methods

• Design: investigator-initiated, multicentre, open-label, randomized superiority trial with blinded adjudication of outcome events; 18 centres in South Korea; 1:1 randomization by a Web-based system with permuted-block design (block sizes 4 or 6), stratified by trial site
• Population: adults ≥19 years with AF and intermediate stroke risk (CHA2DS2-VASc 1 in men, 2 in women). Key exclusions: clinically significant renal disease (serum creatinine ≥3.5 mg/dL or CrCl <30 mL/min) or liver disease (AST/ALT >3× upper limit of normal, or Child–Pugh B/C); anticoagulation already indicated for a mechanical prosthetic valve, moderate-to-severe mitral stenosis, or DVT; and clinically significant structural heart disease (dilated or hypertrophic cardiomyopathy, cardiac amyloidosis)
• Intervention: apixaban 5 mg twice daily or rivaroxaban 20 mg once daily, at the treating physician's discretion, with appropriate dose reductions for prespecified criteria; another DOAC could be substituted for unacceptable side effects. The no-anticoagulant group received no routine oral anticoagulation, though temporary DOAC was permitted around rhythm-control procedures (cardioversion, catheter ablation)
• Antiplatelet therapy was permitted in both groups when clinically indicated (post-coronary intervention or ACS) but routine use was not recommended
• Follow-up: baseline and 3, 6, 12, 18 and 24 months
• Primary end point: composite of stroke, systemic embolism, major bleeding (ISTH definition), or death from cardiovascular causes at 24 months
• Mid-trial end-point amendment: the primary end point originally included death from any cause. In January 2025 the independent data and safety monitoring board, unaware of group assignments, noted that all deaths reported to that point were non-cardiovascular and unlikely treatment-related; to prevent unrelated events diluting the treatment-effect assessment, the primary end point was amended to cardiovascular death only, with all-cause death retained as a secondary outcome. The change was prespecified in the design publication and did not alter the original sample-size or power assumptions
• Power: assuming 2.0% primary events in the DOAC group and 4.5% in the no-anticoagulant group, 1,800 patients (900 per group) gave 80% power to detect a 2.5-percentage-point difference at two-sided alpha 0.05, assuming 12% attrition
• Statistics: intention-to-treat; cumulative incidence by the Aalen–Johansen method with non-cardiovascular death as a competing event; treatment effects by Fine–Gray subdistribution hazard models; death from any cause by Kaplan–Meier. A prespecified sensitivity analysis censored follow-up when the CHA2DS2-VASc score rose to a Class I indication (≥2 in men, ≥3 in women). An additional win-ratio analysis used an unmatched-pairs approach with a hierarchical sequence of CV death, intracranial haemorrhage, ischaemic stroke or systemic embolism, then major bleeding excluding intracranial haemorrhage. No multiplicity adjustment was made for secondary end points or subgroups

Results — Patients and Treatment

• 1,803 patients randomized between 27 July 2020 and 29 June 2023: 902 to DOAC, 901 to no anticoagulation
• Mean age 60.4 ± 7.9 years; 23.7% women; 71.8% paroxysmal AF; median time from AF diagnosis to randomization 21 months (IQR 2–59)
• Baseline risk scores were well balanced: mean CHA2DS2-VASc 1.3 ± 0.5 (median 1), mean CHA2DS2-VA 1.0, mean CHADS2 0.7, mean HAS-BLED 0.5 ± 0.6 (median 0–1) — a population with low bleeding risk as well as intermediate stroke risk
• Treatment actually received: 902/902 (100%) of the DOAC group and 50/901 (5.5%) of the no-anticoagulant group received a DOAC. Most common regimen was apixaban 5 mg (67.4%), then rivaroxaban 20 mg (14.1%) and 15 mg (13.9%). Full doses in 81.6% and reduced doses in 18.4%
• Antiplatelet therapy: 0.4% of the DOAC group vs 36.5% of the no-anticoagulant group — a major between-group asymmetry
• Median follow-up 730 days; 94.6% in both groups completed 24-month follow-up

Results — Primary and Secondary End Points

• Win ratio 3.31 (95% CI 1.15–9.54) favouring DOAC
• The prespecified censoring sensitivity analysis (censoring at the point a Class I anticoagulation indication emerged) was consistent with the primary analysis
• Effects appeared consistent across prespecified subgroups, with wide confidence intervals and no multiplicity adjustment. Notable point estimates: age <65 HR 0.26 (0.05–1.20) vs ≥65 HR 0.37 (0.07–1.91); women 0 events in the DOAC arm vs 2 in the control arm (HR not estimable); non-paroxysmal AF HR 0.15 (0.02–1.27) vs paroxysmal HR 0.44 (0.11–1.69); HAS-BLED 0 HR 0.22 (0.02–1.96) vs ≥1 HR 0.39 (0.11–1.45)
• Proportional hazards and proportional subdistribution hazards assumptions showed no evidence of violation (P=0.18 for the Cox model)

Results — Safety

• Serious adverse events: 80 (8.9%) in the DOAC group vs 84 (9.3%) in the no-anticoagulant group. The most common was hospitalization related to atrial fibrillation — 34 events in the DOAC group vs 15 in the no-anticoagulant group, an asymmetry the article reports without comment on its cause
• Both intracranial haemorrhages in the DOAC group were haemorrhagic strokes. The authors state that given the small number of events no definitive conclusion can be drawn from the numerical imbalance, but that these events "should be interpreted with caution because of their clinical severity and possible prognostic implications"

Discussion

• The untreated event rate was lower than older cohorts suggested. The no-anticoagulant group's 24-month stroke cumulative incidence of 1.1% corresponds to an annual incidence of ~0.6% — lower than Danish (1.55%) and Taiwanese (2.75%) nationwide observational studies, but consistent with a more contemporary multinational European study (0.7%). The authors read this as residual thromboembolic vulnerability that may benefit from targeted risk reduction, while acknowledging the rate is lower than historical series implied
• Cross-trial stroke rates in untreated/control arms: ALONE-AF no-anticoagulation 0.1%, OCEAN aspirin 0.6%, NOAH-AFNET 6 placebo 1.1%, ARTESIA aspirin 1.2%. SINGLE-AF's ~0.6% sits closest to the trials enrolling patients with higher arrhythmia burden (it enrolled ~30% non-paroxysmal AF), and higher than the post-ablation populations of ALONE-AF and OCEAN despite those trials having higher mean CHA2DS2-VASc scores. The authors caution that these cross-trial comparisons are limited by differences in populations, AF burden, and baseline risk
• Bleeding risk is the pivot on which net benefit turns. The 24-month major bleeding incidence of 0.3% (~0.2%/yr) reflects a population with a mean HAS-BLED of 0.5. The authors contrast this directly with ARTESIA and NOAH-AFNET 6 — which enrolled patients at higher clinical risk (median CHA2DS2-VASc 4) and saw DOAC-associated major bleeding of 2.1% and 1.5% per patient-year respectively
• The authors' own bottom line: the lower incidence of primary end-point events in the DOAC group "was based on very few events and may overestimate the true magnitude of benefit. Therefore, although this randomized trial supports the benefit of anticoagulant therapy in this population, confirmatory trials or pooled analyses are needed to refine the precision and magnitude of the treatment effect."

Limitations of the document

Authors' own stated limitations

• Open-label design. End-point events were adjudicated by an independent clinical-events committee unaware of group assignments, but the open-label design "may have introduced bias"
• Imprecision. The lower-than-anticipated incidence of primary end-point events limited the precision of the treatment-effect estimate, and the authors state the results "should be interpreted with caution." The entire primary result rests on 17 events (4 vs 13), and the 95% CI for the absolute difference (−2.0 to −0.1) has a bound touching zero
• Incomplete follow-up. 5.4% of patients in each group did not complete 2-year follow-up, despite similar baseline characteristics among those with and without complete follow-up
• Antiplatelet imbalance. Antiplatelet therapy was used in approximately one third of the no-anticoagulant group versus 0.4% of the DOAC group, which "may have increased the risk of bleeding and put this group at a disadvantage with regard to the safety comparison." The effect of this imbalance on the efficacy comparison is not analysed — aspirin confers partial stroke protection, so the direction of net bias is ambiguous
• Generalisability. The trial included only East Asian patients from a single country; women and patients with non-paroxysmal AF may be underrepresented. The authors cite the literature on racial differences in ischaemic and haemorrhagic stroke and on the "East Asian paradox" in antithrombotic strategy, implying that both the stroke and bleeding rates — and therefore the net benefit — may not transfer to other populations

Additional limitations evident in the design

• The primary end point was amended mid-trial, in January 2025, to remove all-cause death. The change was made by a blinded monitoring board, prespecified in the design publication, and did not alter power assumptions — but it remains an end-point change made after unblinded data had accrued, and it converts the composite into effectively stroke, systemic embolism, or major bleeding, since cardiovascular death occurred in zero patients in both arms
• No multiplicity adjustment was applied across 15 secondary and exploratory end points or across subgroups; the article states explicitly that confidence-interval widths "should not be used to infer definitive treatment effects." The most striking result in the paper — HR 0.10 (0.01–0.77) for ischaemic stroke or systemic embolism — is a post hoc analysis
• The win-ratio analysis was not prespecified as a primary or key secondary analysis and produced an extremely wide confidence interval (1.15–9.54)
• The 'no anticoagulant' arm was not treatment-free: 5.5% crossed over to a DOAC and temporary DOAC was permitted around cardioversion and ablation, which dilutes the contrast between arms and biases the result toward the null
• No placebo, and the safety comparison is confounded by the antiplatelet asymmetry described above
• Two combined industry and government funders (Samjin Pharmaceutical, Hanmi Pharmaceutical alongside the Korean Ministry of Health and Welfare). The article states the funders had no role in design, conduct, analysis, or manuscript preparation
• Publication date. Both guidelines whose recommendation this trial tests (AHA/ACC 2023, ESC 2024) predate it, so no guideline currently incorporates this evidence. The trial was designed explicitly to address a Class IIa evidence gap and will require guideline update to change practice

Key Concepts Mentioned

• concepts/CHA2DS2-VA — the CHA2DS2-VASc 1 (men) / 2 (women) entry criterion is equivalent to CHA2DS2-VA = 1 in both sexes, making SINGLE-AF the first RCT to populate the "consider OAC at score 1" recommendation
• concepts/Subclinical-AF — SINGLE-AF speaks directly to the open question of whether OAC net benefit holds at lower CHA2DS2-VASc scores, and its bleeding-rate contrast with ARTESIA and NOAH-AFNET 6 is the key reconciliation
• concepts/NOAC-AF-Management — dosing reality in an intermediate-risk population (81.6% full dose, 18.4% reduced)
• Intermediate stroke risk (CHA2DS2-VASc 1 men / 2 women) — the population with a Class IIa recommendation and, before this trial, no randomized evidence
• Competing-risk analysis (Aalen–Johansen cumulative incidence, Fine–Gray subdistribution hazards) — non-cardiovascular death modelled as a competing event
• Win ratio — hierarchical unmatched-pairs composite analysis, used here as a supporting analysis

Key Entities Mentioned

• entities/Atrial-Fibrillation — primary entity
• entities/Ischemic-Stroke — the outcome the intervention is intended to prevent
• Apixaban and rivaroxaban — the two DOACs used, at the treating physician's discretion
• Yonsei University College of Medicine, Severance Hospital — lead site
• Ministry of Health and Welfare, South Korea (Korea Health Technology R&D Project) — primary funder, with Samjin Pharmaceutical and Hanmi Pharmaceutical

Wiki Pages Updated

• Created: wiki/sources/single-af-nejm-2026.md
• Updated: wiki/entities/Atrial-Fibrillation.md
• Updated: wiki/concepts/CHA2DS2-VA.md
• Updated: wiki/concepts/Subclinical-AF.md
• Updated: wiki/concepts/NOAC-AF-Management.md
• Updated: wiki/sourceindex.md
• Updated: wiki/wikiindex.md