Anticoagulation After Intracranial Hemorrhage and in Borderline Atrial Fibrillation

Bottom Line:

• Two decisions recur weekly at the bedside — restarting anticoagulation after intracranial hemorrhage (ICH), and anticoagulating "borderline" atrial fibrillation (AF) at CHA₂DS₂-VA of 1. Both are being reshaped by 2025–2026 phase-3 trials, and both remain evidence-limited.
• PRESTIGE-AF (Lancet 2025) showed direct oral anticoagulants (DOACs) markedly reduced ischemic stroke in ICH survivors with AF (HR 0·05, 95% CI 0·01–0·36) but failed non-inferiority for recurrent ICH (HR 10·89, 90% CI 1·95–60·72). Death and other endpoints were not powered.
• ENRICH-AF (ESC Congress 2026 Hot Line, August 2026, publication pending) enrolled 948 survivors of broader intracranial haemorrhage — a data safety monitoring board stopped enrollment of patients with lobar intraparenchymal or convexity subarachnoid haemorrhage in 2023 for safety. Edoxaban vs no anticoagulation: stroke/SE HR 0·88 (95% CI 0·61–1·26); major bleeding HR 2·23 (95% CI 1·39–3·59). Investigators explicitly concluded the trial "does not support" edoxaban in unselected patients; the primary manuscript, including any pre-specified dose-based (60 mg vs 30 mg) subgroup analyses, is awaited.
• BRAIN-AF (Nature Medicine 2026) was stopped for futility in genuinely low-risk AF (ages 30–62, no conventional risk factors — approximately 96% were CHA₂DS₂-VA of 0, not 1). Only 13 participants had stroke as their first qualifying primary event (16 strokes overall); 91·4% of the primary composite was cognitive decline. It does not directly interrogate the CHA₂DS₂-VA of 1 population and does not validate or refute the 2024 ESC/EACTS Class IIa, LOE C recommendation at that score.
• Factor XIa inhibition split by indication in 2025–2026: asundexian 50 mg lost decisively to apixaban in AF (OCEANIC-AF stroke/SE HR 3·79) but reduced recurrent ischemic stroke on top of antiplatelet therapy in non-cardioembolic stroke (OCEANIC-STROKE HR 0·74). The class is not a drop-in DOAC replacement in AF.
• Left atrial appendage occlusion (LAAO) trials fragmented further: OPTION (NEJM 2025) showed non-inferiority to oral anticoagulation after AF ablation; CHAMPION-AF (Doshi et al., NEJM 2026) met non-inferiority against NOACs in general nonvalvular AF; CLOSURE-AF (NEJM 2026) failed non-inferiority to medical therapy in high-bleed-risk AF; LAAOS III (NEJM 2021) tested surgical, concomitant LAA occlusion during cardiac surgery on background OAC and showed benefit that was additive to anticoagulation — not a percutaneous substitute strategy.
• The 2022 AHA/ASA ICH guideline (Greenberg et al., Stroke 2022) frames post-ICH anticoagulation resumption in AF as "may be considered" (Class 2b); the "avoid long-term anticoagulation after lobar ICH" language commonly quoted was actually the 2015 AHA/ASA Class IIa/LOE B statement and applied specifically to warfarin-associated lobar ICH. The 2022 document continues to flag high recurrence risk in lobar/CAA-associated ICH as a factor weighing against restart without carrying the older statement forward at the same class. The 2024 ESC/EACTS AF guideline (Van Gelder et al., Eur Heart J 2024) gives Class IIa, LOE C for anticoagulation at CHA₂DS₂-VA of 1 in either sex and maintains percutaneous LAAO at Class IIb (LOE downgraded from B to C).

The basics

Two overlapping questions dominate this space. First: in an ICH survivor with AF, a mechanical valve, or venous thromboembolism, should anticoagulation be (re)started, when, and with what? Second: in AF at the low end of stroke-risk stratification — a CHA₂DS₂-VA of 1 in either sex under the sex-neutral 2024 European Society of Cardiology (ESC)/European Association for Cardio-Thoracic Surgery (EACTS) nomenclature — is the absolute benefit of anticoagulation large enough to accept the absolute bleeding risk? Mechanical valves, cancer-associated thrombosis, and venous thromboembolism after ICH are deliberately out of scope; each requires distinct evidence and separate treatment.

Definitions and phenotyping

Spontaneous ICH is the bleeding phenotype relevant here — traumatic ICH is out of scope. Location matters because small-vessel biology differs: deep (basal ganglia, thalamic, brainstem, cerebellar) hemorrhage typically reflects hypertensive arteriolopathy, whereas lobar hemorrhage in an older patient triggers cerebral amyloid angiopathy (CAA) evaluation. Probable and possible CAA are defined by the Boston criteria version 2.0 (not "modified 2.0"). Strictly lobar microbleeds were already included in v1.0; cortical superficial siderosis (cSS) was introduced in v1.5. The distinguishing addition of v2.0 is a formalized white-matter-marker pathway: severe visible perivascular spaces in the centrum semiovale or a multispot white-matter-hyperintensity pattern now count as CAA-associated non-hemorrhagic markers. Under v2.0, "possible CAA" can be diagnosed on one strictly lobar hemorrhagic lesion or one qualifying white-matter feature, when other Boston-2.0 conditions are satisfied (Charidimou et al., Lancet Neurology 2022, [DOI](https://doi.org/10.1016/S1474-4422(22)00208-3)). Boston-2.0 categories reflect diagnostic certainty rather than validated recurrence-risk strata; the diagnostic label alone does not translate into an ordinal rebleed risk.

Epidemiology at the bedside

Recurrent-ICH risk is not uniform. In PRESTIGE-AF's no-anticoagulation arm — spontaneous ICH survivors with AF at median age 79 — the ICH event rate was 0·82 per 100 patient-years (95% CI 0·14–2·53). In the DOAC arm, ICH events rose to 5·00 per 100 patient-years (95% CI 2·68–8·39), while ischemic stroke fell from 8·60 to 0·83 per 100 patient-years (Veltkamp et al., Lancet 2025). In BRAIN-AF, the annualized combined rate of stroke, systemic embolism, or motor-deficit/aphasic TIA in the placebo arm was 0·72%; major bleeding was 0·09%/year on rivaroxaban vs 0·21%/year on placebo (Khairy et al., Nature Medicine 2026, [primary paper](https://www.nature.com/articles/s41591-025-04101-y)). These figures do not directly transfer to the AF patient with age 65–74, heart failure, vascular disease, or diabetes — the qualifying CHA₂DS₂-VA-of-1 phenotypes the clinician actually faces.

Pathophysiology at a level that changes decisions

Hypertensive arteriolopathy and CAA share the small-vessel compartment but differ in geography and rebleed physics: CAA lesions are cortical and leptomeningeal, and cortical superficial siderosis marks convexity subarachnoid bleeding into the pia–arachnoid interface. Fibrin-rich, low-flow thrombus in the left atrial appendage is the anatomic rationale for LAAO. Factor XI amplifies thrombin generation and stabilizes pathological thrombi while contributing marginally to physiological hemostasis, as illustrated by the mild bleeding phenotype of congenital factor XI deficiency — the mechanistic premise of "hemostasis-sparing" anticoagulation (Ferrari & Villa, J Clin Pharmacol 2026, [DOI](https://doi.org/10.1002/jcph.70266)).

Diagnostic pointers

MRI with susceptibility-weighted or T2*-weighted sequences is required to grade CAA probability by Boston 2.0 — cerebral microbleed count and topography, cortical superficial siderosis, and now the white-matter-marker pathway all enter the classification (Charidimou et al., Lancet Neurology 2022). The PRESTIGE-AF neuroimaging sub-study identified cortical superficial siderosis, chronic macrohemorrhages, and non-lobar ICH as prognostically important MRI markers of subsequent stroke risk in the randomized cohort as a whole — but only 54·3% of participants had MRI, only 13 recurrent ICHs and 22 ischemic strokes occurred, and interaction with treatment assignment was not pre-specified. These are prognostic associations, not demonstrated effect modifiers of DOAC vs no anticoagulation (Fandler-Höfler et al., Neurology 2025, [DOI](https://doi.org/10.1212/WNL.0000000000214386)).

Where we are

One paragraph of history. For two decades, decisions after ICH were guided by observational cohorts (ICH registries, Danish nationwide data) suggesting that in AF patients who had ICH, restarting vitamin-K antagonists lowered ischemic events at the cost of higher recurrent ICH — net direction uncertain and driven by confounding by indication. Small RCTs followed but were underpowered: RESTART for antiplatelets (n=537, Lancet 2019), SoSTART for anticoagulation (pilot, n=203, Lancet Neurology 2021), APACHE-AF for apixaban vs no OAC (n=101, Lancet Neurology 2021). Guidelines therefore said "individualize" without an RCT-anchored restart date. That is the world PRESTIGE-AF and ENRICH-AF walked into in 2025–2026.

Two evidentiary anchors sit around these ICH-specific trials. First: DOACs are the class-of-choice for stroke prevention in nonvalvular AF requiring anticoagulation. The meta-analysis of RE-LY, ROCKET AF, ARISTOTLE, and ENGAGE AF-TIMI 48 showed a reduction in stroke or systemic embolism (RR 0·81, 95% CI 0·73–0·91) and, notably, in intracranial hemorrhage (RR 0·48, 95% CI 0·39–0·59) versus warfarin (Ruff et al., Lancet 2014). Second: guideline attribution matters, and the prior draft got this wrong. The "avoidance of long-term anticoagulation with warfarin after warfarin-associated spontaneous lobar ICH" recommendation is a 2015 AHA/ASA Class IIa, LOE B statement, not a 2022 statement. The 2022 AHA/ASA ICH guideline instead frames post-ICH anticoagulation resumption in AF as an individualized decision — anticoagulation resumption in nonvalvular AF after spontaneous ICH "may be considered" after weighing benefit and risk (Class 2b), and initiation at approximately 7–8 weeks "may be considered" (Class 2b) (Greenberg et al., Stroke 2022, [DOI](https://doi.org/10.1161/STR.0000000000000407)). The 2022 document does not carry the older lobar-specific avoidance statement forward at Class IIa, but continues to identify high recurrent-ICH risk in lobar/CAA-associated ICH as a consideration weighing against resumption. The class change should not be read as an endorsement of restart in lobar ICH.

For the borderline-AF question, the 2024 ESC/EACTS AF guideline formally moved from CHA₂DS₂-VASc to the sex-neutral CHA₂DS₂-VA, treating sex as a risk modifier rather than a discrete criterion. Anticoagulation at CHA₂DS₂-VA ≥2 is Class I; at CHA₂DS₂-VA of 1 (in either sex) it is Class IIa, LOE C — a shared decision, not a default. Percutaneous LAAO is maintained at Class IIb for patients with contraindications to long-term OAC, with the level of evidence downgraded from B (2020) to C (2024) — this is a cautionary revision, not class strengthening (Van Gelder et al., Eur Heart J 2024, [DOI](https://doi.org/10.1093/eurheartj/ehae176)).

What is new

The two phase-3 post-ICH trials, read together

PRESTIGE-AF (Veltkamp et al., Lancet 2025, [DOI](https://doi.org/10.1016/S0140-6736(25)00333-2)) is a multicentre, open-label, phase 3 trial at 75 hospitals across six European countries. Between 2019 and 2023 it randomly assigned 319 spontaneous-ICH survivors with AF (median age 79, 35% female) to DOAC or no anticoagulation. The trial had two coprimary endpoints tested hierarchically: first ischemic stroke (superiority) and first recurrent ICH (non-inferiority with margin <1·735). At a median follow-up of 1·4 years, first ischemic stroke was reduced with DOAC (HR 0·05, 95% CI 0·01–0·36; log-rank P<0·0001; event rate 0·83 vs 8·60 per 100 patient-years). Non-inferiority for recurrent ICH was not met — HR 10·89 (90% CI 1·95–60·72; P=0·96); event rates 5·00 vs 0·82 per 100 patient-years. Death occurred in 16 (10%) of the DOAC group and 21 (13%) of the no-anticoagulation group — a numerical trend without a powered inference. The authors' verbatim bottom line: DOACs "effectively prevent ischaemic strokes … but a part of this benefit is offset by a substantially increased risk of recurrent intracerebral haemorrhage." PRESTIGE-AF is a demonstration that both endpoint arrows move in the expected directions, with an ICH signal large enough to fail a pre-specified non-inferiority margin; it is not a demonstration of net clinical benefit.

ENRICH-AF (Shoamanesh et al., ESC Congress 2026 Hot Line, August 29, 2026; publication pending; ClinicalTrials.gov NCT03950076) enrolled 948 survivors of broader intracranial haemorrhage — not restricted to spontaneous intracerebral haemorrhage — with AF, across 174 sites in 20 countries. Two design features are essential to the interpretation. First, enrollment included patients with intracranial bleeds "within or around the brain," a definition wider than spontaneous IPH. Second, in 2023 the data safety monitoring board stopped enrollment of patients with lobar intraparenchymal haemorrhage or convexity subarachnoid haemorrhage for safety concerns, so the final population is enriched in deep and non-CAA ICH etiologies. Patients were randomized to edoxaban (60 mg daily; dose-adjusted to 30 mg per label criteria) versus no anticoagulation (no antithrombotic therapy or single antiplatelet, per investigator judgment). The primary composite of stroke or systemic embolism was 11·8% with edoxaban vs 12·8% without (HR 0·88, 95% CI 0·61–1·26). Major bleeding was 11·6% vs 5·2% (HR 2·23, 95% CI 1·39–3·59), with a hemorrhagic-stroke excess offsetting the ischemic-stroke reduction. Investigators concluded that their findings "do not support the use of edoxaban in unselected patients with atrial fibrillation after intracranial haemorrhage" and highlighted the need for an individualized decision-making approach ([ESC press release](https://www.escardio.org/news/press/press-releases/safer-stroke-prevention-strategies-are-needed-for-patients-with-atrial-fibrillation-after-intracranial-haemorrhage/); primary manuscript pending). Dose-based (60 mg vs 30 mg dose-adjusted) subgroup analyses were referenced in the Hot Line presentation but should be interpreted from the published manuscript rather than paraphrased here.

Reading them together. The two trials told the same directional story: anticoagulation lowered ischemic-stroke risk and raised hemorrhagic-stroke risk. What differs is how each trial summarized the trade-off. PRESTIGE-AF measured the two arrows as separate coprimary endpoints and reported the ICH arrow's failure of non-inferiority. ENRICH-AF used a stroke/SE composite that combined ischemic and hemorrhagic events and read out neutral, with a doubled bleeding hazard on the safety side. Populations differed (ENRICH-AF broader intracranial bleed types, PRESTIGE-AF spontaneous IPH), agents differed (edoxaban vs DOAC broadly), and follow-up lengths and CAA-etiology mixes differed. The 2023 safety-driven ENRICH-AF exclusion of lobar IPH and convexity SAH is particularly important: it means the trial under-represents and cannot provide a precise efficacy estimate for the highest-CAA-probability subgroup neurologists most fear. Neither trial pre-specified CAA-based subgroup analyses powered for interaction. An updated patient-level meta-analysis (COCROACH consortium — see below) is anticipated.

Hemorrhage location, cortical superficial siderosis, and CAA as prognostic markers

Lobar ICH and probable CAA carry higher recurrent-ICH risk than deep hypertensive ICH — a consistent observational signal. Within randomized populations, the strongest signal to date comes from the PRESTIGE-AF neuroimaging sub-study, which reported prognostic associations in the cohort as a whole: cortical superficial siderosis (HR 7·7, 95% CI 1·4–42·2 for recurrent ICH), chronic macrohemorrhages on MRI (HR 9·1, 95% CI 1·8–46·8), and non-lobar ICH (HR 9·1, 95% CI 1·2–67·7 for subsequent ischemic stroke) — with only 54·3% of participants imaged, 13 recurrent ICHs, and 22 ischemic strokes across 313 imaged patients, and without pre-specified interaction testing across treatment arms (Fandler-Höfler et al., Neurology 2025). Calling these HRs "effect modifiers of DOAC treatment" or "post-restart risk" overstates what was tested; they are prognostic associations across the whole cohort. The 2022 AHA/ASA ICH guideline itself frames the post-ICH anticoagulation decision as individualized (Class 2b overall), and the "avoidance of long-term anticoagulation after lobar ICH" statement most commonly cited belongs to the 2015 guideline (Class IIa, LOE B, specific to warfarin-associated lobar ICH). The 2022 update discusses lobar/CAA-associated ICH as a high-recurrence phenotype in the accompanying text but does not carry the earlier statement forward at Class IIa.

The RESTART trial (Al-Shahi Salman et al., Lancet 2019) is often cited in this location discussion — but RESTART tested antiplatelet restart vs avoidance in ICH survivors (n=537), not anticoagulation. Recurrent symptomatic ICH occurred in 12 (4%) of 268 antiplatelet-assigned participants vs 23 (9%) of 268 avoiders (adjusted HR 0·51, 95% CI 0·25–1·03, P=0·060); location-subgroup interactions did not reach statistical significance. Extrapolation from RESTART to the anticoagulation decision is not warranted.

Timing of restart

The clinician's question is "how many weeks?" The honest answer in 2026 is that no RCT has established a restart date after spontaneous ICH. The 4-week benchmark originated from the 2015 warfarin-era guideline; the 2022 AHA/ASA update discusses considerations rather than mandating an interval, offering that initiation at approximately 7–8 weeks may be considered (Class 2b). Neither figure is anchored by an RCT-derived cut-point.

A separate — and importantly distinct — timing question sits alongside: after ischemic stroke with AF, when should DOAC be started? The CATALYST prospective individual-participant-data meta-analysis pooled TIMING, ELAN, OPTIMAS, and START (n=5441; mean age 77·7) comparing early (≤4 days) vs later (≥5 days) DOAC. The primary 30-day composite of recurrent ischemic stroke, symptomatic ICH, or unclassified stroke occurred in 2·1% vs 3·0% (OR 0·70, 95% CI 0·50–0·98, P=0·039). Recurrent ischemic stroke: 1·7% vs 2·6% (OR 0·66, 95% CI 0·45–0·96, P=0·029). Symptomatic ICH: 0·4% vs 0·4% (OR 1·02, 95% CI 0·43–2·46, P=0·96) (Dehbi et al., Lancet 2025, [DOI](https://doi.org/10.1016/S0140-6736(25)00439-8)). This is a post-ischemic-stroke timing result and does not transfer to the post-ICH restart question.

LAA occlusion — a more fragmented landscape than a year ago

Five data points, none of which directly answers the ICH-survivor question:

• PROTECT AF long-term follow-up (Reddy et al., JAMA 2014) reported the composite of stroke, systemic embolism, and cardiovascular death of 2·3 vs 3·8 events per 100 patient-years for WATCHMAN vs warfarin at a mean 3·8 years; the Bayesian posterior probability of superiority reached the pre-specified boundary. Nonvalvular AF eligible for warfarin — not the ICH survivor.
• LAAOS III (Whitlock et al., NEJM 2021, [DOI](https://doi.org/10.1056/NEJMoa2101897)) randomized 4811 AF patients undergoing cardiac surgery for another indication, of whom 4770 were included in the primary analysis, to concomitant surgical LAA occlusion or not, on background anticoagulation (76·8% still on OAC at 3 years). Ischemic stroke or systemic embolism occurred in 4·8% vs 7·0% (HR 0·67, 95% CI 0·53–0·85, P=0·001). This was surgical exclusion performed during cardiac surgery for another indication with continued anticoagulation, not percutaneous LAAO substituting for it. The benefit was additive to anticoagulation.
• OPTION (Wazni et al., NEJM 2025, [DOI](https://doi.org/10.1056/NEJMoa2408308)) enrolled 1600 post-ablation AF patients (mean CHA₂DS₂-VASc 3·5) and randomized to WATCHMAN FLX vs oral anticoagulation. At 36 months, non-procedure-related major or clinically relevant non-major bleeding was 8·5% vs 18·1% (P<0·001 superiority); the composite of death, stroke, or systemic embolism was 5·3% vs 5·8% (P<0·001 non-inferiority). Post-ablation — not the ICH survivor.
• CHAMPION-AF (Doshi et al., NEJM 2026, [DOI](https://doi.org/10.1056/NEJMoa2517213); published March 28, 2026 online / June 4, 2026 issue) randomized 3000 nonvalvular AF patients (mean CHA₂DS₂-VASc 3·5) to WATCHMAN FLX vs NOAC. At 3 years, the composite of ischemic stroke, hemorrhagic stroke, cardiovascular death, or systemic embolism was 5·7% (LAAO) vs 4·8% (NOAC) (HR 1·20), meeting the pre-specified non-inferiority margin (P<0·001 for non-inferiority); non-procedural major and clinically relevant non-major bleeding was 10·9% vs 19·0% (HR 0·55, P<0·001); ischemic stroke was numerically higher with LAAO (3·2% vs 2·0%). General nonvalvular AF — this is not an ICH-survivor cohort.
• CLOSURE-AF (NEJM 2026, [DOI](https://doi.org/10.1056/NEJMoa2513310)) tested LAAO vs medical therapy in AF patients at combined high stroke and high bleeding risk (n=888; 446 device vs 442 medical). The trial did not meet non-inferiority for its composite primary endpoint of stroke, systemic embolism, major bleeding, or cardiovascular/unexplained death (16·8 vs 13·3 per 100 patient-years; adjusted RMST difference −0·36 years, 95% CI −0·70 to −0·01; P=0·44 for non-inferiority). Ischemic/hemorrhagic stroke rates were similar (2·6 vs 2·7/100 patient-years); major bleeding was not reduced (7·4 vs 6·2/100 patient-years, with 18 procedure-related bleeds in the device group). This is the read-out that most tempers reflexive LAAO substitution when anticoagulation is refused — but the enrolled population was general high-bleed-risk AF, not a dedicated ICH-survivor cohort.

Where does LAAO sit for the ICH survivor? Still with the ongoing dedicated trials — CLEARANCE, STROKECLOSE, and follow-on work — rather than by extrapolation from OPTION (post-ablation), LAAOS III (surgical exclusion added to OAC), CHAMPION-AF (mixed AF, percutaneous vs NOAC), or CLOSURE-AF (general high-bleed-risk AF, negative for non-inferiority).

Factor XIa inhibitors — an indication-specific class

The 2025–2026 phase-3 read-outs split cleanly by indication:

• AZALEA-TIMI 71 (Ruff et al., NEJM 2025, [DOI](https://doi.org/10.1056/NEJMoa2406674)) enrolled 1287 AF patients at moderate-to-high stroke risk (median age 74). Subcutaneous abelacimab 150 mg or 90 mg monthly was compared with oral rivaroxaban 20 mg daily. Free factor XI was reduced by a median 99% (150 mg) and 97% (90 mg). Major or clinically relevant non-major bleeding was 3·2 (150 mg) and 2·6 (90 mg) versus 8·4 per 100 patient-years with rivaroxaban; HR 0·38 (95% CI 0·24–0·60) and 0·31 (95% CI 0·19–0·51), both P<0·001. The trial was stopped early for greater-than-anticipated bleeding reduction and was not powered for ischemic-stroke efficacy — a phase-2 signal on safety, not a stroke-prevention verdict.
• OCEANIC-AF (Piccini et al., NEJM 2025) randomized 14,810 AF patients (mean age 73·9, mean CHADS-VASc 4·3) to asundexian 50 mg once daily vs standard-dose apixaban. The trial was stopped prematurely. Stroke or systemic embolism occurred in 98 patients (1·3%) on asundexian vs 26 (0·4%) on apixaban (HR 3·79, 95% CI 2·46–5·83). Major bleeding: 17 (0·2%) vs 53 (0·7%) (HR 0·32, 95% CI 0·18–0·55). The 50-mg once-daily asundexian regimen was clinically inferior to apixaban for stroke or systemic embolism prevention.
• OCEANIC-STROKE (Sharma et al., NEJM 2026, [DOI](https://doi.org/10.1056/NEJMoa2513880)) tested asundexian 50 mg on top of antiplatelet therapy after non-cardioembolic ischemic stroke or high-risk TIA. Among 12,327 patients randomized within 72 hours, ischemic stroke occurred in 6·2% (asundexian) vs 8·4% (placebo) — HR 0·74 (95% CI 0·65–0·84), P<0·001 — with no significant excess in major bleeding (1·9% vs 1·7%, HR 1·10, 95% CI 0·85–1·44). First phase-3 win for a factor-XIa inhibitor, and it is in the atherothrombotic secondary-prevention setting, not in AF.
• PACIFIC-Stroke (asundexian dose-finding after non-cardioembolic stroke, phase 2b, Lancet 2022) and AXIOMATIC-SSP (milvexian after non-cardioembolic stroke on DAPT, Lancet Neurology 2024) were neutral on their primary imaging-clinical composites, with reassuring safety profiles, and set up the phase-3 program.

The synthesis is indication-specific rather than class-defining. In AF, the 50-mg asundexian dose was inferior to apixaban — a clinical fact that stands regardless of the mechanistic explanation, which remains contested. In atherothrombotic secondary prevention, adding asundexian to antiplatelet therapy reduced recurrent ischemic stroke without a bleeding penalty. Whether higher-exposure XIa inhibition (LIBREXIA-AF, milvexian in AF) can meet apixaban is the pending pivotal question; whether abelacimab lowers ischemic events in AF patients considered unsuitable for anticoagulation is the pending post-ICH-relevant question (LILAC-TIMI 76).

The borderline AF question — what BRAIN-AF does and does not answer

The borderline AF patient — CHA₂DS₂-VA of 1 in either sex under the sex-neutral 2024 ESC nomenclature — sits where, for some score-1 phenotypes, the estimated absolute stroke benefit may be small and comparable with treatment-related bleeding risk. Both risks vary substantially with the qualifying risk factor, age, renal function, bleeding phenotype, and choice of anticoagulant, so no single population-average comparison covers the space. The 2024 ESC/EACTS guideline gives Class IIa, LOE C for anticoagulation at that score, taking bleeding risk and patient preference into account (Van Gelder et al., Eur Heart J 2024).

The BRAIN-AF trial (Khairy et al., Nature Medicine 2026, [primary paper](https://www.nature.com/articles/s41591-025-04101-y)) is often cited to close this question, but its design does not do that work. BRAIN-AF randomized 1235 AF patients aged 30–62 with low stroke risk (CHA₂DS₂-VASc 0–1 excluding female sex, no conventional risk factors such as prior stroke/TIA, hypertension, diabetes, or heart failure) to rivaroxaban 15 mg daily vs placebo. The primary composite of cognitive decline, stroke, or motor-deficit/aphasic TIA occurred in 130 vs 126 events — HR 1·10 (95% CI 0·86–1·40), P=0·46 — with the trial stopped for futility after median 3·7 years (conditional power 1·2%). Two features of the enrolled population are essential and were misrepresented in the prior draft: (1) only 51 of 1235 participants (4·1%) had vascular disease, the sole qualifying CHA₂DS₂-VA point available under the inclusion criteria, so approximately 96% of enrolled patients had a CHA₂DS₂-VA of 0, not 1; and (2) 91·4% of the primary composite events were cognitive decline. Only 13 participants had stroke as their first qualifying primary event; 16 strokes occurred overall in the trial (9 with rivaroxaban, 7 with placebo). The annualized combined rate of stroke, systemic embolism, or motor-deficit/aphasic TIA in the placebo arm was 0·72%; major bleeding was 0·09%/year on rivaroxaban and 0·21%/year on placebo. BRAIN-AF is therefore a strong argument against a routine anticoagulation default in genuinely low-risk AF (score 0, younger adults, no risk factors), but it does not directly interrogate the CHA₂DS₂-VA-of-1 population most clinicians face at the bedside — the older patient with hypertension, heart failure, vascular disease, or diabetes. The IIa/C recommendation stands on the older evidence base; BRAIN-AF neither validates nor refutes it.

What is coming

Read-outs to watch, with dates only where actually announced:

• LIBREXIA-AF — milvexian 100 mg twice daily vs apixaban in AF, phase 3, event-driven. Earlier program materials had projected a 2027 topline; by November 2025 and BMS's Q1 2026 investor presentation, LIBREXIA-AF was listed among expected 2026 registrational read-outs, subject to the event-driven timeline. The trial remained ongoing in August 2026 ([BMS November 2025 release](https://investors.bms.com/iframes/press-releases/press-release-details/2025/Update-on-Phase-3-Librexia-ACS-Trial/default.aspx); [BMS Q1 2026 investor presentation](https://www.bms.com/assets/bms/us/en-us/pdf/investor-info/doc_presentations/2026/BMY-2026-Q1-Results-Investor-Presentation-with-Appendix.pdf)). This is the definitive AF efficacy test for the XIa class after OCEANIC-AF's dose-dependent failure.
• LILAC-TIMI 76 (NCT05712200) — abelacimab 150 mg SC monthly vs placebo in AF patients deemed unsuitable for anticoagulation; primary endpoint ischemic stroke or systemic embolism. This is the trial with the greatest potential to reshape post-ICH-AF care.
• Ongoing LAAO trials in high-bleed-risk or ICH-survivor AF, including CLEARANCE and STROKECLOSE, exist to fill the gap OPTION, LAAOS III, CHAMPION-AF, and CLOSURE-AF do not.
• Updated COCROACH patient-level meta-analysis. The first prospective individual-participant-data COCROACH meta-analysis was already published in Lancet Neurology in 2023, pooling 412 participants and finding no clear net effect of anticoagulation resumption after ICH in AF ([COCROACH 2023](https://discovery.ucl.ac.uk/id/eprint/10179656/1/1-s2.0-S1474442223003150-main.pdf)). An updated analysis incorporating PRESTIGE-AF and ENRICH-AF is anticipated but the timeline is not confirmed in the cited materials.

An honest note: awaiting these trials is not itself an active management strategy. Nothing in the current evidence base licenses "wait for LILAC-TIMI 76" as a clinical plan.

Open questions

1. CAA-tailored decisions after ICH. No adequately powered RCT has stratified restart decisions by Boston 2.0 CAA probability. PRESTIGE-AF's neuroimaging sub-study is hypothesis-generating (13 recurrent ICHs; MRI in 54·3% of the cohort) and cortical superficial siderosis emerges as a consistent marker of subsequent ICH risk — but as a prognostic association, not as a demonstrated treatment interaction. ENRICH-AF's 2023 safety-driven exclusion of lobar IPH and convexity SAH means the CAA-probable end of the spectrum is under-represented in the phase-3 evidence.
2. Restart timing after ICH. The 4-week and 7–8-week benchmarks are guideline-derived considerations, not RCT-anchored. PRESTIGE-AF's median time from ICH to randomization and the ENRICH-AF entry window inform practice but do not define a threshold.
3. What "unsuitable for anticoagulation" actually means. The definition depends on investigator judgment in LILAC-TIMI 76 and on registry-level criteria in ongoing LAAO trials; the population is not standardized, which limits generalizability.
4. The CHA₂DS₂-VA-of-1 patient. BRAIN-AF's futility read-out is a strong argument against a broad default at CHA₂DS₂-VA of 0 in genuinely low-risk younger AF, but it does not resolve the score-of-1 question — the trial almost entirely enrolled score-0 patients. Heterogeneity among one-point risk factors (age 65–74, heart failure, vascular disease, diabetes) means absolute-risk estimates differ across the score-1 phenotypes.
5. Factor XIa in AF. OCEANIC-AF's failure is a clinical fact at the tested dose; whether the class as a whole meets apixaban at higher exposures awaits LIBREXIA-AF. The 50-mg mechanism explanation is contested and should not be asserted as settled.
6. Reconciling PRESTIGE-AF and ENRICH-AF. Both showed the same directional trade-off. Whether the anticipated COCROACH update, incorporating both trials, can identify an ICH-etiology or imaging subgroup with true net benefit is the pre-registered question.

Guideline comparison

Hemorrhage phenotypes and what the evidence supports

Factor XIa class snapshot

Key clinical trials

What the general neurologist needs to know

Established.

• DOACs are the class-of-choice for stroke prevention in nonvalvular AF requiring anticoagulation; they reduce intracranial hemorrhage compared with warfarin (Ruff meta-analysis, Lancet 2014).
• Early DOAC initiation (≤4 days) after ischemic stroke with AF reduces the 30-day composite without excess symptomatic ICH (CATALYST IPDMA, Lancet 2025). This applies to ischemic stroke, not to post-ICH restart.
• Surgical, concomitant LAA occlusion added to anticoagulation in AF patients undergoing cardiac surgery for another indication reduces ischemic stroke or systemic embolism (LAAOS III, NEJM 2021) — additive, not a percutaneous substitute strategy.
• OPTION (NEJM 2025) and CHAMPION-AF (Doshi et al., NEJM 2026) support percutaneous LAAO as non-inferior to oral anticoagulation in post-ablation AF and in general nonvalvular AF respectively — neither enrolled ICH survivors.

Uncertain.

• Whether to restart anticoagulation after spontaneous ICH in AF. PRESTIGE-AF (Lancet 2025) showed ischemic protection but failed ICH non-inferiority; ENRICH-AF (ESC 2026 Hot Line; publication pending) was neutral on stroke/SE with doubled major bleeding. Guidance remains individualized (2022 AHA/ASA, Class 2b).
• How Boston 2.0 CAA probability should modify the restart decision. Cortical superficial siderosis is the most consistent MRI marker of subsequent ICH risk in the PRESTIGE-AF neuroimaging sub-study, but as a prognostic association in the whole cohort, not a pre-specified treatment interaction. ENRICH-AF's 2023 exclusion of lobar IPH and convexity SAH under-represents the CAA-probable end of the phase-3 evidence.
• Whether factor-XIa inhibition can substitute for DOAC in AF (LIBREXIA-AF pending) or provide benefit in AF unsuitable for anticoagulation (LILAC-TIMI 76 pending).
• Whether percutaneous LAAO is a safe alternative when anticoagulation is refused. CLOSURE-AF (NEJM 2026) failed non-inferiority to medical therapy in high-bleed-risk AF, cautioning against extrapolation to the ICH survivor.
• What to do at CHA₂DS₂-VA of 1. The 2024 ESC/EACTS guideline recommendation (Class IIa, LOE C) stands on older evidence; BRAIN-AF (2026) argues against a routine default in genuinely low-risk (score-0) AF but does not directly test the score-1 population.

Watch for.

• Recurrent ICH signals in any post-ICH DOAC restart — cortical superficial siderosis, new lobar microbleeds, and chronic macrohemorrhages on interval MRI as imaging patterns of progression (not defined triggers).
• The published ENRICH-AF primary manuscript with any pre-specified dose-based (60 mg vs 30 mg) subgroup analyses.
• The updated COCROACH patient-level meta-analysis incorporating PRESTIGE-AF and ENRICH-AF; the 2023 COCROACH publication is the anchor and an update is anticipated.

References

1. Greenberg SM, Ziai WC, Cordonnier C, et al. 2022 Guideline for the Management of Patients With Spontaneous Intracerebral Hemorrhage: A Guideline From the American Heart Association/American Stroke Association. Stroke. 2022;53(7):e282–e361. [DOI](https://doi.org/10.1161/STR.0000000000000407)
2. Hemphill JC, Greenberg SM, Anderson CS, et al. Guidelines for the Management of Spontaneous Intracerebral Hemorrhage: A Guideline for Healthcare Professionals From the American Heart Association/American Stroke Association. Stroke. 2015;46(7):2032–2060. (Source of the Class IIa, LOE B avoidance-of-warfarin-after-lobar-ICH recommendation.)
3. Van Gelder IC, Rienstra M, Bunting KV, et al. 2024 ESC Guidelines for the management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS). Eur Heart J. 2024;45(36):3314–3414. [DOI](https://doi.org/10.1093/eurheartj/ehae176)
4. Charidimou A, Boulouis G, Frosch MP, et al. The Boston criteria version 2.0 for cerebral amyloid angiopathy: a multicentre, retrospective, MRI–neuropathology diagnostic accuracy study. Lancet Neurology. 2022;21(8):714–725. [DOI](https://doi.org/10.1016/S1474-4422(22)00208-3)
5. RESTART Collaboration. Effects of antiplatelet therapy after stroke due to intracerebral haemorrhage (RESTART): a randomised, open-label trial. Lancet. 2019;393(10191):2613–2623.
6. SoSTART Collaboration. Effects of oral anticoagulation for atrial fibrillation after spontaneous intracranial haemorrhage in the UK: the SoSTART RCT. Lancet Neurology. 2021.
7. Schreuder FHBM, van Nieuwenhuizen KM, Hofmeijer J, et al. Apixaban versus no anticoagulation after anticoagulation-associated intracerebral haemorrhage in patients with atrial fibrillation in the Netherlands (APACHE-AF). Lancet Neurology. 2021.
8. STATICH investigators. Study of Antithrombotic Treatment After Intracerebral Haemorrhage — Antiplatelets. Stroke. 2026. PROBE, n=69, antiplatelet vs avoidance of antithrombotic therapy; early-terminated for slow recruitment; no formal statistical inference on the pre-specified primary endpoint.
9. Veltkamp R, Korompoki E, Harvey KH, et al. Direct oral anticoagulants versus no anticoagulation for the prevention of stroke in survivors of intracerebral haemorrhage with atrial fibrillation (PRESTIGE-AF): a multicentre, open-label, randomised, phase 3 trial. Lancet. 2025;405(10482):927–936. [DOI](https://doi.org/10.1016/S0140-6736(25)00333-2)
10. Fandler-Höfler S, Ropele S, Gattringer T, et al. Neuroimaging Markers Associated With Recurrent Stroke in Intracerebral Hemorrhage and Atrial Fibrillation: Secondary Analysis of PRESTIGE-AF. Neurology. 2025;105(11):e214386. [DOI](https://doi.org/10.1212/WNL.0000000000214386)
11. Shoamanesh A, et al. ENRICH-AF: Edoxaban versus no anticoagulation in atrial fibrillation after intracranial haemorrhage. ESC Congress 2026 Hot Line (August 29, 2026). ClinicalTrials.gov NCT03950076. Primary manuscript pending. [ESC press release](https://www.escardio.org/news/press/press-releases/safer-stroke-prevention-strategies-are-needed-for-patients-with-atrial-fibrillation-after-intracranial-haemorrhage/)
12. Khairy P, Amit G, Andrade JG, et al. BRAIN-AF: Anticoagulation to prevent ischemic stroke and neurocognitive impairment in low-risk atrial fibrillation. Nature Medicine. 2026. [Primary paper](https://www.nature.com/articles/s41591-025-04101-y)
13. Dehbi H-M, Fischer U, Åsberg S, et al. Collaboration on the optimal timing of anticoagulation after ischaemic stroke and atrial fibrillation: a systematic review and prospective individual participant data meta-analysis of randomised controlled trials (CATALYST). Lancet. 2025;406(10498):43–51. [DOI](https://doi.org/10.1016/S0140-6736(25)00439-8)
14. Reddy VY, Sievert H, Halperin J, et al. Percutaneous left atrial appendage closure vs warfarin for atrial fibrillation: a randomized clinical trial (PROTECT AF long-term follow-up). JAMA. 2014;312(19):1988–1998.
15. Whitlock RP, Belley-Cote EP, Paparella D, et al. Left Atrial Appendage Occlusion during Cardiac Surgery to Prevent Stroke (LAAOS III). NEJM. 2021;384(22):2081–2091. 4811 randomized, 4770 in primary analysis. [DOI](https://doi.org/10.1056/NEJMoa2101897)
16. Wazni OM, Saliba WI, Nair DG, et al. Left Atrial Appendage Closure after Ablation for Atrial Fibrillation (OPTION). NEJM. 2025;392(13):1277–1287. [DOI](https://doi.org/10.1056/NEJMoa2408308)
17. Doshi SK, Reddy VY, et al. Left Atrial Appendage Closure vs NOACs in Atrial Fibrillation (CHAMPION-AF). NEJM. 2026;394(22) (June 4, 2026 issue; published online March 28, 2026). [DOI](https://doi.org/10.1056/NEJMoa2517213)
18. Landmesser U, et al. Left Atrial Appendage Closure vs Medical Therapy in High-Bleed-Risk AF (CLOSURE-AF). NEJM. 2026. [DOI](https://doi.org/10.1056/NEJMoa2513310)
19. Ruff CT, Giugliano RP, Braunwald E, et al. Comparison of the efficacy and safety of new oral anticoagulants with warfarin in patients with atrial fibrillation: a meta-analysis of randomised trials. Lancet. 2014;383(9921):955–962.
20. Shoamanesh A, Mundl H, Smith EE, et al. Factor XIa inhibition with asundexian after acute non-cardioembolic ischaemic stroke (PACIFIC-Stroke): an international, randomised, double-blind, placebo-controlled, phase 2b trial. Lancet. 2022;400(10357):997–1007.
21. Sharma M, Molina CA, Toyoda K, et al. Milvexian for Secondary Stroke Prevention (AXIOMATIC-SSP). Lancet Neurology. 2024.
22. Ruff CT, Patel SM, Giugliano RP, et al. Abelacimab versus Rivaroxaban in Patients with Atrial Fibrillation (AZALEA-TIMI 71). NEJM. 2025;392(4):361–371. [DOI](https://doi.org/10.1056/NEJMoa2406674)
23. Piccini JP, Patel MR, Steffel J, et al. Asundexian versus Apixaban in Patients with Atrial Fibrillation (OCEANIC-AF). NEJM. 2025. ClinicalTrials.gov NCT05643573.
24. Sharma M, Dong Q, Hirano T, et al. Asundexian for Secondary Stroke Prevention (OCEANIC-STROKE). NEJM. 2026;394(15):1467–1479. [DOI](https://doi.org/10.1056/NEJMoa2513880)
25. Al-Shahi Salman R, Stephen J, Tierney JF, et al. Effects of oral anticoagulation in people with atrial fibrillation after spontaneous intracranial haemorrhage (COCROACH): prospective, individual participant data meta-analysis of randomised trials. Lancet Neurology. 2023;22(12):1140–1149. [Open access repository copy](https://discovery.ucl.ac.uk/id/eprint/10179656/1/1-s2.0-S1474442223003150-main.pdf)
26. Bristol Myers Squibb. Update on Phase 3 Librexia ACS Trial (November 14, 2025). [Sponsor announcement](https://investors.bms.com/iframes/press-releases/press-release-details/2025/Update-on-Phase-3-Librexia-ACS-Trial/default.aspx)
27. Bristol Myers Squibb. Q1 2026 Results — Investor Presentation with Appendix. [BMS Q1 2026 investor presentation](https://www.bms.com/assets/bms/us/en-us/pdf/investor-info/doc_presentations/2026/BMY-2026-Q1-Results-Investor-Presentation-with-Appendix.pdf)
28. Ferrari F, Villa RF. Factor XI Inhibition for Ischemic Stroke Secondary Prevention: Pharmacology, Clinical Evidence, and Future Directions. J Clin Pharmacol. 2026;66(8):e70266. [DOI](https://doi.org/10.1002/jcph.70266)

Trial numbers are quoted from primary publications or, where the primary manuscript is pending, from the sponsor/congress source, with that status noted. Guideline attributions are corrected to the source document (2015 vs 2022 AHA/ASA; 2024 ESC/EACTS) with class and level of evidence preserved. Corrections in this revision: LIBREXIA-AF sponsor timeline chronology; BRAIN-AF age 30–62 and stroke event counting; CHAMPION-AF lead author (Doshi); 2022 AHA/ASA wording ("may be considered"); Boston 2.0 diagnostic-category framing; ENRICH-AF CAA-subgroup language.