EdoxabaN foR IntraCranial Hemorrhage Survivors With Atrial Fibrillation

ENRICH-AF

Patient / Population Intervention / Exposure Comparison Outcome

In 948 patients with atrial fibrillation and a previous intracranial haemorrhage, edoxaban did not reduce stroke or systemic embolism, and major haemorrhage compared with no anticoagulation.

N

948patients

Design

International PROBE RCT (open-label, blinded outcome adjudication), phase 3/4, 1:1, event-driven until 123 adjudicated primary efficacy events; 174 sites in 20 countries, enrolment from September 2019 with follow-up to 2026; after the 2023 DSMB review no further patients with lobar intracerebral or convexity subarachnoid haemorrhage were enrolled

Endpoint

Primary efficacy — stroke (ischaemic, haemorrhagic or unspecified) or systemic embolism; primary safety — ISTH major haemorrhage

Relevance

2Important — one of several pillars.

ResultStroke or systemic embolism 11.8% vs 12.8% (HR 0.88, 95% CI 0.61-1.26; P = 0.48) over a mean 28 months. The composite hides two opposite effects: ischaemic stroke 6.2% vs 10.3% (HR 0.57, 0.36-0.90) against haemorrhagic stroke 5.8% vs 1.9% (HR 2.99, 1.41-6.37). ISTH major bleeding 11.6% vs 5.2% (HR 2.23, 1.39-3.59; P < 0.001), fatal bleeding 4.1% vs 0.8%. Death 25.3% vs 23.7%. Recurrent intracranial haemorrhage on edoxaban ran at 7.4 per 100 patient-years after lobar ICH or convexity SAH, against 2.4 after non-lobar ICH.

Shoamanesh A, Sharma M, Hart RG, et al. Presented at ESC Congress 2026, Hot Line 2, 29 August. Not yet published.

Discussion & critique

The largest randomised trial of anticoagulation after intracranial haemorrhage, and the answer is that the two effects cancel: edoxaban cut ischaemic stroke by 43% and tripled haemorrhagic stroke, leaving the composite flat and major bleeding doubled. The trial found its own contraindication along the way — the safety board removed patients with lobar ICH or convexity subarachnoid haemorrhage in 2023 after recurrent bleeding on edoxaban reached about 11% against 1%, and the final data put that group at 7.4 recurrent haemorrhages per 100 patient-years. It closed with 116 of 123 planned events, so 78% power for the effect it was designed to detect. Read as a selection question, not a yes-or-no one: non-lobar bleeds may be anticoagulated, amyloid-pattern bleeds should not.