PRESTIGE-AF
(2025)Objective
To determine whether DOACs reduce the risk of ischaemic stroke without substantially increasing recurrent intracerebral haemorrhage in survivors of ICH with atrial fibrillation.
Study Summary
• However, DOACs significantly increased recurrent ICH (HR 10.89) and failed to meet the non-inferiority margin for recurrent ICH
• Net clinical benefit and all stroke/systemic embolism composites trended in favour of DOACs but CIs were wide
Intervention
DOAC (apixaban, dabigatran, edoxaban, or rivaroxaban at licensed stroke prevention doses) versus no anticoagulation (antiplatelet or no antithrombotic therapy)
Inclusion Criteria
Age ≥18 years, spontaneous ICH, atrial fibrillation with indication for anticoagulation, mRS ≤4, enrolled 14 days to 12 months after index ICH
Study Design
Arms: DOAC group vs No anticoagulation group
Patients per Arm: 158 DOAC, 161 no anticoagulation
Outcome
• First recurrent ICH: HR 10.89 (90% CI 1.95–60.72; p=0.96) — non-inferiority NOT met; 5.00 vs 0.82 per 100 patient-years
• All stroke and systemic embolism: event rate ratio 0.53 (95% CI 0.27–0.99) favouring DOACs
Bottom Line
DOACs dramatically reduce ischaemic stroke in ICH survivors with atrial fibrillation (HR 0.05), but this benefit is partly offset by a substantially increased risk of recurrent ICH (HR 10.89). Non-inferiority for recurrent ICH was not established. Composite endpoints and mortality trended in favour of DOACs but were not statistically conclusive.
Major Points
- First completed phase 3 RCT of DOACs vs no anticoagulation in ICH survivors with atrial fibrillation
- DOACs reduced first ischaemic stroke with HR 0.05 (95% CI 0.01–0.36; p<0.0001); ischaemic stroke rate was 0.83 vs 8.60 per 100 patient-years (NNT 13 per year)
- Non-inferiority for recurrent ICH was NOT met: HR 10.89 (90% CI 1.95–60.72; p=0.96); ICH rate was 5.00 vs 0.82 per 100 patient-years (NNH 24 per year)
- All stroke and systemic embolism composite favoured DOACs (event rate ratio 0.53, 95% CI 0.27–0.99)
- Net clinical benefit composite (all stroke/SE, MI, CV mortality, major bleeding) trended towards DOACs (19.20 vs 26.52 per 100 patient-years; ratio 0.67, 95% CI 0.33–1.36)
- Cardiovascular mortality point estimates favoured DOACs (2.92 vs 5.73 per 100 patient-years; HR 0.51, 95% CI 0.21–1.27)
- Major bleeding was significantly higher with DOACs (8.75 vs 2.05 per 100 patient-years; ratio 4.27, 95% CI 1.74–12.80)
- Recurrent ICH risk did not appear to differ by lobar vs non-lobar haematoma location, though events were too few for definitive conclusions
- 85% of patients were anticoagulated prior to index ICH; median time from ICH to enrolment was 49 days
- No patients were lost to follow-up; 10% crossover rate was lower than anticipated
Study Design
- Study Type
- Multicentre, open-label, randomised, phase 3 trial
- Randomization
- Yes
- Blinding
- Open-label. Only the events adjudication committee was masked to treatment allocation, drug use, and participant identity via redaction of source documents. Patients, treating clinicians, and investigators were unmasked.
- Sample Size
- 319
- Follow-up
- Median 1.4 years (IQR 0.7–2.3); minimum 6 months, up to 36 months
- Centers
- 75
- Countries
- UK, Germany, Austria, Spain, Italy, France
Primary Outcome
Definition: Coprimary endpoints: (1) First incident ischaemic stroke (tested for superiority); (2) First recurrent intracerebral haemorrhage (tested for non-inferiority with margin HR <1.735)
| Control | Intervention | HR/OR | P-value |
|---|---|---|---|
| Ischaemic stroke: 20 events (8.60 per 100 patient-years); Recurrent ICH: 1 event (0.82 per 100 patient-years) | Ischaemic stroke: 1 event (0.83 per 100 patient-years); Recurrent ICH: 11 events (5.00 per 100 patient-years) | Ischaemic stroke: 0.05; Recurrent ICH: 10.89 (Ischaemic stroke: 0.01–0.36 (95% CI); Recurrent ICH: 1.95–60.72 (90% CI)) | Ischaemic stroke: p<0.0001; Recurrent ICH: p=0.96 (non-inferiority NOT met) |
Limitations & Criticisms
- Low number of primary outcome events and short follow-up (median 1.4 years) resulting in broad confidence intervals limiting interpretability
- Open-label design — only the adjudication committee was masked, potentially introducing bias in reporting and management
- Study was underpowered for secondary endpoints; trends in net benefit and mortality require confirmation in larger trials
- Original target sample size was 654 but revised to ~312 due to slow recruitment (partly COVID-19 related), limiting statistical power
- Exclusion of patients with severe disability (mRS >4), which may have enriched for patients with small haematoma volumes (median 3–4 mL), limiting generalisability
- Predominantly White European population (>95%) limits generalisability to other ethnicities
- Women were under-represented (35% of participants)
- 10% crossover rate between groups, though lower than anticipated and per-protocol results were consistent
- Slight imbalance in lobar ICH between groups (34% vs 26%) due to discrepancies between local and core imaging reads
- Too few events to perform subgroup analyses or stratify by specific DOAC type
- 11 patients underwent left atrial appendage closure during follow-up, potentially confounding results
- Non-inferiority margin of HR <1.735 for recurrent ICH was generous and the trial still failed to meet it
Citation
Lancet 2025; 405: 927–36