CLOSURE-AF: left atrial appendage closure versus medical therapy in high-risk patients

Summary

CLOSURE-AF randomised 912 AF patients at high risk of stroke and bleeding to left atrial appendage closure (LAAC) or physician-directed best medical care, including DOACs when eligible. The primary analysis included 888 patients. Mean age 78, CHA₂DS₂-VASc 5.2 and HAS-BLED 3.0.

The primary endpoint combined stroke, systemic embolism, major bleeding and cardiovascular or unexplained death. The estimator was the difference in restricted mean survival time (dRMST). LAAC did not meet noninferiority: dRMST −0.36 years, P = 0.44 for noninferiority. Median follow-up was 3 years.

Stroke rates were 2.6 and 2.7 per 100 patient-years in the LAAC and medical arms.

Strengths

• Randomised, in a high-risk population where the question matters.
• dRMST is the right estimator when hazards are non-proportional between a procedural and a pharmacological strategy.
• The supplement publishes the data that show the effect reversing over time (Table S11, Figure S2).

Methodological issues

• More procedural complications than expected. 5.7% of patients had a complication within 7 days, including 18 BARC 3 or higher bleeds requiring transfusion out of 421 attempts. In PRAGUE-17 (mean age 73, same devices, same HAS-BLED), early complications were 2.1%. Procedural success was higher in CLOSURE-AF (98.3% vs 96.8%), which makes the complication gap hard to explain.
• DAPT after implantation in an old, frail population. Patients were about 5 years older and frailer than in earlier trials. In ADALA (JAMA Cardiol 2024), the first randomised trial of post-LAAO antithrombotic strategy, low-dose apixaban reduced the composite by 80% against DAPT (HR 0.19; p = 0.02) in 90 patients. The reduction came from less device-related thrombosis and a trend toward less major bleeding. DAPT may have put the LAAC arm at a disadvantage from day one.
• The RMST horizon decides the result. RMST integrates the area under the event-free curve up to τ. When the effect reverses direction, τ sets the verdict. Per Table S11, bleeding before 3 months was 9.0% with LAAC vs 3.9% in the medical arm. After 6 months it was 6.7% vs 10.6%. Schoenfeld residuals (Figure S2) confirm the crossing. Truncated at 3 years, dRMST captures all the early harm and only part of the late benefit.
• Competing mortality. In PRAGUE-17, non-procedural clinically relevant bleeding reached significance only at 3.5 years (sHR 0.55; 95% CI 0.31 to 0.97; p = 0.039). In CLOSURE-AF, median follow-up was 3 years and one third of patients had died by trial end, many before the window where the LAAC benefit appears.