Fauchier et al.: reduced-dose DOAC versus antiplatelet strategies after left atrial appendage occlusion

Summary

The study used the TriNetX network to identify 48,788 AF patients who underwent percutaneous left atrial appendage occlusion (LAAO) between 2010 and 2025. Three independent comparisons were run, with 1:1 propensity score matching on 89 covariates. Mean follow-up was 1.4 to 1.6 years.

The primary endpoint was net clinical benefit (NCB): death, ischaemic stroke, MI and major bleeding.

Strengths

• Largest comparative dataset on antithrombotic therapy after LAAO. The randomised data (ADRIFT, ADALA, ANDES) are small and short.
• The three estimates are coherent. If DOAC and SAPT have HRs of 0.866 and 0.865 against DAPT, DOAC vs SAPT should be about 1.001. The observed value was 1.056, a 5.5% gap.
• Supplementary Appendix 3 holds a more complete limitations section than usual for this design. Supplementary Figure 3 publishes self-critical data.
• The bleeding code list is published (Supplementary Table 1).

Methodological issues

• DOAC vs DAPT does not survive the paper’s own analyses. I applied Benjamini-Hochberg to the 30 tests in the three main tables. For DOAC vs DAPT: NCB q = 0.060, death q = 0.075, bleeding q = 0.075. With time zero at day 28: NCB 0.904 (p = 0.15) and bleeding 0.915 (p = 0.29). In the 2020–2025 cohort the NCB CI is 0.939 to 1.242, which excludes the main estimate and puts the point estimate above 1. For mortality, the post-2020 CI (0.717 to 1.217) still contains 0.789: imprecision, not contradiction.
• Proportional hazards not tested. About half of non-fatal events occurred in the first 90 days (NCB 48% and 51%, major bleeding 50% and 52%, stroke 46% and 49%), against 14% of deaths. Early stroke runs at about 22%/year and then 5.6%/year. The methods state that Schoenfeld residuals were tested, but no result appears in the paper or supplement. RMST would have been informative.
• Code-based major bleeding. It includes R31 (haematuria), R04 (respiratory tract haemorrhage, which contains epistaxis), N93 (abnormal uterine bleeding, in a cohort with mean age 78 and 53% men), M25.0 (haemarthrosis) and S06.4 (traumatic extradural haemorrhage). The observed rate is 13.8 to 15.3%/year. With 515 bleeds against 757 NCB events, this component drives the composite, and coded haematuria weighs the same as death.
• Residual confounding. I calculated E-values (VanderWeele approximation for common outcomes) at the confidence limit nearest 1: 1.20 for mortality in the DOAC comparison, 1.21 for NCB, 1.24 for mortality in the SAPT comparison and 1.43 for bleeding in the SAPT comparison. A confounder associated about 1.2-fold with both treatment and death erases the mortality findings. Frailty, cognition and functional status were unavailable.
• Similar mortality with opposite strategies. DOAC and SAPT, opposites on the thrombotic axis, reduce mortality against DAPT by a similar amount (0.789 and 0.839). The common factor is not being on DAPT, a prescriber choice. Selection of patients judged fit for anticoagulation would push the result in this direction.
• Two thrombotic signals not discussed. With time zero at day 28, device-related or intracardiac thrombosis with DOAC vs DAPT: 2.153 (1.050 to 4.418; p = 0.03). In the post-2020 cohort, ischaemic stroke: 1.325 (1.003 to 1.751; p = 0.05), with 113 vs 88 events. Both are exploratory and share the multiplicity problem. The text describes the comparisons that lost significance in the recent cohort and omits this one, which gained significance in the opposite direction.
• Yearly rates that do not reconcile. Dividing events by the annual rate gives 3,796 person-years for NCB in the DOAC arm, against a ceiling of 2,482 (1,773 × 1.4 years). It also gives more time at risk for non-fatal endpoints than for death. Absolute effects are not reported and crude counts are nearly identical (death 202 vs 206, NCB 757 vs 757, bleeding 515 vs 526). ARR and NNT cannot be calculated. I leave it as a question: how does TriNetX define this denominator?