Left Atrial Appendage Occlusion vs Direct Oral Anticoagulants for Atrial Fibrillation

Abstract

Background

Left atrial appendage occlusion (LAAO) is a treatment option to prevent thromboembolic events in patients with atrial fibrillation. However, it remains unclear whether LAAO offers superior efficacy and safety compared with oral anticoagulation for these patients.

Objectives

This study aimed to assess the efficacy and safety of LAAO vs direct oral anticoagulants (DOACs).

Methods

We reviewed randomized controlled trials comparing oral anticoagulation and LAAO, using MEDLINE and CENTRAL. The primary outcome was a composite of stroke, systemic embolism, and cardiovascular death, whereas the safety outcome was major bleeding. Two network meta-analyses were conducted to assess LAAO vs DOAC, incorporating indirect comparisons: 1 comparing LAAO, DOACs, and vitamin K antagonist (3 arms), and the other comparing Watchman, Amulet, DOACs, and vitamin K antagonist (4 arms).

Results

A total of 67,693 patients from 11 randomized controlled trials were included, of whom 5,735 underwent LAAO. In the three-arm network meta-analysis, LAAO and DOACs demonstrated statistically similar risks for both the primary efficacy outcome (HR: 1.02; 95% CI: 0.88-1.19; I² = 0%) and the safety outcome (HR: 0.98; 95% CI: 0.78-1.23; I² = 70.5%). In the 4-arm network meta-analysis, both Amulet and Watchman were associated with similar risks of the primary outcome, comparable to those of DOACs, as well as the risk of major bleeding.

Conclusions

LAAO devices did not demonstrate statistical superiority over DOACs in reducing thromboembolic events and major bleeding. Further trials are warranted to define patients for whom LAAO may offer the most significant benefit, such as those with high bleeding risk (CRD420251237886).

Central Illustration

Atrial fibrillation (AF) is the most common cardiac arrhythmia, affecting more than 46 million patients worldwide in 2016.¹ Its prevalence has risen markedly over recent decades, with the number of affected patients in the United States in 2020 exceeding projections from the 1990s by more than 3-fold.² As the population continues to age, the clinical importance of effective AF management will continue to grow.

One of the most serious complications of AF is thromboembolism, with reported annual stroke rates ranging from 0.45% to 9.28%.³ This represents the second most common nonfatal adverse event following hospitalization for heart failure.⁴ To reduce the risk of stroke and systemic thromboembolism, current guidelines recommend oral anticoagulation (OAC) for patients with a CHA₂DS₂-VASc score ≥2 in men and ≥3 in women, corresponding to an estimated annual thromboembolic risk of at least 2%.^4,5

Long-term OAC is associated with an increased risk of bleeding, and discontinuation of OAC is often recommended when bleeding events requiring specific management occur.⁶ Historically, surgical exclusion of the left atrial appendage (LAA), the site of origin of more than 90% of thrombi in patients with AF,⁷ was performed during cardiac surgery to reduce the risk of future ischemic events.⁸ In the early 2000s, the Watchman device (Boston Scientific) was introduced as a percutaneous approach to LAA occlusion (LAAO), offering an alternative strategy for stroke prevention in patients who are intolerant of long-term anticoagulation.⁹ More recently, several LAAO devices with different structural designs have been developed, including the Amplatzer Amulet LAA Occluder (Abbott Laboratories), which received approval from the U.S. Food and Drug Administration (FDA) in 2021.

Current American and European guidelines recognize percutaneous LAAO as a treatment option for patients with contraindications to long-term anticoagulation (Class IIa-IIb).^4,5 These guidelines also recommend direct oral anticoagulants (DOACs) over vitamin K antagonists (VKAs), except in patients with rheumatic mitral stenosis or mechanical heart valves. However, a comprehensive comparison of the effectiveness of LAAO vs DOACs in preventing thromboembolism, beyond considerations of bleeding risk, remains limited. Therefore, this study aimed to evaluate the efficacy and safety of LAAO in patients with AF compared with DOACs by incorporating recently published clinical trials.

Methods

Protocol and study design

This study was conducted using a frequentist statistical framework, with results reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses. The study protocol and analysis plan were prospectively registered in the International Prospective Register of Systematic Reviews (CRD420251237886). Ethical approval was not required because only publicly available published data were used.

Study selection and data extraction

We conducted a systematic search of MEDLINE and the Cochrane Central Register of Controlled Trials from database inception through April 1, 2026. Eligible studies met the following criteria: 1) randomized controlled trial (RCT) design; 2) enrollment of patients with AF who were eligible for OAC, defined as having at least 1 ischemic risk factor included in the CHADS₂ or CHA₂DS₂-VASc score, while excluding patients with mechanical heart valves or significant mitral stenosis; 3) comparison of thromboembolism prevention strategies, including percutaneous LAAO and oral anticoagulant therapies; and 4) reporting of clinical outcomes, including mortality, stroke, systemic embolism, or bleeding. LAAO devices were limited to U.S. FDA-approved devices, namely the Watchman and Amplatzer Amulet LAA Occluder. OAC included DOACs (dabigatran, apixaban, rivaroxaban, and edoxaban) and VKA (warfarin). Dose adjustments based on clinical profiles were permitted; however, trials that intentionally randomized patients to low-dose DOAC regimens were excluded.

Two investigators (Y.H. and T.K.) independently screened the titles and abstracts of retrieved studies using the strategies and terms outlined in Supplemental Tables 1 and 2, followed by a full-text review to determine final eligibility. Any discrepancies were resolved by consensus. Data extracted from each RCT included study design, baseline characteristics, event counts, and effect estimates for outcomes. For trials involving LAAO, the incidence of device-related complications was also collected.

Risk of bias

The risk of bias for each trial was assessed using the Cochrane Risk of Bias tool version 2. Publication bias was assessed by examining the symmetry of funnel plots and Egger test. Certainty of the evidence for the primary outcome and major bleeding was assessed according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach.¹⁰

Outcomes

The primary outcome was defined as a composite of cardiovascular death, stroke, or systemic embolism to capture thromboembolic events, including fatal consequences. The safety outcome was major bleeding, defined according to the International Society on Thrombosis and Hemostasis criteria or Bleeding Academic Research Consortium types 3 and above, including procedure-related bleeding. Discontinuation of anticoagulants following LAAO may reduce bleeding events, whereas the LAAO procedure itself can trigger bleeding events. Therefore, nonprocedure-related major bleeding was additionally analyzed. Other clinical outcomes included all-cause death, cardiovascular death, all-cause stroke, and the composite of stroke or systemic embolism.

Statistical analyses

We conducted 2 network meta-analyses: 1 comparing LAAO, DOACs, and VKA, and the other comparing Watchman, Amulet, DOACs, and VKA. The VKA arm enhanced statistical power through indirect comparisons by increasing the number of RCTs available for LAAO devices and DOACs. Study arms were classified according to the treatment received by the substantial majority of participants to retain trials in which only a small minority deviated from the assigned strategy. HRs with 95% CIs were synthesized using a frequentist random effects model. Prediction intervals were also calculated to estimate the expected range of treatment effects in a future study. When only risk ratios (RRs) were reported, these were used because RRs can approximate HRs when the overall incidence is low (≈≤10%).¹¹

Heterogeneity among studies was quantified using the τ² and I² statistics, with thresholds for I² defined as low (<25%), moderate (25% to 50%), and high (>50%). A common heterogeneity variance was assumed for the primary analysis of each outcome and was estimated using the Restricted Maximum Likelihood method to ensure stable assessment given the limited number of studies contributing to each comparison. Comparison-specific heterogeneity variances were additionally estimated using pairwise meta-analyses to evaluate the primary assumption. Global inconsistency between direct and indirect estimates for each clinical outcome was evaluated using the between-design component of the Q statistic. Local inconsistency was also assessed using the node-splitting method to evaluate disagreement between them for each comparison. The relative rankings of treatments were assessed using P-scores. The influence of each trial on each network contrast was assessed using the shortest-path method.¹²

Network meta-regression was performed for the primary outcome and major bleeding, using trial-level mean/median age, follow-up duration, and diabetes mellitus prevalence as continuous covariates available across most included trials to explore potential effect modifiers of the relative treatment effects. Each covariate was evaluated separately in a univariable random-effects network meta-regression using a consistency model, under the assumption of independent regression coefficients for each nonreference treatment relative to DOAC (reference treatment). Regression coefficients (log HR per unit) and their exponentiated coefficients (HR) with 95% CI were reported.

For multiarm trials, no adjustment for within-study correlation was required because only pairwise comparisons were extracted in this study. All statistical analyses were performed using R (version 4.5.1).

Sensitivity analyses

Sensitivity analyses were performed to investigate the influence of differences in baseline characteristics across trials and limitations arising from analysis strategies. First, the target population of the included trials may have changed over time with the evolution of treatment options. Therefore, a sensitivity analysis excluded trials comparing 2 OACs, as these studies were conducted before the widespread adoption of LAAO. Second, bleeding risk may vary across trials irrespective of the study period, potentially influencing both treatment selection and clinical outcomes. We performed a sensitivity analysis by excluding trials with an average HAS-BLED score >3. For trials that did not report HAS-BLED scores, classification was based on whether the study explicitly stated that patients at high bleeding risk were enrolled. Third, trials predominantly using early-generation Watchman devices were excluded, given improvements in device technology with newer generations.

Fourth, the OAC group, which comprised predominantly DOAC prescriptions, was included in analyses as a DOAC group in some trials. To address this approximation, a sensitivity analysis excluding trials in which the DOAC group included other anticoagulants was performed. Fifth, we also performed sensitivity analyses using an exact definition of the primary outcome, defined as a composite of cardiovascular death, stroke, and systemic embolism. For the bleeding outcome, the analysis was restricted to trials using ISTH bleeding criteria. Sixth, we conducted an analysis excluding RCTs with a high risk of bias. Seventh, trials reporting RRs were excluded. Finally, the contribution of each trial to the synthesized estimates was evaluated using leave-one-out analyses for all clinical outcomes.

Results

Study inclusion and patient characteristics

The systematic review identified 11 eligible RCTs (Supplemental Figure 1).13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23 One trial was designed to test superiority,¹⁴ whereas the others were noninferiority trials. A total of 67,693 patients were included, of whom 5,735 underwent LAAO. Patients randomized to dabigatran 110 mg in the RE-LY trial and edoxaban 30 mg in the ENGAGE AF-TIMI 48 trial were excluded, as these groups represented intentional allocation to low-dose therapy. Study populations and treatment compositions for each trial are summarized in Supplemental Tables 3 and 4. For the network meta-analyses, the medication arm in the CLOSURE-AF trial, in which 85% of participants received DOACs, and the OAC arm in the OPTION trial, in which 95% of participants received DOACs, were classified as DOAC groups. The mean or median age of participants across trials ranged from 69.6 to 79.1 years, and the proportion of males ranged from 60% to 77.8%. Follow-up duration ranged from 1.6 to 5 years (Table 1). Definitions of outcomes, including the primary outcome, cardiovascular death, stroke, systemic embolism, and major bleeding, varied across trials and are detailed in Supplemental Tables 5-9.

Table 1.

Baseline Characteristics of Included Trials

	CHAMPION-AF	CLOSURE-AF	SWISS-APERO	Amulet IDE	OPTION	PRAGUE-17	PROTECT AF PREVAIL	ENGAGE AF-TIMI 48a	ARISTOTLE	ROCKET AF	RE-LYa
Year of publication, year	2026	2026	2025	2025	2024	2022	2017	2013	2011	2011	2009
Sample size, n	3,000	888	221	1,833	1,600	402	1,114	14,071	18,201	14,264	12,098
Arm	Watchman	LAAO	Amulet	Amulet	Watchman	LAAO	Watchman	Edoxaban	Apixaban	Rivaroxaban	Dabigatran
Counter arm	DOAC	Medication (DOAC 85%)	Watchman	Watchman	OAC (DOAC 95%)	DOAC	Warfarin	Warfarin	Warfarin	Warfarin	Warfarin
Follow-up period, year	3	3	3	5	3	3.5	4	2.8b	1.8b	1.6	2b
Population
Age, y	71.7	79.1	76.9	75.1	69.6	73.3	72.9	72 [64–78]	70 [63–76]	73 [65–78]	71.5
Male, %	68.1	61.4	70.6	60	65.9	65.7	70.2	77.8	64.7	60.3	63.2
Comorbidities, n (%)
Age ≥75 y	NA	628 (70.7)	NA	NA	NA	164 (40.8)	461 (41.3)	5,668 (40.3)	5,678 (31.2)	NA	NA
Hypertension	NA	834 (93.9)	177 (80.0)	1,701 (92.8)	1,422 (88.9)	372 (92.5)	1,007 (90.4)	13,179 (93.7)	15,916 (87.4)	12,910 (90.5)	9,545 (78.9)
Diabetes	NA	361 (40.7)	58 (26.2)	640 (34.9)	443 (27.7)	163 (40.5)	317 (28.5)	5,080 (36.1)	4,547 (25.0)	5,695 (39.9)	2,812 (23.2)
Prior stroke/TIA	247 (8.2)c	293 (33.0)	NA	499 (27.2)d	172 (10.8)	NA	252 (22.6)	3,967 (28.2)	3,538 (19.4)	7,811 (54.8)d	2,428 (20.0)
Congestive heart failure	NA	NA	10 (4.5)	674 (36.7)	327 (20.4)	178 (44.2)	285 (25.6)	8,145 (57.9)	6,451 (35.4)e	8,908 (62.5)	3,856 (31.9)
Atrial fibrillation, n (%)
paroxysmal	2,066 (68.9)	NA	87 (39.4)	1,015 (55.4)	978 (61.1)	120 (29.9)	501 (45.0)	3,531 (25.1)	2,786 (15.3)	2,514 (17.6)	4,014 (33.2)
persistent/permanent	933 (31.1)	NA	NA	818 (44.6)	622 (38.9)	282 (70.1)	588 (52.8)	NA	15,412 (84.7)	11,548 (81.0)	8,082 (66.8)
Risk scoring
CHA2 DS2-VASc score	3.5 ± 1.3	5.2 ± 1.5	4.3 ± 1.4	4.6	3.5 ± 1.3	4.7 ± 1.5	3.7	NA	NA	NA	NA
HAS-BLED score	1.3 ± 0.8	3.0 ± 0.9	3.1 ± 0.9	3.3	1.2 ± 0.8	3.1	NA	NA	NA	NA	NA

DOAC = direct oral anticoagulant; LAAO = left atrial appendage occlusion; NA = not available; OAC = oral anticoagulation; TIA = transient ischemic attack.

^aIncluding only groups receiving a standard dose of DOAC.

^bShowing the follow-up period of the overall trial.

^cIschemic and hemorrhagic stroke.

^dStroke, systemic embolism, or transient ischemic attack.

^eHeart failure or reduced left ventricular ejection fraction.

Bias assessment

Risk of bias for each trial, assessed using Cochrane Risk of Bias tool version 2, is shown in Supplemental Figure 2. Two trials were rated as having a high risk of bias, and 4 trials were assessed as having some concerns, primarily due to unequal crossover or deviations between treatment arms,¹⁴ unscheduled follow-up,¹³ or ad hoc analyses.¹⁷ Funnel plots for each meta-analysis showed generally symmetrical distributions, suggesting minimal publication bias (Supplemental Figures 3 and 4). Egger test was also not significant for any outcomes; however, these findings should be interpreted cautiously given the limited number of included trials.

Transitivity and network certainty

For each outcome, point estimates and uncertainty from included trials are summarized in Supplemental Table 10. Baseline characteristics were compared across treatment comparisons to evaluate potential effect modifiers for the primary outcome. History of stroke/transient ischemic attack and congestive heart failure were more frequently observed among trials directly comparing DOAC with VKA than trials including LAAO devices (Supplemental Tables 11 and 12).

Common heterogeneity and inconsistency for each outcome are summarized in Supplemental Table 13. Comparison-specific heterogeneity and inconsistency are provided in Supplemental Tables 14-17. Heterogeneity was substantial for major bleeding (I² = 69% to 70.5%), with no apparent pattern according to comparisons. Overall, neither global nor local inconsistency was statistically significant, although inconsistency could not be evaluated for stroke because the network loop was not fully closed. The certainty of evidence was also limited by the small number of studies and events, particularly for major bleeding (Supplemental Tables 14 to 17).

Procedural and device-related complications

For patients undergoing LAAO, follow-up imaging strategies are provided in Supplemental Table 19. The incidence of periprocedural complications, including peridevice leak (PDL) and device-related thrombus (DRT), is summarized in Supplemental Table 20.

Network meta-analysis of LAAO vs DOAC vs VKA

A total of 65,639 patients from 9 RCTs,15, 16, 17, 18, 19, 20, 21, 22, 23 including 3,681 treated with LAAO, 32,304 with DOACs, and 29,654 with VKA, were included (Central Illustration). Network meta-analysis estimates showed that LAAO and DOACs were associated with comparable risks for the primary outcome, all-cause/cardiovascular death, and thromboembolic events (Figure 1). Direct comparisons between LAAO and DOACs similarly showed no significant differences across any outcomes (Supplemental Figure 5). Heterogeneity for these outcomes was low (I² = 0%, except for stroke: I² = 24.8%, τ² ≤0.007), and inconsistency was not significant (between-design P > 0.05, except for stroke: not estimable).

Central Illustration.

Network Meta-Analysis Comparing Left Atrial Appendage Occlusion and Direct Oral Anticoagulants in Patients With Atrial Fibrillation

(A) Included 11 randomized controlled trials comparing LAAO devices and anticoagulants. (B) Network plots and forest plots for the network meta-analyses of the primary outcome, defined as a composite of stroke, systemic embolism, and cardiovascular death. (C) Network plots and forest plots for the network meta-analyses of major bleeding. AF = atrial fibrillation; DOAC = direct oral anticoagulant; LAAO = left atrial appendage occlusion; OAC = oral anticoagulation; VKA = vitamin K antagonist.

Figure 1.

Network Meta-Analysis Comparing Left Atrial Appendage Occlusion, Direct Oral Anticoagulant, and Vitamin K Antagonist

Network estimates comparing LAAO, DOAC, and VKA are shown for (A) the primary outcome, (B) all-cause death, (C) cardiovascular death, (D) stroke, (E) stroke/systemic embolism, and (F) major bleeding. The primary outcome was defined as a composite of cardiovascular death, stroke, or systemic embolism. DOAC = direct oral anticoagulant; LAAO = left atrial appendage occlusion; PI = prediction interval; VKA = vitamin K antagonist.

The risk of major bleeding was also similar among the 3 treatment strategies (I² = 70.5%, τ² = 0.02, between-design P = 0.64). However, the risk of nonprocedure-related major bleeding was significantly lower in the LAAO (Supplemental Figure 6) (I² = 70.9%, τ² = 0.02, between-design P = 0.15). The rankings based on P-scores are shown in Supplemental Figure 7.

Network meta-analysis of Watchman vs Amulet vs DOAC vs VKA

A total of 66,402 patients from 9 RCTs13, 14, 15^,17, 18, 19, 20^,22,23 were included, comprising 4,060 treated with the Watchman device, 1,028 with the Amulet device, 31,660 with DOACs, and 29,654 with VKA (Central Illustration). LAAO devices and DOACs showed comparable efficacy for the primary outcome and thromboembolic events, with most comparisons showing low to moderate heterogeneity (I² = 16.2%, except for stroke: I² = 40.5%, τ² ≤0.008) and insignificant inconsistency (between-design P > 0.05, except for stroke: not estimable).

For the safety outcome, all treatment strategies were associated with similar risks of major bleeding, although heterogeneity was high (I² = 69%, τ² = 0.02, between-design P = 0.37) (Figure 2, Supplemental Figure 8). Meanwhile, nonprocedure-related major bleeding was less frequent with Watchman and Amulet than with DOACs (Supplemental Figure 6) (I² = 74.6%, τ² = 0.02, between-design P = 0.24). Rankogram analyses suggested that VKA was associated with the highest bleeding risk. Treatment rankings based on P-scores are shown in Supplemental Figure 9.

Figure 2.

Network Meta-Analysis Comparing Amulet, Watchman, Direct Oral Anticoagulant, and Vitamin K Antagonist

Network estimates comparing Amulet, Watchman, DOAC, and VKA are shown for (A) the primary outcome, (B) all-cause death, (C) cardiovascular death, (D) stroke, (E) stroke/systemic embolism, and (F) major bleeding. The primary outcome was defined as a composite of cardiovascular death, stroke, or systemic embolism. Abbreviations as in Figure 1.

Sensitivity analyses

Sensitivity analyses accounting for heterogeneity in baseline characteristics in each trial compared the risk of the primary outcome and major bleeding between LAAO devices and DOACs. Excluding trials directly comparing OAC medications18, 19, 20^,23 consistently showed comparable risks between the 2 strategies (Supplemental Figure 10). Similarly, after excluding trials with an average HAS-BLED score >3.0,^13,14,16,21 both endpoints remained comparable between LAAO devices vs DOACs (Supplemental Figure 11). Because the Amulet IDE, PRAGUE-17, and PROTECT AF/PREVAIL trials predominantly used first-generation devices among patients treated with Watchman (Supplemental Table 4), we excluded these trials, which again yielded consistent risk of the primary outcome (Supplemental Figure 12).

Deviations arising from the pooling process and statistical analyses were also assessed in sensitivity analyses. In 2 trials, the medication arm was categorized as DOACs, including 5% to 15% non-DOAC therapy. Exclusion of these trials did not significantly change the results for the primary analysis and bleeding outcomes (Supplemental Figure 13). Restricting analyses to an exact definition of the primary outcome and major bleeding also showed comparable risks between LAAO devices and DOACs (Supplemental Figure 14). The analyses excluding trials with substantial crossover^14,22 and ad hoc analysis shown in Supplemental Figure 15,¹⁷ yielded results consistent with the primary findings. After excluding the effect estimates with RRs,^21,23 both the primary and bleeding findings remained consistent (Supplemental Figure 16).

In the leave-one-out analyses, exclusion of the CLOSURE-AF trial shifted the treatment effect of LAAO vs DOAC further in favor of LAAO for all-cause death. For stroke, omitting the CHAMPION-AF trial markedly attenuated the effect estimate toward the null (Supplemental Figures 17 and 18). A similar pattern was observed in the contribution matrices, which showed a disproportionate contribution of the CLOSURE-AF trial to the direct comparison between LAAO and DOAC, particularly for all-cause and cardiovascular death (49.0% to 51.4%). In the contribution matrix for stroke, the CHAMPION-AF trial showed the highest contribution percentage (40.6%) (Supplemental Figures 19 and 20).

Network meta-regression

In the comparison between DOAC vs VKA, age significantly modified the treatment effect on the risk of major bleeding in both 3-arm (regression coefficient, −0.122; 95% CI: −0.211 to −0.033) and four-arm (regression coefficient, −0.121; 95% CI: −0.215 to −0.026) network meta-regressions, suggesting an increasing relative bleeding risk with DOAC as age increased (Supplemental Figure 21, Supplemental Table 21). However, only 4 trials directly compared DOAC vs VKA with a narrow range of trial-level mean/median age. No significant interaction was observed for the other treatment arms with DOAC.

Discussion

This study synthesized the findings from the latest RCTs and comprehensively evaluated the effectiveness of LAAO devices and DOACs in patients with AF. The main findings are as follows: 1) statistical analyses did not demonstrate a significant difference in the risks of mortality, ischemic events, and major bleeding between LAAO and DOACs; 2) the Watchman and Amulet devices demonstrated similar efficacy and safety profiles; and 3) these findings were consistent across multiple sensitivity analyses with no significant global or local inconsistencies within the treatment networks and were aligned with previous meta-analyses focusing on direct evidence with low heterogeneity.^24,25

The appropriate conditions for LAAO

This study found that DOACs and percutaneous LAAO devices provided statistically similar effectiveness in preventing thromboembolic and bleeding events. The balance between risks and benefits of LAAO, compared to DOACs, may change over time following the procedure. In the early phase after LAAO, the intracardiac procedure carries inherent risk, with procedure-related adverse events reported in 2% to 9% of LAAO cases in this analysis. Although recent nationwide registry studies with contemporary LAAO devices demonstrated lower procedure-related incidence (postprocedural composite safety outcome: 0.45% to 0.9%), front-loaded procedural burden may be significant for high-risk patients.^26,27

Furthermore, device-related complications, such as DRT, which has been associated with a 3.5-fold increased risk of stroke or systemic embolism, and PDL, may also boost the risk of thrombosis.²⁸ Supporting this hypothesis, stroke rates are higher within 45 days following LAA closure, corresponding to the early endothelialization phase.²⁹ On the other hand, a retrospective study reported that ischemic strokes occurring in patients who underwent LAAO were associated with more favorable functional neurological outcomes than those in patients treated with DOACs, suggesting that more granular assessment of clinical outcomes may reveal potential advantages of LAAO.³⁰

Meanwhile, bleeding risk, which was comparable between LAAO and DOACs, whereas nonprocedure-related major bleeding was less frequent with LAAO, may require long-run perspectives. These findings were consistent with the time-dependent divergence in the Kaplan-Meier curves, where the reduction in bleeding risk becomes more evident in the LAAO group.¹⁵ Accordingly, the net benefit of LAAO may accrue over time, supporting its use as a reasonable alternative in patients who are intolerant of long-term anticoagulation.⁵ Even though indefinite aspirin therapy may be required after discontinuation of anticoagulation, a prior meta-analysis demonstrated that antiplatelet therapy was associated with a lower risk of major bleeding compared with DOACs, emphasizing the importance of discontinuing DOACs in high-risk populations.³¹

Ongoing RCTs are expected to further clarify the comparative effectiveness and optimal indications for these therapies, including the CATALYST trial (NCT04226547), which is evaluating the Amulet device vs DOACs; the trial by Gao et al (NCT05353140), assessing the outcomes between the Watchman device and rivaroxaban. The continued accumulation of high-quality clinical data is anticipated to better define which patients are most likely to obtain a net clinical benefit from LAAO.

Choices of LAAO devices

We also compared the effectiveness of individual LAAO devices, and all outcomes were statistically comparable between Watchman and Amulet. However, study designs varied across the included trials, and only 2 randomized trials directly compared the 2 devices, warranting cautious interpretation of the results. The Amulet IDE trial, which directly compared the Watchman 2.5 device vs the Amulet, reported significantly higher rates of PDL and DRT in the Watchman group, with these differences being more pronounced among patients who experienced stroke over 5 years of follow-up. Similarly, the SWISS-APERO trial reported higher rates of PDL and DRT in the Watchman group, more than three-quarters of whom received the Watchman FLX. These findings may reflect structural limitations of the earlier Watchman model, including fewer frame struts and a larger area of exposed metal screw,³² and its periprocedural profile, which has been reported to be improved with newer-generation Watchman devices.³³ In addition, a subanalysis of the Amulet IDE trial showed a distinct temporal pattern of DRT, with 63% of DRT events in the Amulet group occurring within 45 days, whereas 75% of DRT events in the Watchman group occurred after 45 days.³⁴ These findings may be influenced by antithrombotic regimens or device characteristics, highlighting the importance of frequent imaging evaluation. Therefore, further trials with the latest device iterations, adequate follow-up duration, and standardized postprocedural management are warranted to investigate the comparative efficacy of each device.

Overall, our findings indicated that DOACs and LAAO represent viable treatment options, with no statistically significant difference in effectiveness between the 2 U.S. FDA-approved devices. LAAO may serve as an alternative to long-term anticoagulation in selected patients; however, device-related complications may limit its efficacy and underscore the need for standardized postprocedural antithrombotic regimens. Further clinical data evaluating outcomes with DOACs and LAAO, using newer devices, are warranted to better define optimal strategies for preventing ischemic events across diverse populations with AF.

Study Limitations

This study has several limitations. First, antithrombotic regimens following LAAO varied across trials and devices. Trials included in this analysis provided either DOAC plus aspirin or dual antiplatelet therapy therapy during the first 3 months after LAAO. DOAC-included regimens may be beneficial in reducing thromboembolic risk by preventing DRT through inhibition of the coagulation cascade, as demonstrated in prior studies.^35,36 However, the influence of regimens on clinical outcomes remains speculative, as current RCT-based evidence has primarily focused on DOAC monotherapy, dual antiplatelet therapy, and single antiplatelet therapy, and subgroup analysis according to postprocedural antithrombotic regimens was not feasible because of the limited number of trials.³⁷ Second, individual LAAO devices encompassed multiple generations, largely reflecting the time periods during which the trials were conducted, although sensitivity analysis excluding first-generation Watchman demonstrated consistent findings. Notably, the latest Watchman device iteration was not evaluated in any of the included studies, limiting the generalizability of our findings to contemporary clinical practice. Third, baseline ischemic and bleeding risks were not consistently evaluated across trials. Early studies, particularly those conducted in the early 2010s, relied on the CHADS₂ score rather than the CHA₂DS₂-VASc score and did not incorporate the HAS-BLED score. Furthermore, some studies, even those having an LAAO arm, did not restrict the study population based on bleeding risks.^15,22 Although sensitivity analyses restricted to low bleeding risk populations were consistent with the primary findings, bleeding risks of trials without reported HAS-BLED scores were classified based on enrollment criteria, introducing residual heterogeneity. Fourth, other baseline discrepancies, including the prevalence of prior stroke or transient ischemic attack and congestive heart failure, were observed across treatment comparisons, particularly in trials comparing medication therapies. Meanwhile, inconsistency assessments and sensitivity analyses excluding these trials yielded consistent results, supporting the transitivity assumption. Fifth, follow-up duration varied across trials, and long-term outcomes were only sparsely assessed. Bleeding risk, in particular, may increase over time in patients receiving anticoagulants due to attenuation of periprocedural bleeding in the device group and declining adherence in the medical therapy group. We could not perform landmark analyses or a robust evaluation of long-term effects. Instead, nonprocedure-related bleeding outcomes suggested a benefit of LAAO, although limited RCTs reported them. Sixth, each trial used distinct follow-up imaging strategies. Differences in the detection of device-related complications among modalities, the impact of early DRT detection and anticoagulation resumption on DRT resolution, and the increased risk of stroke in patients with persistent DRT suggest that imaging follow-up protocols may influence clinical outcomes.^38,39 Furthermore, variable follow-up durations across trials may have weakened the exchangeability between HRs and RRs. In particular, the CLOSURE-AF trial reported deviations from the proportional-hazards assumption and a high incidence of the primary outcome, further undermining the rationale for using RRs. Although a sensitivity analysis excluding RRs yielded results consistent with the primary analysis, leave-one-out analyses and contribution matrix using the shortest-path method suggested that the CLOSURE-AF trial made a substantial contribution to the pooled estimates for some outcomes. Seventh, some clinical outcomes have variation in their definitions across trials. In particular, some trials used a definition of the primary outcome that deviated from the predefined criteria, although sensitivity analyses using uniform definitions did not significantly change the findings. Finally, this study conducted trial-level analyses based on published data, limiting our ability to harmonize statistical approaches across studies. In particular, the methods used to account for the competing risk of death varied among trials, which may affect the interpretation of nonfatal outcomes in this high-risk AF population.

Conclusions

This study provides a comprehensive comparison of thromboembolism prevention strategies in patients with AF, incorporating data from the most recent RCTs. LAAO devices were not associated with a statistically significant difference in reducing thromboembolic events or major bleeding. Further randomized studies are warranted to better define the optimal role of LAAO in selected patient populations.

Perspectives.

COMPETENCY IN MEDICAL KNOWLEDGE: The network meta-analyses incorporating the latest RCTs showed that LAAO has comparable efficacy in preventing thromboembolism and bleeding but did not demonstrate any statistical superiority over DOACs.

TRANSLATIONAL OUTLOOK: Further exploration of the indication for LAAO regarding bleeding risk, particularly with newer-generation devices, may help future treatment decisions.

Funding support and author disclosures

This work was partly supported by a Grant-in-Aid for Scientific Research (C) from the Japan Society for the Promotion of Science (JSPS KAKENHI grant number 25K10897) and the Nakatani Foundation (to Dr Aikawa). No funding sources were involved in determining the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the article for publication. Dr Slipczuk discloses honoraria from Elsevier/Progress in Cardiovascular Diseases as Editor-in-Chief; he served as site PI for clinical trials for Amgen and Philips; he holds equity in APQ Health Inc; and received past research funding from Amgen and Philips. Dr Aikawa has received grant-in-aid for scientific research (C) from the JSPS KAKENHI (grant number 25K10897) and Kanzawa Medical Research Foundation. Dr Laham has received grants from NIH/NHLBI (P50HL063609) and NIH/NCRR (M01RR001032) outside the submitted work. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.