When it came time to write about the CLOSURE-AF trial of percutaneous left atrial appendage closure (LAAC) vs best medical therapy in patients with atrial fibrillation (AF) at high risk for both stroke and bleeding, I expected to have to explain how the noninferiority design obscured the results. 

That is not necessary. The results are clear: LAAC was worse than best medical therapy. The saddest thing is that the German trial comes not before but after more than a half-million patients have been implanted with Watchman devices worldwide. 

My long-standing argument against percutaneous LAAC stems from signals of inefficacy in the seminal trials against warfarin. CLOSURE-AF investigators found similar signals comparing the devices to mostly direct acting oral anticoagulants (DOAC). Over the past decade, both device and anticoagulants have iterated, yet the results are similar.

The CLOSURE-AF Trial 

Trial results have been presented at AHA, but not yet published in full. 

Slightly more than 900 patients were randomly assigned at 42 centers in Germany to either LAAC or “best medical therapy.” LAAC devices included Watchman (54%), Amulet (42%) and LAmbre (4%). The best medical therapy group were mostly treated with DOACs (85%), vitamin K antagonists were rare (3%), and 10% of patients got no anticoagulant.

CLOSURE-AF enrolled patients with average CHA2DS2VASc and HAS-BLED scores of 5.2 and 3.0, respectively. Their average age was 79 years and 39% were female. These are people at high risk for stroke and bleeding, which is a strength of the trial because in contemporary practice in both the United States and Europe, the patients having this procedure are also higher risk than those enrolled in the seminal trials. 

Trial conduct was excellent. Very few patients with LAAC remained on oral anticoagulation and the procedure was successful in almost 98% of patients, as defined by technical success and peri-device leak

The primary endpoint was a composite of stroke, systemic embolism, cardiovascular or unexplained death, or major bleeding. The trial used a noninferiority design with a margin of 1.3. 

As I wrote in my AHA preview column, the choice of this strategy had two downsides: one is that it combined both thrombotic (efficacy) and bleeding (safety) endpoints, which are expected to go in different directions and cancel each other out, thus biasing the trial toward finding no difference. 

My second criticism of the primary endpoint is that it included CV death — which we now know from the LAAOS-3 trial is unlikely to be affected by either treatment. Including an outcome not affected by the treatment in a primary endpoint adds noise and makes noninferiority easier to reach — as seen in another LAAC trial, OPTION, where death was the most common event and similar in both arms. 

LAAC Implantation Comes With Procedural Complications

Periprocedural complications were notable: of the 414 attempted procedures, five patients had tamponade, 18 had major bleeding requiring transfusion, there was one device embolization, one procedure-related TIA, one peripheral embolism, and two deaths within 7 days of the implant. 

Over 3 years of follow-up, the incidence of a primary outcome event occurred in 16.83 per 100 patient years in the LAAC arm vs 13.27 per 100 patient years in the best medical care arm. The adjusted hazard ratio was 1.28 with a 95% CI ranging from 1.01-1.62. This was the intention-to-treat analysis, which includes everyone assigned to a treatment arm regardless of whether they underwent their assigned treatment. 

The per-protocol analysis includes only patients who actually had their assigned treatment (ie, they got a device or they took a drug); it’s considered the more robust analysis in a noninferiority trial. That hazard ratio was 1.34 and the 95% CI ranged from 1.04-1.72. 

In both cases, the upper bound of the 95% CI, or worst-case scenario, was much higher than the noninferiority margin of 1.3 — indicating that LAAC failed to reach noninferiority against best medical therapy. Additionally, the fact that the lower bound of the 95% CI was greater than 1.0 allows us to declare that LAAC was actually inferior. 

The components of the primary endpoint were interesting. Stroke and systemic embolism occurred in 30 LAAC patients vs 28 patients in the best medical care group. The incidence of major bleeding was 21% higher in the LAAC arm (HR, 1.21; 95% CI, 0.86-1.71). CV death or unexplained death was also 25% higher in the LAAC arm (HR, 1.25; 95% CI, 0.93-1.68). 

Strong Evidence Against LAAC

This is strong evidence against left atrial appendage closure. The trial was conducted in experienced centers in Germany that were doing this procedure well before it was approved in the US. It’s also a contemporary trial; the majority of devices implanted were the latest iteration. The greater than 96% procedural success rate supports the contention that this trial environment is likely a best case scenario. CLOSURE-AF was also non-industry funded, something to consider when we learn results of the coming industry funded trials. 

Yet LAAC not only failed to reach noninferiority but was found inferior to best medical therapy. Worse was that it failed in an endpoint that should have been easy to succeed in — because of the inclusion of efficacy and safety components as well as CV death, which is not likely affected by either strategy. 

I’ve seen some comment that CLOSURE-AF was underpowered, and it is only a “small” datapoint. There is some truth to that. The ongoing CHAMPION-AF trial enrolled 3000 patients and CATALYST enrolled 2600; we look forward to those data. Yet we should not ignore important signals from CLOSURE-AF.

First is the vital importance of including periprocedural events. In CLOSURE-AF, 18 of 70 major bleeding events were periprocedural. And all 18 of these required blood transfusion. The decision by OPTION investigators to exclude periprocedural bleeding in its primary safety endpoint led to biased conclusions of that trial of LAAC post-AF ablation. 

Second, while the total number of stroke events in CLOSURE-AF numbered only 58, it’s similar to the 59 ischemic stroke and systemic embolism events in the two seminal trials, PROTECT AF and PREVAIL, combined. One minor positive in CLOSURE-AF is that at least the stroke rates were not higher — as they were in PROTECT AF and PREVAIL. 

Third, major bleeding was higher in the LAAC arm. Much of it was likely driven by periprocedural bleeds, but not all of it. LAAC requires the use of antiplatelet drugs. And proponents tend to forget that antiplatelet drugs also cause bleeding. 

The lower bleeding rates in the best medical therapy arm are remarkable considering that many enrolled patients had a history of a major bleed. These patients are typically taken off their anticoagulation and referred for LAAC. CLOSURE-AF challenges that strategy. It’s notable that 10% of patients in the best medical therapy arm had no oral anticoagulation. That pragmatic feature of the trial was excellent, because in practice, no device or anticoagulant may be the best choice for such patients. 

Fourth, proponents might argue that the higher rates of CV/unexplained death in the LAAC arm were just noise. Maybe so; it was surely noise in PREVAIL, because the FDA mandated inspection of all deaths, and none were due to either the Watchman device or drug. CLOSURE-AF had two deaths within 7 days of the procedure; what about deaths within 30 days of the procedure? With 18 bleeds requiring transfusion, I wonder if some of the CV deaths were related to the procedure. We shall see when the paper is published.

Periprocedural risk is the Achilles heel of the LAAC strategy: probabilistically, the device has to overcome the up-front risk of the procedure. Future stroke and bleeding events have to be substantially lower to overcome the initial risk a patient undertakes. In CLOSURE-AF, done in experienced labs, within the careful confines of a clinical trial, this was clearly not achieved. 

I conclude with a Bayesian perspective — written on X by Sanjay Kaul, MD, an expert in clinical trial design: PROTECT AF was rejected by the FDA; PREVAIL missed its most important noninferiority primary endpoint, PRAGUE-17 met noninferiority for net adverse cardiac events but not for ischemic outcomes, and OPTION did not “win” using procedural bleeding. Thus, our priors for LAAC should have been extremely pessimistic. CLOSURE-AF data only strengthen that pessimism. 

We wait for the larger trials, but what a reversal this could be. 

John Mandrola practices cardiac electrophysiology in Louisville, Kentucky, and is a writer and podcaster for Medscape. He espouses a conservative approach to medical practice. He participates in clinical research and writes often about the state of medical evidence.