LAAO vs DOAC: PRAGUE-17 Falls Short

John Mandrola, MD


July 09, 2020

Given the rising global burden of atrial fibrillation (AF) plus the difficulty of taking a lifelong anticoagulant, a nonpharmacologic way of preventing stroke would be a massive advance in patient care.

The PRAGUE-17 trial comparing percutaneous left atrial appendage occlusion or closure (LAAO) with direct-acting oral anticoagulants (DOACs) in patients with AF attempted to fill a crucial knowledge gap: Is LAAO as good as DOACs for stroke prevention?  

Trying to square current use of LAAO with its regulatory history is daunting. To earn regulatory approval, the Watchman device (Boston Scientific) was tested against warfarin, the only anticoagulant available at the time. Patients who could not tolerate long-term anticoagulation were excluded from the trials, and yet these are now the patients in whom the device is most often used.

That's peculiar, for two major reasons. One is that the regulatory trials hardly looked great for LAAO. PROTECT-AF, the first LAAO vs warfarin trial, did not pass FDA muster. Then, in the FDA-mandated PREVAIL trial, LAAO did not meet noninferiority for its first co-primary endpoint of stroke, systemic embolism, or cardiovascular death. It did not match warfarin because there were more than twice as many ischemic strokes in the device arm. The second reason our acceptance of LAAO is peculiar is that we now have better and safer anticoagulants in the form of DOACs.

The problem with PRAGUE-17 is that its methods and results do not support the authors' favorable conclusion that LAAO is noninferior to DOAC. In a systematic review, Aberegg and colleagues found that the design and interpretation of noninferiority trials lead to significant and systematic bias in favor of the experimental therapy. They advised that clinicians using evidence from noninferiority trials should "cautiously inspect the actual results rather than rely on authors' interpretations and conclusions." Let's do that with PRAGUE-17.

The Trial

Investigators at 10 centers in the Czech Republic randomly assigned 415 high-risk patients with AF to LAAO or DOAC and followed them for a median of 21 months. The average age was 72 years; the CHA2DS2-VASc score was 4.7, and the HAS-BLED score was 3.0. About a third of patients had had an embolic event, and nearly half had a history of bleeding. 

The authors chose a noninferiority design with a margin of 1.47, which means the device could be 47% worse than a DOAC and still be considered noninferior. This is more rigorous than the Watchman trials, which used 2.00 and 1.75 as noninferiority margins.

The researchers chose an extremely broad composite endpoint of stroke or transient ischemic attack, systemic embolism, clinically significant bleeding, cardiovascular death, or a significant procedure- or device-related complication. This choice is the key flaw in the experiment, as I will discuss later.

There were 47 primary outcome events in the DOAC group vs 38 in the LAAO arm, resulting in a hazard ratio of 0.84 with confidence intervals of 0.53 to 1.31. Because the upper bound of 1.31 was less than the noninferiority margin of 1.47, the authors claimed noninferiority. Each component of the composite occurred in small numbers, leading to wide confidence intervals.

The efficacy endpoints, stroke and transient ischemic attack, were similar in both groups, and the hazard ratio was 1.00. The confidence intervals, however, ranged from 0.40 to 2.5. Components of the composite endpoint were not powered for comparisons, but the upper bound of 2.5 here would have far exceeded the noninferior margin of 1.47.

The safety endpoint of bleeding was also similar in each group. The hazard ratio of 0.81 had a lower bound of 0.44 and upper bound of 1.52. In other words, the device could reduce bleeding by 56% or worsen it by 52%. That is a wash: no advantage. Another safety concern was that nine patients in the device arm (about 5%) experienced a serious adverse even, including two deaths directly related to the procedure.

Dropout was not uniform: Twelve patients left the LAAO arm, whereas only one from the DOAC arm was excluded (because of withdrawal of consent).

The Problems

Please allow a brief diversion into noninferior trial design. One way to make progress in medicine is to develop superior therapies. Another way to improve patient care is to find therapies of similar efficacy that are more convenient, less invasive, or less toxic than the established treatment.

For instance, in selected patients, a transcatheter aortic valve can be nearly equal to surgery for relieving aortic stenosis but has the advantage of being less invasive. A DOAC could be just as good as warfarin for stroke prevention in patients with AF but has the added benefit of less bleeding risk and more convenience.

The putative mechanism of benefit for LAAO is that mechanical plugging of the appendage prevents embolic events. Thus, the goal of a noninferiority trial would be to show that the device is not worse than a DOAC for stroke prevention but has other advantages, such as less bleeding (safety) and less need for long-term oral anticoagulation (convenience).

The DOAC vs warfarin trials serve as a guide. Three trials tested dabigatran, rivaroxaban, and apixaban for noninferiority to warfarin on efficacy, narrowly defined as stroke or systemic embolism. The authors then asked whether the new drugs were superior for safety (bleeding). You don't need a trial to show DOACs are more convenient.

But that is not what the PRAGUE-17 investigators did. Instead, they muddied the primary endpoint by combining efficacy and safety outcomes in the composite outcome.

Clinician-researchers Ricky Turgeon, PharmD, and William McIntyre, MD, explained on Twitter that safety is the justification for noninferiority (e.g., "I am okay if the device is slightly worse, as long as it is safer."). Turgeon continues: "putting safety in a composite outcome leads to circular logic and biases toward no difference." McIntyre added: "If you include events that you expect to trend in opposite directions (thrombosis and bleeding), you bias to the null and make noninferiority easier to obtain." 

A composite endpoint can be acceptable in a noninferiority trial, but it should be a composite of efficacy, say stroke and systemic embolism. Then you test safety with a superiority design. In the case of LAAO, superiority in safety is key because the invasive procedure comes with front-loaded risk.

Turgeon, McIntyre, and I were not the only ones bothered by the circular logic of this primary outcome. Sanjay Kaul, MD, from Cedars-Sinai in Los Angeles, who has offered guidance on interpreting noninferiority trials, wrote in an email that the choice of NACE (net adverse clinical events) for a noninferiority trial is troubling. Kaul believes this decision was due to sample size considerations. Indeed, in the supplement to the article, the authors say they would have needed to enroll 7000 patients to have adequate power to determine noninferiority on efficacy alone.


When PRAGUE-17 was presented, I met with primary investigator, Pavel Osmancik, MD, PhD. I found him to be a warm and caring person. He wanted his trial to answer an important question and was deeply concerned about the adverse events. This critique is not meant to be personal.

Because of the small number of total events in the trial, the specific outcomes in each of the components of the composite have wide confidence intervals. That said, LAAO offered no efficacy advantage over DOACs in terms of preventing thrombotic events and no safety advantage in bleeding reduction. Nearly 5% of patients in the LAAO arm had a serious complication, and 1% of patients died from the procedure. Electrophysiologist Jason Andrade, MD, from the University of British Columbia, wrote on Twitter that "it was amazing to me that LAAO didn't reduce bleeding over [DOACs] in patients at high risk of bleeding."

I do not find it that surprising because there are bleeding complications associated with the procedure and LAAO requires short-term oral anticoagulation (and/or dual antiplatelet therapy) and long-term antiplatelet therapy.

PRAGUE-17, therefore, does not change the pessimistic priors we ought to have from the Watchman regulatory trials. To learn whether it is wise to expose patients to an upfront risk of an invasive procedure, we should await larger trials.

Two aspects about the evidence surrounding this procedure make me sad for our field. One is that months ago, a US registry reported results on 38,000 similar high-risk patients who had undergone LAAO. If a fraction of these patients had been randomly assigned in a PRAGUE-17-like trial, we would have had our answer. The other is that publication of an inconclusive study will now be cited for years to come as evidence that LAAO is similar to DOACs.

Editor’s Note: In an earlier version of this column, the dropout numbers for the DOAC and LAOO arms in PRAGUE-17 were mistakenly reversed. We have corrected this error.

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. 

Follow John Mandrola on Twitter

Follow | Medscape Cardiology on Twitter

Follow Medscape on Facebook, Twitter, Instagram, and YouTube


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.