John Mandrola, MD


November 10, 2017

Percutaneous left atrial appendage closure (LAAC) is a proposed stroke-prevention strategy for patients with nonvalvular atrial fibrillation (AF). Five-year results of the PREVAIL trial, which compared the Watchman device with warfarin, were presented at the TCT 2017 and published in the Journal of the American College of Cardiology.[1]

Watchman trialists combined PREVAIL and PROTECT-AF data in a patient-level meta-analysis. Their conclusions were positive. You will hear that Watchman has now made noninferiority in the late-ischemic efficacy end point of PREVAIL.

These new data do not change my take on LAAC as a stroke-prevention strategy. I remain unconvinced.

The following arguments parallel an editorial I wrote with Drs Andrew Foy and Gerald Naccarelli (Penn State University, Hershey, PA) published in Heart Rhythm.[2]

1. No
change in the direction of the trends

Nothing has changed in the direction of the data at 5 years: ischemic stroke rates are still higher with the device; the lower hemorrhagic stroke signal (in PROTECT-AF) remains an extreme outlier compared with contemporary trials (see table), and the lower cardiovascular death with Watchman is also likely to be spurious. Note that FDA reviewers mandated inspection of all deaths in PREVAIL and so far not one of the 15 deaths related to either warfarin or Watchman (as reported at the October 2014 FDA meeting).[3]

This latter point on the components of the primary end point is important. Trialists in the direct oral anticoagulant (DOAC)–vs-warfarin trials[4,5,6,7] used only stroke and systemic embolism as the end point. The Watchman trialists added CV/unexplained death to their composite. Adding an end point unlikely to be affected by either therapy makes noninferiority easier to reach.

Hemorrhagic Stroke Rates in the Warfarin Arm of Anticoagulation Trials Compared With the Rate in PROTECT-AF

Trial Warfarin (n) CHADS2 Risk Score Hemorrhagic Stroke Rate (%)
BAFTA 488 NA 0.5
ACTIVE W 3371 2.0 0.36
RELY 11763 2.1 0.4
ARISTOTLE 9081 2.1 0.5
ROCKET-AF 7090 3.4 0.4
ENGAGE-AF 7036 2.8 0.5
PROTECT-AF 244 2.3 1.1
2. No change in the net-benefit calculation

These new data do not change the net-benefit calculation for LAAC. The average rate of procedural complication in PROTECT AF[8], PREVAIL,[9] ASAP,[10] and CAP[11] is 6%. Newer LAAC devices look no better. A just-published series of 83 patients who were implanted with the Amplatzer Amulet device (Abbott Vascular) at an experienced center also found a 6% major procedural complication rate. This means a patient starts the LAAC stroke-prevention benefit/risk gamble with a 6% chance of being harmed for a debatable promise of "noninferiority" in the future. (See points 6 and 7.)

3. No change in the dubious pathophysiologic basis for LAAC

The 5-year data do not change the uncertain pathophysiologic basis for appendage occlusion. We don't know the relationship between stroke and left atrial clot. The frequently cited statistic that "the LAA has been implicated as the source of emboli in ∼90% of patients with nonvalvular AF" usually references a 1996 paper[12] that did not study patients with stroke. It was a review of published reports detailing the frequency and site of thrombus in patients with AF.

Most concerning is that LAAC may be least effective in the group of patients it is being used for now—those ineligible for anticoagulation. Patients who are ineligible for anticoagulation get that way because of comorbidities. A recent meta-analysis found higher rates of nonappendage LA clot in sicker patients—specifically, those not on anticoagulation or those with left ventricular dysfunction or prior stroke.[13]

4. No change in the concerns over antiplatelet drugs

Percutaneous LAAC leaves a foreign body in the heart. Antithrombotic drugs are used to counter device thrombus. These new data don't change the fact that both AVERROES[14] and BAFTA[15] found bleeding risk in patients with AF on antiplatelet drugs nearly identical to the bleeding risk for anticoagulation therapy with apixaban (Eliquis, Bristol-Myers Squibb/Pfizer) and warfarin, respectively.

This is a crucial point, because one of the proposed benefits of LAAC is getting off anticoagulation. The problem with that thinking is that the best evidence suggests aspirin is no safer than anticoagulants for patients with AF. Notably, in AVERROES, intracranial hemorrhage rates were numerically higher with aspirin than with apixaban.

What's more, two registry studies[16,17] presented as late-breaking science at European Heart Rhythm Association (EHRA) EUROPACE-CARDIOSTIM 2017 reported a high percentage of patients on dual antiplatelet therapy (DAPT) after LAAC. One of the studies, a multicenter French registry (n=453), found nearly 60% of patients who had percutaneous LAAC remained on either oral anticoagulation or DAPT nearly a year later.

5. Stroke is a systemic disease

The PREVAIL 5-year data do not change the fact that atrial fibrillation associates with structural changes in the atria and blood components.[18,19,20] Virchow's triad teaches us that thrombosis isn't dependent only on stasis. A percutaneous plug in one part of the atria does not change the hypercoagulability or endothelial dysfunction seen in patients with AF. Stroke is a systemic disease. Focal treatments don't perform well in systemic diseases.

The final two points delve into statistics—but are important because they support the arguments above.

6. Flawed Bayesian priors

Watchman trialists chose a Bayesian statistical analysis plan for the two trials—PROTECT AF[8] and PREVAIL.[9] That is reasonable. But Bayesian analyses depend on prior information, which three researchers recently described in a JAMA paper as "evidence or beliefs about something that exist prior to or independently of the data to be analyzed."[21] (Emphasis mine.)

Let's focus on said evidence and beliefs: When readers of the medical literature study the PROTECT AF trial, they find positive conclusions and mostly supportive editorials. But as a regulatory trial, it did not pass FDA review. Instead of approving the device after PROTECT AF, the FDA mandated a second trial—PREVAIL.

Was PROTECT AF as positive as the authors conclude or as uncertain as the FDA rejection implies?

Watchman trialists propose that the FDA rejected PROTECT AF on the basis of procedural complications, which improved in PREVAIL. I state this because they wrote exactly that in the introduction of the most recent JACC paper.[1]

Yet procedural complications were only one of many problems with PROTECT AF. Other flaws include overlap of antithrombotics (27% of the Watchman group remained on warfarin 45 days later); no signal of reduction in ischemic strokes with the device; spuriously high rates of hemorrhagic stroke in the warfarin arm; and concerns over the low rate of CV death in the Watchman arm.[22] The CV death issue prompted the FDA-mandated inspection of deaths in PREVAIL.[3] (See point 3.)

Thus, PROTECT AF results were at best supportive of feasibility but not pivotal for approval.

7. Noninferiority in PREVAIL met using less stringent standards

The final reason I remain skeptical pertains to the claim that Watchman met noninferiority for the second primary end point of ischemic stroke and systemic embolism more than 7 days after implant. In the planning of the trial, it was agreed that noninferiority could be met if the 95% upper credible interval was below 2.75% for the absolute risk difference or below 2.0 for the rate ratio.

At 5 years, the upper bound of the 95% credible interval for Watchman hit 2.75% for the absolute risk difference (exactly at the cutoff for noninferiority), but the upper-bound for the rate ratio hit 4.9—much higher than the cutoff for noninferiority of <2.0.

In the selection of the noninferiority margin based on risk difference, the investigators expected an event rate of 4% for the active-control arm. As it turns out, the observed event rate in the warfarin arm was 2% at the January 2013 follow-up and 1.35% at the 5-year follow-up. According to Dr Sanjay Kaul (Cedars-Sinai Medical Center), fixing the noninferiority margin as a risk difference rather than a risk ratio biases toward noninferiority when the observed event rate in the active control is less than the expected rate—as it was here.[23]

Bottom line, if you infer noninferiority based on the risk-difference margin, you must also admit that Watchman is up to nearly fivefold worse than warfarin.


I agree that there is an unmet need for better stroke prevention in patients deemed ineligible for anticoagulation. But we have no RCT-level evidence that LAAC is better than no therapy. This, despite the fact that tens of thousands of patients have had this procedure. To me, the prior information (evidence and beliefs) for such a trial do not bode well for LAAC.

I often observe a strong intervention bias among clinicians. The urge to intervene in times of uncertainty often stems from beneficent intent. But to stretch the available evidence from Watchman vs warfarin and conclude that it is worth exposing anticoagulation-ineligible patients to an invasive, unproven, preventive procedure is misguided.

I have heard clinicians say we owe it to these patients to try to help them. I say what we owe these patients is to study whether LAAC bests no therapy before accepting it as truth.


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.