In Defense of Left Atrial Appendage Closure

Vivek Y. Reddy, MD


November 29, 2016

To the Editor:

As an investigator in several left atrial appendage (LAA) closure trials, I read with interest Dr Mandrola's column, "Left Atrial Appendage Closure Should Stop Now" (posted November 10, 2016). On the one hand, I believe that the proper medium for scientific discourse is through the peer-reviewed process of scientific publications in medical journals and abstract presentations at national/international conferences. However, I recognize that Medscape is a valuable source of information and opinion reaching a large, varied audience, and is relied upon by many to digest an extensive literature. A number of important issues were raised in Dr Mandrola's column that warrant a more detailed examination, and I wish to thank Medscape for the opportunity to comment in this (admittedly lengthy) response.

Vivek Y. Reddy, MD

PROTECT AF was an FDA trial that randomized AF patients to either warfarin or LAA closure with Watchman (Boston Scientific; Marlborough, Massachusetts). Dr Mandrola's column references the 2009 manuscript of the initial analysis of PROTECT AF, at 1.5 years mean follow-up.[1] However, we have also published two subsequent analyses with extended follow-up—in 2013 and 2014 at 2.3 and 3.8 years follow-up, respectively.[2,3] The latest analysis was consistent with the initial analysis, showing annualized ischemic stroke rates of 1.4% and 1.1% in the Watchman and warfarin arms, respectively. First and foremost, these differences in ischemic stroke were not statistically significant. Second, several ischemic strokes in the Watchman arm were procedure-related strokes—that is, operator technical errors during the device implantation procedure resulting in inadvertent release of air or thrombus into the circulation. This is relevant from a mechanistic perspective because once the LAA was closed, the postprocedure ischemic stroke rates were the same (1.1%) in both the Watchman and warfarin arms. Furthermore, with improved training and education for the implanters, procedure-related strokes have become quite uncommon in subsequent clinical trials, decreasing from 1.1% of patients in PROTECT AF to ~0.1% in the remaining studies.[4,5]

Finally, in PROTECT AF, the primary endpoint was not simply ischemic stroke, but rather a composite of ischemic stroke, hemorrhagic stroke, and cardiovascular death—all "hard" endpoints of clinical relevance (systemic embolism was also included but occurred infrequently, thereby contributing little to the composite endpoint). Importantly, Watchman demonstrated statistically significant reductions in both of these remaining components of the primary endpoint: 85% reduction in hemorrhagic stroke and a 60% reduction in cardiovascular death.[3] Taken together, there was a statistically significant 40% reduction in the primary endpoint, again favoring Watchman.

PREVAIL was also a randomized trial with similar inclusion/exclusion criteria to PROTECT AF, but with a substantially different statistical plan. PREVAIL was not statistically powered to "stand on its own"; rather, it employed a Bayesian statistical design with an informative prior—that is, it "borrowed" data from PROTECT AF. This resulted in a smaller number of new patients being enrolled in PREVAIL. The prespecified plan was to conduct the analysis when the new PREVAIL patients were followed for a minimum of 6 months and the endpoints were determined using the modeled 18-month event rates against prespecified performance boundaries.[6] The first efficacy endpoint (stroke, systemic embolism, or cardiovascular death) outcome was a rate ratio of 1.07 (95% confidence interval (CI), 0.57-1.89), which did not meet the criteria for noninferiority due to overly wide confidence intervals. The second efficacy endpoint (postprocedure ischemic stroke or systemic embolism) outcome used a risk difference of 0.0053% (95% CI, –0.0190 to 0.0273), and this did meet the prespecified criteria for noninferiority.[6] (Please note that this is not statistical obfuscation but rather the prespecified statistical methodology under which this trial was designed.)

As noted in Dr Mandrola's column, additional follow-up in PREVAIL revealed a differential accrual of ischemic strokes in the Watchman arm—that is, more ischemic strokes with Watchman than with warfarin. But it is critical to recognize that PREVAIL was never powered to examine for differences in ischemic stroke. This is not simply an academic issue but a case example of the danger of overinterpretation of underpowered data. The reason for this differential rate was elucidated by the FDA-requested imputed placebo analyses: After adjustment for the CHA2DS2-VASc score, the ischemic stroke rate in the Watchman group was 65% lower than would be expected in untreated AF patients. This is right in line with the effect predicted with warfarin.

So if Watchman performed as well as warfarin would be expected to perform, why the different ischemic stroke rates? The reason was an extremely well-performing warfarin group in PREVAIL: The ischemic stroke rate was 0.3%, a rate far superior to that observed in any warfarin group in any randomized or nonrandomized clinical trial (for example, the ischemic stroke rates in the warfarin arms of RELY, ROCKET AF, ARISTOTLE, ENGAGE-AF, and PROTECT-AF were 1.2, 1.4, 1.1, 1.3, and 1.1, respectively). Why is this? It is not because these patients were at lower risk; the mean CHA2DS2-VASc score in PREVAIL was 4.0 ± 1.2. Rather, the answer is found in the wide 95% CI for this ischemic stroke rate: 0.03%-2.4%. That is, the relatively small number of patients in the warfarin arm of PREVAIL (n = 138) resulted in a point estimate with low confidence—ie, chance. Indeed, if warfarin was truly able to reduce ischemic stroke to 0.3% in a CHA2DS2-VASc = 4.0 AF population, there is no known therapeutic alternative that would be equivalent, including dabigatran, rivaroxaban, apixaban, and edoxaban. (Finally, even looking at this PREVAIL-only data, at the time of the third FDA panel, the increased ischemic stroke rate in the Watchman arm was offset by decreases in both hemorrhagic stroke, by 2.9-fold, and in cardiovascular death, by 1.5-fold, in the Watchman arm.[7])

But it is also not appropriate to simply ignore the data from these new PREVAIL patients. This is the reason that, within months of the FDA panel when this analysis was conducted, we submitted a manuscript in which we conducted a patient-level meta-analysis that included all randomized patients in both PROTECT AF and PREVAIL.[8] This was not a biased analysis; it included both the procedure-related strokes occurring in the Watchman arm of PROTECT AF (which have clearly since declined in subsequent trials) as well as the overperforming warfarin arm of PREVAIL. Also, because this was a patient-level meta-analysis, neither over- nor underemphasis of either trial could occur. Furthermore, this was a post hoc analysis with no complicated statistics, so the data could be simply presented as hazard ratios and confidence intervals.

The meta-analysis revealed that the overall stroke rate between Watchman and warfarin was similar (HR 1.02; 95% CI, 0.62-1.69); thus, mechanical LAA closure performed as well as systemic anticoagulation with warfarin for the endpoint of stroke.[8] As partly discussed in Dr Mandrola's column, it is also true that there was a divergence in the relative rates of ischemic and hemorrhagic strokes: The former favored the warfarin arm (HR 1.95; 95% CI, 1.0-3.8) and the latter the Watchman arm (HR 0.22; 95% CI, 0.08-0.61). These data should be interpreted from two perspectives: the intention-to-treat patient-centered perspective and a mechanistic perspective.

  • Patient perspective: Beyond the divergence of stroke mechanism, it is known that: (1) the overall stroke rate is the same between groups (HR 1.02, as noted above); (2) stroke severity favored Watchman—there were 56% more disabling/fatal strokes in the warfarin arm[3,7]; and (3) there were fewer cardiovascular and all-cause deaths in the Watchman arm (HR 0.48; 95% CI, 0.28-0.81 vs HR 0.73; 95% CI, 0.52-1.03).[8] And this ~50% reduction in cardiovascular mortality really does need to be emphasized; there are few cardiovascular interventions that have actually demonstrated mortality benefits.

  • Mechanistic perspective: To ask the question of whether mechanical LAA closure can recapitulate the benefit of systemic oral anticoagulation, one must look at ischemic strokes beyond those that are procedure-related (HR 1.56; 95% CI, 0.78-3.09).[8] Again, the wide confidence intervals indicate no statistical difference in ischemic stroke rates between Watchman and warfarin. That is, the reduction in ischemic stroke expected with mechanical closure is statistically similar to that expected for warfarin anticoagulation. And is it reasonable to think that operators can place the Watchman device without causing procedure-related strokes? The answer is an unequivocal "yes." As previously mentioned, PROTECT AF was the only trial with a high procedure-related stroke rate (1.1%). With increased awareness of this issue, in all subsequent FDA and large non-FDA studies, procedure-related strokes have infrequently been observed (average rate of ~0.1%).[4,5,9] (At the TCT 2016 Scientific Sessions, we reported in an abstract that with additional follow-up in PREVAIL, the postprocedure ischemic stroke difference has narrowed further to HR 1.35.[10] After the 5-year follow-up is complete in mid-2017, we expect to be able to report the final outcomes.)

One final point regarding hemorrhagic stroke. It is indeed true that the statistically significant ~80% decrease is hemorrhagic stroke with Watchman is related to these patients not taking anticoagulants. But this is exactlythe point. The problem with oral anticoagulants is that systemic anticoagulation comes with a cost—bleeding, including intracranial bleeding. This is not a minor issue. It is well known that compared with ischemic strokes, hemorrhagic strokes have a worse impact on patients: more disability and death. Indeed, the major benefits observed with novel oral anticoagulants (NOACs) over warfarin are related to the lower hemorrhagic stroke rates, which drove a 10% reduction in mortality.[11] Similarly, the major advantage of mechanical LAA closure is the ability to provide a similar reduction in ischemic stroke with a lower bleeding penalty.

As noted in the column, there was a high complication rate in PROTECT AF (the most frequent complication, cardiac tamponade, occurred in 4.3% of patients), which decreased in the subsequent FDA trials (tamponade rates were 1.9% in PREVAIL, 1.4% in CAP, and 1.9% in CAP2).[2,5] The complication rates in real-world practice are of critical importance. It would be reasonable to expect complications to increase when the procedure is performed beyond the careful oversight expected in clinical trials. However, it is interesting to note that complications rates were quite low in the European real-world registry, EWOLUTION (N = 1021; tamponade rate of 0.3%).[9]

It was in this context that we were interested in the outcomes of Watchman implantation after FDA approval in March 2015. However, the CMS-mandated National Cardiovascular Data Registry (NCDR) was only certified in August 2016. So while our center and others are submitting data from before this timepoint, it is expected that the data from many of the initially implanted patients will not be captured. However, as part of the initial rollout plan for the Watchman device that the manufacturer devised in conjunction with the FDA, every Watchman case was attended by a trained manufacturer clinical specialist who, in addition to providing guidance for the implant, was mandated to collect deidentified procedure-related data, including acute complications, on a standardized form. So instead of not having any information on these initially implanted patients, we decided to collect and report the data, imperfect as it is. The fact that an industry representative collected the data was indicated multiple times in the manuscript, including the abstract, introduction, methods, and discussion. Briefly, between March 2015 and May 2016, a total of 3822 consecutive patients underwent Watchman implantation at 169 centers by 382 operators with a relatively low complication rate (eg, tamponade rate of 1.02%).[5] We certainly would have preferred a prospective registry with independent monitoring, a core lab, etc. But in the absence of this, we did our best: We separately adjudicated the events (two of the authors) and independently corroborated the data with that reported to the FDA MAUDE database. While the data have limitations (that we acknowledged in the manuscript), we believed, and still believe, that it adds to the literature and our understanding of the therapy. Of course, we are looking forward to seeing the outcomes in the comprehensive NCDR registry, which hopefully will be available next year.

In addition to the above referenced manuscripts, I would direct interested clinicians to other peer-reviewed manuscripts examining different aspects of Watchman therapy, including: (1) the favorable effect on quality of life[12] (2) reduction of late bleeding[13]; and (3) economic benefits.[14,15,16] Furthermore, while I have largely limited my comments to the Watchman device because it is the only one with FDA approval, there are published, nonrandomized data with other LAA closure devices (such as the Amplatzer™ Amulet™ device; St. Jude Medical; St. Paul, Minnesota), much from outside the United States.[17,18] While not definitive, it is reassuring that different LAA closure device technologies all show favorable concordant outcomes. Of course, more data will eventually become available as many of these devices are currently being studied in randomized FDA trials (mostly compared with Watchman).

Despite these favorable outcomes with Watchman (vs warfarin), the advent of the NOACs—drugs with substantial advantages over warfarin—changes the playing field. This is one of the primary reasons that the clinical trial investigators have, to a person, advocated that LAA closure should be considered for patients who are poor candidates for long-term oral anticoagulation. This is also the position taken by the device manufacturer. In the 2014 FDA panel, the manufacturer's chief medical officer stated: "...the device should not be viewed as a broad first line replacement for oral anticoagulants. Patients and physicians should view Watchman as an alternative for those who are eligible for warfarin, but who have individual reasons to seek another long-term therapeutic option." The fact is: oral anticoagulants work very well for many patients, but there is a need for this alternative for those in whom drugs are not a good option. Unless future prospective randomized trials demonstrate equivalence of LAA closure with NOACs, this therapy should not be considered an equivalent first-line alternative to oral anticoagulants.

Why not limit LAA closure to anticoagulation-ineligible patients? In the randomized clinical trials, after Watchman implantation, patients received 6 weeks of oral anticoagulation with warfarin. Thus, patients receiving the Watchman device must have been eligible for oral anticoagulation. It would not have been scientifically sound for either the FDA or CMS to advocate, or even permit, Watchman implantation in anticoagulant-ineligible patients. On the other hand, there are nonrandomized European data indicating that the device can be placed without the 6-week oral anticoagulant transition.[19,20] Accordingly, the device manufacturer has recently commenced a multicenter, multinational randomized FDA clinical trial of Watchman implantation in contraindicated patients. In this trial, called ASAP-TOO,[21] eligible patients are randomized to either an antiplatelet agent or to Watchman implantation, with only a 12-week postimplant regimen of dual antiplatelet agents. This study is expected to commence first enrollments before the end of 2016. But in the interim, Watchman implantation in the United States must be limited to those patients who are poor candidates for long-term oral anticoagulation but can tolerate at least a short-term exposure to oral anticoagulants.

In conclusion, the totality of the scientific evidence indicates that, as compared with warfarin, LAA closure with the Watchman device is:

  • Equivalent for total stroke

  • Inferior for ischemic stroke, in large part because of procedure-related strokes (that have since decreased with operator experience)

  • Superior for hemorrhagic stroke

  • Superior for disabling/fatal strokes

  • Superior for cardiovascular mortality

  • Like all procedures, the LAA closure procedure is associated with risks, but with increased experience, their frequency has diminished substantially

And again, it is very hard to not be impressed by a therapy that delivers a mortality benefit. These are the reasons that the device has received regulatory approval in the United States and over 75 other countries. This device should not be considered first-line therapy for AF stroke prevention. Instead, it is being considered for patients who are poor candidates for long-term oral anticoagulation.

One final editorial observation: As I read the column and the corresponding comments, I was struck by the apparent ease with which insinuations of impropriety have been levied against the trial investigators, device manufacturers, and even the FDA. Of course, everyone agrees on the importance of transparency regarding potential conflicts of interest that may consciously, or even subconsciously, influence scientific discourse. (For example, the readers should know that I have received grant support and have served as a consultant to Boston Scientific, the manufacturer of the Watchman device, as well as to Coherex and St. Jude Medical, manufacturers of other LAA closure devices that are being/will be studied in FDA trials.) And certainly it would be preferable for clinical trials such as these to be funded by the NIH or other nonindustry sources. (Though anyone who has tried to gain government funding for device studies knows the difficulty of this prospect; without industry sponsorship, there would be very few cardiovascular trials conducted indeed.) However, I would submit that while thoughtful, scientifically sound, and statistically supported debate is both appropriate and necessary, indiscriminate and unsubstantiated accusations of bias are neither helpful, respectful, or fair to the numerous investigators dedicated to completing these difficult-to-conduct clinical trials over the past decade-plus, the patients who have willingly participated in them (and to whom we are grateful), and even the numerous industry-related clinical scientists integral to their conduct. I would encourage anyone interested in the care of AF patients and wanting to advance the science to participate in ongoing clinical trials or to conduct other independent investigations to further our understanding.


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.