The timing of the withdrawals and loss to follow-up is described in eTable 1 in Supplement 2. AF indicates atrial fibrillation; LV, left ventricular. aNo device was implanted because of embolization, explant, or aborted procedure for 9 patients, 2 patients had other medical conditions that did not allow for follow-up, and 1 patient developed amyloidosis.bSix patients discontinued warfarin therapy, 3 had other medical conditions that did not allow for follow-up, and 1 patient relocated overseas.
The primary efficacy outcome (A) was stroke, systemic embolization, or cardiovascular death. The primary safety outcome (B) was a composite of major bleeding events and procedure-related complications. Incident probabilities for the intention-to-treat analysis are shown. HR indicates hazard ratio.
HR indicates hazard ratio; RR, rate ratio.
Results are from Cox proportional hazards models, with each subgroup examined in a separate model. AF indicates atrial fibrillation; HR, hazard ratio; LAA, left atrial appendage; LV, left ventricular; TIA, transient ischemic attack.
eTable 1: Interim Patient Accountability
eTable 2: All Stroke Event Rates, According to Secondary Analyses
eTable 3: Cardiovascular/Unexplained Death Event Rates, According to Secondary Analyses
eTable 4: Baseline Characteristics of All Patients Withdrawing Consent or Lost to Follow-up
eTable 5: Conservative Sensitivity Analysis For Primary Efficacy Endpoint
eTable 6: Vital Status Update
eTable 7: Hemorrhagic Stroke Rates in the Warfarin Arms of Controlled Clinical Trials
eTable 8: Mortality Rates in the Warfarin Arms of Controlled Clinical Trials
eTable 9: Types of Cancers, According to Treatment Group
Vivek Y. Reddy, Horst Sievert, Jonathan Halperin, Shephal K. Doshi, Maurice Buchbinder, Petr Neuzil, Kenneth Huber, Brian Whisenant, Saibal Kar, Vijay Swarup, Nicole Gordon, David Holmes, . Percutaneous Left Atrial Appendage Closure vs Warfarin for Atrial FibrillationA Randomized Clinical Trial. JAMA. 2014;312(19):1988–1998. doi:10.1001/jama.2014.15192
While effective in preventing stroke in patients with atrial fibrillation (AF), warfarin is limited by a narrow therapeutic profile, a need for lifelong coagulation monitoring, and multiple drug and diet interactions.
To determine whether a local strategy of mechanical left atrial appendage (LAA) closure was noninferior to warfarin.
Design, Setting, and Participants
PROTECT AF was a multicenter, randomized (2:1), unblinded, Bayesian-designed study conducted at 59 hospitals of 707 patients with nonvalvular AF and at least 1 additional stroke risk factor (CHADS2 score ≥1). Enrollment occurred between February 2005 and June 2008 and included 4-year follow-up through October 2012. Noninferiority required a posterior probability greater than 97.5% and superiority a probability of 95% or greater; the noninferiority margin was a rate ratio of 2.0 comparing event rates between treatment groups.
Left atrial appendage closure with the device (n = 463) or warfarin (n = 244; target international normalized ratio, 2-3).
Main Outcomes and Measures
A composite efficacy end point including stroke, systemic embolism, and cardiovascular/unexplained death, analyzed by intention-to-treat.
At a mean (SD) follow-up of 3.8 (1.7) years (2621 patient-years), there were 39 events among 463 patients (8.4%) in the device group for a primary event rate of 2.3 events per 100 patient-years, compared with 34 events among 244 patients (13.9%) for a primary event rate of 3.8 events per 100 patient-years with warfarin (rate ratio, 0.60; 95% credible interval, 0.41-1.05), meeting prespecified criteria for both noninferiority (posterior probability, >99.9%) and superiority (posterior probability, 96.0%). Patients in the device group demonstrated lower rates of both cardiovascular mortality (1.0 events per 100 patient-years for the device group [17/463 patients, 3.7%] vs 2.4 events per 100 patient-years with warfarin [22/244 patients, 9.0%]; hazard ratio [HR], 0.40; 95% CI, 0.21-0.75; P = .005) and all-cause mortality (3.2 events per 100 patient-years for the device group [57/466 patients, 12.3%] vs 4.8 events per 100 patient-years with warfarin [44/244 patients, 18.0%]; HR, 0.66; 95% CI, 0.45-0.98; P = .04).
Conclusions and Relevance
After 3.8 years of follow-up among patients with nonvalvular AF at elevated risk for stroke, percutaneous LAA closure met criteria for both noninferiority and superiority, compared with warfarin, for preventing the combined outcome of stroke, systemic embolism, and cardiovascular death, as well as superiority for cardiovascular and all-cause mortality.
clinicaltrials.gov Identifier: NCT00129545
Oral anticoagulation with warfarin has been the mainstay of treatment for prevention of cardioembolic stroke in atrial fibrillation (AF).1 Although effective, warfarin is limited by a narrow therapeutic profile, a need for lifelong coagulation monitoring, and multiple medication and food interactions.2 Indeed, approximately 40% of eligible patients do not receive anticoagulation, leaving them at substantial risk for stroke.3
Echocardiographic and autopsy studies suggested that the left atrial appendage (LAA) is the major source of thromboembolism in patients with AF, leading to development of mechanical approaches to close the LAA.4- 9 A self-expanding device composed of a nickel-titanium frame and permeable polyester fabric covering was designed for percutaneous delivery via a transseptal sheath to the LAA ostium.8 After the device is deployed, patients are treated with aspirin and warfarin for 45 days to facilitate endothelial coverage, followed by aspirin and clopidogrel and, after 6 months, aspirin monotherapy.
“Local” therapy by device-based LAA closure was compared with systemic therapy with warfarin in the randomized PROTECT AF clinical trial in patients with nonvalvular AF and at least 1 CHADS2 risk factor. The trial revealed noninferiority of LAA closure to warfarin for prevention of stroke, systemic embolism, or cardiovascular or unexplained death.10,11 At 2 years, there were 3.0 and 4.3 events per 100 patient-years in the LAA closure and warfarin groups, respectively (relative rate ratio, 0.71; 95% credible interval, 0.44-1.30), meeting the criteria for noninferiority (posterior probability, >99.9%), but not superiority (posterior probability, 88%).
These findings validated the hypothesis that in nonvalvular AF, ischemic stroke occurs largely as a result of embolism of thrombus from the LAA. Results from this study at 18 months and 2.3 years of follow-up have been previously reported.10,11 But there remained uncertainty surrounding these early results based on the breadth of the prespecified noninferiority margin and frequency of procedure-related complications. Accordingly, we examined the long-term (4-year) efficacy and safety of LAA closure for stroke prophylaxis.
The design and conduct of the trial have been described (see the trial protocol in Supplement 1).12 This unblinded randomized study was conducted at 59 centers in the United States and Europe. Patients were enrolled from February 2005 to June 2008; the planned ending date for analysis of outcomes after 4 years was in October 2012. The protocol was approved by each participating site’s institutional review board, and written informed consent was obtained from each patient prior to enrollment. All safety and efficacy events were independently adjudicated by an unblinded clinical events committee (CEC). An independent data and safety monitoring board oversaw the conduct of the trial.
As previously described, the major inclusion criteria were age 18 years or older; paroxysmal, persistent, or permanent nonvalvular AF; 1 or more CHADS2 risk factors (age ≥75 years, hypertension, diabetes, heart failure or left ventricular [LV] systolic dysfunction, prior transient ischemic attack [TIA] or stroke); and eligibility for long-term anticoagulation with warfarin. Race and ethnicity data were self-reported. Major exclusion criteria were patent foramen ovale with atrial septal aneurysm, an atrial septal defect, mechanical valve prosthesis, LV ejection fraction less than 30%, mobile aortic atheromata, and symptomatic carotid disease. Eligible patients underwent neurological examination by a neurologist, and those with a history of prior thromboembolism underwent baseline magnetic resonance imaging or computed tomography (CT) neuroimaging. Potential study candidates were identified in both the outpatient and inpatient setting by physician or self-referrals in response to approved study announcements.
Patients were randomized by computer in 2:1 allocation to either WATCHMAN device implantation (Boston Scientific) or warfarin therapy, stratified by clinical center in a block size of 6 (eMethods in Supplement 2).12 The device was implanted under transesophageal echocardiographic (TEE) guidance. After transseptal puncture, an LAA angiogram was performed. An appropriately sized device (21-33 mm in diameter) was advanced to the LAA ostium using a 12-French transseptal sheath. Once proper positioning and stabilization were verified, the device was released.
Device implantation included concomitant antithrombotic medication to facilitate device endothelialization: warfarin and aspirin (81-325 mg) for 45 days. To assess for device stability, peridevice leaks, and device-related thrombus, TEE imaging was performed at 45 days, 6 months, and 12 months. When the 45-day TEE revealed minimal residual peridevice flow (jet width ≤5 mm) and no device-related thrombus, warfarin was stopped and replaced by clopidogrel, 75 mg daily, until the 6-month visit, after which only aspirin was continued. If an adequate seal was not obtained or a thrombus was detected, patients continued taking warfarin until an adequate seal was attained or thrombus was resolved before transitioning to aspirin. For patients taking warfarin, international normalized ratio (INR) monitoring was performed at least every 2 weeks for 6 months and at least monthly thereafter, targeting an INR between 2 and 3. Follow-up visits occurred twice annually after the first year, with neurological assessments at 12 months and yearly thereafter or whenever a neurological event was suspected.
The trial was designed to establish whether a device-based strategy was noninferior to warfarin for a composite primary efficacy end point of stroke, systemic embolism, or cardiovascular death; unexplained death was presumed cardiovascular. The primary composite safety end point included procedure-related events in the device group (pericardial effusion requiring intervention or prolonged hospitalization, procedure-related stroke, or device embolization) and major bleeding in both groups (intracranial or bleeding requiring transfusion). Rates were calculated as the number of events per 100 patient-years of follow-up. The time in therapeutic range during warfarin anticoagulation was calculated according to the linear interpolation method of Rosendaal et al.13
The study design and primary results have been reported at various time points of study follow-up.10,12 The derivation of the required sample size has been previously described (eMethods in Supplement 2).12,14 The sample size was based on an expected primary efficacy end point event rate of 6.15 per 100 patient-years in the warfarin group, calculated by use of data from the Stroke Prevention in Atrial Fibrillation studies database and requiring a probability of noninferiority of 97.5% or greater, using a noninferiority margin of 2.0 for the difference in event rates between treatment groups expressed as a rate ratio (RR). Simulations were performed to ensure 80% power and a 5% false-positive (type I error) rate under a group sequential analysis plan that included a first interim analysis after 600 patient-years of follow-up and subsequent analyses after each additional 150 patient-years, up to a maximum of 1500. A probability criterion of 95% was defined as required for superiority. Noninferiority of the primary efficacy end point was achieved at the first prespecified interim analysis after 600 patient-years of exposure. A prespecified noninferiority margin was not set for the primary safety end point. The protocol also required follow-up for 5 years after enrollment with annual regulatory reporting. Statistical results are reported nominally without adjustment for multiple comparisons over time because the sampling scheme for extended long-term follow-up was beyond that defined by the original protocol.
The analysis reported here is based on 4 years of follow-up. Efficacy and safety analyses are based on a Bayesian Poisson model assuming a constant hazard rate and event occurrence following a Poisson distribution.14 For Bayesian analyses, credible intervals and posterior probabilities are reported. Analyses were based on intention-to-treat, censoring data from patients without events at the time of the last known status. Consistency was assessed using a Cox proportional hazards model with confidence intervals for the difference in rates. The Kaplan-Meier method was used for graphical assessment of time-dependent events.
To isolate the effect of LAA closure, a postprocedure analysis excluded patients in whom the device could not be successfully implanted and excluded events directly related to device implantation. Additional secondary analyses were performed using frequentist methods to demonstrate the effect of the postimplant drug regimen on efficacy and safety outcomes. A per-protocol analysis was performed to assess the outcomes of patients with the device after completing the requisite 45-day period of anticoagulation after implantation (day 0 was designated the day after stopping warfarin). A terminal therapy analysis assessed outcomes after discontinuation of clopidogrel, thereby defining efficacy and safety of the device during long-term therapy with aspirin alone. No α-level adjustment was made for these sensitivity analyses. The P values for these secondary analyses were nominal, 2-sided, and nominally significant at P < .05 with no adjustment for multiple comparisons.
The study cohort included 707 patients at 59 sites: 463 randomized to the device group and 244 patients to warfarin (Figure 1). Data quality concerns resulted in censoring data from 1 site (eMethods in Supplement 2). There were no significant differences in principal baseline characteristics between the groups (Table 1). Most were white, and approximately four-fifths of the cohort was enrolled in the United States. The mean CHADS2 scores in the device and warfarin groups were 2.2 and 2.3, respectively, and approximately equivalent thirds of the patient cohort had CHADS2 scores of 1, 2, or 3 or greater. The most frequent stroke risk factor was hypertension, and slightly less than a fifth of the patient cohort had previously sustained a stroke or TIA. Approximately two-fifths of the patients had paroxysmal AF, and in most, AF had been present for 1 year or longer. Prior to entry, approximately half the patients had taken warfarin for 1 year or longer.
This analysis reflects a mean (SD) follow-up of 3.8 (1.7) years (range, 0-6.5 years), in aggregate 2621 patient-years. By group, the mean (SD) follow-up was 3.9 (1.9) years in the device group and 3.7 (1.6) years in the warfarin group. A higher proportion of patients in the warfarin group (22.5%) than the device group (6.3%) actively withdrew consent to continue in the study, but the proportion of patients lost to follow-up was similar in the groups: 4.5% vs 2.8% of the warfarin and device groups, respectively.
Adherence was high in the warfarin group (time in therapeutic range, 70%). Implant success has been previously reported at 88% (408/463 patients).11 After the 45-day, 6-month, and 12-month TEE evaluations, warfarin was discontinued for 348 of 401 patients (86.8%), 355 of 385 patients (92.2%), and 345 of 370 patients (93.2%), respectively.
There were 39 primary efficacy events among 463 device patients (8.4%, 2.3 events per 100 patient-years, 95% credible interval 1.7-3.2) vs 34 events among 244 warfarin patients (13.9%, 3.8 events per 100 patient-years, 95% credible interval, 2.5-4.9) (RR, 0.60 favoring device; 95% credible interval, 0.41-1.05), meeting criteria for both noninferiority (posterior probability, >99.9%) and superiority (posterior probability, 96.0%) (Table 2 and Figure 2). Similar results were obtained using the Cox proportional hazards model: 2.3 events per 100 patient-years in the device group vs 3.8 in the warfarin group (hazard ratio [HR], 0.61; 95% CI, 0.38-0.97; P = .04). Fewer hemorrhagic strokes occurred in the device group (3/463 patients, 0.6%) than in the warfarin group (10/244 patients, 4.0%; RR, 0.15; 95% credible interval, 0.03-0.49), and fewer cardiovascular deaths occurred in the device group (17/463 patients, 3.7%) than in the warfarin group (22/244 patients, 9.0%; RR, 0.40; 95% credible interval, 0.23-0.82). Although rates of all stroke (26/463 [5.6%] device patients vs 20/244 [8.2%] warfarin patients; RR, 0.68; 95% credible interval, 0.42-1.37) and ischemic stroke (24/463 [5.2%] device patients vs 10/244 [4.1%] warfarin patients; RR, 1.26; 95% credible interval, 0.72-3.28) did not differ between the 2 groups (Figure 3A), there were fewer fatal or disabling strokes in the device group (RR, 0.37; 95% credible interval, 0.15-1.00).
Mortality over time is depicted in Figure 3. There were 57 deaths from any cause among 463 patients (12.3%, 3.2 events per 100 patient-years, 95% CI, 2.5-4.2) in the device group and 44 deaths from any cause among 244 patients (18.0%, 4.8 events per 100 patient-years, 95% CI, 3.6-6.4) in the warfarin group (HR, 0.66; 95% CI, 0.45-0.98; P = .04), favoring the device-based strategy. In the warfarin group, there were a greater number of cardiovascular, pulmonary, and hemorrhagic stroke-related deaths—although only the last of these reached statistical significance (Table 3). The frequency of ventricular arrhythmic death or sudden death was similar in the device (0.9%) and warfarin groups (1.2%, P = .70). The rate of cancer-related deaths was not statistically different between groups.
The efficacy results were consistent across a number of subgroups, including those based on sex, age (dichotomized at 75 years), pattern of AF, CHADS2 score, LAA morphology, and prior warfarin use (dichotomized at 1 year) (Figure 4). Among patients who had sustained a stroke or TIA prior to enrollment (82/463 [17.7%] in the device group, 49/244 [20.1%] in the warfarin group), the results were directionally favorable toward the device strategy (HR, 0.66; 95% CI, 0.30-1.45) but were not statistically significant.
The rate of composite primary safety events was similar in the 2 groups. Based on the Bayesian-Poisson model, there were 3.6 events per 100 patient-years in the device group vs 3.1 in the warfarin group (RR, 1.17; 95% credible interval, 0.78-1.95), meeting the criterion for noninferiority (posterior probability, 98.0%) (Table 2). Similar results were obtained using the Cox proportional hazards model: 3.6 events per 100 patient-years in the device group vs 3.1 in the warfarin group (HR, 1.21; 95% CI, 0.78-1.94, P = .41).
The most frequent adverse events were serious pericardial effusions and major bleeding in the device and warfarin groups, respectively (Table 4). Although the overall rate of adverse events was similar, Kaplan-Meier analysis revealed differences in the temporal distribution of these events between the groups (Figure 2B). Unlike the linear evolution of events over time during warfarin therapy, a bimodal distribution was observed after device implantation with the predominating adverse events—pericardial effusion, procedure-related stroke, and device embolization—occurring in the periprocedural period.
Because the device strategy involves not only device implantation but concomitant antithrombotic therapy, prespecified secondary analyses assessed the extent to which benefit of the device strategy was derived from adjunctive therapy rather than LAA closure. With regard to the primary efficacy end point (Table 5), evaluations consisted of the postprocedure analysis, which included only patients successfully receiving the device; the per-protocol analysis, which included patients after completing warfarin therapy; and the terminal therapy analysis, which included patients completing both warfarin and clopidogrel therapy. All revealed risk reductions of approximately 50% with the device. Similar results were seen for both stroke alone and cardiovascular or unexplained death (eTables 2 and 3 in Supplement 2). Hazard ratios for these analyses dropped below unity for the safety end points, consistent with the observation that adverse events in patients undergoing device deployment were predominantly procedural. Taken together, these findings suggest that the benefit of the device strategy cannot simply be attributed to the transient antithrombotic treatment.
The long-term (3.8 years) follow-up of patients randomized in the PROTECT AF trial revealed that among patients with nonvalvular AF, LAA closure reduced the relative risk of the composite end point of stroke, systemic embolism, and cardiovascular death by 40% (1.5% absolute reduction) compared with warfarin anticoagulation. Furthermore, the device-based strategy was associated with a 60% relative risk (1.4% absolute reduction) of cardiovascular death and 34% relative reduction (5.7% absolute reduction) in all-cause death. However, it should be noted that these mortality end points are secondary end points, and due to multiplicity of data analysis, there is some uncertainty in the confidence of this conclusion. Although the device implantation procedure was associated with early complications, the accumulation of complications related to chronic anticoagulation resulted in similar safety profiles for the 2 modalities.
The patient cohort was representative of AF populations enrolled in other stroke prevention trials.15- 17 The mean CHADS2 score was 2.2, and approximately half of the patients were warfarin naive (<1 year). It is noteworthy, however, that more than 90% of enrolled patients were white, and other ethnic groups were not well represented.
The majority of randomized patients received the intended therapy, with 99% of patients in the warfarin group receiving the anticoagulant. Adherence with therapy (time in therapeutic range, 70%) was high compared with contemporary trials15- 17 although not surprising as the time in therapeutic range may be expected to steadily improve with greater warfarin experience. Among patients randomized to receive the device, deployment was successful in 88%, although previous reports show higher success rates as operators gained experience.18 Secondary analyses excluding patients not receiving the device did not negatively affect the efficacy comparison.
The mean follow-up in this trial (3.8 years) exceeded that of most contemporary stroke prophylaxis trials, such as RELY (2.0 years), ROCKET-AF (1.9 years), and ARISTOTLE (1.8 years).15- 17 However, this extended follow-up resulted in a relatively large number of voluntary withdrawals, especially among those randomized to receive warfarin (22.5%), although the number of patients lost to follow-up was similar in the 2 groups (4.5% vs 2.8%). The principal reasons for withdrawal from warfarin were a desire for alternate treatment such as a novel oral anticoagulant and perceptions by patients in the warfarin group of little benefit from continued participation. Differential withdrawal could bias interpretation of event rates, but truncated exposure preventing detection or registration of events would be more costly to the device group if patients ceased participation after complications from warfarin therapy. Furthermore, additional sensitivity analyses revealed that as assessed by time in therapeutic range prior to dropout, the higher-risk warfarin patients withdrew from the study, thereby biasing the study against the device group. When a vital status update was obtained, the overall mortality even more strongly favored the device group (HR, 0.63; 95% CI, 0.45-0.90; P = .01) (eMethods and eTables 4-6 in Supplement 2); patients who did not comply with the follow-up requirements of the protocol may have been less adherent to warfarin dose regulation, raising the risk of ischemic or hemorrhagic events.
Earlier analyses of trial data at 1065 and 1588 patient-years of exposure (mean follow-up, 1.5 and 2.3 years, respectively) found the device strategy noninferior to warfarin for the primary efficacy end point.10,11 Although the RR ranged from 0.6 to 0.7, wide credible intervals had precluded establishment of superiority. But the current scheduled data analysis yielded superiority, thereby representing a notable change in the conclusion as compared with previously reported outcomes. With the additional follow-up detailed herein, whether analyzed according to the Bayesian-Poisson model or the frequentist Cox proportional hazards model, the device was associated with a 40% relative reduction in the risk of the primary end point when compared with warfarin. Secondary analyses indicated that this advantage was unexplained by the antithrombotic therapies required early after device deployment.
The beneficial outcomes with the device were driven largely by lower rates of hemorrhagic stroke and cardiovascular death. Cardiovascular mortality in the warfarin group (2.4% annually) was similar to that in other trials (range, 2.0%-2.7%), but the rate of intracranial hemorrhage (ICH) in the warfarin group (1.1%/year) was higher than in other trials (typically 0.4%-0.5%/year), although confidence intervals overlap (eTable 7 in Supplement 2). While the average CHADS2 score was similar in other trials, and the difference in ICH could represent chance variation, patients enrolled in this study may have been at higher risk based on characteristics not reflected in the CHADS2 score. There may also have been differences in definitions of stroke in the various studies; half the reported events occurred after falls, making it difficult to determine whether ICH was cause or consequence.
There was a statistically significant decrease in the relative risk of all-cause mortality, favoring the device-based strategy. In absolute terms, the all-cause mortality rate in the warfarin group was 21.5% compared with 14.5% in the device group for an absolute reduction at 5 years of 7.0%. The death rate in the warfarin group (3.2%/year) was similar in other trials (range, 3.2%-4.9%/year) (eTable 8 in Supplement 2) and driven by differential rates of ICH, cardiovascular mortality, and death due to pulmonary causes. Conversely, the HR for cancer-related mortality in the device group was increased. This increase did not reach statistical significance (eTable 9 in Supplement 2) but is consistent with studies suggesting a potential protective effect of warfarin in the pathogenesis and mortality of urological or gastrointestinal tumors, perhaps related to earlier detection.19
The ischemic stroke rate in the device group (1.4%/year, 24/463 patients) was not significantly greater than in the warfarin group (1.1%/year, 10/244 patients, P = .49). The rates were also identical between groups in the postprocedure analysis (6 procedure-related strokes among 463 patients), reflecting technical complications of device implantation and suggesting that stroke prevention after LAA closure accrues over time.
Although there was no significant difference in composite safety outcome between groups, among patients in the device group there was a time-dependent distribution of safety events, with a high initial accumulation of procedure-related complications followed by a lower rate of adverse events over time. Procedure-related pericardial effusion with tamponade was relatively frequent but not fatal,18 while procedure-related stroke was presumably due to embolism of air or thrombus during device deployment. Analysis considering the weighted risks of stroke, ICH, death, major bleeding, and pericardial tamponade found the net clinical benefit to favor the device over warfarin.20 The high up-front procedural risk is further mitigated by a decrease in the complication rate with operator experience.18 Furthermore, in the registry that followed this trial, the cardiac tamponade rate decreased by more than 50%, and no procedure-related strokes occurred.18 Although this residual pericardial tamponade rate of approximately 2% is not trivial, tamponade rates during other similarly invasive cardiac procedures such as catheter ablation of AF have been reported to range between 1% and 6%.21
While evaluating the safety of the device, it must be noted that the LAA closure device used in this trial (WATCHMAN), is not just an implant, but a therapeutic strategy involving a procedure plus 6 months of anticoagulation or antithrombotic pharmaceutical intervention (or both) to fully protect a patient from the defined primary efficacy end points; thus, this study does not address patients with absolute contraindications to warfarin unable to tolerate this initial anticoagulant transition. Effectiveness might improve if postimplant anticoagulation were abbreviated because there were several bleeding events during the first 6 months after deployment; indeed, a nonrandomized study of device implantation in warfarin-ineligible patients found implantation without a warfarin transition to be safe.22 Additional studies are needed to determine the optimum antithrombotic therapy for use in conjunction with the device.
While the protocol did prespecify follow-up to 5 years, this analysis is still limited by the fact that the statistical plan in PROTECT AF was only prespecified to 1500 patient-years of follow-up. This is somewhat mitigated by concordance of the current analysis strategy with that used for the earlier analyses (Bayesian and frequentist approaches for the primary and secondary analyses, respectively).
It is important to recognize that patients and physicians in this trial were not blinded to treatment assignment. This could result in bias with regard to both clinical event adjudication and the possibility that patients in the device group may have received better overall cardiovascular care, thereby affecting mortality. But unlike in pharmaceutical trials, blinding is difficult when you have a device with an upfront risk that may be self-revealing (ie, pericardial effusion, cardiac tamponade) as well as a drug whose use is revealed by simple blood tests (ie, prothrombin time, INR). Furthermore, the safety end point required unblinding for assessment of procedure relationship. Finally, stroke events required unblinding for assessment of potential thrombus on the device. But for hard end points like death, stroke, and systemic embolism, the potential for bias due to unblinding is low. Furthermore, the trial did use an independent CEC to adjudicate adverse events and operated separately from any implanters, and only adjudicated events are included in all primary end point analyses.
The clinical generalizability of this trial has several limitations. This trial randomized LAA closure against warfarin, but the advent of novel Factor II and Xa inhibitors raises questions about their comparative safety and efficacy vs LAA closure; at this point, it would be inappropriate to generalize these results to the new oral anticoagulants. However, the relevancy of these results is highlighted by warfarin’s current role as the most commonly used agent for stroke prophylaxis in AF.23 It would be similarly inappropriate to generalize the results of this trial to other LAA closure devices: these should be compared directly with either the device or anticoagulation before concluding relative efficacy and safety. Second, patients were likely referred for this trial because of certain clinical characteristics indicating they may not ideal candidates for long-term anticoagulation; thus, the applicability of these data needs to be corroborated by carefully conducted “real-world” registries. Furthermore, while the inclusion and exclusion criteria for this trial are similar to those of the other major stroke prevention trials, one important exception is the exclusion of patients with an LV ejection fraction less than 30%, because unlike anticoagulants, LAA closure would not prevent thromboembolism from the left ventricle.
After 3.8 years of follow-up in patients with nonvalvular AF at elevated risk for stroke, percutaneous LAA closure met criteria for both noninferiority and superiority, compared with warfarin therapy, for preventing the combined outcome of stroke, systemic embolism, and cardiovascular death, as well as superiority for cardiovascular mortality and all-cause mortality.
Corresponding Author: Vivek Y. Reddy, MD, Helmsley Electrophysiology Center, Mount Sinai School of Medicine, One Gustave L. Levy Place, PO Box 1030, New York, NY 10029 (firstname.lastname@example.org).
Author Contributions: Dr Reddy had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Reddy, Buchbinder, Swarup, Gordon, Holmes.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Reddy, Holmes.
Critical revision of the manuscript for important intellectual content: Sievert, Halperin, Doshi, Buchbinder, Neuzil, Huber, Whisenant, Kar, Swarup, Gordon, Holmes.
Statistical analysis: Reddy.
Obtained funding: Holmes.
Administrative, technical, or material support: Reddy, Huber, Gordon, Holmes.
Study supervision: Reddy, Sievert, Doshi, Buchbinder, Neuzil, Whisenant, Kar, Holmes.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Reddy reported having received research grant support and consultant fees from Boston Scientific. Dr Sievert reported having received research grant support and consultant fees from Boston Scientific. Dr Doshi reported having received research grant support and consultant fees from Boston Scientific. Dr Buchbinder reported having received research grant support and consultant fees from Boston Scientific. Dr Neuzil reported having received research grant support from Boston Scientific. Dr Whisenant reported having received research grant support and consultant fees from Boston Scientific. Dr Kar reported having received research grant support and consultant fees from Boston Scientific. Dr Swarup reported having received research grant support and consultant fees from Boston Scientific. Ms Gordon reported being an employee of Boston Scientific. Dr Holmes reported having received research grant support from Boston Scientific. Also, the LAA closure technology has been licensed to Boston Scientific, and both Mayo Clinic and Dr Holmes have contractual rights to receive future royalties from this license. But to date, no royalties have been received. No other disclosures were reported.
Funding/Support: This study was funded by the manufacturer of the device, Atritech (now owned by Boston Scientific), which provided the LAA closure device used in this trial.
Role of the Funder/Sponsor: The sponsor designed the study protocol in collaboration with the US Food and Drug Administration (FDA) and participating investigators. The study was intended to support market approval of the LAA closure device, which is manufactured by the sponsor. The sponsor had the overall responsibility for the conduct of the study, including assurance that the study met the regulatory requirements of the FDA. The sponsor’s general duties consisted of submitting the Investigational Device Exemption application to the FDA, obtaining FDA and institutional review board approvals before shipping the devices, approving the investigators, ensuring proper clinical site monitoring, and ensuring patient informed consent was obtained. The sponsor was responsible for providing quality data that satisfied federal regulations, informing proper authorities of serious unanticipated adverse events and deviations from the protocol, training all participating investigators on the study device and protocol, and monitoring the study for data integrity throughout the duration of the investigation. In addition, the sponsor was responsible for data collection and basic data analysis. The sponsor participated in additional data analysis, data interpretation, and the drafting of the manuscript in conjunction with the principal and other investigators, as well as the decision to submit the manuscript for publication.
The PROTECT AF Trial Investigators:Steering Committee: David R. Holmes, Mayo Clinic College of Medicine, Rochester, Minnesota (P.I.); Maurice Buchbinder, Foundation for Cardiovascular Medicine, La Jolla, California; Larry Chinitz, New York University Medical Center, New York; Pam Douglas, Duke COResearch Core Laboratory, Raleigh, North Carolina; John Gurley, University of Kentucky, Lexington; Vivek Y. Reddy, Mount Sinai School of Medicine, New York; Peter Sick, Krankenhaus der Barmherzigen Bruder, Regensburg, Germany; Zoltan G. Turi, Cooper Hospital, Camden, New Jersey. Data and Safety Monitoring Board: Stephen T. Hustead, DO, Metropolitan Cardiology Consultants, Fridley, Minnesota (Chair); Richard W. Asinger, MD, Hennepin County Medical Center, Minneapolis, Minnesota; Wendy Shear, MD, North Memorial Heart and Vascular Institute, Minneapolis; Chris Pulling, MS, NAMSA, Minneapolis (Statistician). Clinical Events Committee: Brian Lew, MD, Minnesota Heart Clinic, Minneapolis (Chair); D. Anthony Plucinski, MD, Hennepin County Medical Center, Minneapolis; Eve Rogers, MD, Columbus, Ohio. Clinical Investigators (Site [No. of Enrollments]): Europe: Petr Neuzil (Na Homolce Hospital, Prague, Czech Republic ); Horst Sievert (Cardiovascular Center Frankfurt, Sankt Katharinen, Frankfurt, Germany ); Sven Mobius-Winkler (Herzzentrum Leipzig, Leipzig, Germany ); Peter Sick (Krankenhaus der Barmherzigen Bruder, Regensburg, Germany ); United States: Kelly Tucker (Orange County Heart, Orange County, California ); Shephal Doshi (Pacific Heart Institute/St John’s Health Center, Santa Monica, California ); Vijay Swarup, Marwan Bahu (Arizona Arrhythmia Consultants, Scottsdale ); Ramon Quesada (Baptist Hospital of Miami, Miami, Florida ); Kenneth Huber (St Luke’s Hospital, Kansas City, Missouri ); Vivek Reddy (Massachusetts General Hospital, Boston ); Maurice Buchbinder (Foundation for Cardiovascular Research, La Jolla, California ); Brian Whisenant (Intermountain Medical Center, Murray, Utah ); Steven Almany (William Beaumont Hospital, Royal Oak, Michigan ); David R. Holmes (Mayo Clinic, Rochester ); Robert M. Siegel, Ashok Garg (Advanced Cardiac Specialists, Gilbert, Arizona ); Gregory Mishkel (PERC/St John’s Hospital, Springfield, Illinois ); Stephen Ramee (Ochsner Clinic, New Orleans, Louisiana ); Saibal Kar (Cedars-Sinai Medical Center, Los Angeles, California ); Brijeshwar Maini (Moffitt Heart and Vascular Group, Wormleysburg, Pennsylvania ); Ray Matthews, Steven Burstein (Los Angeles Cardiology Associates, Los Angeles ); Rodney Horton (Texas Cardiac Arrhythmia Research, Austin ); Paul Mahoney, John Onufer (Sentara Norfolk General Hospital, Norfolk, Virginia ); Kenneth Baran, Stuart Adler (St Paul Heart Clinic, St Paul, Minnesota ); Kimberly Skelding (Geisinger Medical Center, Danville, Pennsylvania ); John Gurley (University of Kentucky, Lexington ); Miland Shah (Marshfield Clinic, Marshfield, Wisconsin ); Steven J. Yakubov (Riverside Methodist Hospital, Columbus ); Angel Leon (Crawford Long Hospital, Atlanta, Georgia ); Peter C. Block (Emory University School of Medicine, Atlanta ); Peter Fail, Richard Abben (Terrebonne General Medical Center, Houma, Louisiana ); Mark Reisman (Swedish Cardiovascular Research, Seattle, Washington ); Gery Tomassoni (Lexington Cardiology at Central Baptist, Lexington ); Vishwajeth Bhoopalam (Nebraska Heart Institute, Lincoln ); William Anderson (UPMC [Presbyterian University Hospital], Pittsburgh, Pennsylvania ); Robert A. Pickett, Douglas Wolfe (Baptist Medical Center, Jackson, Mississippi ); Reginald Low (UC Davis Medical Center, Sacramento, California ); Ted Feldman, Michael Sallinger (Evanston Northwestern Healthcare, Evanston, Illinois ); James Irwin (Bay Heart Group, Tampa, Florida ); John Lopez, Bradley Knight (University of Chicago Medical Center, Chicago, Illinois ); Scott Lim (University of Virginia School of Medicine, Charlottesville ); Larry Chinitz (New York University Medical Center, New York ); Mehdi Razavi (St Luke’s Episcopal Hospital, Houston, Texas ); David Wilbur, Ferdinand Leya (Loyola University Medical Center, Maywood, Illinois ); Zoltan G. Turi (Cooper Hospital, Camden, New Jersey ); Bryan Raybuck (INOVA Research Center, Falls Church, Virginia ); Ron Waksman, Horst Sievert (Washington Hospital Center, Washington, DC ); Steven Kalbfleisch (Ohio State University, Columbus ); Michael Mooney (Abbott Northwestern Hospital, Minneapolis ); William Gray (Columbia University Medical Center, New York ); Geoffrey Kunz (New Mexico Heart, Albuquerque ); Malcolm Foster (Baptist Heart Institute, Knoxville, Tennessee ); Eric Good (University of Michigan, Ann Arbor ); Murat Tuzcu (Cleveland Clinic Foundation, Cleveland, Ohio ); Fred St. Goar (El Camino Hospital, Mountain View, California ); Richard Josephson (Summa Health System, Akron, Ohio ); W. Carl Jacobs (Piedmont Hospital, Atlanta ); Rajesh Dave (Harrisburg Hospital, Harrisburg, Pennsylvania ); John Young (Lindner Clinical Trial Center, Cincinnati, Ohio ); David Lasorda (Allegheny General Hospital, Pittsburgh ).
Correction: This article was corrected online February 4, 2015.