Left Atrial Appendage Closure for Stroke Prevention in AF
Left Atrial Appendage Closure for Stroke Prevention in AF
Local institutional review board approval was obtained for each dataset. All clinical trials were registered on ClinicalTrials.gov. The specific Watchman device was identical throughout; a self-expanding nitinol framed structure positioned at LAA ostium with diameter ranges from 20 to 33 mm and fixation barbs to prevent embolization. Implant protocols were identical. As previously described, after implantation, patients were treated with warfarin with an international normalized ratio (INR) goal of 2.0 to 3.0 and aspirin (81 mg) for 45 days; at that time, transesophageal echocardiography was performed. If there was complete LAAC or if residual peridevice flow was <5 mm in width, warfarin was discontinued and the patient was treated with clopidogrel 75 mg and aspirin 81 to 325 mg until 6 months following implantation, at which time clopidogrel was discontinued and 325 mg of aspirin was used indefinitely. In the control (warfarin) limb, INR was monitored every 2 weeks to achieve an INR of 2 to 3. All patients had follow-up visits at 45 days, and at 6, 9, and 12 months, and then twice yearly.
All 4 datasets included a primary efficacy composite of all cause stroke (both hemorrhagic and ischemic), systemic embolization, and CV death. Any death of unknown origin was included as CV. Safety endpoints varied slightly between the 2 randomized trials and included bleeding as well as all cause stroke (ischemic and hemorrhagic). To be a candidate for the device, all patients had to be able to take warfarin for 45 days following implantation.
Because of the identical efficacy endpoint definitions in both randomized trials, the data from both trials were pooled for this meta-analysis.
Eligibility criteria included ≥18 years of age with paroxysmal persistent or permanent AF with a CHADS2 risk score ≥1. Exclusion criteria included absolute contraindications to warfarin, LAA thrombus, a patent foramen ovale with an atrial septal aneurysm and right to left shunt, mobile aortic atheroma, or symptomatic carotid disease. The primary composite safety endpoint included excessive bleeding or procedural-related complications (e.g., pericardial effusion requiring intervention, device embolization, or procedural-related stroke). A Bayesian statistical model was used for noninferiority trial design, and pre-specified endpoint analysis of observed event rates after 600 follow-up PY and after each additional 150 PY, up to 1,500.
Selection criteria were modified to include higher risk patients: a CHADS2 score ≥2 or a CHADS2 score ≥1 with more than 1 of the following higher risk characteristics; female ≥75 years of age, baseline ejection fraction ≥30% but <35%, 65 to 74 years of age with either diabetes or coronary artery disease, and ≥65 years of age with heart failure. Exclusion criteria were similar to the PROTECT AF trial, except patients in whom clopidogrel therapy was indicated were excluded because of the potential confounding influence of this drug on efficacy outcome.
There were 3 coprimary endpoints: 1) a composite efficacy endpoint identical to the PROTECT AF trial; 2) a second, "late ischemic efficacy endpoint"— a composite of ischemic stroke or systemic embolization excluding the first 7 days post-randomization, which aimed to isolate the mechanism of LAAC and its potential effect on outcomes without the confounding influence of procedural complications; and 3) a composite of early safety—death, ischemic stroke, systemic embolization, device-/procedural-related events requiring open CV surgery or a major endovascular intervention between randomization and 7 days post-procedure or during the index hospitalization.
Inclusion of new operators was mandated to assess procedural performance. Trial design recognized the utility of the available efficacy data from the PROTECT AF trial, and used Bayesian statistical methodology that allowed "borrowing" some data on PREVAIL-eligible PROTECT AF trial patients at 1,500 PY of follow-up as an informative prior. Incorporation of the informative prior allowed for a smaller confirmatory PREVAIL trial, but results in the fact that the confirmatory trial by itself is not powered to reach robust conclusions. The PROTECT AF trial results were combined with the small sample size of the PREVAIL trial to confirm device safety and collect additional efficacy data on LAAC. The resulting study requirements were such that follow-up with the new PREVAIL trial patients was limited to 6 months minimum in order for the endpoint to occur using the modeled 18-month event rates against pre-specified performance boundaries.
The registries were designed to treat patients with similar baseline characteristics according to the same protocols after the specific trial enrollment had been completed—CAP following PROTECT AF and CAP2 following the PREVAIL trial. Procedural performance and adjunctive medications were identical in each registry and the respective randomized trials except registries did not mandate 1-year neurological assessment required in both randomized trials.
For all data sets, the CHADS2 score was used for patient entry. Because the CHA2DS2-VASc score has largely supplanted CHADS2, risk prediction was reported using both. Assessment of bleeding risk was a clinical site determination. Though not prospectively captured in the Watchman studies, many components of the HAS-BLED score were captured as a part of routine data collection. A conservative bleeding score was determined using the available case report form data and points were assigned per the HAS-BLED score. Abnormal liver function and labile INR were not captured and were consequently assigned a score of zero. An independent Clinical Events Committee adjudicated strokes in all trials and registries. Ischemic stroke was defined as sudden onset of a focal neurological deficit with symptoms and/or signs persisting >24 h, or symptoms <24 h with computed tomography or magnetic resonance imaging evidence of tissue loss without hemorrhage. Although hemorrhagic stroke definitions varied slightly from the PROTECT AF trial to the PREVAIL trial, they included sudden onset of a focal neurological deficit with computed tomography or magnetic resonance imaging evidence of tissue loss with evidence of blood vessel and/or intracranial hemorrhage with symptomatic focal neurological deficit. Subdural hematomas with evidence of parenchymal involvement such as parenchymal extensions or contusion were adjudicated as a hemorrhagic stroke, unless confined exclusively to the subdural space.
This patient level meta-analysis had 3 components: 1) assessment of the device outcomes including all primary efficacy components, all-cause mortality, and major bleeding versus warfarin control in patients randomized in both the PROTECT AF and PREVAIL trials; 2) the primary composite efficacy rates of patients receiving the Watchman device in both randomized controlled trials and registries; and 3) comparison of the device performance in each of the 4 trials.
Both randomized trials were designed to establish noninferiority of a device-based strategy versus warfarin for a composite primary efficacy endpoint. To evaluate benefit-risk ratio, the PROTECT AF and PREVAIL data sets were combined with all available follow-up and analyzed as a traditional patient-level meta-analysis. Fully utilizing the data from both trials while accounting for the fact that they are different studies facilitates more robust exploration of the role of covariates.
This meta-analysis of these 2 studies is appropriate because both studies randomized subjects to the same treatment strategies (Watchman vs. warfarin) and primary efficacy endpoint definitions. The baseline risk profile in the combined the PROTECT AF and PREVAIL trial population included a somewhat higher risk profile for the PREVAIL trial; subgroup analyses by CHADS2 and CHA2DS2-VASc scores were performed to assess differences in outcome by baseline risk profile. Further, results were stratified with Cox proportional hazards modeling, adjusting for potential baseline risk differences. In addition, the device patients from the registry data were combined with all randomized data in the same fashion. Though registries had no control group, the same criteria around device and primary efficacy definitions apply. Given the variable initiation of recruitment, the PROTECT AF trial patients have the longest follow-up, but registry patients provide the largest number of patients.
Analyses were intent-to-treat, censoring data from patients without events at the time of the last known status. A Cox proportional hazards model with confidence intervals (CIs) was used for comparison of event rates. For the analysis including the randomized trials, this model was stratified by study (PROTECT AF or PREVAIL) in order to account for differences in risk profiles. For the analysis including all 4 data sets, studies were treated as "clusters" in order to account for correlation among patients within studies. The Kaplan-Meier method was used for graphical assessment of time-dependent events. Results are presented using frequentist statistics and 2-sided p values nominally significant at p < 0.05 with no adjustments for multiple comparisons.
Methods
Local institutional review board approval was obtained for each dataset. All clinical trials were registered on ClinicalTrials.gov. The specific Watchman device was identical throughout; a self-expanding nitinol framed structure positioned at LAA ostium with diameter ranges from 20 to 33 mm and fixation barbs to prevent embolization. Implant protocols were identical. As previously described, after implantation, patients were treated with warfarin with an international normalized ratio (INR) goal of 2.0 to 3.0 and aspirin (81 mg) for 45 days; at that time, transesophageal echocardiography was performed. If there was complete LAAC or if residual peridevice flow was <5 mm in width, warfarin was discontinued and the patient was treated with clopidogrel 75 mg and aspirin 81 to 325 mg until 6 months following implantation, at which time clopidogrel was discontinued and 325 mg of aspirin was used indefinitely. In the control (warfarin) limb, INR was monitored every 2 weeks to achieve an INR of 2 to 3. All patients had follow-up visits at 45 days, and at 6, 9, and 12 months, and then twice yearly.
All 4 datasets included a primary efficacy composite of all cause stroke (both hemorrhagic and ischemic), systemic embolization, and CV death. Any death of unknown origin was included as CV. Safety endpoints varied slightly between the 2 randomized trials and included bleeding as well as all cause stroke (ischemic and hemorrhagic). To be a candidate for the device, all patients had to be able to take warfarin for 45 days following implantation.
Because of the identical efficacy endpoint definitions in both randomized trials, the data from both trials were pooled for this meta-analysis.
PROTECT AF Study
Eligibility criteria included ≥18 years of age with paroxysmal persistent or permanent AF with a CHADS2 risk score ≥1. Exclusion criteria included absolute contraindications to warfarin, LAA thrombus, a patent foramen ovale with an atrial septal aneurysm and right to left shunt, mobile aortic atheroma, or symptomatic carotid disease. The primary composite safety endpoint included excessive bleeding or procedural-related complications (e.g., pericardial effusion requiring intervention, device embolization, or procedural-related stroke). A Bayesian statistical model was used for noninferiority trial design, and pre-specified endpoint analysis of observed event rates after 600 follow-up PY and after each additional 150 PY, up to 1,500.
PREVAIL Trial
Selection criteria were modified to include higher risk patients: a CHADS2 score ≥2 or a CHADS2 score ≥1 with more than 1 of the following higher risk characteristics; female ≥75 years of age, baseline ejection fraction ≥30% but <35%, 65 to 74 years of age with either diabetes or coronary artery disease, and ≥65 years of age with heart failure. Exclusion criteria were similar to the PROTECT AF trial, except patients in whom clopidogrel therapy was indicated were excluded because of the potential confounding influence of this drug on efficacy outcome.
There were 3 coprimary endpoints: 1) a composite efficacy endpoint identical to the PROTECT AF trial; 2) a second, "late ischemic efficacy endpoint"— a composite of ischemic stroke or systemic embolization excluding the first 7 days post-randomization, which aimed to isolate the mechanism of LAAC and its potential effect on outcomes without the confounding influence of procedural complications; and 3) a composite of early safety—death, ischemic stroke, systemic embolization, device-/procedural-related events requiring open CV surgery or a major endovascular intervention between randomization and 7 days post-procedure or during the index hospitalization.
Inclusion of new operators was mandated to assess procedural performance. Trial design recognized the utility of the available efficacy data from the PROTECT AF trial, and used Bayesian statistical methodology that allowed "borrowing" some data on PREVAIL-eligible PROTECT AF trial patients at 1,500 PY of follow-up as an informative prior. Incorporation of the informative prior allowed for a smaller confirmatory PREVAIL trial, but results in the fact that the confirmatory trial by itself is not powered to reach robust conclusions. The PROTECT AF trial results were combined with the small sample size of the PREVAIL trial to confirm device safety and collect additional efficacy data on LAAC. The resulting study requirements were such that follow-up with the new PREVAIL trial patients was limited to 6 months minimum in order for the endpoint to occur using the modeled 18-month event rates against pre-specified performance boundaries.
Registries
The registries were designed to treat patients with similar baseline characteristics according to the same protocols after the specific trial enrollment had been completed—CAP following PROTECT AF and CAP2 following the PREVAIL trial. Procedural performance and adjunctive medications were identical in each registry and the respective randomized trials except registries did not mandate 1-year neurological assessment required in both randomized trials.
Definitions
For all data sets, the CHADS2 score was used for patient entry. Because the CHA2DS2-VASc score has largely supplanted CHADS2, risk prediction was reported using both. Assessment of bleeding risk was a clinical site determination. Though not prospectively captured in the Watchman studies, many components of the HAS-BLED score were captured as a part of routine data collection. A conservative bleeding score was determined using the available case report form data and points were assigned per the HAS-BLED score. Abnormal liver function and labile INR were not captured and were consequently assigned a score of zero. An independent Clinical Events Committee adjudicated strokes in all trials and registries. Ischemic stroke was defined as sudden onset of a focal neurological deficit with symptoms and/or signs persisting >24 h, or symptoms <24 h with computed tomography or magnetic resonance imaging evidence of tissue loss without hemorrhage. Although hemorrhagic stroke definitions varied slightly from the PROTECT AF trial to the PREVAIL trial, they included sudden onset of a focal neurological deficit with computed tomography or magnetic resonance imaging evidence of tissue loss with evidence of blood vessel and/or intracranial hemorrhage with symptomatic focal neurological deficit. Subdural hematomas with evidence of parenchymal involvement such as parenchymal extensions or contusion were adjudicated as a hemorrhagic stroke, unless confined exclusively to the subdural space.
Statistical Analysis for Meta-analysis
This patient level meta-analysis had 3 components: 1) assessment of the device outcomes including all primary efficacy components, all-cause mortality, and major bleeding versus warfarin control in patients randomized in both the PROTECT AF and PREVAIL trials; 2) the primary composite efficacy rates of patients receiving the Watchman device in both randomized controlled trials and registries; and 3) comparison of the device performance in each of the 4 trials.
Both randomized trials were designed to establish noninferiority of a device-based strategy versus warfarin for a composite primary efficacy endpoint. To evaluate benefit-risk ratio, the PROTECT AF and PREVAIL data sets were combined with all available follow-up and analyzed as a traditional patient-level meta-analysis. Fully utilizing the data from both trials while accounting for the fact that they are different studies facilitates more robust exploration of the role of covariates.
This meta-analysis of these 2 studies is appropriate because both studies randomized subjects to the same treatment strategies (Watchman vs. warfarin) and primary efficacy endpoint definitions. The baseline risk profile in the combined the PROTECT AF and PREVAIL trial population included a somewhat higher risk profile for the PREVAIL trial; subgroup analyses by CHADS2 and CHA2DS2-VASc scores were performed to assess differences in outcome by baseline risk profile. Further, results were stratified with Cox proportional hazards modeling, adjusting for potential baseline risk differences. In addition, the device patients from the registry data were combined with all randomized data in the same fashion. Though registries had no control group, the same criteria around device and primary efficacy definitions apply. Given the variable initiation of recruitment, the PROTECT AF trial patients have the longest follow-up, but registry patients provide the largest number of patients.
Analyses were intent-to-treat, censoring data from patients without events at the time of the last known status. A Cox proportional hazards model with confidence intervals (CIs) was used for comparison of event rates. For the analysis including the randomized trials, this model was stratified by study (PROTECT AF or PREVAIL) in order to account for differences in risk profiles. For the analysis including all 4 data sets, studies were treated as "clusters" in order to account for correlation among patients within studies. The Kaplan-Meier method was used for graphical assessment of time-dependent events. Results are presented using frequentist statistics and 2-sided p values nominally significant at p < 0.05 with no adjustments for multiple comparisons.
