Meta-analysis: Antithrombotic Therapy to Prevent Stroke in Patients Who Have Nonvalvular Atrial Fibrillation

Context Thirteen new randomized, controlled trials are available since a 1999 meta-analysis of antithrombotic agents for stroke prevention in patients with atrial fibrillation. Contribution This updated meta-analysis shows that, compared with placebo, adjusted-dose warfarin reduces stroke risk by 64% (6 trials) and antiplatelet agents reduce stroke risk by 22% (8 trials). Meta-analysis of 12 trials shows that adjusted-dose warfarin is more effective than antiplatelet therapy, but it doubles the risk for major extracranial and intracranial hemorrhage. However, absolute rates of these adverse events were only 0.2% per year. Implication Additional trials are unlikely to substantially change current estimates of the effectiveness of vitamin K antagonists and antiplatelet agents in stroke prevention in patients with atrial fibrillation. The Editors Nonvalvular atrial fibrillation is an important cause of disabling stroke whose incidence can be reduced by using antithrombotic prophylaxis. Our meta-analysis of the initial 16 randomized clinical trials that tested antithrombotic therapies for stroke prevention included approximately 10000 participants (1). Since then, 13 randomized trials that included 18140 additional patients with atrial fibrillation have been reported (Table 1 [231]). Results of single, relatively small trials are sometimes difficult to interpret and often conflict, and meta-analysis is useful to assess the totality of trial evidence. We present an updated meta-analysis of all currently available randomized trials that extends observations about the efficacy and safety of antithrombotic therapies for preventing stroke in patients who have atrial fibrillation. Table 1. Randomized Trials of Antithrombotic Therapy for Patients with Nonvalvular Atrial Fibrillation* Methods Search and Selection Process We identified unconfounded randomized trials that tested long-term (12 weeks) use of antithrombotic agents in patients who had nonvalvular atrial fibrillation. We did a computerized search of the OVID and MEDLINE databases (from 1966 to March 2007, unrestricted by language) and of the Cochrane Stroke Group Trials Register and queried investigators working in the field (1). Trials that included patients who have prosthetic cardiac valves or mitral stenosis were not considered; double-blind and open-label trial designs were eligible. Two physician-reviewers independently extracted data from published sources and determined whether the trials met the inclusion criteria. Disagreements were resolved by joint review and consensus. We included 29 of 41 randomized trials that tested antithrombotic therapies in patients who had atrial fibrillation (Table 1), including 2 trials that reported results for subgroups of patients with atrial fibrillation from a larger number of patients without atrial fibrillation (13, 18). We identified 4 randomized trials that are ongoing (3234) or have not been published (Appendix Table 1) and excluded 8 randomized trials: 5 in which the treatment duration averaged 3 to 8 weeks (3539), 1 in which the results for the subset of patients who had atrial fibrillation were not reported separately (40), 1 that included patients who had mitral stenosis (41), and 1 because of potential confounding in which 51% of randomly assigned participants were excluded from the reported analysis because of ill-defined nonadherence (42). The 5 short-term trials were considered only in the safety analyses (3539). Appendix Table 1. Unpublished or Ongoing Randomized Trials of Antithrombotic Therapies for Nonvalvular Atrial Fibrillation* Data Extraction Two physician-reviewers independently extracted data from published sources regarding methodological features, the number of treated patients, total follow-up exposure, and the occurrence of the following 5 outcomes according to the intention-to-treat paradigm: all stroke (ischemic and hemorrhagic), ischemic stroke, intracranial hemorrhage, all-cause mortality, and major extracranial hemorrhage. Disagreements were resolved by joint review and consensus. The criteria for each outcome were those used by the individual trial. The fraction of participants with strokes who had neuroimaging or autopsy that reliably distinguished between ischemic or hemorrhagic stroke varied and was not consistently reported. Consequently, all stroke (ischemic and hemorrhagic) was chosen as the primary outcome. Undefined strokes in patients who did not have neuroimaging were considered ischemic. Intracranial hemorrhages included subdural hematomas and were considered with all strokes; transient ischemic attacks (TIAs) were not considered. Data Synthesis Two physician-reviewers abstracted key methodological features for each trial (Appendix Table 2). Intention-to-treat results were used for the main analyses. Primary prevention refers to patients without previous stroke or TIA. Secondary prevention refers to patients with previous stroke or TIA. Three trials (11, 14, 23) that combined aspirin with very low, fixed doses of warfarin that did not substantially prolong international normalized ratios (INRs) were considered with those that tested aspirin alone because of the negligible effect of warfarin at these dosages (4345). Appendix Table 2. Key Methodological Features of the 29 Included Randomized Trials* Meta-analyses of the trial results are presented as relative risk reductions and absolute risk reductions for treatment groups compared with control groups. To estimate relative risk reduction, the combined odds ratio was computed by assuming a random-effects model (DerSimonian and Laird method) (46), and the estimate was then subtracted from 1. The absolute risk reduction is a weighted estimate of the difference in annualized event rates (46). Before estimating risk reduction, we tested the assumption of the statistical homogeneity of the treatment effect (across trials and within a specific scenario) by using the QL statistic for the relative odds scale or the QW statistic with unequal weights for the absolute risk scale (46). Lack of homogeneity across trials precluded estimation of the overall treatment effect as noted, and the exact P value was reported for all analyses in which the P value was less than 0.20. We computed estimates of relative risk reduction in individual trials by subtracting the estimated odds ratio from 1. A P value less than 0.05 was considered statistically significant; all tests and CIs were 2-sided. Calculations were done by using MedCalc for Windows, version 9.2.0.2 (MedCalc Software, Mariakerke, Belgium), and SPSS software (SPSS, Chicago, Illinois). Role of the Funding Source This study was not funded. Results Twenty-nine published randomized trials included 28044 patients with nonvalvular atrial fibrillation (Table 1). Methodological features varied; larger trials were more thoroughly reported and were of higher quality (Appendix Table 2). Nine trials were double-blind designs to compare antiplatelet therapy (2, 5, 8, 13, 18, 20) and anticoagulation therapy (6, 7, 24) with control or with each other. Only outcomes during assigned treatment were published for 4 relatively small trials (13, 26, 30, 31). Total follow-up exposure was approximately 42450 patient-years (mean follow-up, 1.5 years per patient). The average age of the patients was 71 years, and 35% were women. Most trials were done in Europe (16 trials, 7390 patients) (2, 8, 9, 1218, 20, 21, 23, 25, 27, 31) or North America (7 trials, 8349 patients) (47, 10, 11, 24), 2 were done in Japan (986 patients) (19, 26), 1 was done in China (704 patients) (30), and 3 were intercontinental (10615 patients) (22, 28, 29). Eight studies enrolled more than 1000 patients; the average number of patients was 423 (range, 45 to 916 patients) in the remaining studies. Most trials studied oral vitamin K inhibitors or aspirin in varying dosages and intensities, but other anticoagulants (low-molecular-weight heparin, ximelagatran, and dabigatran) and other antiplatelet agents (clopidogrel, dipyridamole, indobufen, and triflusal) were also tested. There were 3003 participants assigned to placebo or control groups in 12 trials. The average stroke rate for these untreated participants was 13% per year in trials that were restricted to those who had previous stroke or TIA (secondary prevention trials) and 4.1% per year for those in primary prevention trials. Adjusted-Dose Warfarin Compared with Placebo or No Treatment To our knowledge, no new trial data have been added since our 1999 meta-analysis (1), and abridged results are reported here. In 6 randomized trials, 2900 participants (mean age at study entry, 69 years; 29% were women; and 20% had previous stroke or TIA) were included who had had 186 strokes during a mean follow-up of 1.6 years per participant (Table 2 [2, 48]). The mean achieved INR ranged from 2.0 to 2.6 among patients who were assigned warfarin in the 5 primary prevention trials (2 were double-blinded) and was 2.9 in the only secondary prevention trial. The average stroke rate was 4.5% per year for primary prevention and 12% per year for secondary prevention among patients assigned to the placebo or control groups. Table 2. Adjusted-Dose Warfarin Compared with Placebo or No Treatment* According to meta-analysis, adjusted-dose warfarin was associated with a 64% (95% CI, 49% to 74%) reduction in stroke (Table 2 and Figure, top). When we considered trials that stratified stroke severity, 60% of strokes were disabling and reductions in disabling (60%) and nondisabling (60%) strokes with anticoagulation therapy were similar. The absolute risk reduction in all strokes was 2.7% per year (number needed to treat [NNTB] for 1 year to prevent 1 stroke was 37) for primary prevention and 8.4% per year (NNTB, 12) for secondary prevention. When only ischemic strokes were considered, adjusted-dose warfarin was associated with a 67% (CI, 54% to 77%) relative risk reduction. Figure. Relative effects of antithrombotic t