Oral Anticoagulants vs Aspirin in Nonvalvular Atrial Fibrillation: An Individual Patient Meta-analysis

Context Patients with nonvalvular atrial fibrillation (AF) have an increased risk of stroke and other vascular events.

Objective To compare the risk of vascular and bleeding events in patients with nonvalvular AF treated with vitamin K –inhibiting oral anticoagulants or acetylsalicylic acid (aspirin).

Design Pooled analysis of patient-level data from 6 published, randomized clinical trials.

Patients A total of 4052 patients with AF randomly assigned to receive therapeutic doses of oral anticoagulant or aspirin with or without low-dose oral anticoagulants.

Main Outcome Measures Ischemic and hemorrhagic stroke, other cardiovascular events, all-cause death, and major bleeding events. Person-year incidence rates were calculated to provide crude comparisons. Relative efficacy was assessed using proportional hazards modeling stratified by study. The variation of the oral anticoagulant's relative effect by pertinent patient factors was explored with interaction terms. All analyses were conducted using the intention-to-treat principle.

Results Patients receiving oral anticoagulant and aspirin were balanced for important prognostic factors. There was no significant heterogeneity between trials in the relative efficacy of oral anticoagulant vs aspirin for any outcome. Patients receiving oral anticoagulant were significantly less likely to experience any stroke (2.4 vs 4.5 events per 100 patient-years; hazard ratio [HR], 0.55; 95% confidence interval [CI], 0.43-0.71), ischemic stroke (HR, 0.48; 95% CI, 0.37-0.63), or cardiovascular events (HR, 0.71; 95% CI, 0.59-0.85) but were more likely to experience major bleeding (2.2 vs 1.3 events per 100 patient-years; HR, 1.71; 95% CI, 1.21-2.41). The reduction in ischemic stroke risk was similar in patients with paroxysmal AF (1.5 vs 4.7 events per 100 patient-years; HR, 0.32; 95% CI, 0.16-0.61; P<.001). Treating 1000 patients with AF for 1 year with oral anticoagulant rather than aspirin would prevent 23 ischemic strokes while causing 9 additional major bleeds. Overall all-cause survival did not differ but appeared to improve for oral anticoagulant patients 3 years after therapy was started.

Conclusions Compared with aspirin, oral anticoagulant significantly decreases the risk of all strokes, ischemic strokes, and cardiovascular events for patients with nonvalvular chronic or paroxysmal AF but modestly increases the absolute risk of major bleeding. The balance of benefits and risks varies by patient subgroup.

Nonvalvular atrial fibrillation (AF) more than quadruples the risk of stroke.¹ Stroke prophylaxis in patients with atrial fibrillation has been examined extensively in randomized trials,^2-16 analyses of individual patient-level data,^17-19 traditional meta-analyses,^20-31 and review articles.^32-42 The clinical trials, and their quantitative reviews, show that treatment with vitamin K–dependent oral anticoagulants significantly decrease the risk of stroke in AF patients by more than 50%. Acetylsalicylic acid (aspirin) also appears to decrease stroke risk but to a lesser extent.²⁰ Because oral anticoagulant use is more troublesome⁴³ and has a greater likelihood of complications,⁴⁴ it is important to determine the relative efficacy and safety of oral anticoagulant vs aspirin overall and in relevant patient subgroups.

Six published clinical trials have randomized AF patients to oral anticoagulant or a regimen containing aspirin.^2-7,9 Three of these studies found that oral anticoagulant was significantly more efficacious than aspirin at reducing thromboembolic outcomes.^2,5,9 However, the conclusions from quantitative reviews of randomized trials comparing oral anticoagulant and aspirin-containing regimens have varied extensively from oral anticoagulants being "substantially more efficacious than aspirin"²⁰ to finding "moderate evidence for fewer strokes among patients on oral anticoagulants than on aspirin"³¹ to "considerable uncertainty about the value of long-term anticoagulation compared to antiplatelet treatment."⁴⁵ This uncertainty of the relative benefit of oral anticoagulant vs aspirin in AF might explain the relatively low use of oral anticoagulant in usual practice.⁴⁶

All prior meta-analyses have been limited to using summary data available in published trial reports. We analyzed the pooled individual patient data from all published randomized trials comparing oral anticoagulant and aspirin for AF. These data are distinctive by permitting true time-to-event analyses, exploration of subgroup effects, and comparison of the efficacy of oral anticoagulant vs aspirin on a number of important cardiovascular outcomes in addition to ischemic stroke.

This study included patient-level data from every published clinical trial in which patients with AF were randomly assigned to full-dose oral anticoagulant or aspirin. These trials included the Atrial Fibrillation, Aspirin, Anticoagulation (AFASAK) Studies 1² and 2,³ Primary Prevention of Arterial Thromboembolism in patients with Nonrheumatic Atrial Fibrillation in Primary Care (PATAF),⁴ the European Atrial Fibrillation Trial (EAFT),⁵ and the Stroke Prevention in Atrial Fibrillation (SPAF) Studies 1,⁶ 2,⁷ and 3.⁹ Since SPAF 1 patients randomized to receive oral anticoagulant or aspirin all continued into SPAF 2, SPAF 1 data are combined with SPAF 2 as in the original publication.⁷ To our knowledge, there are no other published studies in which AF patients were randomized to receive a full-dose of oral anticoagulant or aspirin.

All patients were adults with nonvalvular AF. The exclusion criteria varied only slightly between the studies. In general, patients were excluded if they had clinical indications for, or contraindications to, oral anticoagulation or aspirin. These contraindications included pregnancy, alcoholism, renal or hepatic failure, thrombocytopenia, or bleeding disorders. With the exception of the AFASAK studies,^2,3 patients were also excluded if they had a recent acute coronary event or cardiac revascularization. All studies also excluded patients with thyrotoxicosis except for AFASAK 1, which included only 28 such patients.

Patients were randomly assigned to receive a full-dose of oral anticoagulant or aspirin with or without low-dose oral anticoagulant. Patients in the oral anticoagulant group were treated primarily with coumarin derivatives including warfarin sodium and 4-hydroxycoumarin. The target international normalized ratio (INR) varied slightly in each trial (Table 1). The daily aspirin dose ranged between 75 and 325 mg (Table 1).

In our analyses, we combined patients treated with aspirin alone with patients treated with aspirin plus low-dose warfarin because there were no significant differences in outcomes between such patients.³ We also reasoned that, if low-dose warfarin has a previously unmeasured effect on cardiovascular outcomes, including such patients would likely decrease the differences between the oral anticoagulant and aspirin groups for cardiovascular outcomes. The AFASAK 2 and SPAF 3 trials were the only ones in which patients receiving aspirin also received low-dose warfarin in dosages of 1.25 mg/d and a median of 2.0 mg/d, respectively, and with negligible prolongation of the INR. To ensure that adding low-dose warfarin to aspirin did not bias our study, we repeated the analyses after excluding patients taking aspirin plus low-dose warfarin.

Patient clinical features were collected by research coordinators and physicians before the initiation of therapy (see eTable 1). These included previous stroke or transient ischemic attack, hypertension, congestive heart failure, diabetes, and coronary artery disease. All studies classified patients as hypertensive if they were taking medications given to lower blood pressure. In the AFASAK studies, patients were considered to have congestive heart failure if it was graded as moderate or severe. We used the Atrial Fibrillation Investigators risk stratification to classify patient stroke risk.¹⁸ Using these data, we labeled the patients with hypertension, diabetes, or prior cerebral ischemia as high risk. Patients without these risk factors were considered low risk if they were younger than 65 years. All others were considered moderate risk.

We compared the relative efficacy of oral anticoagulant and aspirin for 6 outcomes: ischemic or hemorrhagic stroke; ischemic stroke alone; hemorrhagic stroke alone (including subarachnoid and subdural hemorrhage); aggregate cardiovascular events (including ischemic stroke, myocardial infarction, systemic embolism, or cardiovascular death); major bleeding (including intracranial and systemic bleeding); and all-cause death. Cardiovascular deaths included those due to stroke, myocardial infarction, congestive heart failure, pulmonary embolism, or systemic embolism. Ischemic stroke and major bleeds were classified as lethal if death occurred within 30 days of the event. (For outcome criteria for each trial, see eTable 2.)

Patients were followed up for a mean of 1.9 years (Table 1). Patients were routinely seen every 3 to 6 months by study investigators or when an outcome event was suspected. With the exception of patients in AFASAK 1, a central events committee that was blinded to therapeutic group reviewed and confirmed all events. Of all strokes, 97% were assessed by neuroimaging, almost exclusively by computed tomography.

Crude event rates were calculated as events per 100 person-years of observation and compared using the normal approximation of the Poisson distribution as described by Rothman and Greenland.⁴⁷ Kaplan-Meier curves were generated to compare time to each outcome, and the log-rank statistic determined the significance of differences between the curves. We used proportional hazards regression modeling to estimate the relative effect of oral anticoagulant on all 6 outcomes. Each model was stratified by study, thereby allowing different hazard functions to be estimated for each study and then a pooled hazard ratio (HR) across all studies was estimated. Such modeling helps adjust for factors that might be unique to a particular study.^48,49 We tested the proportional hazards assumption for all Cox models using an interaction term between the predictor variable and time.⁴⁹ If the P value for this term was less than .05, we concluded that the assumption of proportional hazards was not met over time.

We assessed heterogeneity between studies for each outcome by visually comparing HRs for each study with the overall estimate and by calculating the DerSimonian and Laird Q statistic for heterogeneity.⁵⁰ To measure interactions between treatment group and baseline factors, appropriate interaction terms were included in the model. All analyses were conducted using SAS 8.1 software (Cary, NC).

Table 1 describes the patients according to the study in which the patients were enrolled. Atrial fibrillation was nonparoxysmal in 83% and was present for more than a year in 67% of patients. Sixty-five percent of patients were classified as having high, 27% as having moderate, and 8% as having low risk of stroke. Reflecting different patient selection criteria between studies, the prevalence of patient characteristics varied between trials (Table 1). After pooling patients from all 6 trials, patients receiving oral anticoagulant and aspirin were well matched (Table 2).

Compared with aspirin, oral anticoagulant significantly decreased the rate of all stroke, ischemic stroke, and cardiovascular events (Table 3). The decrease in the rate of all stroke was due to a large decrease in ischemic stroke with only a small absolute increase in hemorrhagic stroke. Of all ischemic strokes, 12.5% were lethal. Oral anticoagulants also decreased the lethal ischemic stroke rate (0.5 vs 0.2 events per 100 patient-years; P = .01). The significant decrease in cardiovascular events seen in the oral anticoagulant group was due primarily to decreased rates of ischemic stroke and myocardial infarction.

The use of oral anticoagulant significantly increased the rate of major bleeding (Table 3), with 15.3% of all major bleeding episodes being lethal. Hemorrhagic stroke accounted for 21.9% of all major bleeding and 52.4% of fatal bleeding events. Patients receiving oral anticoagulant had a nonsignificantly increased rate of lethal hemorrhages. Overall mortality did not differ between patient groups.

Life-table analyses (Figure 1) confirmed the crude-rate comparisons and demonstrated, with the exception of all-cause death, a consistent relative beneficial effect of oral anticoagulant over time. All-cause death was the only outcome for which the assumption of proportionality was not satisfied. This is reflected in the survival plots for death showing a difference between treatment groups starting 2½ years into observation (Figure 1).

The outcomes for each trial are presented in Figure 2. The relative effect of oral anticoagulant for each outcome was similar across studies with the exception of major bleeds. For major bleeding, EAFT and SPAF 3 provided particularly high and low HRs, respectively. However, the DerSimonian and Laird Q statistic for heterogeneity⁵⁰ in treatment effect was not significant for any outcome, including major bleeding.

To determine whether our conclusions changed without patients assigned to receive aspirin plus low-dose warfarin, we repeated the analysis after excluding such patients. This analysis excluded SPAF 3 and the combined aspirin-low-dose warfarin group of AFASAK 2, resulting in 2837 patients. We noted similar results for all stroke, 2.4 events among those taking oral anticoagulant vs 3.8 taking aspirin per 100 patient-years (HR, 0.65; 95% CI, 0.49-0.86; P = .003); ischemic stroke, 2.0 vs 3.7 events per 100 patient years (HR, 0.56; 95% CI, 0.41-0.76; P<.001), cardiovascular events, 5.3 vs 6.9 events per 100 patient-years (HR, 0.76; 95% CI, 0.62-0.93; P = .02); and death, 4.7 vs 5.1 (HR, 0.92; 95% CI, 0.74-1.15; P = .41). There was a slight increase of the difference between the 2 groups in hemorrhagic stroke, 0.5 vs 0.2 events per 100 patient-years (HR, 2.26; 95% CI, 0.93-5.50; P = .06) and major bleeding, 2.2 vs 1.2 events per 100 patient-years (HR, 1.93; 95% CI, 1.30-2.88; P = .001). The HRs excluding those who received combined aspirin-low-dose warfarin are similar to HRs attained with the entire cohort (Table 3). When the patients taking aspirin and low-dose warfarin were excluded, cardiovascular outcome rates did not change extensively in the oral anticoagulant group but were noticeably lower in the aspirin group. This is likely due to the exclusion of high-risk patients enrolled in the SPAF 3 study.

We explored the interaction of treatment effect with relevant patient features on ischemic stroke and major bleeding (Figure 3). In almost all patient subgroups, oral anticoagulant was significantly more efficacious than aspirin for reducing the risk of ischemic stroke. Although we noted some variability in the relative benefit of oral anticoagulant between patient subgroups, most of these interactions were not significant with P>.10. There were 2 possible exceptions. The relative benefit of oral anticoagulant vs aspirin in ischemic stroke prevention appeared greater for patients younger than 75 years vs those who were 75 years or older (P for interaction = .08) and for women vs men (P for interaction = .04). The increased risk of major bleeding for patients taking oral anticoagulants appeared consistent in all patient subgroups and none of the interaction terms had a P<.10.

The absolute rate reduction in ischemic stroke by oral anticoagulant vs aspirin varied with stroke risk (Figure 3). The absolute risk reduction in ischemic stroke achieved with oral anticoagulant was greater in patients at highest baseline risk of stroke. Patients with a previous history of stroke or transient ischemic attack had an absolute risk reduction of 6.0% per year (number needed to treat [NNT], 17) while those patients without previous cerebrovascular disease had an absolute risk reduction of 1.2% per year (NNT, 83). Patients who were classified as being at high risk of ischemic stroke by the AF investigators' criteria had an absolute risk reduction of 3.3% per year (NNT, 30), whereas those at low risk of ischemic stroke only had an absolute risk reduction of 0.4% per year (NNT not significant). In contrast to ischemic stroke, the absolute risk increase of major bleeding with oral anticoagulant vs aspirin was more consistent between patient subgroups.

Compared with aspirin, oral anticoagulant significantly decreased the risk of all strokes, ischemic strokes, and cardiovascular events in patients with nonvalvular paroxysmal or chronic AF. The absolute risk reductions of these events were not substantially offset by small but statistically significant increases in major hemorrhage. In these patients, the overall risk of stroke or cardiovascular events was substantially higher than that of major bleeding.

These results support the use of oral anticoagulant as first-line preventive therapy for high-risk patients with nonvalvular AF. Among trial participants, the absolute rate reduction for all stroke (2.1 events per 100 patient-years), ischemic stroke (2.3 events per 100-patient-years), and cerebrovascular events (2.2 events per 100 patient-years) was considerable. Since the absolute rate increase of major bleeding with oral anticoagulant is 0.9 events per 100 patient-years, treating 1000 AF patients for 1 year with oral anticoagulant rather than aspirin would prevent 23 ischemic strokes while causing 9 additional major bleeding episodes.

We believe that our results both highlight and clarify the trade-off between potential harms and benefits of oral anticoagulant prophylaxis for AF patients. Compared with aspirin, oral anticoagulant decreased the risk of ischemic stroke but increased the risk of major bleeding. We also found that the overall absolute risk of ischemic stroke was approximately twice that for major bleeding. Several studies show that patients with AF perceive stroke as a much more serious outcome than bleeding. Hing et al⁵¹ also found that patients had a strong aversion to strokes and would take warfarin if it reduced the absolute risk of stroke by 1% annually. The AF patients that Gage et al⁵² interviewed gave a median utility value of 0.03 for moderate strokes and 0.0 for major stroke, with death having a utility value of 0.⁵² In a study by Solomon et al,⁵³ patients equated the utility of severe stroke with that of death. Devereaux et al⁵⁴ found that the average patient would take warfarin if it reduced the absolute risk of stroke by at least 0.9 events per year but would only stop the drug if it increased the risk of bleeding by 8.7 events per year. These findings show that patients fear the consequences of having a stroke more than they fear the consequences of being prone to severe bleeding, which supports the use of oral anticoagulant treatment for such patients.

However, our results also highlight the importance of determining baseline risk to identify AF patients most likely to benefit from oral anticoagulant. There were some notable variations in the absolute difference of stroke rates (Figure 3). For patients at higher risk of ischemic stroke, such as the elderly or those with previous cerebrovascular disease, the absolute risk reduction in ischemic stroke with oral anticoagulant supercedes the associated bleeding risk. In contrast, patients at low risk of ischemic stroke have only a small absolute risk reduction in ischemic stroke. Therefore, proper risk stratification is essential to identify patients who have the best chance of benefiting from oral anticoagulant. This is especially true for patients without prior cerebrovascular disease for whom the absolute benefit of oral anticoagulant is less (Figure 3). Several risk stratification schemes have been published^18,43,55,56 and further research is required to better refine our estimates of stroke risk in AF patients.

Three meta-analyses have compared oral anticoagulant with antiplatelet therapy for patients with AF.^20,31,45 These analyses have summarized reported aggregate rates of trial outcome events. Our analysis used individual patient data, which allows a more complete and accurate comparison of outcome risk between oral anticoagulant and aspirin for patients with AF. Individual patient data permitted full survival analyses of a variety of relevant outcome events, not just ischemic stroke. We were able to provide absolute as well as relative comparisons of person-year rates of outcomes. Finally, analysis of pooled individual patient data allowed the exploration of treatment effects in relevant patient subgroups, such as those with paroxysmal AF. Because we pooled individual patient data, we had the opportunity to study 672 patients with paroxysmal AF. It is noteworthy that these patients had a similar reduction in stroke risk with oral anticoagulant (Figure 3).

Our results differ most from the meta-analysis by Taylor et al,³⁰ which concluded that "the evidence for current clinical practice in long-term anticoagulation for patients with nonrheumatic atrial fibrillation is not strong." In addition to the methodological differences between the studies, other factors explain why our findings were considerably different. Taylor et al excluded the EAFT study⁵ and the high-risk component of the SPAF 3 study,⁹ both of which showed considerable advantage of oral anticoagulant over aspirin for all strokes, ischemic strokes, and cardiovascular events (Figure 2). The EAFT study was excluded from the analysis by Taylor et al because data were not presented to allow direct comparison of patients randomized to oral anticoagulant and aspirin.⁵⁷ Our access to individual patient data avoided this problem. The SPAF 3 study was excluded because patients taking low-dose warfarin were included with those taking aspirin. We believe that it was preferable to include these patients in our analysis because including patients taking low-dose warfarin in an analysis with those taking aspirin patients would, if anything, decrease differences in cardiovascular events. Excluding these patients from the analysis did not alter our results. The only study that randomized patients to receive aspirin or low-intensity oral anticoagulant with aspirin found no significant differences in outcomes when low-dose warfarin was added to aspirin for patients with AF.³ Finally, their meta-analysis dichotomized outcomes into exclusive fatal and nonfatal events. This greatly decreases their statistical power to identify differences between treatment groups.

A potential limitation of our analyses stems from their uncertain applicability to AF patients outside of clinical trials. When compared with other patients, those enrolled in clinical trials tend to be healthier and more compliant. In addition, study investigators might more effectively monitor oral anticoagulant therapy than usual clinical caregivers. As a consequence, both bleeding and stroke rates might be higher among patients anticoagulated in general clinical practice. However, a recently published systematic review of anticoagulated AF patients in actual clinical practice found stroke and bleeding rates that were very similar to those found in randomized trials,⁵⁸ supporting the external validity of our conclusions.

In summary, oral anticoagulant is more effective than aspirin in decreasing the risk of stroke and other cardiovascular events in patients with nonvalvular AF. Although oral anticoagulant also increases the risk of major bleeding, the frequency of strokes and the gravity of their consequences exceed that of therapy-associated bleeding events in most patients with AF. As a result, oral anticoagulant should be the preferred prophylactic treatment for patients with AF at a significant risk of thromboembolism.