Mark Crowther, Rm L301, St Joseph’s Hospital, 50 Charlton Ave East, Hamilton, L8N 4A6, ON, Canada. Tel.: +1 905 521 6024; fax: +1 905 521 6090. E-mail: email@example.com
Summary. Background: Warfarin and aspirin (acetylsalicylic acid [ASA]) are the most commonly used anticoagulant and antiplatelet drugs in the treatment of cardiovascular disease.Objectives: To provide a pooled estimate of the bleeding risk from randomized controlled trials (RCTs) comparing warfarin and ASA at the dose ranges recommended in evidence-based guidelines.Patients/Methods: Ovid MEDLINE, Embase and the Cochrane Library, up to September 2011, were searched for RCTs comparing bleeding rates in adult patients randomized to warfarin, target International Normalized Ratio (INR) 2.0–3.5, and ASA, 50–650 mg daily, with at least 3 months of follow-up. Pooled odds ratios (ORs) and associated 95% confidence intervals (CIs) were calculated with the inverse variance method and the random effects model.Results: Four thousand four hundred and forty-two abstracts were screened, resulting in eight included studies for final analysis. A pooled estimate derived from the 2904 patients enrolled indicated a trend towards an increase in major bleeding risk in those randomized to warfarin (OR 1.27; 95% CI 0.83–1.94). The pooled OR for intracranial hemorrhage in patients treated with warfarin vs. ASA was 1.64 (95% CI 0.71–3.78), and that for extracranial major bleeding was 1.03 (95% CI 0.61–1.75). Minor bleeding, from a 1748-patient sample, was more common in warfarin patients (OR 1.50; 95% CI 1.13–2.00).Conclusions: This meta-analysis failed to find a statistically significant difference in major bleeding between warfarin, target INR 2.0–3.5, and ASA, 50–650 mg daily. The trend towards increased bleeding with warfarin appears to be explained by an excess of intracranial bleeding in warfarin patients.
Anticoagulant and antiplatelet drugs are highly effective for the prevention and treatment of thrombotic cardiovascular diseases. Warfarin and aspirin (acetylsalicylic acid [ASA]) are the most commonly used anticoagulant and antiplatelet agents for long-term prophylaxis in patients with atrial fibrillation, myocardial infarction, peripheral artery disease, cerebrovascular disease, heart failure, and heart valve replacement. Although they are highly effective, both ASA and warfarin can result in bleeding. Data on the risk of bleeding are surprisingly limited, perhaps because the reported rates vary as a result of study design and population, definition, site of bleeding, and drug dosage. The annual incidence of major bleeding in trials and cohort studies has been reported to be between 1.1% and 2.3% [1–5] in patients treated with warfarin to achieve an International Normalized Ratio (INR) of 2.0–3.0 and between 1.1% and 1.5% in ASA-treated patients [2,3,6,7]. These rates are derived from carefully selected patients; rates in ‘unselected’ outpatients are likely to be higher, as demonstrated in a recent nationwide registry, where the risks of bleeding were found to be 4.3% and 2.6% in patients receiving warfarin and ASA, respectively . Taken together, these studies suggest that bleeding rates with warfarin are higher than those seen with ASA, but the difference between these bleeding rates is less than expected, given that ASA is a less potent anticoagulant than warfarin.
In clinical practice, the risk of bleeding related to warfarin is usually considered to be significantly higher than the risk associated with ASA. However, a pooled estimate of the magnitude of this difference from direct comparison of bleeding risk in trials that have randomized patients to warfarin vs. ASA within current therapeutic ranges has not been published, to our knowledge. To address this gap, we performed a systematic review and meta-analysis of randomized controlled trials (RCTs) that enrolled adult patients with any indication for long-term antithrombotic therapy and randomized them to warfarin or ASA. Our objective was to compare the bleeding rates between these two groups of patients. Many studies have used ASA doses or warfarin INR target ranges that are not used in modern clinical practice, so we sought to assess bleeding risks by use of the dose ranges recommended in evidence-based guidelines: ASA 50–650 mg day−1 and INR range 2.0–3.5 [9–11]. We were also interested in, as a secondary question, what the bleeding risk was in trials that randomized patients to warfarin vs. warfarin plus ASA (at equivalent warfarin targets in both treatment arms, and within the same therapeutic ranges specified already). Prespecified secondary analyses of this study included an investigation of bleeding risk according to age (< 70 and ≥ 70 years old) and clinical indication for therapy.
A protocol for assessment of study eligibility was drafted, and is available upon request from the authors. No study review protocol is available for data extraction and data analysis.
Data sources and searches
We searched the Ovid MEDLINE (1948 to August week 4 2011) and Ovid MEDLINE In-Process and Other Non-Indexed Citations (2 September 2011), Embase (1980–2011 week 35) and the Cochrane Central Register of Controlled Trials (2011, issue 3) databases. The search strategies for these three databases were formulated by A.E.W., following consultation with a professional librarian. For a complete description of search terms that were applied to each of these three databases, see Appendix S1. The electronic search strategy was complemented by manually reviewing the reference lists of articles that were identified for full-text review, as well as through contact with content experts.
The included studies, for the purposes of full-text review, satisfied the following criteria: (i) the study considered patients prescribed warfarin with a target INR of 2.0–3.5 as compared with those precribed ASA at a dose of 50–650 mg day−1, or warfarin vs. warfarin plus ASA (with the same warfarin INR ranges and ASA dosages already specified); (ii) the study was an RCT; (iii) adult patients were enrolled; (iv) patient follow-up was carried out for at least 3 months; and (v) bleeding was an outcome ascertained in the study. The study titles and abstracts were screened for study inclusion, and if any of the five criteria were not satisfied, the study was excluded. If there was ambiguity as to whether each criterion was satisfied, the abstract under consideration was noted for full-text review. In addition, we excluded articles that were not published in English, if the full-text article could not be retrieved (online, from the library, or via interlibrary loan), and if other oral anticoagulants were used in addition to warfarin. If data from the same patients were published in multiple articles (i.e. duplicate data), data from the most recent publication were used, unless the data were not extractable. The title and abstract review, and full-text review, were performed in duplicate by two reviewers (A.E.W. and M.P.D.). If there was disagreement in the initial title and abstract selection phase, the article was discussed by the two reviewers. If consensus could not be reached, a third reviewer (M.C.) could be consulted. Consultation with a third reviewer could also be utilized for disagreements during the full-text review.
We included all studies meeting our inclusion criteria, irrespective of the efficacy outcome of the individual studies and the indication for antithrombotic therapy, as it has been previously reported that the therapeutic indication for antithrombotic therapy does not significantly impact on bleeding rates with long-term treatment .
Data extraction and study bias assessment
Two reviewers (A.E.W. and M.P.D.) independently performed the data extraction, using an extraction sheet drafted for the current study, and completed a bias assessment of included articles utilizing the Cochrane Handbook for Systematic Reviews of Interventions . Data pertaining to study inclusion criteria, such as target INR, ASA dosage, and duration of follow-up, were extracted, as were data pertaining to factors such as number of patients enrolled in the study, patient characteristics, clinical indication for antithrombotic medication, bleeding definitions, bleeding outcomes in all patients and in specific subgroups, and number of patients assigned to the treatment arms of interest to the current review (for a complete description, data extraction forms are available from the authors upon request). Owing to heterogeneity in across-study definitions for bleeding, major bleeding events were counted as they were defined in each study.
Discrepancies in the extracted data were resolved by consensus between the two reviewers.
The bias assessment of studies was conducted in accordance with that specified in the Cochrane Handbook for Systematic Reviews of Interventions . Studies were assessed for selection bias, performance bias, detection bias, attrition bias, and reporting bias. Selection bias assessment involved analysis of random sequence generation and allocation concealment. Performance bias assessment involved the blinding of study participants and personnel. Detection bias assessment involved the blinding of study outcome assessors. Attrition bias assessment involved an analysis of the completeness of study outcome data. Reporting bias assessment involved an analysis of selective reporting of study outcomes . For each component of the analysis, studies could receive a label of: low risk of bias; high risk of bias; or unclear risk of bias. In an effort to avoid a possible selection bias, all studies that were retrieved and subsequently analyzed with this bias assessment were ultimately included in our final analysis.
Data synthesis and analysis
Pooled odds ratios (ORs) and associated 95% confidence intervals (CIs) were calculated with the inverse variance method for both major and minor bleeding in patients receiving warfarin vs. those receiving ASA. The analysis was performed with Review Manager statistical software (RevMan version 5.0; Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2008).
To explore between-study variability and the appropriateness of pooling the results from individual studies, the I2-test for heterogeneity was used. The I2-value expresses the percentage of between-study variability that is attributable to heterogeneity rather than chance. An I2-value of ≥ 60% suggests significant heterogeneity, indicating that a formal meta-analysis would be of limited value, given the amount of heterogeneity. To further reduce the impact of heterogeneity, we chose to use a random effects model that would allow for heterogeneity and incorporate its impact in the meta-analysis. Funnel plots of effect size against standard error were used to display small-study effects that may arise from reporting bias, differences in methodological quality, or true heterogeneity among studies. Prespecified subgroup analyses were performed to evaluate the bleeding risk: (i) according to age (< 70 and ≥70 years); and (ii) according to the clinical indication for antithrombotic therapy.
A secondary analysis, specified a priori, was a comparison of bleeding risk in studies that randomized patients to warfarin plus ASA vs. warfarin alone (with the same target INR in both treatment arms, and the same prespecified INR range and ASA dosages specified previously).
We identified 4442 studies with the electronic search strategy, and 47 of these were ultimately identified for full-text evaluation (Fig. 1). Of these 47 full-text articles, 40 studies were excluded for one or more of the following reasons: arms of treatment and/or INR range and/or ASA dosage were not appropriate (36 studies); duplicate data (two studies); and study was not an RCT (two studies). Therefore, seven studies remained for data extraction and bias assessment. One additional article was identified from a manual review of study bibliographies. There were no additional eligible studies identified after communication with four content experts. Therefore, eight studies were included in the systematic review and meta-analysis [14–21].
For the 47 full-text articles screened, there was perfect agreement between reviewers regarding inclusion and exclusion, except for one article that was deemed to be ambiguous by one of the reviewers. For this study, a third party (M.C.) was consulted to resolve the ambiguity. This article was ultimately excluded. 
The main characteristics of the included studies are described in Table 1. The patient population included atrial fibrillation in three studies [16–18], chronic heart failure in three studies [19–21], acute coronary syndrome in one study , and heart valve replacement in one study . The study sample sizes varied from 69 to 1587 patients. A total of 2948 patients were enrolled in the therapeutic arms relevant to the systematic review. ASA dosage varied from 75 to 325 mg day−1; target INR ranges were 2.0–2.5 in one study, 2.0–3.0 in six studies, and 2.0–3.5 in one study. The mean age varied from 62 to 83 years. Duration of follow-up varied among the included studies.
Table 1. Study and patient characteristics
First author, year (study acronym)
Indication for antithrombotic therapy
Total no. of patients (no. of patients in arms of interest)
Mean age (years) (patients in arms of interest)
Female/male (%) (patients in arms of interest)
ASA dose (mg)
Mean follow-up duration (months)
Patients lost to follow-up, no. (%)
Withdrawal/dropout*, no. (%)
ASA, acetylsalicylic acid; CABG, coronary artery bypass graft; INR, international normalized ratio; NA, not available; NSTEMI, non-ST-elevation myocardial infarction; UA, unstable angina. *For withdrawal/dropout, all patients who stopped using the study drug or switched group were counted. †Seventy-five patients were initially enrolled, six of whom were withdrawn from final statistical analysis in the primary article. ‡The study was stopped prematurely, after 42 months. The mean duration of follow-up is not provided in the article describing the study. §This value represents the mean of the reported median age in the primary article. ¶Only one of three patients belongs to one of the arms of interest. **Fifty patients (4.7%) were lost to follow-up in the arms of interest. ††Two hundred and ten patients (19.8%) were withdrawn from study medication in the arms of interest.
Only one study included an additional arm assigned to warfarin plus ASA at the same prespecified INR range and ASA dosage . In four studies, other drugs or no treatment were used in additional arms not considered for the purpose of this meta-analysis: placebo or no treatment in two studies [19,20]; clopidogrel in one study ; and fixed-dose warfarin and fixed-dose warfarin plus ASA in one study .
Major bleeding was defined in six studies, and differed across the studies (Table 2). Among the criteria that defined major bleeding, a need for transfusion was present in five studies [14,17–19,21]; a fall in hemoglobin (Hb) level of ≥ 2 g dL−1 in three studies [14,17,21]; a fatal outcome in three studies [16,18,21]; intracranial bleeding in two studies [17,18]; a need for surgical intervention in three studies [16,18,21]; an outcome of disability in one study ; an outcome of cardiopulmonary arrest or irreversible damage in one study ; and the presence of at least two characteristics including a request for more than three red blood cell units, a systolic blood pressure of < 90 mmHg or a fall in Hb level below 9.7 g dL−1 in one study .
Table 2. Major bleeding definitions in the included studies
Components of major bleeding definition
CP, cardio-pulmonary; Hb, hemoglobin. *Myocardial infarction, stroke, blindness. †Defined in the presence of at least two of the following: > 3 U of red blood cells required; systolic blood pressure of < 90 mmHg; Hb < 6 mmol L−1 (9.7 g dL−1). ‡Bleeding event defined as ‘life-threatening’. §Surgery or angiographic intervention required. ¶At least two units of packed red blood cells or whole blood.
Five studies reported the frequency of minor bleeding outcomes [14,16,17,19,21]. A formal minor bleeding definition was, however, present in only two studies [14,16]. One study defined minor bleeding as all non-major bleeding events reported by the physician or patient , and one study defined it as all non-major and non-threatening bleeds .
For the purpose of the meta-analysis, we made an attempt to differentiate intracranial from non-intracranial major bleeding events. However, data on these events were reported only in four studies [16,18,19,21]. After the authors of the other four studies had been contacted, the relevant information was made available for only one study .
Only one of the eight included studies  was identified as low risk of bias among all categories, as shown in Table 3. One additional study was identified as low risk in all but the random sequence generation and allocation concealment categories, in which insufficient information was provided to classify these items as low risk . The remaining six studies all contained elements of high risk of bias with respect to blinding of participants/personnel [15–19,21]. This was because of the unblinded nature of warfarin delivery in these trials. All studies were identified as low risk in terms of selective reporting, and all but two studies [14,15] were low risk for selection biases. Owing to conservative estimation of risk in incomplete outcome data, six of eight studies [15–19,21] were classified as unclear risk. Fully detailed bias assessments can be made available upon request to the authors.
Major bleeding occurred in 69 of 1455 (4.7%; 95% CI 3.8–6.0) patients treated with warfarin and in 54 of 1449 (3.7%; 95% CI 2.9–4.8) patients treated with ASA (Fig. 2). The pooled OR for major bleeding showed a non-statistically significant increase in major bleeding for patients receiving warfarin vs. those receiving ASA (OR 1.27; 95% CI 0.83–1.94).
The pooled OR for minor bleeding events (data available for five studies [14,16,17,19,21], 1748 patients) showed a significantly increased risk in patients treated with warfarin vs. ASA (OR 1.50; 95% CI 1.13–2.00) (Fig. 3).
A distinction between intracranial hemorrhage and non-intracranial major bleeding events was possible for only five studies [16–19,21] (2630 patients). The pooled OR for intracranial hemorrhage in patients treated with warfarin vs. ASA was 1.64 (95% CI 0.71–3.78) (Fig. 4), and that for extracranial major bleeding was 1.03 (95% CI 0.61–1.75) (Fig. 5).
A sensitivity analysis was performed in two groups of trials according to ASA dosage, in order to assess the robustness of the results. Considering only the RCTs that used ASA dosages of < 300 mg [14,15,18,21], the pooled OR for major bleeding confirmed a non-statistically significant increased risk for patients receiving warfarin vs. those receiving ASA (OR 1.25; 95% CI 0.83–1.86). A similar result for the same comparison was also obtained with consideration of the RCTs that used ASA dosages equal to or greater than 300 mg [16,17,19,20] (pooled OR 1.45; 95% CI 0.31–6.88).
Funnel plots of effects size against standard error were created for major and minor bleeding. The funnel plots appeared symmetric, suggesting the absence of publication bias or other reasons for small-study effects (Figs S1 and S2).
Two prespecified subgroup analyses were performed according to: (i) the mean age of the study participants; and (ii) the clinical indication for antithrombotic therapy. With regard to the secondary analysis based on patient age, the pooled ORs for major bleeding for patients receiving warfarin vs. ASA in studies that included patients with a mean age of < 70 years (four studies [14,19–21], 1492 patients) or ≥ 70 years (four studies [15–18], 1456 patients) were 1.71 (95% CI 0.98–2.98) and 0.96 (95% CI 0.58–1.59), respectively. With regard to the secondary analysis based on the clinical indication for antithrombotic therapy, and the risk of major bleeding related to warfarin vs. ASA in studies that enrolled patients with atrial fibrillation (three studies [16–18], 1387 patients) or with congestive heart failure (three studies [19–21], 1357 patients), the pooled ORs were 0.91 (95% CI 0.54–1.52) and 2.08 (95% CI 0.81–5.36), respectively.
Finally, the bleeding risk of warfarin plus ASA vs. warfarin alone was only evaluated in one study that met our inclusion criteria . In this study two of 44 (5%; 95% CI 0–16) patients receiving warfarin and ASA experienced an episode of major bleeding, as compared with one of 45 (2%; 95% CI 0–13) patients receiving warfarin (OR 2.10; 95% CI 0.18–23.98). For the same comparison, minor bleeding events were experienced by nine of 44 (20%; 95% CI 11–35) patients and 10 of 45 (22%; 95% CI 12–36) patients, respectively (OR 0.90; 95% CI 0.33–2.48).
This systematic review provides a pooled analysis of the risk of bleeding in RCTs that directly compared warfarin with ASA using the INR and ASA dose ranges commonly used and recommended by recent clinical practice guidelines [9–11]. Interestingly, our meta-analysis showed a non-statistically significant increased risk of major bleeding in patients treated with warfarin as compared with those treated with ASA (OR 1.27; 95% CI 0.83–1.94). This observation may have clinical significance, insofar as clinicians will frequently switch patients with a perceived increased risk of bleeding from warfarin therapy to ASA; the findings of this meta-analysis, along with considerations of the relative efficacy of ASA and warfarin, should be considered by clinicians when balancing the risks and benefits of warfarin or ASA treatment.
The results of our analysis of intracranial hemorrhage and non-intracranial major bleeding should be considered with caution, as data were available for only five studies, and only three of these studies [16,18,21] had estimable ORs for intracranial hemorrhage, owing to the lack of reported bleeding events in the other two studies [17,19]. In addition, the CIs of the pooled ORs are wide. However, the results seem to suggest that the trend towards increased major bleeding in patients receiving warfarin was entirely attributable to an increased risk of intracranial bleeding. The frequency of extracranial major bleeding was equivalent between warfarin and ASA.
The results of the secondary analysis based on age deserve some consideration. We found a tendency for a higher risk of bleeding related to warfarin than to ASA in patients younger than 70 years (OR 1.71; 95% CI 0.98–2.98), but not in those older than 70 years (OR 0.96; 95% CI 0.58–1.59). These findings coming from secondary analysis should be considered with caution. Indeed, they may be attributable to chance, given the small number of studies included in this secondary analysis. Age has been described previously as a risk factor for bleeding in both patients treated with warfarin and those treated with ASA [3,4,8]. Additional data are needed to confirm or refute this finding.
This review does have limitations. First, there were relatively few RCTs comparing ASA, 50–650 mg day−1, and warfarin within an INR range of 2.0–3.5. Four of these studies were prematurely interrupted, because of slow enrollment of patients in all three congestive heart failure studies [19–21], and for ethical reasons in the AFASAK-2 trial . We excluded many older RCTs because they did not fulfill the prespecified criteria including the use of an INR target range of 2.0–3.5. These issues resulted in fewer patients being available for the meta-analysis than we would have hoped, but we are confident that we identified all eligible studies. We excluded observational studies, because we wished to present the most accurate estimate of bleeding risks by using a direct comparison of warfarin and ASA. Second, the definition for major bleeding differed between studies. This heterogeneity, which limits our ability to provide reliable estimates of bleeding risk, argues strongly for the use of a standardized definition of major bleeding. The ISTH has established a definition for major bleeding in both medical and surgical patients [23,24]; however, all of the studies included in this systematic review were initiated before those definitions were published. We considered major bleeding and other types of bleeding as they were defined in each study and the meta-analysis compared the ORs calculated within each study, thus allowing for a meaningful pooled estimate. This approach was also utilized in a similar meta-analysis . Third, we were particularly interested in intracranial hemorrhage, because of its high case-fatality rate  and its unique association with antithrombotic therapies [26,27]. However, separate data for intracranial and non-intracranial hemorrhage were available for only five studies. Fourth, the risk of bleeding may vary across different vitamin K antagonists, as a possible consequence of significant differences in half-life. For this reason, we decided to focus on warfarin only, even though we acknowledge that the results may be less generalizable. However, our choice was made in order to provide more accurate and precise results. Finally, the patients enrolled in our analysis were selected because they were participants in RCTs. Such patients probably have a lower risk of bleeding or other complications than unselected patients in the community.
As noted previously, our results may be useful as a guide for clinical practice. Clinicians frequently switch patients from warfarin to ASA as a result of a perception of increased risk of bleeding; this switch is made on the assumption that ASA is associated with a lower risk of hemorrhage. Switching from warfarin to ASA has been shown to be associated with reduced efficacy in many clinical situations, particularly in those patients with atrial fibrillation and additional risk factors for ischemic stroke . The inability to demonstrate any difference in extracranial major bleeding (OR 1.03; 95% CI 0.61–1.75) and the failure to identify a reduced risk of bleeding in those patients over 70 years of age (OR 0.96; 95% CI 0.58–1.59) suggest that the practice of switching from warfarin to ASA because ASA is less likely to be associated with bleeding should be re-evaluated.
In conclusion, this systematic review and meta-analysis comparing the bleeding risk associated with warfarin vs. ASA in RCTs did not show a significant difference in major bleeding risk between these two agents. Further study of this observation is warranted, given this finding.
J. McKinnell acted in a guidance capacity for how to properly search the databases utilized in this investigation. She is a professional librarian affiliated with the McMaster Health Science Library (McMaster University).
Disclosure of Conflict of Interests
M. Crowther has sat on advisory boards for Leo Pharma, Pfizer, Bayer, BI, Alexion, and Artisan, has prepared educational materials for Pfizer, Octapharm and CSL Behring, has provided expert testimony for Bayer, and holds a Career Investigator award from the Heart and Stroke Foundation of Ontario, and the Leo Pharma Chair in Thromboembolism Research at McMaster University. M. Crowther and F. A. Spencer are Career Investigators of the Heart and Stroke Foundation of Ontario. W. Lim is the recipient of the EJ Moran Campbell Internal Career Award (McMaster University). F. A. Spencer is affiliated with the Thrombosis and Atherosclerosis Research Institute (McMaster University). The other authors state that they have no conflict of interest.