Summary. Background: Rate of major bleeding is generally accepted as a good measure of the risks associated with anticoagulant therapy, but this may not be true if the proportion of major bleeds with the most serious consequences differs according to the indication for anticoagulant therapy. Objective: To determine whether the indication for long-term oral anticoagulant therapy influences the proportion of major bleeds that are intracranial and fatal. Patients/Methods: Two authors abstracted intracranial and fatal bleeds from randomized trials of patients who received anticoagulant therapy for a minimum of 6 months for atrial fibrillation, ischemic heart disease, venous thromboembolism, prosthetic heart valves and ischemic stroke. Results: There were 877 major bleeds among 23 518 patients in 39 studies. The proportion of bleeds that were intracranial was significantly higher in patients with ischemic stroke (36%; 95% CI, 22–52%; P = 0.02) compared with patients with venous thromboembolism (10%; 95% CI, 5–20%). The difference in the proportion of bleeds that were intracranial among atrial fibrillation, ischemic heart disease, venous thromboembolism and prosthetic heart valves was not statistically significant; however, the estimates varied from 10% to 27%. The proportion of bleeds that were fatal did not differ significantly according to indication, but varied from 8% to 20%. For all indications for anticoagulation, intracranial bleeds were much more likely to be fatal than extracranial major bleeds (44% vs. 4% overall). Conclusions: In current practise, the indication for oral anticoagulant therapy has limited influence on the proportion of major bleeds that are intracranial or fatal.
The benefits of oral anticoagulant therapy must always be weighed against the risk of bleeding. Serious bleeding episodes during anticoagulant therapy are usually referred to as ‘major bleeds’ , and the rate of major bleeds during treatment is used to quantify the risk of bleeding. It is often assumed that the consequences of a major bleed, particularly the probability that the bleed will be intracranial or cause death, are unrelated to the indication for anticoagulant therapy. Provided this is true, the rate of major bleeding is a good generalizable measure of the risks associated with anticoagulant therapy. However, if the proportion of major bleeds that are intracranial or fatal is higher with some indications for anticoagulant therapy compared with others, the rate of major bleeding alone will not provide an accurate reflection of the severity of anticoagulant-related bleeding.
There is evidence in the literature to suggest that these assumptions may not be true. For example, investigators have observed that intracranial bleeding occurs more frequently in anticoagulated patients who have had a previous non-cardioembolic ischemic stroke compared with those with previous stroke and atrial fibrillation . This observation suggests that the indication for anticoagulation might influence the site and severity of major bleeding. Similarly, biological differences between patient populations with other indications for anticoagulant therapy, such as those with venous thrombosis or prosthetic heart valves, may influence the site and severity of bleeding.
To determine whether the proportion of major bleeds that are intracranial or fatal varies according to the indication for oral anticoagulant therapy, we performed a meta-analysis of studies in which patients received long-term oral anticoagulant therapy for five indications.
We searched the MEDLINE and Cochrane Controlled Trials Registry electronic databases (English-language articles only) to identify randomized trials published from January 1989 through to April 2007 that involved patients who received oral anticoagulant therapy for: (i) prevention of stroke in patients with atrial fibrillation; (ii) prevention of recurrent transient ischemic attack or stroke in patients with transient ischemic attack or ischemic stroke; (iii) secondary prevention of acute coronary syndrome in patients with ischemic heart disease; (iv) treatment of venous thromboembolism; and (v) prevention of stroke in patients with prosthetic heart valves. The key words used for the search strategy were atrial fibrillation, stroke, angina, myocardial infarction, deep vein thrombosis, pulmonary embolism, prosthetic heart valve, warfarin, vitamin K antagonist, and anticoagulants. We also manually reviewed citations in retrieved articles and in reviews to identify articles that may have been overlooked in the database search. We did not include articles published before 1989 because most clinical centres were using non-standardized prothrombin time ratios instead of International Normalization Ratio (INR) measurements to monitor oral anticoagulant therapy at that time.
Studies were included if they satisfied all of the following criteria: (i) it was a randomized controlled trial; (ii) at least one of the study arms consisted of patients who received a vitamin K antagonist for a minimum of 6 months, administered to achieve a mid-range INR of at least 1.5; (iii) major bleeds and intracranial bleeds were reported, and the definition for major bleeds was provided. Studies that enrolled fewer than 100 patients per arm were excluded to reduce the potential publication bias associated with preferential publication of small studies with extreme findings. Studies in which patients routinely received an antiplatelet agent in addition to a vitamin K antagonist were also excluded.
The methodological quality of the included studies was assessed using the Jadad scale , which allocates points for randomization, blinding, and completeness of follow-up.
Two authors (L. Linkins and M. O’Donnell) independently reviewed the studies to assess for eligibility and performed data extraction. Disagreements were resolved by consensus between the two authors. When data were missing or unclear, study investigators were contacted for clarification. For each study, data were extracted on: (i) number of patients enrolled, (ii) number and type of major bleeds (i.e. intracranial, extracranial, fatal or non-fatal), and (iii) definition of major bleeding. All reported major bleeds were accepted as major bleeds regardless of how major bleeds were defined in the study. However, all intracranial bleeds and fatal bleeds were counted as major bleeds, whether or not they were classified as major bleeds in the original reports. Major bleeds that occurred in patients who had stopped taking their anticoagulant therapy were also counted and analysed as part of the group to which the patients were randomly assigned.
Data synthesis and analysis
The results of individual studies were pooled using random intercept mixed logistic models to determine summary estimates and associated 95% confidence intervals (CIs). The following estimates were calculated for each of the five indications for anticoagulant therapy: (i) rate of major bleeding per 100 patient-years; (ii) rate of intracranial bleeding per 100 patient-years; (iii) rate of fatal bleeding per 100 patient-years; (iv) proportion of major bleeds that were intracranial; (v) proportion of major bleeds that were fatal (case-fatality); (vi) proportion of major bleeds that were extracranial; (vii) proportion of intracranial bleeds that were fatal, and (viii) proportion of extracranial bleeds that were fatal. Lastly, the mixed effects logistic models were extended to determine if differences among studies in mean patient age, mid-point of target INR and number of units of blood required to meet the definition for major bleeding influenced outcomes. If the model identified a statistically significant effect of indication for anticoagulation therapy on an outcome (P < 0.05), pair-wise comparisons of indications were conducted using Tukey’s method to adjust for multiple testing. Proc GLIMMIX in sas version 9.1 (SAS, Cary, NC, USA) was used.
Of 187 identified studies, 39 studies satisfied all eligibility criteria: 15 atrial fibrillation studies [4–19]; four ischemic stroke studies [20–23]; five ischemic heart disease studies [24–28]; nine venous thromboembolism studies [29–37]; and six prosthetic heart valve studies [38–43] (Fig. 1). Of the included studies, six had more than one group of patients who were eligible to be analysed and, therefore, a total of 45 study arms were included in the analysis (Table 1). Data on the number of intracranial bleeds in one study , and the number of fatal intracranial bleeds in another , were not available. Mean age and mid-point of target INR according to indication for anticoagulant therapy are given in Table 2. Quality of the included studies is given in Table 1.
Table 1. Study characteristics
No. of patients receiving oral anticoagulants
Months of follow-up
No. of major bleeds
No. of fatal bleeds
No. of ICHs
% Loss to follow-up
*One point if randomized and one additional point if randomization method considered appropriate.
†One point if double blind and one additional point if method of blinding considered appropriate.
Table 2. Age and target INR for study arms by indication for anticoagulant therapy
Indication for anticoagulant therapy
No. of study arms
No. of patients
*Mid-point of target INR.
†Mean of included studies (not weighted according to number of patients in each study).
Ischemic heart disease
Prosthetic heart valve
Baseline risk factors for bleeding on anticoagulant therapy varied across the studies: four studies [9,20,24,28] excluded patients older than 75 years of age and two studies [22,35] excluded patients older than 85 years of age. Patients with a past history of stroke were excluded from 14 studies: any previous stroke in four studies [9,18,25,26], stroke within 1–6 months in seven studies [5–8,10,12,29], and stroke within 1–2 years in three studies [11,13,14]. Twenty-six studies included general statements about excluding patients with ‘any disorder that contraindicated anticoagulant therapy’ [4–6,8–12,14,15,17,20,21,23–25,27–29,31,34,35,38,39,41,42].
Timing of initiation of anticoagulant therapy relative to enrollment also varied across the studies: in 18 studies [5,9,10,14–16,22–29,34–37], enrollment was restricted to patients who were starting long-term oral anticoagulant therapy; in 10 studies [4,6–8,11–13,17,18,39], both patients who were starting long-term anticoagulant therapy and patients who were already established on long-term anticoagulant therapy were enrolled; in five studies [30–33,40], only patients who had completed a predefined initial period of anticoagulant therapy without experiencing a major bleed were enrolled; and in six studies [20,21,38,41–43], the anticoagulant status of the patients at enrollment was unclear.
There was a total of 877 major bleeds among 23 518 patients who were followed for 51 915 patient-years. The rates of major bleeding differed according to the indication for anticoagulant therapy (P = 0.0002) (Table 3 and Fig. 2A). The rate of major bleeding was highest in the ischemic stroke studies (3.3 per 100 patient-years) and lowest in the ischemic heart disease studies (0.6 per 100 patient-years). In pair-wise comparisons adjusted for multiple testing, the ischemic heart disease estimate was statistically significantly lower than each of the other four indications.
Table 3. Summary of major bleeding events by indication for anticoagulant therapy
Indication for anticoagulant therapy
Major bleeds per 100 pyrs (95% CI)
ICH per 100 pyrs (95% CI)
Fatal bleeds per 100 pyrs (95% CI)
Fatal bleeds as percentage of major bleeds (95% CI)
In the extended mixed model, the rate of major bleeding increased by 7% with each year of age (P < 0.0001) and increased by 37% with each 1.0 increase in target INR (P = 0.003). The number of units of blood required to meet the definition for major bleeding did not influence the rate of major bleeding (P = 0.54).
Rate of intracranial bleeding
There was a total of 188 intracranial bleeds among 22 415 patients who were followed for 49 709 patient-years. The rates of intracranial bleeding differed according to the indication for anticoagulant therapy (P = 0.001) (Table 3 and Fig. 2B). Ischemic stroke studies had the highest rate of intracranial bleeding (1.3 per 100 patient-years) while the other indications for anticoagulant therapy showed rates of intracranial bleeding of 0.2–0.5 per 100 patient-years. In pair-wise comparisons adjusted for multiple testing, the estimates for venous thromboembolism and ischemic heart disease studies were statistically significantly lower than the estimate for ischemic stroke studies, but not significantly different from the estimates for atrial fibrillation or prosthetic heart valves studies.
In the extended mixed model, the rate of intracranial bleeding increased by 9% with each year of age (P = 0.0001) and increased by 72% with each 1.0 increase in target INR (P = 0.0006).
Intracranial bleeding as a proportion of major bleeding
Of 877 major bleeds, 188 (21%) were intracranial (the number of intracranial bleeds was not available for one study ). The proportion of major bleeds that were intracranial differed according to indication for anticoagulant therapy (P = 0.02) (Table 3 and Fig. 2C). In pair-wise comparisons adjusted for multiple testing, the proportion of major bleeds that were intracranial in the ischemic stroke studies was statistically significantly higher than in the venous thromboembolism studies. In the extended mixed model, mean age (P = 0.84), mid-point target INR (P = 0.35) and number of units of blood required to meet the definition of major bleeding (P = 0.51) did not influence the proportion of major bleeds that were intracranial (Table 3 and Fig. 1D).
Fatal bleeding as a proportion of major bleeding
Of the 877 major bleeds, 118 were fatal bleeds (13%). The proportion of major bleeds that were fatal did not differ according to indication for anticoagulant therapy (P = 0.32) (Table 3 and Fig. 2D). In the extended mixed model, mean age (P = 0.48) and number of units of blood required to meet the definition of major bleeding (P = 0.65) did not influence the proportion of major bleeds that was fatal, but as the mid-point target INR increased, there was a tendency for more major bleeds to be fatal (P = 0.06).
Case-fatality of intracranial and extracranial major bleeds
Of 188 intracranial bleeds, 82 were fatal (44%). The proportion of intracranial bleeds that were fatal was 35% (95% CI, 24–47) for the atrial fibrillation studies, 54% (95% CI, 38–70) for the ischemic stroke studies, 50% (95% CI, 32–69) for the ischemic heart disease studies, 40% (95% CI, 15–71) for the venous thromboembolism studies and 56% (95% CI, 32–78) for the prosthetic heart valve studies (P = 0.29).
Of 689 major extracranial bleeds, 27 were fatal (4%). The proportion of extracranial bleeds that were fatal was 6% (95% CI, 3–12) for the atrial fibrillation studies, 5% (95% CI, 1–19) for the ischemic stroke studies, 2% (95% CI, 0.4–12) for the ischemic heart disease studies, 5% (95% CI, 1–14) for the venous thromboembolism studies and 1.0% (95% CI, 0.1–8) for the prosthetic heart valve studies (P = 0.56).
Of 118 fatal bleeds, 82 were intracranial (69%) (the site of bleeding was not reported for nine fatal bleeds). The proportion of fatal bleeds that were intracranial was 66% (95% CI, 50–79) for the atrial fibrillation studies, 87% (95% CI, 69–95) for the ischemic stroke studies, 84% (95% CI, 61–95) for the ischemic heart disease studies, 45% (95% CI, 17–75) for the venous thromboembolism studies and 90% (95% CI, 53–99) for the prosthetic heart valve studies (P = 0.35).
This analysis found that, when major bleeding occurred, the proportion of bleeds that were intracranial or fatal was not significantly different between patients who received long-term anticoagulant therapy for atrial fibrillation, ischemic heart disease, venous thromboembolism or prosthetic heart valves. While the proportion of major bleeds that were intracranial appeared to be higher in patients with ischemic stroke, this observation is of lesser clinical importance because non-cardioembolic stroke is no longer considered an indication for oral anticoagulant therapy. For all indications for anticoagulation, intracranial bleeds were about 10 times more likely to be fatal than extracranial major bleeds.
Previous studies and analyses have reported the proportion of major bleeds that were fatal or intracranial in patients with a single indication for long-term anticoagulant therapy [44–48], but our study is the first to perform a standardized comparison of these proportions across indications for anticoagulant therapy. Reliable estimates of intracranial and fatal bleeding are difficult to obtain from individual trials because they occur infrequently. The greatest strength of the current analysis is that it included over 23 000 patients from methodologically rigorous trials who experienced 188 intracranial bleeds and 118 fatal bleeds.
The most important finding of this analysis was the similarity in the proportion of major bleeds that were intracranial and fatal among the patient groups receiving anticoagulant therapy for atrial fibrillation, venous thromboembolism, ischemic heart disease and prosthetic heart valves. Although the proportion of major bleeds that were intracranial appeared to be higher for atrial fibrillation and ischemic heart disease than for venous thromboembolism and prosthetic heart valves, the difference was not statistically significant. However, due to the broad range of the number of patient-years of exposure across these indications, the possibility that a real difference exists could not be excluded. The mean age of patients and the number of transfused units of blood that defined a major bleed in different studies also did not appear to influence the proportion of the major bleeds that were intracranial or fatal, although there was a trend toward a higher proportion of fatal bleeds with increasing mid-point target INR. These findings suggest that a low rate of major bleeding in patients receiving long-term oral anticoagulant therapy (e.g. ischemic heart disease) does not necessarily result in a lower likelihood that a major bleed will be fatal or intracranial.
There are limitations to the current analysis. The included studies used a variety of definitions for major bleeding. Although meta-regression did not identify that the number of units of blood required to meet the definition for major bleeding influenced outcomes, we cannot exclude that other differences in the definition of major bleeding may have influenced the findings of our analysis. The timing of initiation of anticoagulant therapy also differed across the studies. Due to front-loading of anticoagulant-related bleeding, studies that enrolled patients who were just starting anticoagulant therapy are likely to have higher overall bleeding rates than studies that enrolled patients who were already taking anticoagulant therapy . The time in therapeutic range, which may vary according to the indication for anticoagulation or year of the study, could not be factored into the analysis because it was not consistently reported . Lastly, for the same reason, we are unable to comment on the potential influence of comorbidities, such as diabetes and hypertension, on our results.
In patients receiving long-term anticoagulant therapy for contemporary indications, our analysis found that the indication for anticoagulation appeared to have limited influence on the proportion of major bleeds that were intracranial or fatal.
Disclosure of Conflict of Interests
The authors state that they have no conflict of interest.