Background & Aims A new generation of oral anticoagulants (nOAC), which includes thrombin and factor Xa inhibitors, has been shown to be effective, but little is known about whether these drugs increase patients’ risk for gastrointestinal bleeding (GIB). Patients who require OAC therapy frequently have significant comorbidities and may also take aspirin and/or thienopyridines. We performed a systematic review and meta-analysis of the risk of GIB and clinically relevant bleeding in patients taking nOAC.
Methods We queried MEDLINE, EMbase, and the Cochrane library (through July 2012) without language restrictions. We analyzed data from 43 randomized controlled trials (151,578 patients) that compared nOAC (regardless of indication) with standard care for risk of bleeding (19 trials on GIB). Odds ratios (ORs) were estimated using a random-effects model. Heterogeneity was assessed with the Cochran Q test and the Higgins I2test.
Results The overall OR for GIB among patients taking nOAC was 1.45 (95% confidence interval [CI], 1.07–1.97), but there was substantial heterogeneity among studies (I2, 61%). Subgroup analyses showed that the OR for atrial fibrillation was 1.21 (95% CI, 0.91–1.61), for thromboprophylaxis after orthopedic surgery the OR was 0.78 (95% CI, 0.31–1.96), for treatment of venous thrombosis the OR was 1.59 (95% CI, 1.03–2.44), and for acute coronary syndrome the OR was 5.21 (95% CI, 2.58–10.53). Among the drugs studied, the OR for apixaban was 1.23 (95% CI, 0.56–2.73), the OR for dabigatran was 1.58 (95% CI, 1.29–1.93), the OR for edoxaban was 0.31 (95% CI, 0.01–7.69), and the OR for rivaroxaban was 1.48 (95% CI, 1.21–1.82). The overall OR for clinically relevant bleeding in patients taking nOAC was 1.16 (95% CI, 1.00–1.34), with similar trends among subgroups.
Conclusions Studies on treatment of venous thrombosis or acute coronary syndrome have shown that patients treated with nOAC have an increased risk of GIB, compared with those who receive standard care. Better reporting of GIB events in future trials could allow stratification of patients for therapy with gastroprotective agents.
Gastrointestinal bleeding (GIB) is a serious medical condition that causes considerable morbidity and mortality (5%–15%) and poses an enormous burden on global health care use. The mean hospital costs are reported to range from $2500 to $7300 for upper GIB, $4800 for lower GIB, and around $40,000 for small-bowel bleeding. The expanding indications and increasingly intensive treatment with antithrombotic agents have increased the burden of GIB related to these agents. Antiplatelet agents (eg, aspirin and thienopyridine derivatives) can give rise to GIB by producing ulcers and erosions throughout the gastrointestinal tract. Anticoagulants (ie, vitamin K antagonists [VKA]) and heparins might precipitate bleeding from pre-existing lesions. The relative risk of GIB varies from 1.5 for low-dose aspirin compared with nonuseand more than 5 for the combination of aspirin and VKA. In light of their efficacy, the increased risk of bleeding induced by the therapy is acceptable. Two important limitations of the traditional antithrombotic agents comprise the need for international normalized ratio monitoring with tailored VKA dosing, or subcutaneous administration of low-molecular-weight heparins (LMWH).
New oral anticoagulants (nOAC) (eg, factor IIa [thrombin] or factor Xa inhibitors) have been developed and theoretically lack these limitations.[6–8] These drugs are as effective as current therapy. Some randomized controlled trials (RCTs) reported an isolated higher GIB risk,[9,10] which is potentially fatal, costly, and avoidable. It is therefore important to carefully review the literature on GIB risk attributable to use of nOAC. This is particularly relevant because patients on nOAC often use concomitant low-dose aspirin and/or thienopyridines, which may add substantially to the as yet unknown GIB risk. Furthermore, in contrast with the traditional OAC, no clinically tested antidote is currently available for the novel agents, hampering therapeutic options in case of GIB. For these reasons, we conducted a systematic review focusing on the risk of GIB of all nOAC. Because not all trials separately reported GIB risk, we also reviewed the evidence on risk of clinically relevant bleeding associated with nOAC use.
The exposure of interest was defined as the (approximated) indication-specific recommended daily dose of the nOAC either by the European Medicines Agency or the Food and Drug Administration for registered nOAC. When nOAC was not registered for the indication for which it was studied, the indication-specific daily dose was defined according to the pharmaceutical manufacturer.
Standard care was defined as either low-molecular-weight heparin, vitamin K antagonist, antiplatelet therapy, or no (additional) therapy/placebo, depending on the (inter)national guidelines regarding antithrombotic therapy for the concerning indication.
The primary outcome of this systematic review was the risk of GIB. GIB was considered as at least one episode of clinically apparent hematemesis (frank blood or coffee-ground material that tested positive for blood), melena, or spontaneous rectal bleeding (if more than a few spots) or endoscopically confirmed bleeding, and was judged as major or clinically relevant nonmajor depending on the severity.
The secondary outcome was the risk of clinically relevant bleeding (encompassing both major bleeding and clinically relevant nonmajor bleeding). Major bleeding and clinically relevant nonmajor bleeding in the included studies were defined by the following: (1) the International Society on Thrombosis and Haemostasis, [15,16] (2) the Thrombolysis In Myocardial Infarction, or (3) an adjustment of the International Society on Thrombosis and Haemostasis definition (see Table 1 for exact definitions).
Data Sources and Searches
A comprehensive literature search was conducted to identify RCTs reporting GIB or clinically relevant bleeding in patients receiving nOAC compared with standard treatment. Medline with PubMed as interface, EMbase, and the Cochrane Central Register of Controlled Trials were searched from inception to July 2012. Medical subject heading terms and keywords used to identify RCTs included “apixaban,” “rivaroxaban,” “dabigatran,” “edoxaban,” “betrixaban,” “humans,” and “randomized controlled trial.” No language restrictions were applied. The electronic search strategy was complemented by a manual review of reference lists of included articles. References of recent reviews on nOAC also were examined.[11,18-23]
Search results were combined and duplicates were removed. Studies were first screened based on title and abstract for relevance, after which the full text was reviewed. This was performed independently by 2 reviewers (I.L.H. and V.E.V.). Inter-rater agreement was assessed using the k statistic. Any discrepancies were resolved by consensus, contacting a third author (E.T.T.L.T.). Studies had to meet the following inclusion criteria: (1) the study compared nOAC with the current standard care in a randomized setting; (2) results included bleeding events as a safety outcome; (3) the study was conducted in the target population of the drug and not in healthy volunteers; and (4) it was published as a full-text article. If any of the 4 criteria were not met, the study was excluded. If data from the same study were published in multiple languages, data from the English article were extracted. In case of suspicion of double reporting of the same patient populations, data from the main publication were extracted.
The included studies were divided by clinical indication of anticoagulant therapy into the following indication groups: (1) prevention of stroke and systemic embolism in patients with atrial fibrillation (AF); (2) prevention of venous thromboembolism after orthopedic surgery (OS); (3) prevention of venous thromboembolism in medically ill patients; (4) treatment of acute deep vein thrombosis (DVT) or pulmonary embolism (PE); and (5) treatment of acute coronary syndrome (ACS). For each included study, we recorded the number of trial participants, follow-up period, and the number of patients who developed the primary safety end points for both treatment arms. The mean age at baseline and the percentage of males were assessed, as well as other characteristics of the study population such as relevant concomitant medications that may affect bleeding risk. This was performed independently by 2 authors (I.L.H. and V.E.V.). Finally, we contacted the main investigator for missing data. Furthermore, given the heterogeneity of the studies, an individual patient data analysis was attempted. All authors were contacted and requested to provide individual patient data. We received responses from 7 of 23 authors (covering 12 of 43 studies). Unfortunately, no one agreed to share this information.
The quality of included studies was assessed according to the Cochrane Reviewers’ Handbook. Both manuscript and protocol, if available online, were scanned for relevant information on quality.
Data Synthesis and Analysis
Odds ratios (ORs) and associated 95% confidence intervals (CIs) were calculated for each RCT and were the bases for the meta-analyses. To include studies with null events in either the active treatment arm or the standard care arm, 0.5 events were added to all cells with study results. In case of null events in both arms, no OR was calculated. To quantify how many patients needed to be exposed to nOAC therapy to cause one additional GIB compared with standard care the number needed to harm (NNH) was assessed.
To explore between-study variability the Cochran Q test and the Higgins I2 test for heterogeneity were used. Significant heterogeneity was assumed when the Cochran Q P value was less than .10 and the I2 was greater than 50%. To reduce the impact of heterogeneity, we used a random-effects model in these cases.
To account for possible sources of heterogeneity, we performed prespecified subgroup analyses according to type of nOAC and indication. Heterogeneity between subgroups was evaluated further by a post hoc meta-regression analysis by indication, type of nOAC, and comparator. Comprehensive meta-analysis v2.0 (Biostat, Englewood, NJ) was used to perform the meta-analysis. Meta-regression was performed using PASW statistics 20.0 for Windows (SPSS, IBM, Armonk, New York).
Sensitivity analysis was performed to exclude studies that compared the bleeding risk of nOAC use with the use of placebo as standard care because this intervention is unlikely to increase bleeding risk. Because we only included published data, publication bias was quantified with the Egger regression test, with the results considered to indicate publication bias when the P value was less than .10. In addition, funnel plots were examined for asymmetry.
Our initial search identified 375 records (Figure 1A). A total of 42 studies were eligible for inclusion. The agreement between reviewers for trial inclusion was excellent (κ, 0.94). The clinical indication comprised AF in 8 studies,[9,10,25–30] OS in 21 studies, [31–51]medically ill patients in 2 studies,[52,53] DVT/PE in 6 studies (reporting on 7 trials),[54–59] and ACS in 5 studies (Figure 1B).[60-64]
To gain insight into the performance per drug, the information on bleeding risk was summarized per individual drug (Figure 1C). Rivaroxaban was studied most frequently (15 studies reporting on 16 trials),[10,32–35,39–41,44,54,55,58,59,61,63] followed by apixaban (12 trials),[28–30,38,45,48,49,52,53,56,60,62] dabigatran (10 trials),[9,25,31,36,37,42,46,51,57,64] edoxaban (4 trials),[26,27,47,50] and betrixaban (1 trial). The main characteristics of the 43 included trials are summarized in Supplementary Tables 1–5.
A total of 151,578 patients were included in the 43 trials. Duration of follow-up evaluation ranged from 3 weeks to 31 months, with shorter durations of follow-up evaluation for the OS studies and longer durations for AF studies. Patients with a recent history of peptic ulcer disease or patients with an otherwise increased risk of GIB (eg, patients with a thrombocytopenia or coagulation disorder) were excluded in all 43 trials. Concomitant use of any co-medication affecting coagulation was prohibited in 19% of trials, only low-dose aspirin (<160 mg) was allowed in 14%, only short-acting nonsteroidal anti-inflammatory drugs (NSAIDs) (<17 hours) were allowed in 16%, and short-acting NSAIDs/cyclooxygenase-2 inhibitors and/or low-dose aspirin and/or thienopyridines was allowed in 44%, mostly with the addition that it was discouraged. Information on the allowance of antithrombotic co-medication was absent in 7% of trials (Supplementary Tables 1–5).
First, the risk estimates from each study were pooled by indication because the registered/recommended dose for each individual nOAC differs per indication (Supplementary Table 6). A total of 125,354 patients (83%) were enrolled in the therapeutic arms relevant to this review. Of the 8 trials on AF, 7 trials compared one of the novel agents with dose-adjusted warfarin. Of the 21 trials on thromboprophylaxis after OS, 19 compared a nOAC with LMWH (Supplementary Tables 1–5). All trials, except one trial on DVT/PE treatment, compared a nOAC with LMWH followed by VKA. The trials on treatment of ACS compared nOAC with placebo, in addition to standard (double) antiplatelet therapy.
The result of the Egger regression test for publication bias was not significant (intercept, 0.7; 95% CI, –0.4 to 1.7; P = .20) and no funnel plot asymmetry was observed (Supplementary Figure 1), indicating no evidence of publication bias.
Methodologic Quality of Included Studies
Supplementary Table 7 presents an overview of the methodologic quality of included RCTs. The majority of trials mentioned the method used for randomization (93%) and adequate concealment of allocation (72%). Seventy percent of studies applied a double-blind design, 23% had a single-blind design, and 7% followed an open-label design. An independent blinded committee identified all suspected outcome events in each study. Ninety-three percent of studies used an intention-to-treat analysis at least for the safety analysis. The number of patients lost to follow-up evaluation varied between 0.1% and 2.5%, but were reported in only 53% of studies.
Nineteen trials (44%) reported separate data on GIB. Two small trials yielded null events in both groups and therefore were excluded from the GIB analyses.[35,38] A total of 1101 GIB events in 75,081 patients were reported (1.5%) (Supplementary Table 8). These GIBs were predominantly major bleeds (89%). The percentage of GI bleeds per trial in the nOAC group was low in the trials on OS (nOAC, 0.1%; control, 0.2%), intermediate in the trials on AF (nOAC, 2.1%; control, 1.6%) and DVT/PE (nOAC, 3.0%; control, 1.9%), and high in the trials on ACS (nOAC, 5.3%; control, 1.0%). The NNH was 500 (95% CI, -10,000 to 200), meaning that if 1000 patients were treated with the nOAC instead of standard care, this would result in 2 additional GIBs.
Four of 17 studies showed an increased risk, 12 a comparable risk, and 1 a lower risk of GIB when the nOAC was administered compared with the standard care. After pooling the results of 17 RCTs, the nOAC were found to be associated with a higher risk of GIB compared with standard care (pooled OR, 1.45; 95% CI, 1.07–1.97), but with substantial heterogeneity (I2, 61%). First, a considerable part of the increased risk could be attributed to the 2 trials on ACS (pooled OR, 5.21; 95% CI, 2.58–10.53; I2, 0%). To illustrate, the NNH was 24 (95% CI, 17–42), meaning that per 24 patients treated with the nOAC on top of standard care for ACS, 1 extra GIB would occur. Second, the risk of GIB with nOAC was increased for the 2 trials on DVT/PE (pooled OR, 1.59; 95% CI, 1.03–2.44; I2 27%), but not for other indications for nOAC. The calculated OR (95% CI) of each trial is shown in Figure 2A and 2B. With post hoc meta-regression, we studied the effect of indication of use (therapeutic use of nOAC vs prophylactic use). This showed no difference between therapeutic or prophylactic use when adjusted for comparator (placebo vs antithrombotic agent).
Forrest plot of GIB with subgroup analysis by indication. (B) Forrest plot of GIB with subgroup analysis by drug. Data are presented as OR (95% CI) using a random-effects model and I2 test for heterogeneity. Api, apixaban; bet, betrixaban; dab, dabigatran; edo, edoxaban; riv, rivaroxaban; ACS, acute coronary syndrome; AF, atrial fibrillation; DVT, deep vein thrombosis; Med ill, medically ill; OS, orthopedic surgery; PE, pulmonary embolism.
In a subgroup analysis of individual drugs, dabigatran (3 studies;[9,57,64] I2, 36%), and rivaroxaban (5 studies;[10,33,39,41,58] I2, 0%) were associated with a significant increase in risk of GIB, whereas apixaban (8 studies; I2, 0%) and edoxaban (1 study) were not. The pooled OR of GIB associated with dabigatran use was 1.58 (95% CI, 1.29–1.93) (Figure 2B). Expressed in terms of NNH: per 83 patients treated with dabigatran compared with standard care, 1 additional GIB would occur (95% CI, 59–143). The GIB risk associated with use of rivaroxaban had an OR of 1.48 (95% CI, 1.21–1.82). When adjusting for indication of use (therapeutic vs prophylactic), the risk of rivaroxaban remained significantly higher than that of apixaban (OR, 1.77; 95% CI, 1.32–2.38). Analysis by comparator, adjusted for indication of use, revealed no significant differences between different comparators.
In the sensitivity analysis, we excluded studies that compared nOAC with placebo therapy. No major deviations were seen, except for the risk of GIB during DVT/PE treatment, which reduced and became inconclusive (OR, 1.53; 95% CI, 0.99–2.36). Complete results of this sensitivity analysis are shown in Supplementary Table 8.
Clinically Relevant Bleeding
Because GIB is a substantial component of clinically relevant bleeding, we also included this in our analysis. All 43 trials reported on clinically relevant bleeding. The overall risk of clinically relevant bleeding was significantly higher with the use of nOAC compared with standard care (OR, 1.16; 95% CI, 1.00–1.34). Considerable overall heterogeneity, however, was observed (I2, 83%).
In a subgroup analysis in which different indications for nOAC therapy were considered, we found that patients treated for ACS have an increased risk of bleeding (OR, 2.06; I2, 22%) in contrast to patients receiving thromboprophylaxis during OS (OR, 1.05; I2, 36%). The other indications did not show a significantly increased risk, but this may be hampered by the substantial heterogeneity. Subgroup analysis by individual drug showed a slightly increased risk of rivaroxaban compared with standard care (OR, 1.31; 95% CI, 1.04–1.64), but likewise was marked by heterogeneity (I2, 85%), limiting a solid conclusion on the risk of clinically relevant bleeding (Figure 3 and Supplementary Table 8). The risk of clinically relevant bleeding did not differ by drug when adjusted for indication of use.
Forrest plot of clinically relevant bleeding summarized by indication and by drug. Data are presented as OR (95% CI) using a random effects model and an I2 test for heterogeneity. ACS, acute coronary syndrome; AF, atrial fibrillation; DVT, deep vein thrombosis; Med ill, medically ill; OS, orthopedic surgery; PE, pulmonary embolism.
In the sensitivity analysis, excluding studies comparing with placebo, the overall clinically relevant bleeding risk was not increased (OR, 0.98; 95% CI, 0.88–1.10; I2, 65%) .
This systematic review and meta-analysis on 43 trials shows that the nOACs are associated with a modest, but significantly higher, risk of GIB compared with current standard care. This risk is the highest in patients treated for thrombosis (ACS and DVT/PE). In ACS, nOACs were administered on top of other antithrombotic medication, increasing the well-known cumulative risk of GIB. The risk of GIB in patients treated for DVT/PE or receiving thromboprophylaxis for AF is higher than in patients receiving thromboprophylaxis after OS, this might suggest a dose and/or duration effect on top of difference in risk caused by patient characteristics in the different indication groups. However, within the subgroup of AF patients, only patients treated with dabigatran and rivaroxaban carry a higher GIB risk, but not with apixaban. Because head-to-head studies between nOAC in AF have not been performed, it is not possible to determine the drugs with the lowest GIB risk in AF without applying statistically indirect comparisons. A network meta-analysis on overall safety was conducted by others on OS patients and showed no significant differences.
The major strength of this meta-analysis was its focus on GIB. We provide a complete review of 43 trials with a total of 151,578 patients. Given the implementation of nOAC on a large scale, all currently available types of nOAC and all present indications were included because GI physicians will have to deal with GIB complications, irrespective of drug or indication. The data conveyed are corroborated by 2 small meta-analyses with GIB as a secondary safety outcome and in which in total only 3 studies for AF were reviewed. [66,67] For optimal clinical relevance, we included only data obtained with the indication-specific registered/recommended dose per drug, instead of combining all levels of dosages per trial, which was performed in meta-analyses assessing overall risk/benefit of thromboprophylaxis after OS.[8,65]
Two limitations of the current study need to be addressed: (1) study design and GIB report of included studies, and (2) heterogeneity between studies. First, all included studies have been designed for showing noninferior or superior efficacy of nOAC vs current standard care. As a consequence, GIB is not reported as a safety outcome in the majority of studies and will have to be assessed by future studies or by a critical assessment of published studies. A large number of included studies reported only on the composite end point of bleeding outcomes in general. Although the use of this end point has the advantage of increased power, a difference in GIB risk therefore cannot be investigated. However, when studies separately reported on GIB, this was performed for major bleedings, but mostly not for clinically relevant nonmajor bleedings. This led to an underestimation of the risk of all clinically relevant GIBs (ie, composed of both major and clinically relevant nonmajor GIBs). In addition, for GIB there was no standard definition according to a scientific commission, but most trials reported used a uniform definition to identify GIB. Regarding heterogeneity, which is inevitable with current available data, we applied a random-effects model and excluded observational cohorts, healthy volunteer studies, nonregistered drugs, and unpublished data. Furthermore, we addressed all perceived sources of heterogeneity by prespecified subgroup analysis and meta-regression by indication, type of nOAC, and comparator. Analysis by concomitant use of antiplatelet therapy was not feasible owing to lack of stratification of outcome by use of antiplatelet therapy.
Some statistical issues merit clarification. First, we calculated risk estimates per study by means of ORs. Although it would have been preferable to calculate hazard ratios, the rationale to compute ORs was that the mean follow-up time until GIB was not reported per treatment arm for any study. The OR can be interpreted as an estimate of the relative risk because the overall occurrence of GIB is rare (1.5%). Second, for the analysis on GIB, following standard practice, we excluded 2 studies that had no events in both arms. This exclusion was performed because such studies do not provide any indication of either the direction or magnitude of the relative treatment effect, whereas exclusion of the 2 trials would not affect the point estimate. Both studies were of relatively small size (and thus would have had a low weight in the meta-analyses, together equaling approximately 2%).
As evidence of the superior efficacy of nOAC accumulates,[8,65,67] it is important to consider 2 crucial issues. First, most trials used extensive exclusion criteria to enroll only those patients with a presumed low risk of GIB complications attributable to anticoagulants. It is estimated that when these drugs are marketed for daily clinical practice, almost 25%–40% of future users are high-risk patients and the risk of hemorrhage can be as much as 3- to 15-fold increased. It is tempting to speculate that the balance between efficacy and safety will shift unfavorably in these patients because the bleeding risk increases to a much greater extent than the risk of thromboembolism. Second, data on concomitant proton pump inhibitor (PPI) use was not available, except for one trial. A recent consensus guideline states that PPIs should be considered in any person with a risk factor for GIB receiving any type of antithrombotic agent because PPIs have proven to reduce the risk of upper GIB among both traditional NSAID users, low-dose aspirin users, and among patients taking clopidogrel. Future trials, investigating whether gastroprotective agents could increase NNH in patients on nOAC, are warranted. This is of importance because many patients may use nOAC for a considerable duration of time and mostly have significant comorbidity.
In conclusion, we have shown that the gastrointestinal bleeding risk associated with nOAC use might be higher compared with standard care. The current evidence, however, is based on a highly selected patient group with a low bleeding risk, disallowing a true reflection of future patients in daily clinical practice. We recommend that future studies specifically report on the gastrointestinal bleeding risk to further elucidate the true incidence and associated risk. Subsequently, co-administration of gastroprotective agents could be beneficial and warrants further investigations.