Bleeding Risk of Patients With Acute Venous Thromboembolism Taking Nonsteroidal Anti-Inflammatory Drugs or Aspirin

Importance  Combined anticoagulant and aspirin therapy is associated with increased bleeding risk in patients with atrial fibrillation, but the bleeding risk of combined use of anticoagulant and nonsteroidal anti-inflammatory drugs (NSAIDs) is poorly documented.

Objective  To estimate the bleeding risk of combined anticoagulant (rivaroxaban or enoxaparin–vitamin K antagonist [VKA]) and NSAID or aspirin therapy in patients with venous thromboembolism.

Design, Setting, and Participants  Prospective analysis of observational data from the EINSTEIN deep vein thrombosis and pulmonary embolism clinical trials comparing rivaroxaban with enoxaparin-VKA treatment, trials performed in hospitals and clinics in 8246 patients enrolled from 2007 to 2009.

Exposure  Bleeding event rates during exposure to NSAID and aspirin therapy were compared to time without exposure.

Main Outcomes and Measures  Days of NSAID or aspirin use and nonuse, clinically relevant bleeding event and major bleeding event rates by patient-years, and hazard ratios.

Results  During NSAID-anticoagulant concomitant treatment, clinically relevant bleeding occurred with an event rate of 37.5 per 100 patient-years vs 16.6 per 100 patient-years during anticoagulant use only (hazard ratio [HR], 1.77 [95% CI, 1.46-2.14]). Major bleeding during NSAID-anticoagulant treatment occurred with an event rate of 6.5 per 100 patient-years, compared to 2.0 per 100 patient-years during nonuse (HR, 2.37 [95% CI, 1.51-3.75]). For aspirin-anticoagulant concomitant treatment, clinically relevant bleeding occurred with an event rate of 36.6 per 100 patient-years, compared to 16.9 per 100 patient-years during aspirin nonuse (HR, 1.70 [95% CI, 1.38-2.11]). Major bleeding in aspirin-anticoagulant–treated patients occurred with an event rate of 4.8 per 100 patient-years, compared to 2.2 per 100 patient-years during aspirin nonuse (HR, 1.50 [95% CI, 0.86-2.62]). Increases in risk for clinically relevant and major bleeding were similar for rivaroxaban and enoxaparin-VKA anticoagulation regimens.

Conclusions and Relevance  Among patients with venous thromboembolism receiving anticoagulant therapy, concomitant use of an NSAID or aspirin is associated with an increased risk of clinically relevant and major bleeding.

Patients with acute deep vein thrombosis (DVT) or pulmonary embolism (PE) require anticoagulation to prevent recurrent venous thromboembolism. Although anticoagulant treatment is highly effective, it is associated with increased bleeding risk. Bleeding is most often trivial but can be clinically relevant or even fatal. The incidence of DVT and PE increases with age; thus, DVT and PE commonly coexist with comorbid conditions such as atherosclerotic vascular disease and arthritis. A substantial proportion of these patients will receive anticoagulant therapy combined with aspirin and/or nonsteroidal anti-inflammatory drugs (NSAIDs).^1,2 Combined anticoagulant therapy and aspirin use has been associated with an approximate 1.5-fold to 2.5-fold increased bleeding risk in patients with atrial fibrillation and a bleeding risk increase ranging from none to 2.3-fold in patients with ischemic heart disease and prosthetic heart valves.^3-13 Combined anticoagulant and NSAID therapy has been evaluated to a very limited extent in those cardiac conditions, mainly in retrospective analyses, which showed risk increases for major bleeding ranging from none to 3-fold.^14-16 The bleeding risks for anticoagulation combined with either aspirin or NSAID use have not been reported for patients receiving anticoagulation treatment for DVT or PE. Therefore, we used the EINSTEIN DVT and PE^17,18 study cohort to compare the incidence of clinically relevant bleeding and major bleeding in patients with venous thromboembolism receiving study anticoagulant therapy combined with either NSAID or aspirin compared to patients receiving anticoagulant therapy only.

The EINSTEIN DVT and PE program evaluated the efficacy and safety of a 3-month, 6-month, or 12-month course of oral rivaroxaban vs subcutaneous enoxaparin overlapping with and followed by international normalized ratio–adjusted vitamin K antagonists (VKAs) (ie, warfarin or acenocoumarol) in patients with acute symptomatic DVT, PE, or both. The studies were conducted in 39 countries at 336 sites. Patients were observed for the occurrence of suspected symptomatic DVT and/or PE and overt bleeding, which had to be reported to the central independent adjudication committee. Overt bleeding events were classified as major if they were fatal, occurred at a critical site, or were associated with a decrease in hemoglobin concentration of more than 2 g/dL (to convert to grams per liter, multiply by 10) and/or the need for transfusion of at least 2 units of red blood cells. Clinically relevant nonmajor bleeding was defined as bleeding that was not major but associated with medical intervention, an unscheduled contact with a physician, (temporary) cessation of study treatment, or discomfort for the patient such as pain or impairment of activities of daily living. Together, these 2 bleeding subgroups comprised clinically relevant bleeding. Use of NSAIDs and platelet aggregation inhibitors was discouraged. However, if indicated, aspirin up to 100 mg/d and clopidogrel up to 75 mg/d were advised dosages. Concomitant medications taken during the study were collected at each site on an electronic case record form, and source verification was done by a clinical research associate. Concomitant medication was coded according to the WHO Drug Dictionary, version 2005/Q3.¹⁹ The EINSTEIN study patient data were obtained after written informed consent and institutional review board approvals. Data from the EINSTEIN studies’ databases had been entirely delinked from personal health information when accessed for this observational study. This study is consistent with the principles of the Declaration of Helsinki.

The objective of the present analysis was to assess the effects of rivaroxaban or enoxaparin-VKA use combined with NSAID or aspirin therapy on the risk of clinically relevant bleeding and major bleeding. Nonsteroidal anti-inflammatory drugs were defined via the ATC (Anatomical Therapeutic Chemical) code M01A (anti-inflammatory/antirheumatic products, nonsteroids). Drugs containing aspirin with or without other ingredients were included.

The use of NSAID or aspirin therapy was categorized as either “with” or “without” for each day between the day of randomization and the end of the patient’s at-risk period, which was defined as last intake of rivaroxaban or enoxaparin-VKA plus 2 days, or the onset date of first clinically relevant bleeding, major bleeding, or recurrent DVT and/or PE if any of these occurred. The NSAID or aspirin exposure period also included an additional period of 7 days added to the stop date to reflect a possible continuing effect on subsequent outcome risk. Because the onset of a first clinically relevant nonmajor bleeding event was defined to end the risk period for that type of bleeding but often did not end a patient’s use of NSAIDs or aspirin and study medication, the risk periods for major bleeding were generally slightly longer.

A person-time–based approach was used to characterize the rate per 100 patient-years of a first clinically relevant bleeding event or major bleeding event while a patient was taking NSAID or aspirin. Person-time was accumulated per patient from the day of randomization until the end of the at-risk period. Patients who were sometimes taking and sometimes not taking NSAIDs or aspirin during the at-risk period contributed person-time to both categories of exposure. Event rates per 100 patient-years and corresponding 95% confidence intervals (CIs) were computed for the at-risk periods. The resulting point estimate and CI was then transformed into a rate per 100 patient-years. Then, a single Cox regression model with NSAID or aspirin use (no vs yes) as a time-dependent variable was fitted to derive the hazard ratio (HR) and its corresponding 95% CI for rivaroxaban vs enoxaparin-VKA use. P values for interaction between anticoagulant treatment and use or nonuse of NSAID or aspirin medication were calculated. The Cox regression model was stratified according to the study entry index event (ie, DVT only vs PE with or without DVT) to address potential differences between these populations. Additional Cox regression models were fitted to generate HRs for use vs nonuse of NSAIDs or aspirin overall and by treatment group, including stratification for index event and intended treatment duration, with adjustment for age (≥60 vs <60 years), sex, and plasma creatinine clearance (<50 vs ≥50 mL/min/1.73 m²; to convert to milliliters per second per square meter, multiply by 0.0167).

The group that was used to evaluate safety in the EINSTEIN DVT and PE studies comprised 8246 patients (54.6% male; mean age, 57 years), with 4130 patients receiving rivaroxaban and 4116 patients receiving enoxaparin-VKA, of whom 1884 (22.8%) and 1202 (14.6%), respectively, received concomitant NSAID or aspirin therapy at any time during their study treatment.^17,18 A first clinically relevant bleeding event occurred in 388 rivaroxaban-treated patients (9.4%), compared to 412 enoxaparin-VKA–treated patients (10.0%), whereas a first major bleeding event occurred in 40 (1.0%) and 72 (1.7%) patients, respectively. Table 1 shows characteristics of patients who took NSAIDS and took aspirin at any time, as well as patients who did not. Fewer than 45 patient-years altogether were available for analysis for patients taking a study anticoagulant with nonaspirin antiplatelet therapy, with minimal bleeding among them, so bleeding incidences with these co-medications were not further analyzed.

Users of NSAIDs were more likely than nonusers to be women, so all analyses were adjusted for sex. For the treatment groups combined, 127 clinically relevant bleeding occurred during 339 patient-years of NSAID use (37.5 per 100 patient-years) and 673 events occurred during 4054 patient-years of nonuse of NSAIDs (16.6 per 100 patient-years), for an adjusted HR of 1.77 (95% CI, 1.46-2.14). Major bleeding occurred in 24 patients during 369 patient-years of NSAID use (6.5 per 100 patient-years) and 88 events occurred during 4295 patient-years of nonuse of NSAIDs (2.0 per 100 patient-years), for an adjusted HR of 2.37 (95% CI, 1.51-3.75).

In NSAID-rivaroxaban–treated patients, clinically relevant bleeding occurred with an event rate of 37.6 per 100 patient-years (95% CI, 29.0-48.1 per 100 patient-years), compared to 15.8 per 100 patient-years (95% CI, 14.1-17.6 per 100 patient-years) during NSAID nonuse (HR, 1.90 [95% CI, 1.45-2.49]) (Table 2 and Table 3). During NSAID use in enoxaparin-VKA–treated patients, clinically relevant bleeding occurred with an event rate of 37.3 per 100 patient-years (95% CI, 28.7-47.7 per 100 patient-years), compared to 17.4 per 100 patient-years (95% CI, 15.7-19.4 per 100 patient-years) when NSAIDs were not used (HR, 1.65 [95% CI, 1.26-2.17]).

In NSAID-rivaroxaban–treated patients, major bleeding occurred with an event rate of 4.7 per 100 patient-years (95% CI, 2.2-9.0 per 100 patient-years), compared to 1.4 per 100 patient-years (95% CI, 1.0-2.0 per 100 patient-years) during NSAID nonuse (HR, 2.56 [95% CI, 1.21-5.39]) (Table 2 and Table 3). In NSAID-enoxaparin-VKA–treated patients, major bleeding occurred with an event rate of 8.4 per 100 patient-years (95% CI, 4.7-13.8 per 100 patient-years), compared to 2.7 per 100 patient-years (95% CI, 2.0-3.5 per 100 patient-years) during nonuse (HR, 2.28 [95% CI, 1.28-4.04]).

Gastrointestinal bleeding accounted for 18 of 127 clinically relevant bleeding events, with remaining bleeding events divided between multiple sites. Among 1593 patients who used NSAIDs but not antiplatelet agents, there were 110 clinically relevant bleeding events (6.9%). In another 248 patients who simultaneously took an antiplatelet agent, there were 17 clinically relevant bleeding events (6.9%). Very few patients received a COX-2 NSAID, and there were no major bleeding events among them. Clinically relevant and major bleeding occurred with similar frequency at any time in patients taking NSAID co-medication, rather than requiring a long period of exposure before a bleeding event.

Compared to aspirin nonusers, aspirin users were older and had a lower plasma creatinine clearance rate, variables for which all analyses were adjusted. For the treatment groups combined, 104 clinically relevant bleeding events occurred during 284 patient-years of aspirin use (36.6 per 100 patient-years) and 696 events occurred during 4109 patient-years of nonuse of aspirin (16.9 per 100 patient-years), for an adjusted HR of 1.70 (95% CI, 1.38-2.11). Fifteen major bleeding events occurred during 310 patient-years of aspirin use (4.8 per 100 patient-years) and 97 major bleeding events occurred during 4354 patient-years of nonuse of aspirin (2.2 per 100 patient-years), for an adjusted HR of 1.50 (95% CI, 0.86-2.62).

During rivaroxaban treatment, the event rate for clinically relevant bleeding with aspirin use was 34.9 per 100 patient-years, compared to 16.1 per 100 patient-years during aspirin nonuse (HR, 1.81 [95% CI, 1.36-2.41]) (Table 2 and Table 3). For major bleeding during rivaroxaban treatment, the event rate with aspirin use was 3.3 per 100 patient-years, compared to 1.6 per 100 patient-years for aspirin nonuse (HR, 1.50 [95% CI, 0.63-3.61]) (Table 2 and Table 3). During enoxaparin-VKA treatment, the clinically relevant bleeding rate during concomitant aspirin therapy was 39.1 per 100 patient-years, compared to 17.8 per 100 patient-years during aspirin nonuse (HR, 1.59 [95% CI, 1.17-2.17]) (Table 2 and Table 3). For major bleeding during enoxaparin-VKA treatment, the event rate was 6.9 per 100 patient-years, compared to 2.9 per 100 patient-years during aspirin nonuse (HR, 1.50 [95% CI, 0.74-3.05]) (Table 2 and Table 3). The rivaroxaban–enoxaparin-VKA HRs for clinically relevant bleeding and for major bleeding during aspirin use and aspirin nonuse were similar (Table 3). Twelve of 104 clinically relevant bleeding events during aspirin treatment were gastrointestinal. Like the NSAID clinically relevant and major bleeding events, the aspirin bleeding events were distributed sparsely and relatively evenly over the duration of use; ie, there was no apparent increased risk of bleeding with longer duration of concomitant use.

For major bleeding, there were only 37 patient-years of exposure altogether and 2 events among patients taking both NSAIDs and aspirin. For clinically relevant bleeding, there were 33 patient-years of exposure altogether and 16 clinically relevant bleeding events. The point estimates of bleeding rates per 100 patient-years were somewhat higher than for NSAID and aspirin use individually, but the 95% CIs were wide and overlapping. When the odds of bleeding while taking either NSAIDs or aspirin was compared to the odds while taking neither (“neither” rows in Table 2 and Table 3), the point estimates for each comparison show a higher risk, but the CIs overlap with those of the co-medication being compared. To investigate the timing of bleeding, we compared in Table 4 the number of bleeding events of each category for each drug by the quartile of days’ duration of taking either NSAIDs or aspirin. From 22% to 46% of major and clinically relevant bleeding events occurred within the first 8 days of either NSAID or aspirin use for both anticoagulant treatment groups, and relatively fewer occurred in patients who had more than 109 (for NSAID) or more than 182 (for aspirin) days of use concomitant with 1 of the study anticoagulants, but sporadic late bleeding did occur.

Our study, for the first time to our knowledge, showed a clinically important and statistically significant increase in clinically relevant bleeding (ie, 1.8-fold) and major bleeding (ie, 2.4-fold) in patients co-administered NSAIDs with anticoagulation. As has been previously shown for patients with atrial fibrillation, ischemic heart disease, and prosthetic heart valves,^3-13 this study confirms the increased bleeding risk with aspirin used together with anticoagulants in patients treated for venous thromboembolism, although unlike the finding for NSAIDs, this increased risk was less pronounced and not significant for major bleeding. The potential contribution of NSAIDs to bleeding in patients taking an anticoagulant has not been previously rigorously quantified. Although aspirin and NSAID therapy were associated with an approximately 2-fold increase in gastrointestinal bleeding, this could not explain the overall increase in clinically relevant bleeding.

In our study, 22% of patients took NSAIDs at some time, although the protocol discouraged NSAID use. In a recently published registry study of patients with atrial fibrillation discharged from hospitals in Denmark, approximately 22% also disclosed NSAID use.² We wonder whether it is widely appreciated that NSAIDs, available over the counter in most places, put patients receiving anticoagulant therapy at nearly double the risk of clinically important bleeding. We found that a short duration of exposure to NSAIDs and aspirin confers similar risk to longer exposure. We could neither confirm nor deny that COX-2 NSAIDs reduce risk and cannot comment on NSAID dosage.

What are the clinical implications of our findings? The doubled risk for bleeding should urge physicians to combine anticoagulation with either NSAID or aspirin therapy with caution and only if genuinely indicated, with no similarly effective and safer alternative treatment available. If indicated, patients treated with VKAs should be monitored carefully to avoid prolonged periods in the supratherapeutic zone, especially in those with an a priori enhanced bleeding risk. For NSAID therapy, we could not address whether the use of selective COX-2 NSAIDs may be superior because our analysis was underpowered for this subgroup. The medical literature is limited to but 3 relevant retrospective studies that reported conflicting results.^14-16 A recent comparison of rivaroxaban vs placebo in patients with acute coronary syndrome reported a bleeding HR of 3.8 for rivaroxaban-treated patients who took both aspirin and a thienopyridine antiplatelet drug, as well as rivaroxaban at either 2.5 or 5 mg twice daily.²⁰ Such a treatment regimen differs distinctly from that taken by patients after DVT and PE.

Some methodological aspects and possible limitations of our study warrant comment. First, this study was a post hoc analysis. However, the use of concomitant medication with NSAID and aspirin therapy was prospectively collected, intensively monitored, and centrally coded. Bleeding events were prospectively collected as well and were reported using standardized forms. In addition, bleeding events were adjudicated centrally using internationally accepted criteria without knowledge of treatment assignment and use of concomitant medication. Second, use of concomitant NSAIDs or aspirin occurred in patients with differing demographic characteristics and comorbid conditions between users and nonusers. Although the HRs from users vs nonusers were adjusted for confounders, the true additive effect on bleeding of NSAID and aspirin therapy to comorbid illness would optimally be addressed in a randomized trial. Third, the participating physicians were discouraged from using NSAID and aspirin therapy and instructed to use the lowest possible dose if such medications were indicated. Moreover, the EINSTEIN study protocols required that patients with a clearly increased bleeding risk be excluded from participation, and physicians might have been selective in prescribing or continuing NSAIDs and aspirin in those patients with an estimated lower bleeding risk. Thus, the observed bleeding incidences in our studies likely reflect underestimates compared to what would be encountered in community practice and the HR estimates for bleeding that we found in our reduced bleeding risk population may be underestimates for the overall population of patients with DVT and PE who are receiving anticoagulation treatment.

Combined anticoagulant and NSAID or aspirin therapy is associated with an increased risk of clinically relevant bleeding. The increase was similar in rivaroxaban-treated and enoxaparin-VKA–treated patients. Physicians should inform patients about the potential for increased bleeding with these readily available commonly used drugs and advise patients to curtail their casual use.

Accepted for Publication: February 18, 2014.

Corresponding Author: Bruce L. Davidson, MD, MPH, 12209 Shorewood Dr SW, Burien, WA 98146 (brucedavidson@pobox.com).

Published Online: April 14, 2014. doi:10.1001/jamainternmed.2014.946.

Author Contributions: Drs Davidson and Gebel had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Davidson, Lensing, Gebel, Prins.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Davidson, Verheijen, Lensing, Brighton, Lyons, Prins.

Critical revision of the manuscript for important intellectual content: Davidson, Lensing, Gebel, Brighton, Lyons, Rehm, Prins.

Statistical analysis: Davidson, Lensing, Gebel, Prins.

Administrative, technical, or material support: Davidson, Lensing, Prins.

Study supervision: Davidson, Lensing.

Conflict of Interest Disclosures: Drs Davidson and Prins were paid by Bayer for steering committee and related work for the EINSTEIN studies. Drs Brighton, Lyons, and Rehm, site principal investigators in the EINSTEIN studies, were paid for time and costs of patient recruitment and follow-up, and Dr Rehm has been a paid speaker for a sponsor, Janssen Pharmaceuticals. All attended EINSTEIN-related meetings with travel-related costs paid by Bayer. No other disclosures are reported.

Funding/Support: Drs Lensing and Gebel are paid employees of Bayer and devoted salaried time to this work; the other authors were unsupported.

Role of the Sponsors: Bayer Healthcare reviewed the manuscript, but Bayer Healthcare and Janssen Pharmaceuticals had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation of the manuscript; and decision to submit the manuscript for publication. Bayer Healthcare and Janssen Pharmaceuticals sponsored the 2 EINSTEIN clinical trials, collected and maintained the data, and performed the analyses that the authors requested. The sponsors’ other roles are described in the original EINSTEIN publications.

Previous Presentation: This study was presented at the 24th Congress of the International Society on Thrombosis and Haemostasis; July 2, 2013; Amsterdam, the Netherlands.