Bleeding risk using non-steroidal anti-inflammatory drugs with anticoagulants after venous thromboembolism: a nationwide Danish study

Abstract

Background and Aims

The bleeding risk of using non-steroidal anti-inflammatory drugs (NSAIDs) in patients treated with oral anticoagulants for venous thromboembolism (VTE) remains unclear.

Methods

A nationwide cohort study of 51 794 VTE patients initiating oral anticoagulants between 1 January 2012 and 31 December 2022 was conducted. Time-dependent multivariate cause-specific Cox regression was used to compute adjusted hazard ratios between NSAID use and hospital-diagnosed bleeding episodes.

Results

Event rates for any bleeding per 100 person-years were 3.5 [95% confidence interval (CI), 3.4–3.7] during periods without NSAID use and 6.3 (95% CI, 5.1–7.9) during periods with NSAID use (number needed to harm = 36 patients treated for 1 year). Compared with non-use, the adjusted hazard ratios for any bleeding associated with NSAID use were 2.09 (95% CI, 1.67–2.62) overall, 1.79 (95% CI, 1.36–2.36) for ibuprofen, 3.30 (95% CI, 1.82–5.97) for diclofenac, and 4.10 (95% CI, 2.13–7.91) for naproxen. Compared with non-use, the adjusted hazard ratios associated with NSAID use were 2.24 (95% CI, 1.61–3.11) for gastrointestinal bleeding, 3.22 (95% CI, 1.69–6.14) for intracranial bleeding, 1.36 (95% CI, .67–2.77) for thoracic and respiratory tract bleeding, 1.57 (95% CI, .98–2.51) for urinary tract bleeding, and 2.99 (95% CI, 1.45–6.18) for anaemia caused by bleeding. Results were consistent for anticoagulant and VTE subtypes.

Conclusions

Patients treated with oral anticoagulants for VTE had a more than two-fold increased bleeding rate when using NSAIDs. This increased bleeding rate was not restricted to the gastrointestinal tract.

Structured Graphical Abstract

Methods and summary results in the study investigating the bleeding risk associated with NSAID use in patients treated with oral anticoagulants for VTE. CI, confidence interval; NSAID, non-steroidal anti-inflammatory drug; VTE, venous thromboembolism.

Venous thromboembolism, Anticoagulants, Anti-inflammatory agents, Non-steroidal, Haemorrhage, Epidemiology

See the editorial comment for this article Oral anticoagulation and non-steroidal anti-inflammatory drugs: a recipe for bleeding', by W.A.E. Parker and R.F. Storey, https://doi.org/10.1093/eurheartj/ehae795.

Introduction

Venous thromboembolism (VTE) affects approximately 1 in 12 individuals.¹ Direct oral anticoagulants (DOACs) are increasingly preferred over vitamin K antagonists for VTE treatment and secondary prophylaxis due to their comparable efficacy and lower bleeding risk.^2–4 However, all oral anticoagulants still increase bleeding risk.^2–4

Non-steroidal anti-inflammatory drugs (NSAIDs) are used to treat pain, fever, and inflammation.⁵ As such conditions are also prevalent among patients with VTE, a substantial proportion of VTE patients frequently use NSAIDs.⁶ Concerns remain about the bleeding risk of using NSAIDs concomitantly with oral anticoagulants in patients with VTE. Due to NSAIDs’ effect on platelet function and the gastric mucosa, gastrointestinal bleeding is a particular concern.⁷ Hence, by inhibiting cyclooxygenases 1 and 2, NSAIDs compromise the synthesis of prostaglandins and thromboxane, thus impeding platelet aggregation and gastric cytoprotection.⁸

Previous studies examining the bleeding risk of concomitant use of NSAIDs and oral anticoagulants were limited by a small sample size⁹ or by including only one type of DOAC (rivaroxaban).¹⁰ Thus, it is unclear if the bleeding risk depends on the individual types of DOACs and NSAIDs, and whether the increased bleeding risk, if present, is limited to the gastrointestinal tract or applies to other organ systems. We therefore investigated the bleeding risk associated with NSAID use in patients treated with oral anticoagulants for VTE.

Methods

Setting

The study was set in Denmark. The Danish National Health Service provides tax-financed healthcare, including free access to general practitioners, specialists in private practice, and hospitals, along with partial reimbursement for prescription medication costs to all inhabitants.¹¹ In Denmark, low-dose ibuprofen (200 mg) in limited quantity (20 tablets per package) is the only NSAID available over the counter.¹² The proportion of total ibuprofen sales sold over the counter is approximately 25%.¹² Danish residents receive, at birth or upon immigration, a unique Civil Personal Register number.¹³ This number allows individual-level linkage across Danish health registries and virtually complete long-term follow-up with accurate censoring at death or emigration.¹³

Study design and cohort

We conducted a population-based cohort study of patients with first-time VTE who initiated oral anticoagulant treatment. Supplementary data online, Table S1, presents all codes used in the study. First, we identified all patients with a first-time primary or secondary, inpatient or outpatient discharge diagnosis of VTE recorded in the Danish National Patient Registry from 1 January 2012 to 31 December 2022. Second, among these patients, we identified those with a redeemed prescription of a DOAC (rivaroxaban, apixaban, dabigatran, or edoxaban) or a vitamin K antagonist (warfarin) after their VTE. Third, we restricted the cohort to individuals aged 18 years or older. Lastly, we excluded patients with previous bleeding diagnoses (see Supplementary data online, Figure S1).

The Danish National Patient Registry contains nationwide information on all non-psychiatric inpatient contacts since 1977 and on all non-psychiatric outpatient contacts, psychiatric inpatient and outpatient contacts, and emergency room contacts since 1995.¹⁴ Each contact is recorded in the registry with a primary and, when relevant, secondary diagnoses classified according to the International Classification of Diseases.¹⁴ The coding of first-time VTE has been validated in the Danish National Patient Registry with a positive predictive value of 88%.¹⁵ We classified VTE as either deep vein thrombosis (DVT) or pulmonary embolism (PE). Simultaneous DVT and PE diagnoses were classified as PE because of its higher mortality.¹⁶ Patients with a malignancy within 5 years before VTE, or pregnancy/delivery, trauma/fracture, or surgery within 90 days before VTE, were classified as having possibly provoked VTE. In the absence of these provoking factors, patients were classified as having primary VTE.

Oral anticoagulants

We identified prescription redemptions for oral anticoagulants from the Danish National Prescription Registry, which contains information on all redeemed prescriptions at Danish community pharmacies since 1995.¹⁷ Guidelines recommend oral anticoagulant treatment for at least 3 months, but longer therapy may be required based on VTE severity, lack of provoking factors, and recurrence risk.^3,4 The tentative treatment duration is set at VTE discharge and re-evaluated at a planned outpatient visit after 3–6 months.^3,4 Patients who begin oral anticoagulant treatment for VTE during hospital admission are discharged with enough pills to continue treatment until they can fill their prescription. The date of the first redeemed oral anticoagulant prescription served as the index date. Each prescription was assumed to last the number of days corresponding to the number of tablets in a package of rivaroxaban or edoxaban (used once daily) or half the number of tablets in a package of apixaban or dabigatran (used twice daily). For warfarin, the length of each prescription corresponded to the number of tablets in a package divided by 1.5.¹⁸ If a patient redeemed a new prescription for an oral anticoagulant within the treatment period plus a 30-day grace period, the treatment period was extended by the number of days provided by the new prescription. Otherwise, a patient was considered as having discontinued treatment and was censored.

Non-steroidal anti-inflammatory drugs

We identified NSAID use via the Danish National Prescription Registry and modelled it in a time-varying manner (Figure 1). A use period was defined from the time of a redeemed NSAID prescription until the number of days corresponding to half the number of tablets in a package (used twice daily). If a patient redeemed a new prescription of an NSAID within a use period plus a 14-day grace period, the use period was extended by the number of days provided by the new prescription. If a new prescription was redeemed more than 14 days after a use period, a new use period started on the date of the new prescription.

Figure 1

Illustration of the study design capturing time-varying non-steroidal anti-inflammatory drug use in patients treated with oral anticoagulants for venous thromboembolism. Patients were exposed to oral anticoagulants from the first redeemed oral anticoagulant prescription after venous thromboembolism and until a bleeding event, emigration, death, discontinuation of oral anticoagulant treatment, or end of study period, whichever came first. Non-steroidal anti-inflammatory drug use was then recorded in a time-varying fashion as illustrated by Patients A–E. Patients A, B, C, and E redeemed non-steroidal anti-inflammatory drug prescriptions and thus alternated between non-steroidal anti-inflammatory drug use and no non-steroidal anti-inflammatory drug use. Patient E was censored at the time of discontinuation of oral anticoagulant treatment, defined as not redeeming a new oral anticoagulant prescription within the treatment period covered by the preceding prescription plus a 30-day grace period. NSAID, non-steroidal anti-inflammatory drug; OAC, oral anticoagulant

Bleeding

The primary outcome was a bleeding composite defined as hospital-diagnosed gastrointestinal bleeding, intracranial bleeding, thoracic and respiratory tract bleeding, urinary tract bleeding, or anaemia caused by bleeding. Secondary outcomes included the individual components of the primary outcome. All outcomes were identified from primary or secondary diagnoses registered in the Danish National Patient Registry. Inpatient or outpatient contacts were used for all outcomes except for intracranial bleeding, for which only inpatient contacts were used. The diagnosis codes for bleeding were aligned with previous validation studies showing positive predictive values between 92% and 99%.^19,20

Covariables

Table 1 presents all covariables. Comorbidities were identified from the Danish National Patient Registry based on received diagnoses in the 10 years before the index date. Each patient’s comorbidity burden was categorized according to the Danish Comorbidity Index for Acute Myocardial Infarction.²¹ Concomitant medications were identified from the Danish National Prescription Registry based on prescriptions redeemed within 180 days before the index date.

Table 1

Characteristics of the total study population and of patients treated with direct oral anticoagulants vs. warfarin for venous thromboembolism

Characteristics	Oral anticoagulant drugs overall, N (%)a	Direct oral anticoagulants, N (%)	Warfarin, N (%)
Total patients	51 794	37 991	13 789
Sex, female	24 748 (47.8)	18 000 (47.4)	6739 (48.9)
Median age, years (IQR)	69.4 (56.8–78.3)	70.1 (57.7–78.6)	67.7 (54.2–77.3)
NSAID use before anticoagulant initiationb
None	46 180 (89.2)	33 933 (89.3)	12 235 (88.7)
Any	5614 (10.8)	4058 (10.7)	1554 (11.3)
NSAID use after anticoagulant initiationc
None	46 153 (89.1)	34 136 (89.9)	12 228 (88.7)
Any	5641 (10.9)	3855 (10.1)	1561 (11.3)
Ibuprofen	4570 (8.8)	3139 (8.3)	1243 (9.0)
Diclofenac	329 (.6)	228 (.6)	86 (.6)
Naproxen	582 (1.1)	388 (1.0)	161 (1.2)
Venous thromboembolism risk factor
Primary	17 959 (34.7)	13 484 (35.5)	4471 (32.4)
Possibly provokedd	33 835 (65.3)	24 507 (64.5)	9318 (67.6)
Calendar year
2012–13	7590 (14.7)	1293 (3.4)	6296 (45.7)
2014–15	8731 (16.9)	4545 (12.0)	4182 (30.3)
2016–17	8977 (17.3)	7172 (18.9)	1802 (13.1)
2018–19	9866 (19.0)	9008 (23.7)	855 (6.2)
2020–22	16 630 (32.1)	13 360 (35.2)	654 (4.7)
Comorbidity burdene
None	21 535 (41.6)	16 134 (42.5)	5398 (39.1)
Low	15 625 (30.2)	11 448 (30.1)	4170 (30.2)
Moderate	5041 (9.7)	3615 (9.5)	1423 (10.3)
Severe	9593 (18.5)	6794 (17.9)	2798 (20.3)
Cardiovascular disease
Heart failure	2993 (5.8)	1967 (5.2)	1024 (7.4)
Intermittent arterial claudication	1352 (2.6)	937 (2.5)	414 (3.0)
Aortic disease	838 (1.6)	607 (1.6)	231 (1.7)
Valvular heart disease	1909 (3.7)	1369 (3.6)	538 (3.9)
Stroke	2789 (5.4)	2032 (5.3)	756 (5.5)
Hypertension	13 341 (25.8)	9522 (25.1)	3814 (27.7)
Atrial fibrillation or flutter	5153 (9.9)	3506 (9.2)	1643 (11.9)
Cancer
High-risk cancer	2645 (5.1)	1998 (5.3)	647 (4.7)
Low-risk cancer	8671 (16.7)	6542 (17.2)	2129 (15.4)
Active cancerf	5150 (9.9)	3920 (10.3)	1230 (8.9)
Haematologic disease
Coagulopathy	3486 (6.7)	2272 (6.0)	1214 (8.8)
Endocrine disease
Diabetes, uncomplicated	4132 (8.0)	2888 (7.6)	1241 (9.0)
Diabetes with end-organ damage	2039 (3.9)	1346 (3.5)	693 (5.0)
Hyperthyroidism	795 (1.5)	582 (1.5)	213 (1.5)
Neurologic disease
Dementia	1316 (2.5)	1082 (2.8)	232 (1.7)
Hemiplegia	222 (.4)	164 (.4)	58 (.4)
Neurodegenerative disorder	1034 (2.0)	828 (2.2)	206 (1.5)
Epilepsy	850 (1.6)	608 (1.6)	242 (1.8)
Psychiatric disease
Alcohol and drug abuse	2399 (4.6)	1747 (4.6)	652 (4.7)
Schizophrenia	274 (.5)	193 (.5)	81 (.6)
Affective disorder	1532 (3.0)	1060 (2.8)	472 (3.4)
Kidney disease
Chronic kidney disease	2016 (3.9)	1176 (3.1)	839 (6.1)
Pulmonary disease
Chronic pulmonary disease	6695 (12.9)	4785 (12.6)	1909 (13.8)
Gastrointestinal disease
Ulcer disease	688 (1.3)	472 (1.2)	216 (1.6)
Mild liver disease	653 (1.3)	449 (1.2)	204 (1.5)
Moderate to severe liver disease	159 (.3)	98 (.3)	61 (.4)
Chronic pancreatitis	177 (.3)	122 (.3)	54 (.4)
Conditions with chronic pain
Inflammatory rheumatic disease	4039 (7.8)	2905 (7.6)	1133 (8.2)
Degenerative rheumatic disease	15 167 (29.3)	11 322 (29.8)	3842 (27.9)
Soft tissue disorders	6950 (13.4)	5337 (14.0)	1612 (11.7)
Osteoporosis	3696 (7.1)	2804 (7.4)	892 (6.5)
Headache	652 (1.3)	482 (1.3)	170 (1.2)
Other comorbidities
Obesity	3977 (7.7)	2778 (7.3)	1195 (8.7)
Concomitant medications
ACE inhibitors	7011 (13.5)	4888 (12.9)	2121 (15.4)
Angiotensin II receptor blockers	6226 (12.0)	4989 (13.1)	1234 (8.9)
Antidiabetic drugs	4962 (9.6)	3640 (9.6)	1319 (9.6)
Antiplatelets	9619 (18.6)	6771 (17.8)	2845 (20.6)
Beta-blockers	10 522 (20.3)	7350 (19.3)	3165 (23.0)
Calcium channel blockers	9414 (18.2)	6881 (18.1)	2529 (18.3)
Diuretics	14 604 (28.2)	10 161 (26.7)	4435 (32.2)
Heparins	1842 (3.6)	573 (1.5)	1269 (9.2)
Proton pump inhibitors	14 290 (27.6)	10 488 (27.6)	3796 (27.5)
Statins	11 954 (23.1)	8774 (23.1)	3173 (23.0)
SSRIs	4445 (8.6)	3133 (8.2)	1309 (9.5)
Systemic glucocorticoids	6108 (11.8)	4320 (11.4)	1787 (13.0)

ACE, angiotensin converting enzyme; DOAC, direct oral anticoagulant; IQR, interquartile range; N, number; NSAID, non-steroidal anti-inflammatory drug; OAC, oral anticoagulant; SSRIs, selective serotonin reuptake inhibitors; VTE, venous thromboembolism.

^aThe total counts of direct oral anticoagulants and warfarin are not consistent with the numbers for oral anticoagulant drugs overall because DOAC initiators were censored if they redeemed a prescription for warfarin, and warfarin initiators were censored if they redeemed a prescription for a DOAC. Patients who redeemed prescriptions for two different OACs as of the index date were included in the overall analysis.

^bIndividuals with a redeemed prescription for an NSAID in the 90 days before the index date.

^cIndividuals with at least one redeemed prescription for an NSAID during follow-up.

^dDefined as VTE with a malignancy recorded within 5 years before VTE diagnosis or VTE associated with pregnancy/delivery, trauma/fracture, or surgery within 90 days before VTE diagnosis.

^eCategorized according to the continuous DANCAMI score as none (score: 0), low (score: 1–3), moderate (score: 4–5), or severe (score: 6 or more).

^fDefined as cancer diagnosed within 6 months before the VTE diagnosis.

Statistical analyses

We reported categorical variables as counts with percentages and continuous variables as medians with interquartile ranges. To illustrate trends in oral anticoagulant initiation over time, we plotted the proportion of patients initiating individual oral anticoagulants in 3-month increments.

Using a person-time–based approach, we computed event rates as number of bleeding events per 100 person-years. Person-time was accumulated per patient from the index date until the end of follow-up. Patients who alternated between use and non-use of NSAIDs contributed person-time to both exposure categories accordingly. We calculated the rate difference of bleeding per 1000 patients receiving oral anticoagulants for VTE over 1 year of NSAID treatment compared with no NSAID treatment. We calculated the number needed to harm by taking the reciprocal of the rate difference per person-year. We used cause-specific Cox regression to estimate crude and adjusted hazard ratios (HRs) with 95% confidence intervals (CI), comparing bleeding hazards between use and no use of NSAIDs.^22,23 The adjusted model included sex, age, comorbidity, and comedications listed in Table 1. We followed patients until a bleeding event, emigration, death, discontinuation of oral anticoagulation, or end of study period (31 December 2022), whichever occurred first.

We repeated the analyses for individual NSAIDs (ibuprofen, diclofenac, or naproxen) and oral anticoagulants (DOACs overall, warfarin, rivaroxaban, apixaban, dabigatran, or edoxaban). When we examined DOAC or warfarin initiators separately, we censored patients who initiated DOACs if they redeemed a prescription for warfarin and patients who initiated warfarin if they redeemed a prescription for a DOAC. When we separately examined initiators of individual DOACs, we also censored patients who redeemed a prescription for a DOAC other than the one they had initiated.

Subgroup analyses

We performed the analyses in subgroups according to (i) VTE type (i.e. DVT only vs. PE with or without DVT); (ii) sex and age (18–59, 60–74, or ≥75 years); (iii) calendar year; (iv) presence of a provoking risk factor for VTE (primary or possibly provoked); and (v) active cancer (malignancy recorded within 6 months before VTE diagnosis), as cancer is associated with a higher risk of bleeding in patients with VTE.^24–26

Sensitivity analyses

We performed three sensitivity analyses. First, we excluded individuals who redeemed an NSAID prescription in the 90 days before the index date, as prevalent and new NSAID users may have different bleeding risks.²⁷ Second, we excluded individuals with a prescription for an oral anticoagulant in the 90 days before the index date to increase the likelihood that patients received anticoagulant treatment specifically for VTE. Third, we extended the NSAID grace period from 14 to 30 days, as increasing grace periods increases treatment duration.²⁸ Data management and statistical analyses were performed using SAS version 9.4 (SAS Institute Inc.) and Stata version 17.0 (StataCorp). Illustrations were created using Stata version 17.0 (StataCorp) and R version 4.3.1.

Results

Cohort characteristics

The study cohort consisted of 51 794 patients initiating oral anticoagulant treatment after first-time VTE (see Supplementary data online, Figure S1). Table 1 and Supplementary data online, Table S2, present the baseline characteristics of these patients at the index date. The proportion of females was 48%, the median age was 69 years (interquartile range, 57–78 years), 65% had possibly provoked VTE, 9.9% had active cancer, and 9.9% had atrial fibrillation or flutter. We observed 1897 (3.7%) bleeding events and 4396 (8.5%) deaths during a total follow-up of 52 481 years (median, .6 years; interquartile range, .4–1.0 years).

In 2012, warfarin was the most frequently initiated oral anticoagulant, whereas rivaroxaban and apixaban were the most frequently initiated in 2022 (Figure 2). Dabigatran and edoxaban were rarely initiated. Patients generally redeemed their prescription for oral anticoagulant treatment immediately after hospital discharge for VTE (median, 3 days; interquartile range, 1–18 days). During follow-up, 6.9% switched to an oral anticoagulant other than the one initiated (see Supplementary data online, Table S3), and 11% redeemed at least one prescription for an NSAID. The median duration from oral anticoagulant initiation to the first redeemed NSAID prescription was 100 days (interquartile range, 27–247 days).

Figure 2

Trends in initiation of oral anticoagulants in patients with venous thromboembolism (2012–22). The figure shows the proportion of patients initiating warfarin, rivaroxaban, apixaban, dabigatran, or edoxaban in 3-month increments in relation to the total number of patients initiating oral anticoagulant therapy

Overall bleeding risk

Figure 3 shows the bleeding risk associated with use vs. no use of NSAIDs in patients receiving oral anticoagulants for VTE. The event rates for any bleeding per 100 person-years were 3.5 (95% CI, 3.4–3.7) during periods without NSAID use and 6.3 (95% CI, 5.1–7.9) during periods with NSAID use, corresponding to a crude HR associated with NSAID use of 2.03 (95% CI, 1.62–2.54), which did not change substantially after adjustment (adjusted HR, 2.09; 95% CI, 1.67–2.62). Compared with non-use, the number of extra bleeding events per 1000 patients over 1 year of NSAID use was 28 (95% CI, 14–42). The number needed to cause one additional bleeding event during 1 year with NSAID use vs. no use was 36 (95% CI, 24–72).

Figure 3

Risk of bleeding associated with non-steroidal anti-inflammatory drug use in patients treated with oral anticoagulants for venous thromboembolism. CI, confidence interval; DOAC, direct oral anticoagulant; NSAID, non-steroidal anti-inflammatory drug; OAC, oral anticoagulant. *Adjusted for sex, age, comorbidity burden [categorized according to the continuous DANCAMI score as none (score: 0), low (score: 1–3), moderate (score: 4–5), or severe (score: 6 or more)], additional comorbidities not included in DANCAMI, and comedications listed in Table 1

Drug-specific bleeding risk

Compared with non-use, the adjusted HRs for any bleeding were 1.79 (95% CI, 1.36–2.36) for ibuprofen use, 3.30 (95% CI, 1.82–5.97) for diclofenac use, and 4.10 (95% CI, 2.13–7.91) for naproxen use. The adjusted HRs of any bleeding associated with NSAID use were 2.27 (95% CI, 1.73–3.00) for patients on DOAC and 1.79 (95% CI, 1.16–2.76) for patients on warfarin. Similar trends were observed when examining use of specific NSAIDs. Compared with non-use, use of NSAIDs was associated with an increased bleeding rate for the most commonly used individual DOACs (see Supplementary data online, Table S4). Data were inconclusive for dabigatran and edoxaban due to the low number of initiators.

Bleeding sites

Compared with non-use, the adjusted HRs associated with NSAID use were 2.24 (95% CI, 1.61–3.11) for gastrointestinal bleeding, 3.22 (95% CI, 1.69–6.14) for intracranial bleeding, 1.36 (95% CI, .67–2.77) for thoracic and respiratory tract bleeding, 1.57 (95% CI, .98–2.51) for urinary tract bleeding, and 2.99 (95% CI, 1.45–6.18) for anaemia caused by bleeding (Figure 4). These results were consistent for both DOAC and warfarin.

Figure 4

Site-specific risk of bleeding associated with non-steroidal anti-inflammatory drug use in patients treated with oral anticoagulants for venous thromboembolism. CI, confidence interval; DOAC, direct oral anticoagulant; NSAID, non-steroidal anti-inflammatory drug; OAC, oral anticoagulant. *Adjusted for sex, age, comorbidity burden [categorized according to the continuous DANCAMI score as none (score: 0), low (score: 1–3), moderate (score: 4–5), or severe (score: 6 or more)], additional comorbidities not included in DANCAMI, and comedications listed in Table 1

Patient subgroups

Compared with non-use, NSAID use was associated with an increased bleeding rate in patients treated with oral anticoagulants for DVT (adjusted HR, 2.33; 95% CI, 1.65–3.31) and PE (adjusted HR, 1.97; 95% CI, 1.46–2.65) (Figure 5). In patients treated with oral anticoagulants for VTE, the risks of bleeding did not substantially differ after stratifying by sex, age, calendar year, presence of a provoking risk factor for VTE, and presence of cancer (see Supplementary data online, Tables S5 and S6).

Figure 5

Risk of bleeding associated with non-steroidal anti-inflammatory drug use in patients treated with oral anticoagulants for deep vein thrombosis or pulmonary embolism. CI, confidence interval; DOAC, direct oral anticoagulant; NSAID, non-steroidal anti-inflammatory drug; OAC, oral anticoagulant. *Adjusted for sex, age, comorbidity burden [categorized according to the continuous DANCAMI score as none (score: 0), low (score: 1–3), moderate (score: 4–5), or severe (score: 6 or more)], additional comorbidities not included in DANCAMI, and comedications listed in Table 1

Sensitivity analyses

Our findings were consistent after excluding patients who redeemed an NSAID prescription within 90 days before the index date, after excluding patients who redeemed an oral anticoagulant prescription within 90 days before the index date, and after extending the NSAID grace period from 14 to 30 days (see Supplementary data online, Tables S7 and S8).

Discussion

In patients treated with oral anticoagulants for VTE, we found that NSAID use, compared with no use, was associated with a more than two-fold increased bleeding rate. This increased rate was observed both in patients treated with DOACs and warfarin and was independent of the type of NSAID used. Although limited by imprecision, diclofenac and naproxen use seemed associated with a higher bleeding rate than ibuprofen use, and the increased bleeding rate associated with NSAID use was not restricted to gastrointestinal bleeding but also included other organ systems (Structured Graphical Abstract).

Comparison with other studies

Our study is the first to examine the bleeding risk associated with NSAID use in a large cohort of patients receiving oral anticoagulation for VTE, with all DOACs included. In addition, our study also provides bleeding risk estimates across the use of specific NSAIDs, several organ systems, and VTE type.

Consistent with our findings, a post hoc analysis of the EINSTEIN study found that NSAID use was associated with increased rates of clinically relevant bleeding and major bleeding among patients receiving rivaroxaban or enoxaparin–vitamin K antagonist treatment for VTE.¹⁰ The EINSTEIN study excluded patients with a clearly elevated bleeding risk, potentially leading to conservative bleeding rates.¹⁰ Similar to our findings in patients with active cancer, a Japanese single-centre study found that in 147 patients with advanced cancer, NSAID use during DOAC treatment for VTE was related to major and non-major bleeding.⁹ Our finding that the associated bleeding rate appeared higher for diclofenac and naproxen use than for ibuprofen use is consistent with the previous data.²⁹

Strengths and limitations

Our study had some limitations. First, although our large sample size yielded precise estimates for most individual DOACs, NSAIDs, and bleeding sites, the results remained imprecise for some individual DOACs and bleeding outcomes. However, in all subgroup analyses, the point estimates consistently indicated an increased bleeding rate. Thus, using NSAIDs concomitantly with oral anticoagulants in patients with VTE seemed to increase the rate of bleeding independent of the type of DOAC and NSAID used and across various organ systems. A recent cohort study suggested that patients with VTE who initiated apixaban had a lower rate of gastrointestinal and intracranial bleedings compared with patients who initiated rivaroxaban.³⁰ Thus, future studies should explore whether NSAIDs pose a lower bleeding risk when used in combination with specific oral anticoagulants compared with others. Second, including prevalent NSAID users and assuming all individuals were unexposed to NSAIDs on the index date may underestimate the actual harm from initiating NSAIDs.³¹ However, our findings were robust after excluding prevalent NSAID users. Third, as we lacked information on over-the-counter and in-hospital NSAID use, limited misclassification of NSAID exposure may have occurred.¹² However, previous data have shown that such misclassification of NSAID exposure due to non-prescription NSAID use only has a minor impact on effect estimates.¹² Fourth, we lacked information on NSAID adherence after a redeemed prescription. However, because data collection was prospective, any misclassification due to non-adherence would likely be non-differential regarding future outcomes and therefore cannot explain the results. Fifth, being observational, we cannot exclude unmeasured confounding. While lacking data on potential confounders such as creatinine clearance, smoking, and body mass index, we did indirectly adjust in part for these through hospital-diagnosed and/or prescription-defined chronic kidney disease, chronic pulmonary disease, and obesity. More importantly, the comparisons were made within the cohort of patients with VTE, which likely reduced the potential for confounding in the comparison between NSAID use and no use. The comparable crude and adjusted HRs confirmed little measured confounding. Confounding by indication could not explain our results. If anything, NSAIDs would be less likely to be prescribed to patients with a high bleeding risk (providing conservative estimates),⁷ and we adjusted for the most common chronic pain conditions treated with NSAIDs.

The nationwide, population-based design of the study likely makes the results generalizable to populations and healthcare settings similar to those in Denmark. We included patients who switched to an oral anticoagulant other than the one initiated, making our findings applicable to routine clinical practice.

Implications

The common use of NSAIDs in patients with cardiovascular disease⁶ emphasizes the importance of being aware of the bleeding risks associated with their use. Treatment of pain and inflammation may, for some patients, be worthwhile to improve their quality of life despite an increased bleeding risk. Our findings emphasize the need for additional awareness when prescribing NSAIDs to patients treated with oral anticoagulants for VTE. The need for concomitant proton pump inhibitor use to prevent gastrointestinal bleeding should always be considered.³² However, when assessing the benefits and risks of initiating NSAIDs in these patients, clinicians should assess the patient's total bleeding risk, considering not only the gastrointestinal tract. Safer alternatives such as physiotherapy and paracetamol should always be considered before NSAIDs. If NSAIDs are needed, ibuprofen could be considered over diclofenac and naproxen in VTE patients with high bleeding risk. While the bleeding rate appeared highest for naproxen, as expected due to its cyclooxygenase 1 selectivity, the thromboembolic properties of diclofenac are an additional concern also to be considered.²⁹

Conclusion

In patients treated with oral anticoagulants for VTE, we found that NSAID use, compared with no use, was associated with a more than two-fold increased bleeding rate. This increased rate was observed both in patients treated with DOACs and warfarin and was independent of the type of NSAID used. The increased bleeding rate was not restricted to the gastrointestinal tract.

Supplementary data

Supplementary data are available at European Heart Journal online.

Declarations

Disclosure of Interest

E.L.G. has received speaker honoraria or consultancy fees from AstraZeneca, Bayer, Bristol-Myers Squibb, Pfizer, Novo Nordisk, Lundbeck Pharma, and Organon. He is an investigator in clinical studies sponsored by AstraZeneca, Idorsia, and Bayer and has received unrestricted research grants from Boehringer Ingelheim. All other authors report no conflict of interest.

Data Availability

It is not permitted by Danish law to share the data used for this study.

Funding

M.S. and S.R.P. were supported by the Novo Nordisk Foundation (grant NNF19OC0054908). K.B. was supported by a research grant from the Danish Diabetes and Endocrine Academy, which is funded by the Novo Nordisk Foundation (grant NNF22SA0079901). The funding organizations did not have any role in the design and conduct of the study; in the collection, management, and analysis of the data; or in the preparation, review, and approval of the manuscript.

Ethical Approval

Ethical Approval was not required.

Pre-registered Clinical Trial Number

None supplied.