Risk of Upper Gastrointestinal Hemorrhage in Warfarin Users Treated With Nonselective NSAIDs or COX-2 Inhibitors

Background  Little is known about the risk of upper gastrointestinal (GI) hemorrhage during the concomitant use of warfarin and selective cyclooxygenase (COX)-2 inhibitors. We examined the association between the concomitant use of warfarin and nonselective nonsteroidal anti-inflammatory drugs (NSAIDs) or selective COX-2 inhibitors in older adults hospitalized for upper GI hemorrhage.

Methods  This nested case-control analysis of multiple linked health care databases conducted over 1 year identified a cohort of patients in Ontario, Canada, who were older than 66 years and continuously prescribed warfarin. Case patients were those admitted to the hospital with upper GI hemorrhage while taking warfarin. We compared their prescription records prior to hospitalization with those of age- and sex-matched controls who were also receiving warfarin (the control-case ratio was 4:1). Odds ratios (ORs) for the risk of hospitalization for upper GI hemorrhage while concomitantly using warfarin and celecoxib, rofecoxib, or nonselective NSAIDs were determined.

Results  During the study period, we identified 98 821 elderly patients continuously receiving warfarin. Of those, 361 (0.3%) were admitted to the hospital with upper GI hemorrhage. After adjusting for other potential confounders, case patients were significantly more likely to be also taking nonselective NSAIDs (OR, 1.9; 95% confidence interval [CI], 1.4-3.7), celecoxib (OR, 1.7; 95% CI, 1.2-3.6), or rofecoxib (OR, 2.4; 95% CI, 1.7-3.6) prior to hospitalization relative to controls.

Conclusions  Patients taking warfarin concomitantly with selective COX-2 inhibitors have an increased risk of hospitalization for upper GI hemorrhage. The risk appears similar to that of patients simultaneously taking warfarin and nonselective NSAIDs.

Warfarin is an anticoagulant commonly used for the management of patients with a variety of thromboembolic conditions¹ but it is implicated in many drug-drug interactions. Some interactions alter warfarin pharmacokinetics and others involve drugs that directly influence hemostasis or gastrointestinal integrity.^2,3 Nonsteroidal anti-inflammatory drugs (NSAIDs) are often coprescribed to patients receiving warfarin, potentially resulting in a significantly increased risk of upper gastrointestinal (GI) hemorrhage.⁴ This increased risk may be mediated through several pathways. First, many NSAIDs are substrates for the cytochrome P450 2C9 isoenzyme. Their use may interfere with the oxidative metabolism of S-warfarin, the more active enantiomer of commercially available racemic warfarin, thereby increasing the hypoprothrombinemic response to warfarin.⁵ Second, the nonspecific inhibition of cyclooxygenase (COX) enzymes by nonselective NSAIDs leads to significant inhibition of COX-1–generated thromboxane A ₂, impairing platelet aggregation,^6,7 which may be further compounded by the concomitant use of warfarin.⁸ Third, traditional NSAIDs can cause gastric erosion, thereby further increasing the risk of GI bleeding in patients treated with warfarin.⁹

The introduction of the selective COX-2 inhibitors has been met with wide acceptance¹⁰ given their potentially lower risk of adverse GI events relative to nonselective NSAIDs.^11-13 This perception of greater safety with the selective COX-2 inhibitors may result in a higher rate of coprescription with warfarin among users of selective COX-2 inhibitors relative to users of nonselective NSAIDs.¹⁰ While recent pharmacodynamic evidence suggests that at least 1 of the selective COX-2 inhibitors may slightly augment the pharmacodynamic effect of warfarin by increasing prothombin time,^14,15 information on the clinical consequences of this combination of drug therapy is limited.¹⁶ We conducted a population-based study of older adults to examine the association between the concomitant use of warfarin with these drugs and the risk of hospitalization for upper GI hemorrhage.

We conducted a nested case-control analysis of multiple linked health care databases over 1 year (April 1, 2000, to March 31, 2001) in Ontario, Canada. This research was approved by the ethics review board of Sunnybrook and Women’s College Health Sciences Center.

Prescriptions were identified from the Ontario Drug Benefit Program, which records prescription medications dispensed to Ontario residents who are 65 years and older. Admissions for upper GI hemorrhage were identified from the Canadian Institute for Health Information Discharge Abstract Database (CIHI DAD), which contains a detailed record of all hospital admissions in Ontario. Basic demographic information for each Ontario resident was obtained from the Ontario Registered Persons Database. The databases are linked through encrypted numbers identifying patients.

We studied elderly patients whose prescription records allowed us to define a period of uninterrupted warfarin use. This observation period began with the first prescription for warfarin after their 66th birthday and ended with hospital admission for upper GI hemorrhage, the end of the study period, the patient’s death, or discontinuation of warfarin, whichever occurred first. We chose not to study patients during their first year of eligibility for prescription drug coverage, ie, during their 66th year, to avoid incomplete medication records. Patients were deemed to have discontinued warfarin treatment if more than 120 days had elapsed between prescriptions. In these cases, we extended the observation period to 90 days after the last prescription to include admissions for GI hemorrhage that may have led to cessation of therapy. A similar method has previously been used to define periods of continuous drug use.^17-19

Within the cohort of continuous warfarin users, we defined as case patients those registered in the CIHI DAD as having been admitted to the hospital with any diagnosis of upper GI hemorrhage between April 17, 2000, and March 31, 2001. The most commonly found International Classification of Diseases, Ninth Revision (ICD-9), Clinical Modification diagnostic codes were 531.0, 531.2, 531.4, 531.6, 532.0, 532.2, 532.4, 532.6, 533.0, 533.2, 533.4, 533.6, 534.0 534.2, 534.4, 534.6, 578.0, 578.1, and 578.9),²⁰ and these codes have been shown to have a positive predictive value of 86% for upper GI hemorrhage.²¹ These specific dates were chosen because both celecoxib and rofecoxib were made available on the provincial formulary on April 17, 2000, and our hospital database, the CIHI DAD, was complete until March 31, 2001, at the time of the study. The date of admission for upper GI hemorrhage served as the index date for all analyses.

From the cohort of continuous warfarin users, we randomly selected 50 controls for each case, matching for age (born within 1 year of the case patient) and sex. When numerous controls existed for each case patient, we randomly selected 4 matched controls for analysis. When fewer than 4 potential controls were available, we analyzed only those controls and did not alter the matching process. A case patient could serve as a control for a different case patient prior to his or her admission date, and each control could be used for multiple case patients.^22,23

We identified prescriptions for any COX-2 inhibitor and any prescription for an NSAID in the 90 days prior to the index date. As a test of robustness, we also examined any prescriptions for ocular antibiotics in our cohort of warfarin users to test if no association was observed where none was expected.

To adjust for potential confounders we included several comorbidities in the analysis, as outlined in Table 1. We included hospitalizations for upper GI hemorrhage in the 3 years prior to the index date using the CIHI DAD. For procedures such as endoscopy and upper GI radiologic studies, we examined the data of the Ontario Health Insurance Plan, which records all physician claims for diagnostic and therapeutic procedures. We also examined drug use patterns during the 120 days prior to the index date as markers for conditions that may be associated with an increased risk for upper GI hemorrhage or as drugs that could interfere with the metabolism of warfarin (Table 1).^24,25 We also identified residence in a long-term care institution, since this may be an indicator of comorbidities.²⁶ Finally, we adjusted for the number of different drugs prescribed in the year prior to the index date as a recently validated measure of comorbidity.²⁷

Conditional logistic regression was used to estimate the odds ratios (ORs) for the association between hospital admission for upper GI hemorrhage and receipt of COX-2 inhibitors or nonselective NSAIDs. The 95% confidence intervals (CIs) were constructed using nonparametric bootstrap methods with resampling of controls over 1000 iterations.²⁸ To avoid the extreme OR estimates that may occur with low exposure rates,²⁹ for each interaction we report the point estimate of the median univariate OR from bootstrap analysis. In the multivariate analysis we adjusted for use of COX-2 inhibitors or NSAIDs when they were not the primary exposure of interest, in addition to the covariates outlined in Table 1.

We identified 98 821 elderly patients treated continuously with warfarin for a total of 65 064 patient-years of therapy. Their mean (SD) age at entry was 77 (6.8) years and approximately 50% were women.

During the study period, 361 elderly patients were admitted to the hospital with GI hemorrhage. Their mean (SD) age was 79.8 (6.9) years and 51.5% were women (Table 2). Of these 361 case patients, 24 (6.6%) had received a prescription for a nonselective NSAID within the preceding 90 days, while 23 (6.1%) had received celecoxib and 25 (6.9%) had received rofecoxib. After adjusting for other potential interacting medications and comorbidities, we found that patients receiving warfarin who were admitted with GI hemorrhage were more likely to have been treated with celecoxib (OR, 1.7; 95% CI, 1.2-3.6), nonselective NSAIDs (OR, 1.9; 95% CI, 1.4-3.7), or rofecoxib (OR, 2.4; 95% CI, 1.7-3.6) in the preceding 90 days (Table 3). As expected, no significant association was observed for ocular antibiotics (OR, 0.9; 95% CI, 0.7-1.3).

The findings of this study suggest that increases in the risk of upper GI hemorrhage are similar in warfarin users concomitantly taking either nonselective NSAIDs or selective COX-2 inihibitors compared with patients not treated with these drugs. While the concomitant use of warfarin with NSAIDs is a recognized risk factor for GI hemorrhage,^4,5 to our knowledge our study is the first to examine the comparative safety of the COX-2 inhibitors celecoxib and rofecoxib in patients receiving warfarin. Previous studies of this potential interaction have been conducted in healthy volunteers and have relied on small samples and surrogate outcome rather than actual bleeding episodes.^14,15 Furthermore, in the Celecoxib Long-term Arthritis Safety Study (CLASS), the comparison of patients receiving celecoxib and aspirin with patients receiving standard NSAID therapy demonstrated no difference in ulcer complications. This may suggest that a COX-2 inhibitor taken with an agent that increases the risk of bleeding may not be safer than NSAIDs alone. A small nested case-control study recently observed differences in the results of exposure to nabumetone or meloxicam when compared with nonselective NSAIDs in patients with or without bleeding complications who were concomitantly using coumarine.¹⁶ However, this study was questionnaire-based with a significant differential in response rates between case patients (approximately 70%) and controls (approximately 31%) and it lacked sufficient information on celecoxib and rofecoxib. Furthermore, patients in this study received phenprocoumon or acenocoumarol, whereas in our study patients only received warfarin. These differences would preclude direct comparisons between this study and ours.

Several limitations of our study deserve mention. First, although we controlled for many important confounders, we were unable to account for some potentially important factors such as smoking, alcohol consumption, and the use of nonprescription analgesics. In our study, case patients presented with a higher prevalence of GI comorbidity. Previous research also suggests that COX-2 inhibitor users also typically present with greater GI comorbidity relative to control populations or nonselective NSAID users.¹³ The implications of these discrepancies in comorbidity profiles between groups could result in an overestimation of true risk. Second, the low absolute number of events in the study groups precluded reliable subgroup analyses such as comparisons among users of specific NSAIDs. Third, the generalizability of our findings to younger patients or settings with different drug policies over longer durations of follow-up is uncertain. Fourth, we were not able to examine international normalized ratios in any patients, making it difficult to determine whether patients with GI hemorrhage had supratherapeutic international normalized ratios. Furthermore, the incidence of GI hemorrhage in patients taking warfarin in our study was surprisingly low (0.3%) compared with a prediction model based on different risk factors (high, middle, and low) in which the proportion of bleeding events was 7%, 4%, and 1%, respectively, after 3 months of therapy.³⁰ Our lower incidence of GI hemorrhage may be secondary to intermittent dosing of anti-inflammatory agents or other patient-level factors. Finally, we did not examine the interaction between COX-2 inhibitors and aspirin use.

Our findings suggest that the risk of upper GI hemorrhage is similarly heightened in warfarin users treated with either selective COX-2 inhibitors or nonselective NSAIDs. Given the observational nature of this study, our findings should be viewed as preliminary and require confirmation through well-designed clinical trials. Regardless, physicians and pharmacists who care for elderly patients taking warfarin should be aware of the potential risks of concomitant therapy with NSAIDs or COX-2 inhibitors, particularly because the latter are among the fastest-growing class of prescription medications and have rapidly gained acceptance in clinical practice.

Correspondence: Marisa Battistella, BScPhm, PharmD, University Health Network–Toronto General Hospital, 200 Elizabeth St, ENG Room 260, Pharmacy Department, Toronto, Ontario, Canada M4G2C4 (Marisa.battistella@uhn.on.ca).

Accepted for Publication: July 19, 2004.

Financial Disclosure: Dr Rabeneck has received speaking honoraria from Pfizer Inc.

Funding/Support: Dr Mamdani is supported by a New Investigator Award from the New Emerging Teams of the Canadian Institutes of Health Research (CIHR). Dr Juurlink is supported by a New Investigator Award from the CIHR and the University of Toronto Drug Safety Research Group. Dr Laupacis is a senior scientist of the CIHR.