Johnsen SP, Larsson H, Tarone RE, McLaughlin JK, Nørgård B, Friis S, Sørensen HT. Risk of Hospitalization for Myocardial Infarction Among Users of Rofecoxib, Celecoxib, and Other NSAIDsA Population-Based Case-Control Study. Arch Intern Med. 2005;165(9):978-984. doi:10.1001/archinte.165.9.978
It remains uncertain if the excess cardiovascular risk of rofecoxib and celecoxib reported in clinical trials is present in routine practice and whether the use of other nonaspirin nonsteroidal anti-inflammatory drugs (NSAIDs) also carries an increased cardiovascular risk. We performed a population-based case-control study to examine the risk of myocardial infarction (MI) among users of various categories of nonaspirin NSAIDs.
Using data from hospital discharge registries in the counties of North Jutland, Viborg, and Aarhus, Denmark, and the Danish Civil Registration System, we identified 10 280 cases of first-time hospitalization for MI and 102 797 sex- and age-matched non-MI population controls. All prescriptions for nonaspirin NSAIDs filled before the date of admission for MI were identified using population-based prescription databases. Relative risk estimates for MI were adjusted for a history of cardiovascular disease, hypertension, diabetes mellitus, chronic bronchitis or emphysema, alcoholism, liver cirrhosis, upper gastrointestinal bleeding, rheumatoid arthritis, systemic lupus erythematosus and the use of high-dose aspirin, platelet inhibitors, insulin or oral hypoglycemic drugs, antihypertensive drugs, lipid-lowering drugs, oral anticoagulants, nitrates, penicillamine, gold, oral glucocorticocoids, and hormone therapy before the date of admission for MI.
Current users of rofecoxib had an elevated risk estimate for hospitalization for MI compared with nonusers of any category of nonaspirin NSAIDs (adjusted relative risk [ARR], 1.80; 95% confidence interval [CI], 1.47-2.21). Increased risk estimates were also found among current users of celecoxib (ARR, 1.25; 95% CI, 0.97-1.62), other cyclooxygenase-2 selective inhibitors (ARR, 1.45; 95% CI, 1.09-1.93), naproxen (ARR, 1.50; 95% CI, 0.99-2.29), and other conventional nonaspirin NSAIDs (ARR, 1.68; 95% CI, 1.52-1.85). The highest ARRs were found among new users of all examined drug categories.
Current and new users of all classes of nonaspirin NSAIDs had elevated relative risk estimates for MI. Although the increased risk estimates may partly reflect unmeasured bias, they indicate the need for further examination of the cardiovascular safety of all nonaspirin NSAIDs.
The withdrawal of rofecoxib based on an increased risk of myocardial infarction (MI) and stroke among patients treated for longer than 18 months in the Adenomatous Polyp Prevention on Vioxx (APPROVe) trial marks a dramatic event in the history of the cyclooxygenase-2 (COX-2) selective nonsteroidal anti-inflammatory drugs (NSAIDs).1 This has further been underscored by the decision by the US National Cancer Institute to halt the ongoing Adenoma Prevention with Celecoxib (APC) trial owing to increased cardiovascular risk among patients receiving celecoxib.2
Concerns about the cardiovascular safety of these drugs, however, have been raised for years.3- 6 It became clear in the mid-1990s that COX-2 selective inhibition might affect the hemostatic balance and favor thrombosis by selectively inhibiting the generation of COX-2–derived vascular prostacyclin while not affecting the COX-1–mediated generation of thromboxane A2.3,4 The concerns were further stimulated by data from the Vioxx Gastrointestinal Outcomes Research (VIGOR)7,8 trial, in which rofecoxib was associated with a 5-fold increase in the risk of MI compared with naproxen. However, the lack of a placebo arm and the possible antithrombotic effects of naproxen9- 11 made it difficult to interpret the results.12,13 Several questions remain unanswered. First, data on the risk associated with COX-2 selective inhibitors from well-designed observational studies may provide more relevant information on the safety of these drugs in clinical practice than the data from clinical trials,14- 19 since the number of high-risk patients (eg, patients with a history of cardiovascular disease) was small in most trials.6,14- 19 Second, the association between length of COX-2 selective inhibitor use and cardiovascular risk needs to be further examined because a recent meta-analysis found no association between duration of treatment and cardiovascular risk.6 Third, it is unclear whether the cardiovascular risk found for rofecoxib and celecoxib is specific or reflects a broader class effect.14- 21 Fourth, the unexpected cardiovascular risk associated with naproxen in an ongoing clinical trial on prevention of Alzheimer disease has extended concerns about safety to conventional nonaspirin NSAIDs.22 We therefore examined the risk of MI among users of COX-2 selective inhibitors and other nonaspirin NSAIDs in a Danish case-control study.
This study was conducted within North Jutland, Viborg, and Aarhus counties, Denmark (approximately 1.4 million persons) from January 1, 2000, to December 31, 2003. The National Health Service provides tax-supported health care for all inhabitants including costs of prescribed drugs. Unambiguous linkage between various registers can be performed using the civil registry number.
Hospital discharge registries in the counties retain data on all discharges from all nonpsychiatric hospitals since 1972 (Viborg County) or 1977 (North Jutland and Aarhus counties). The files include information on the civil registry number, date of discharge, and up to 20 discharge diagnoses and procedures,23 coded according to the International Classification of Diseases, Eighth Revision (ICD-8), until the end of 1993, and 10th revision (ICD-10) thereafter.23
Based on the discharge history, we identified patients with a first diagnosis of MI between ICD-10 codes: I21.0-I21.9 (n = 10 343). We excluded patients who had an address outside the counties, had been living within the counties for less than 1 year, or were younger than 20 years (n = 63). In total, 10 280 cases were available for analyses (Table 1).
Using the Civil Registration System, which retains data on vital status, address, and emigration for the Danish population since 1968, we aimed to select 10 controls for each case matched by age and sex (n = 102 797) using the risk set sampling24 (ie, the controls had to be alive and at risk of a first hospitalization for MI according to their discharge history at the time the corresponding case was diagnosed [index date]).
Data on prescriptions for nonaspirin NSAIDs were obtained from prescription databases maintained in the counties, which retain information on refundable drugs, including type of drug and date of prescription (date of dispensing of the drug). Data were available from 1991 (North Jutland County), 1996 (Aarhus County), and 1998 (Viborg County), respectively. Thus, complete coverage was ensured from 2000 through 2003.
All types of nonaspirin NSAIDs, except low-dose ibuprofen (200 mg per tablet), were available only by prescription. Although low-dose ibuprofen was available without prescription, pensioners and regular users of this drug are typically all registered in the databases because the cost is partly refunded when ibuprofen is prescribed by a physician.
Use of rofecoxib, celecoxib, other COX-2 selective inhibitors (ie, the COX-2 selective conventional NSAIDs etodolac, meloxicam, and nabumetone), naproxen, and other nonaspirin NSAIDs was assessed. We classified individuals according to their most recent use: current users (having filled a prescription within 0-30 days), new users (a subset of current users, who were defined as having filled their first prescription within 0-30 days), recent users (having filled a prescription within 31-90 days), former users (having filled a prescription >90 days before index date), or nonusers (no recorded prescriptions for any nonaspirin NSAID before index date). The minimum length of prescription history was 1 year. Persons with prescriptions for more than 1 of the drug categories within the most recent period of use were grouped in a separate category (data not presented). Users were also categorized according to the total number of filled prescriptions for each type of NSAID (ie, no prescriptions, 1-3 prescriptions, 4-9 prescriptions, and ≥10 prescriptions).
Data were obtained from the discharge and prescription registries based on the available history for each case and control. The discharge diagnoses included coronary heart disease (other than MI), stroke, peripheral arterial disease and coronary revascularization procedures (percutaneous coronary intervention or coronary by-pass operation), hypertension, diabetes mellitus, chronic bronchitis or emphysema (a proxy measure of smoking), alcoholism, liver cirrhosis, upper gastrointestinal track bleeding, rheumatoid arthritis, and systemic lupus erythematosus registered before the index date.
The prescription data included prescriptions for high-dose aspirin (200-500 mg), low-dose aspirin, dipyridamole, clopidogrel bisulfate, angiotensin-converting enzyme inhibitors, β-blockers, calcium antagonists, diuretics and angiotensin II receptor antagonists, insulin and oral hypoglycemic drugs, lipid-lowering drugs, warfarin, phenprocoumon, nitrates, penicillamine, gold, oral glucocorticoids, and postmenopausal hormone therapy (women) filled before the index date.
We used conditional logistic regression to compute relative risk estimates with 95% confidence intervals (CIs) for MI among each drug exposure category (see table footnotes for possible confounding factors included in the analyses). Nonusers were used as the reference group in all analyses.
For comparison, the risk of hospitalization for MI was also estimated among users of high-dose aspirin. The likelihood ratio test was used to compare the adjusted risk estimates. Furthermore, separate analyses were performed among new users and among persons at low and high risk of MI. Low risk was defined as no previous hospitalizations for cardiovascular disease, hypertension, diabetes mellitus or coronary revascularization, and no previous prescriptions for cardiovascular drugs, insulin, or oral hypoglycemic drugs. High risk was defined as at least 1 previous hospitalization for cardiovascular disease, hypertension, diabetes mellitus, or coronary revascularization or at least 1 previous prescription for cardiovascular drugs, insulin, or oral hypoglycemic drugs.
Finally, the association between length of use of the nonaspirin NSAIDs and risk of MI was examined by comparing the risk of MI according to the total number of filled prescriptions (ie, no prescriptions, 1-3 prescriptions, 4-9 prescriptions, and ≥10 prescriptions) separately for persons who had filled prescriptions for 1 (exclusive users) or more than 1 category (combined users) of nonaspirin NSAIDs. All analyses were conducted using SAS version 8.00 (SAS Institute Inc, Cary, NC).
Table 1 gives the characteristics of the cases and controls. As expected, cases were characterized by a more adverse cardiovascular risk profile compared with controls. Table 2 presents risk estimates for MI among users of rofecoxib, celecoxib, other COX-2 selective inhibitors, naproxen, and other nonaspirin NSAIDs. Current and new use of rofecoxib was associated with an increased risk estimate of MI compared with the nonuse of any category of nonaspirin NSAIDs with adjusted relative risks (ARRs) of 1.80 (95% CI, 1.47-2.21) and 2.52 (95% CI, 1.74-3.64), respectively. No increased risk estimates were apparent among recent or former rofecoxib users.
Similar patterns were seen for celecoxib, other COX-2 selective inhibitors, naproxen, and other nonaspirin NSAIDs. Adjusted relative risks estimates ranged from 1.25 for celecoxib to 1.80 for rofecoxib among current users (P = .03) and from 1.65 for naproxen to 3.37 for other COX-2 selective inhibitors among new users (P = .25) (Table 2). For new users, the risk estimates were similarly elevated for all nonaspirin NSAID categories except naproxen. Although not statistically significant, the risk appeared lower for recent and former users in all drug categories except celecoxib.
For comparison, elevated risk estimates were found among both current (ARR, 1.34; 95% CI, 1.18-1.52), recent (ARR, 1.26; 95% CI:1.13-1.40), and former users (ARR, 1.29; 95% CI, 1.13-1.48) of high-dose aspirin (data not shown).
The highest ARRs in both the low- and high-risk MI groups were found among current users of rofecoxib (2.77 and 1.58, respectively), whereas the lowest ARRs were found among current users of celecoxib (1.68 and 1.15, respectively) and other COX-2 selective inhibitors (1.80 and 1.13, respectively) (Table 3). We found no association between the number of filled prescriptions and risk of MI for any of the examined drug categories (Table 4).
We found elevated risk estimates for MI among current and in particular new users of rofecoxib and celecoxib. Elevated risk estimates were also found among current and new users of other COX-2 selective inhibitors, naproxen, other conventional nonaspirin NSAIDs, and high-dose aspirin. For current users, the lowest risk estimates were found for celecoxib and the highest for rofecoxib and other nonaspirin NSAIDs. For new users, the risk estimates were similarly elevated for all nonaspirin NSAID categories evaluated, except naproxen.
The strengths of our study are its size, the population-based design, and the ability to examine specific types of nonaspirin NSAIDs. Furthermore, the data on exposures and possible confounding factors were prospectively recorded, and extensive efforts were made to adjust the risk estimates for influence from possible confounding factors, in particular underlying inflammatory conditions.
The limitations include the use of discharge diagnoses, since it is well-known that these are not entirely accurate. However, the validity of an MI discharge diagnosis in Denmark is high because less than 10% of the cases are misclassified.25 Although we adjusted for a wide range of possible confounding factors, our results may still be confounded by diet and lifestyle factors, by channeling bias26 (eg, patients prescribed rofecoxib or celecoxib may have an elevated baseline risk of cardiovascular disease compared with patients being prescribed other categories of nonaspirin NSAIDs),27,28 or by the use of proxy measures. However, our findings appeared consistent after stratification according to baseline risk of MI.
The elevated risk estimates for all types of NSAIDs may to some extent reflect the existence of a protopathic bias, which occurs in observational studies when the indication for being prescribed a drug may be an unrecognized clinical manifestation related to the outcome in question (eg, angina pectoris misdiagnosed as chest pain of a musculoskeletal nature). Such a bias may have inflated the association between the studied drugs and the risk of MI and may thus partly underlie the pattern of clearly increased risk estimates found among current and new users of the examined categories of nonaspirin NSAIDs, including naproxen, vs a lower risk among recent and former users.
Other limitations included the lack of data on over-the-counter sale of NSAIDs (low-dose ibuprofen and aspirin), compliance, and duration of actual use of the prescribed drugs. However, the sales of low-dose ibuprofen constituted only approximately 14% of the total nonaspirin NSAID sales during the study period,29 and the fact that part of the costs of the prescribed drugs is paid by the patients is likely to have improved compliance. Moreover, we had no data on doses and were therefore unable to examine whether high doses of rofecoxib were associated with a particular high risk, as suggested by previous findings.7,14,15,19 Finally, the precision of several of our risk estimates was admittedly moderate owing to relatively low proportions of exposed cases and controls, so caution should be taken before drawing conclusions.
The excess overall risk of MI among users of rofecoxib is in accordance with the data from APPROVe,3 from a meta-analysis of trials of rofecoxib in patients with chronic musculoskeletal disorders,6 and from 3 observational studies.14,15,19 The particular high-risk estimates found among new users were also in accordance with previous studies and do not support the hypothesis of no excess risk within the first 18 months of use as reported from the APPROVe trial.3 The differences in the risk estimates between new users and all current users could be explained by early events in the patients most susceptible to MI, different induction periods of beneficial and adverse effects of the studied drugs, physiologic adaptation that may occur during prolonged periods of treatment, or selection factors, including adherence bias, which vary according to treatment duration.28
Because rofecoxib has already been withdrawn from the market, the focus should be shifted to the safety of the other nonaspirin NSAIDs, in particular the other COX-2 selective NSAIDs. This interest has been further fueled by the recent publication of the Therapeutic Arthritis Research and Gastrointestinal Event Trial (TARGET), in which lumiracoxib, the most selective COX-2 inhibitor available, was compared with ibuprofen or naproxen.21 A nonsignificant increased MI risk associated with use of lumiracoxib compared with naproxen (relative risk, 1.77; 95% CI, 0.82-3.84) but not ibuprofen (relative risk, 0.66; 95% CI: 0.21-2.09) were seen. Until recently, to our knowledge, no studies had found an increased risk of MI or other cardiovascular events among users of celecoxib compared with users of placebo, users of other nonaspirin NSAIDs, or nonusers of nonaspirin NSAIDs.14- 20,29- 34 The excess risk among celecoxib users compared with nonusers of nonaspirin NSAIDs in our study is, however, in accordance with the recently halted APC trial, in which a significant increase in the risk of cardiovascular events (a composite end point of cardiovascular death, MI, and stroke) has been found among patients randomized to a daily dose of 400 or 800 mg of celecoxib.2 However, it should be noted that the doses in the APC trial are high compared with currently approved dosing regimens for the most common indications (eg, osteoarthritis and rheumatoid arthritis) and that no increased cardiovascular risk so far has appeared in 2 other ongoing trials of celecoxib, similar in size and duration to the APC trial.2
A number of traditional nonaspirin NSAIDs (including etodolac, meloxicam, and nabumetone) also demonstrate a high degree of COX-2 selectivity, and the cardiovascular safety data are sparse. However, the similarities in pharmacology with rofecoxib and the particular high risk of MI among new users of these drugs found in our study raises new concerns.
Until recently, naproxen has not been associated with an elevated cardiovascular risk compared with either no use or the use of other nonaspirin NSAIDs.5,8- 10,35,36 However, a National Institutes of Health review recently revealed an unexpected increased risk of cardiovascular events among patients randomized to naproxen in the Alzheimer Disease Anti-inflammatory Prevention Trial (ADAPT).22 Furthermore, an elevated risk of MI (adjusted odds ratio, 1.14; 95% CI, 1.00-1.30) among current users of naproxen was also reported in a recent case-control study.19 Although our risk estimates for naproxen were not significantly elevated, our estimates were higher than previously reported and thus provide some support for the concerns regarding naproxen.
Aspirin, in particular small doses, is considered a cardioprotective agent. The elevated risk estimates found in our study for use of high-dose aspirin may therefore reflect uncontrolled confounding because the risk was increased among former users. However, the cardiovascular safety of persistent high-dose aspirin use, which inhibits both COX-1 and COX-2, has not been well examined, and at least 1 study had found that high-dose aspirin use was associated with an increased risk of ischemic heart disease.37 A causal association can therefore not be excluded.
In conclusion, we found that rofecoxib and celecoxib were associated with an excess risk of MI. However, risk estimates were also increased among users of other COX-2 selective inhibitors and other categories of nonaspirin NSAIDs. The risk estimates appeared lower for celecoxib compared with rofecoxib. The overall increased relative risk of MI associated with use of nonaspirin NSAIDs in general may to some extent reflect a protopathic bias; however, continued attention to the cardiovascular safety of all nonaspirin NSAIDs appears warranted.
Correspondence: Søren P. Johnsen, MD, PhD, Department of Clinical Epidemiology, Aarhus Hospital, Aarhus University Hospital, Ole Worms Allé 150, DK-8000 Aarhus C, Denmark (firstname.lastname@example.org).
Accepted for Publication: February, 2, 2005.
Financial Disclosure: None.
Funding/Support: The study was supported by a grant from the Western Danish Research Forum for Health Sciences, Aarhus, Denmark, and funding from the International Epidemiology Institute, Rockville, Md.