OUP user menu

Cyclooxygenase-2 inhibitors and cardiovascular risk in a nation-wide cohort study after the withdrawal of rofecoxib

Magnus Bäck, Li Yin, Erik Ingelsson
DOI: http://dx.doi.org/10.1093/eurheartj/ehr421 1928-1933 First published online: 21 November 2011


Aims The use of selective cyclooxygenase (COX)-2 inhibitors (coxibs) has been associated with an increased cardiovascular risk. The aim of the present study was to evaluate the association of coxib use and future risk of cardiovascular events in a population-based cohort followed after the warnings concerning the cardiovascular safety of this class of drugs were issued.

Methods and results A nation-wide, population-based cohort of 7 million subjects, integrating data from the Prescribed Drug, Patient, Cause of Death, Income, Educational and Emigration Registers, was followed from 1 July 2005 to 31 December 2008. Analyses were performed for different cardiovascular outcomes in the whole population after exclusion of individuals with prior cardiovascular diagnosis (incident primary cardiovascular events; sample size, n = 6 991 645). Cox proportional hazard ratios (HRs) revealed no significant association of coxib use with risk for myocardial infarction, ischaemic stroke, or heart failure. In contrast to these findings, coxib use was associated with an increased risk for a first episode of atrial fibrillation [HR 1.16; 95% confidence interval (CI) 1.05–1.29]. A post hoc analysis for different coxibs revealed a significant association with incident atrial fibrillation for etoricoxib (HR 1.35; 95% CI 1.19–1.54) but not for celecoxib (HR 0.94; 95% CI 0.79–1.11).

Conclusion Whereas safety measures appear to have limited serious cardiovascular consequences of COX-2 inhibitors, the risk of developing atrial fibrillation may have been overlooked and may necessitate consideration and precautions.

  • Atrial fibrillation
  • Coronary heart disease
  • Cerebrovascular disease
  • Eicosanoids
  • Heart failure
  • Inflammation
  • Myocardial infarction
  • Prostaglandins


Non-steroidal anti-inflammatory drugs (NSAIDs) are inhibitors of cyclooxygenase (COX), the enzyme responsible for the initial step in the metabolism of arachidonic acid into prostaglandins and thromboxane. Whereas the cardioprotective effects of aspirin are well established, the cardiovascular safety of other NSAIDs, and especially for the selective inhibitors of the COX-2 isoform (coxibs, cyclooxygenase-2 inhibitors), has remained controversial. Several randomized controlled trials (RCTs) of long-term use of selective coxibs indicated an increased risk of myocardial infarction, stroke, and heart failure compared with either other NSAIDs1 or placebo.24 Subsequent meta-analyses5,6 and registry studies710 extended those findings and also raised a notion of differential risk profiles within the NSAID group of medications.

Although previous studies of the cardiovascular safety of coxibs focused on myocardial infarction, stroke, and heart failure, the role of these drugs in atrial fibrillation has received less attention. Two recent case–control studies of patients with atrial fibrillation have suggested an increased risk associated with the use of NSAIDs.11,12 However, the follow-up periods of these studies included time both before and after the cardiovascular side effects of COX-2 inhibitors got widely accepted.

The above-mentioned studies of increased cardiovascular risk associated with selective COX-2 inhibitors led to the world-wide withdrawal of rofecoxib on 30 September 2004, and to FDA recommendations on 7 April 2005 for precautions in terms of the cardiovascular safety of coxibs and other NSAIDs. Importantly, all hitherto published studies on coxib-associated cardiovascular risk and atrial fibrillation have included data collected before these dates. Therefore, the aim of the present study was to determine the cardiovascular risk associated with the use of COX-2-selective NSAIDs in a cohort studied after the cardiovascular risk profile of coxibs got widely accepted. To this end, a nation-wide, population-based cohort of 7 million subjects, containing data on dispensed drugs, diagnosis (hospitalization or out-patient care at hospital), and causes of death was used to monitor the associations between coxib exposure and cardiovascular events between 1 July 2005 and 31 December 2008.



The cohort consisted of all Swedish residents above the age of 18 on 1 July 2005, identified through their unique personal identification number. The follow-up started on 1 July 2005 and the end of follow-up was on 31 December 2008 or the date of first occurrence of the following: the respective cardiovascular event (as defined in what follows), emigration from Sweden, or death. Data were collected from the following national registries: the Swedish Prescribed Drug Register, Patient Register, Cause of Death Register, Educational and Income Registers and the Registry of Emigrations and Immigrations, maintained at the National Board of Health and Welfare (http://www.socialstyrelsen.se/) and Statistics Sweden (http://www.scb.se). The study was approved by the Ethics Committee of Karolinska Institutet, Stockholm, Sweden.

According to the a priori analysis plan, the present study addressed the associations of coxibs as a group with cardiovascular events. A post hoc analysis was subsequently performed for the individual drugs within this class. Individuals without prior diagnosis were followed for incident events, and individuals with a prior diagnosis were characterized for coxib exposure and for calculations of prevalence (n = 154 027 for prior myocardial infarction; n = 132 291 for prior ischaemic stroke; n = 73 233 for prior heart failure; n = 133 423 for prior atrial fibrillation). Prior events were assessed from the Patient Register for either of the following diagnoses: myocardial infarction [International Classification of Diseases (ICD) -8 and ICD-9 code 410, ICD-10 codes I21]; ischaemic stroke (ICD-8: 432–437; ICD-9: 4330–4379; ICD-10: I63); heart failure (ICD-8: 42700; ICD-9: 428; ICD-10: I50); or atrial fibrillation (ICD-8: 42792; ICD-9: 427D; ICD-10: I48). The Patient Register has nation-wide coverage of all hospitalizations since 1987 and all outpatient visits to specialist care from 2001 and on. The total sample sizes for each outcome are shown in Table 1.

View this table:
Table 1

Baseline characteristics of samples according to the outcome

OutcomeMyocardial infarctionIschaemic strokeHeart failureAtrial fibrillation
n6 771 109139 8136 792 406140 1716 850 263141 3826 792 112139 323
Mean age (years)49.7755.0349.8055.0650.0155.1849.8554.98
Females (%)51.1060.2750.7859.9150.7759.9150.8760.23
Educational level (%)
Income (%)
Rheumatoid arthritis (%)0.612.340.622.340.622.360.622.35
Drug prescriptions (%)
 Low-dose ASA11.3214.7511.8615.2512.4215.8412.2315.52
 Oral steroids9.8523.249.9123.249.8823.219.8523.15
 TNFα inhibitors0.170.980.170.980.170.980.170.98
 Calcium antagonists8.2912.418.3012.428.6012.688.3512.35
 Lipid-lowering drugs11.0417.9711.768.6612.0619.0311.9118.77
  • ASA, acetyl salicylic acid; NSAIDs, non-coxib, non-aspirin, non-steroidal anti-inflammatory drugs; ACE-inh/ARB, angiotensin-converting enzyme-inhibitors and/or angiotensin receptor blockers.

Medication exposure

From the national Prescribed Drug Register, all claimed prescriptions of COX-2 inhibitors [Anatomical Therapeutical Chemical (ATC) code M01AH] were identified in the study cohort between 1 July 2005 and 31 December 2008. In addition, data on the use of cardiovascular drugs and drugs used for the treatment of rheumatoid arthritis were collected and included in the analysis as described in what follows. Coxib use was analysed as a time-dependent variable, splitting the risk time according to non-use (unexposed) and use (exposed). Calculated from the dispense date, individuals were considered exposed for 4 months based on the prescription length of 3 months of use.

Follow-up and outcome

The following outcomes were monitored in separate analyses: acute myocardial infarction (ICD-10 codes I21), ischaemic stroke (I63), heart failure (I50), and atrial fibrillation (I48). These diagnoses were defined according to the primary diagnosis or the primary cause of death, as recorded in the Patient and Cause of Death Registers. The Patient Register includes hospitalized cases, as well as outpatient visits, but not visits to the primary care. The positive predictive values (i.e. validity) of the myocardial infarction and ischaemic stroke diagnoses in the Swedish Patient register have been demonstrated to be ∼95% when only primary diagnoses are considered.13,14

Co-morbidity and socioeconomic status

Information about diagnosis of rheumatoid arthritis was collected from the Patient Register as a proxy for disease severity, which could be considered a confounder of studied associations between coxibs and cardiovascular disease. Information on the use of different drugs for rheumatoid arthritis and cardiovascular disease was also collected from the Prescribed Drug Register, and used as proxies for these respective diseases which also could be considered potential confounders. Data were collected on the use of traditional non-selective NSAIDs (ATC-code M01A except M01AH), low-dose acetyl salicylic acid (B01AC06), oral steroids (H02), TNFα inhibitors (L04AB01, L04AB02, L04AB04), ACE-inhibitors or angiotensin receptor antagonists (C09), beta-blockers (C07), calcium blockers (C08), diuretics (C03), and lipid-lowering drugs (C10). These variables (diagnoses and drug use) were modelled as time-dependent covariates, where individuals where considered non-affected until first occurrence of respective covariate, but thereafter affected.

Educational level and total yearly income were collected for year 2005 and used as a proxy for socioeconomic status. Educational level was categorized into three groups (high = college or university studies, middle = 2 or 3 years of high school, low = 0–9 years of primary and secondary school), and yearly income (from employment, capital, real estate, and business activities) was categorized into tertiles.

Statistical analysis

The associations between coxib use and the four outcomes (myocardial infarction, ischaemic stroke, heart failure, or atrial fibrillation) were examined by calculating unadjusted incidence rates [i.e. number of events per 1000 person-years at risk (PYAR)] with 95% confidence intervals (CIs) based on the Poisson distribution. Cox proportional hazard models were used to estimate the hazard ratios (HRs) of the drug use, with either partial adjustment for age, gender, income, and education, or with full adjustment, which also included the presence of rheumatoid arthritis and the use of drugs for rheumatoid arthritis and cardiovascular disease. Two-tailed 95% CIs and P-values were given, with P < 0.05 regarded as significant. The statistical software package SAS 9.2 (SAS Institute, Inc., Cary, NC, USA) was used for all analyses.


The baseline characteristics by exposure are shown in Table 1 for the myocardial infarction, ischaemic stroke, heart failure, and atrial fibrillation study samples. Based on the information of prior diagnosis from the Patient Registry, the prevalence of cardiovascular disease in the whole adult Swedish population was 2.2% for previous myocardial infarction, 1.9% for previous ischaemic stroke, 1.0% for previous heart failure, and 1.9% for previous atrial fibrillation.

The dispensed COX-2 inhibitors were 41.8% celecoxib (ATC-code M01AH01) and 58.2% etoricoxib (ATC-code M01AH05). No other coxibs were available for prescription during the study period. The average daily dose was 187 ± 213 mg for celecoxib, and 39 ± 48 mg for etoricoxib. The average length of treatment was 8.14 ± 8.34 months for celecoxib, and 6.32 ± 5.66 months for etoricoxib. The coxib exposure was significantly lower in subjects with a prior cardiovascular event (prior myocardial infarction 1.69%; prior ischaemic stroke 1.68%; prior heart failure 1.32%; prior atrial fibrillation 1.65%) compared with those not having had a prior event (2.0%; P< 0.0001). Rheumatoid arthritis, a common indication for coxib use, was significantly more prevalent in subjects with previous events in all study samples (prior myocardial infarction 1.60%; prior ischaemic stroke 1.32%; prior heart failure 1.91%; prior atrial fibrillation 1.45%) compared with those not having had a prior event (0.65%; P< 0.0001).

The event rates for incident events are shown in Table 2. These analyses revealed an association of coxib use with an increased risk for myocardial infarction, ischaemic stroke, heart failure, and atrial fibrillation when data were adjusted only for age, sex, and socioeconomic status (Table 2). However, in the fully adjusted analysis, taking into account potential confounders as described in Methods and listed in Table 1, only the association of coxib use and an increased risk for atrial fibrillation remained significant (Table 2).

View this table:
Table 2

Hazard ratios for incident myocardial infarction, ischaemic stroke, heart failure, and atrial fibrillation according to cyclooxygenase-2 inhibitor use in subjects without prior diagnosis (non-use was used as reference level)

Number of events (rate/1000 PYAR)Partially adjustedaFully adjustedb
HR (95% CI)P-valueHR (95% CI)P-value
Myocardial infarction (total number of incident events 78 843)
 3.26 (3.24–3.28)1.190 (1.075–1.317)<0.011.055 (0.953–1.168)0.30
Ischaemic stroke (total number of incident events 61 723)
 2.54 (2.52–2.56)1.154 (1.030–1.294)0.011.106 (0.986–1.240)0.08
Heart failure (total number of incident events 60 848)
 2.48 (2.46–2.50)1.146 (1.018–1.291)0.021.028 (0.913–1.158)0.64
Atrial fibrillation (total number of incident events 67 684)
 2.79 (2.77–2.81)1.275 (1.149–1.416)<0.011.161 (1.046–1.289)0.01
  • aAdjusted for age, sex, and socioeconomic status.

  • bAdjusted for all parameters listed in Table 1.

In the fully adjusted analysis, the increased risk of incident atrial fibrillation was also associated with the use of either traditional non-coxib, non-aspirin NSAIDs (HR 1.11, 95% CI 1.09–1.13) or oral steroids (HR 1.48, 95% CI 1.44–1.51). A post hoc analysis of individual coxibs revealed an increased risk for incident atrial fibrillation and heart failure associated with etoricoxib exposure in the fully adjusted analysis, whereas no significant associations were observed for celecoxib (Table 3).

View this table:
Table 3

Post hoc analysis showing fully adjusted (i.e. adjusted for all parameters listed in Table 1) hazard ratios for incident myocardial infarction, ischaemic stroke, heart failure, and atrial fibrillation for celecoxib and etoricoxib in subjects without prior diagnosis (non-use was used as reference level)

HR (95% CI)P-valueHR (95% CI)P-value
Myocardial infarction1.075 (0.926–1.248)0.34011.034 (0.899–1.188)0.6399
Ischaemic stroke1.138 (0.962–1.346)0.13061.084 (0.928–1.267)0.3066
Heart failure0.843 (0.697–1.019)0.07681.205 (1.035–1.402)0.0116
Atrial fibrillation0.936 (0.786–1.114)0.45431.353 (1.189–1.540)<0.0001


Two major observations emerge from the present study. First, in this large nation-wide population-based cohort, initiated after the increased cardiovascular risk of selective COX-2 inhibitors was revealed, coxib use was not associated with incident events of either myocardial infarction or ischaemic stroke. Second, and in contrast to those findings, the results provide evidence for the association of selective COX-2 inhibitors with incident atrial fibrillation.

Although the partially adjusted analysis in the present study indicated a significantly increased risk for incident myocardial infarction and ischaemic stroke, these observations did not persist after adjustment for rheumatoid and cardiovascular co-morbidities. The significantly increased prevalence of rheumatoid arthritis in the subjects with a medical history of cardiovascular disease in the present study supports an increased cardiovascular risk associated with rheumatoid arthritis,15 a patient group commonly prescribed COX-2 inhibitors. The latter findings underline the importance of fully adjusting the data for rheumatoid co-morbidity and drug use (including oral steroids). Furthermore, the concomitant use of coxibs and aspirin in some but not in other RCTs of different coxibs has been suggested as one possible explanation for the diverging cardiovascular outcomes between different studies.1,16

In contrast to these neutral effects of coxibs on myocardial infarction and ischaemic stroke, an increased incidence of atrial fibrillation was established in the present study being significant both in the partially and fully adjusted analyses. Previous reports have associated the use of rofecoxib, but not celecoxib, with arrhythmia events, when addressing mainly ventricular arrhythmias and cardiac arrest.4,17 Recently, a nested case–control study reported that the use of traditional NSAIDs was associated with chronic, but not paroxysmal atrial fibrillation.11 In order to assess whether the observed effect of coxibs on atrial fibrillation was specific for COX-2-selective NSAIDs in the present study, fully adjusted HRs were calculated also for the association of traditional non-selective NSAIDs with atrial fibrillation. The latter analysis replicated the previous findings11 by revealing a significantly increased risk of incident atrial fibrillation associated with the use of traditional NSAIDs, although with a lower HR than that observed for coxib users. However, an accurate comparison between unselective, traditional NSAIDs and selective COX-2 inhibitors is limited by the large dispensation of traditional NSAIDs without prescription, and which therefore cannot be monitored in the registries used for the present study. Nevertheless, the post hoc analysis of individual coxibs in the present study supported that the highest risk for developing atrial fibrillation was associated with etoricoxib, which is more COX-2-selective compared with celecoxib.

Since heart failure is an important risk factor for developing atrial fibrillation, coxib use could potentially reflect an increased incidence of heart failure, which has been previously described for coxibs.4 However, coxib use was not associated with an increased risk of incident episodes of heart failure, which was in contrast to findings for atrial fibrillation and in line with the findings for myocardial infarction and stroke. Nevertheless, the post hoc analysis indicated that etoricoxib, but not celecoxib, exposure moderately increased the risk of heart failure.

Although systemic inflammation has been associated with an increased risk for atrial fibrillation,18,19 the results of the present study indicate that coxibs were not beneficial, despite their anti-inflammatory effects. The mechanisms behind the cardiovascular side effects of selective COX-2 inhibition have been proposed to involve a disturbed balance between pro-aggregatory thromboxane and anti-aggregatory prostacyclin,20 altered renal sodium and water re-absorption,17,21 unfavourable alterations of vascular reactivity,22 and recently also immuno-modulatory effects aggravating atherosclerosis development.23 However, the exact mechanisms behind the pro-arrhythmic consequences of inhibiting the COX-2 pathway, as suggested by the association of coxib use with atrial fibrillation in the present study, remain to be established. Nevertheless, experimental studies have indicated prostaglandin production in the coronary sinus24 and pericardial fluid25,26 after experimental myocardial ischaemia and reperfusion. In the latter context, prostacyclin, which represents one major COX-2-derived prostaglandin, was demonstrated to act as an endogenous anti-arrhythmic through direct inhibition of epicardial sympathetic nerve activity.2527 Recently, selective deletion of cardiomyocyte COX-2 expression in mice was in addition reported to induce interstitial and perivascular fibrosis, associated with an enhanced susceptibility to induced arrythmias.28 Taken together, those experimental studies support a causal relationship between COX-2 inhibition and atrial fibrillation. Nevertheless, coxibs may in addition induce direct on ion channels and intracellular signalling pathways which are unrelated to their COX-2 inhibition.29,30

The strengths of the present study include the very large size and population-based nature of this nation-wide cohort of more than 7 million subjects—the prospective design with a large number of events allowing studies of different types of cardiovascular disease and the full coverage of the registries giving unbiased measures of exposure and confounders as well as outcomes. Several limitations should, however, be acknowledged. In these national registry data, we are not able to completely adjust for all possible confounding factors, such as blood pressure, lipid levels, or smoking. However, the most plausible confounders have been taken into account through adjustment for treatment with anti-hypertensive and lipid-lowering drugs in the present analysis. Likewise, because previous studies have established the correlation between smoking and socioeconomic status,31 the adjustment for educational level and income in the present study served also as a proxy for smoking. Furthermore, the outcomes of the present study were based on the Patient Register, which includes specialist in- and outpatient care, but not diagnoses from the primary care. The result of including only individuals with a diagnosis by a hospital specialist is likely to increase the specificity and positive predictive value of the diagnosis (with lower sensitivity and negative predictive value). However, the type of atrial fibrillation (paroxysmal, persistent, or permanent) cannot be defined. Since atrial fibrillation may be commonly under-diagnosed, as suggested by studies of prolonged ECG recordings through, for example, Holter monitoring or pacemaker,32 the generalizability of our findings to individuals not seeking hospital care is hence unknown. Finally, another missing variable is the drug exposure before the start of study, making the length of treatment difficult to assess, and that previously exposed subjects may have been considered unexposed. However, in this context, it is important to stress that previous studies have indicated highest risks associated with current coxib use compared with remote use10 and that the present time-dependent analysis addressed the effective period of influence due to current use of coxibs. Even if we acknowledge that these kinds of registry data have limitations, it should be emphasized that they also serve as a complement to randomized clinical trials and longitudinal cohort studies given the very large sample size drawn from the general population. This means high generalizability to the whole population, and a large sample size that makes it possible to detect small effects of uncommon exposures. However, that said, the non-randomized, observation design of the present study means that no causality can be definitely attributed and that associations could be driven by unknown confounding.

In summary, the present recent nation-wide cohort, initiated after the cardiovascular safety of coxibs was taken into consideration, indicates no significant association of coxibs with increased risk of myocardial infarction or ischaemic stroke. In contrast, the recent notion of increased risk of atrial fibrillation by NSAIDs11 was replicated and extended to indicate that COX-2 inhibitors are associated with incident atrial fibrillation. Additional studies including patients with milder forms of atrial fibrillation not seeking specialist care are needed to further characterize this observation. One possible interpretation of the present data is that the increased awareness of the serious cardiovascular side effects of coxibs after the withdrawal of rofecoxib and the warnings issued for NSAIDs as a class of drugs have led to precautions of their use in patients with, or with risk factors for, coronary and cerebrovascular disease. In contrast, a possible increased risk of atrial fibrillation may have been overlooked, and may necessitate consideration and precautions.


This work was supported by The Craaford Foundation and The Swedish Society of Medicine. M.B. was supported by a grant from the Swedish Heart and Lung Foundation. E.I. was supported by grants from the Swedish Research Council and the Swedish Foundation for Strategic Research.

Conflicts of interest: none declared.


View Abstract