- Morten Schmidt, junior research fellow1,
- Christian F Christiansen, senior registrar1,
- Frank Mehnert, biostatistician1,
- Kenneth J Rothman, professor23,
- Henrik Toft Sørensen, professor1
- 1Department of Clinical Epidemiology, Aarhus University Hospital, 8200 Aarhus N, Denmark
- 2RTI Health Solutions, Research Triangle Institute, Research Triangle Park, NC, USA
- 3Department of Epidemiology, Boston University School of Public Health, Boston, MA, USA
- Correspondence to: M Schmidt
- Accepted 4 March 2011
Objectives To examine the risk of atrial fibrillation or flutter associated with use of non-selective non-steroidal anti-inflammatory drugs (NSAIDs) or selective cyclo-oxygenase (COX) 2 inhibitors.
Design Population based case-control study using data from medical databases.
Setting Northern Denmark (population 1.7 million).
Participants 32 602 patients with a first inpatient or outpatient hospital diagnosis of atrial fibrillation or flutter between 1999 and 2008; 325 918 age matched and sex matched controls based on risk-set sampling.
Main outcome measures Exposure to NSAID use at the time of admission (current use) or before (recent use). Current use was further classified as new use (first ever prescription redemption within 60 days before diagnosis date) or long term use. We used conditional logistic regression to compute odds ratios as unbiased estimates of the incidence rate ratios.
Results 2925 cases (9%) and 21 871 controls (7%) were current users of either non-selective NSAIDs or COX 2 inhibitors. Compared with no use, the incidence rate ratio associating current drug use with atrial fibrillation or flutter was 1.33 (95% confidence interval 1.26 to 1.41) for non-selective NSAIDs and 1.50 (1.42 to 1.59) for COX 2 inhibitors. Adjustments for age, sex, and risk factors for atrial fibrillation or flutter reduced the incidence rate ratio to 1.17 (1.10 to 1.24) for non-selective NSAIDs and 1.27 (1.20 to 1.34) for COX 2 inhibitors. Among new users, the adjusted incidence rate ratio was 1.46 (1.33 to 1.62) for non-selective NSAIDs and 1.71 (1.56 to 1.88) for COX 2 inhibitors. Results for individual NSAIDs were similar.
Conclusions Use of non-aspirin NSAIDs was associated with an increased risk of atrial fibrillation or flutter. Compared with non-users, the association was strongest for new users, with a 40-70% increase in relative risk (lowest for non-selective NSAIDs and highest for COX 2 inhibitors). Our study thus adds evidence that atrial fibrillation or flutter needs to be added to the cardiovascular risks to be considered when prescribing NSAIDs.
Non-steroidal anti-inflammatory drugs (NSAIDs) are widely used to treat inflammatory conditions and pain.1 By inhibiting cyclo-oxygenase (COX)-1 mediated production of prostaglandins,1 non-selective NSAIDs are known to cause gastrointestinal toxicity1 and a variety of nephrotoxic syndromes.2 An alternative is selective COX 2 inhibitors, available in the form of older or newer agents.3 The newer COX 2 inhibitors, introduced into clinical practice in 1998, were developed as NSAIDs with an improved gastrointestinal side effect profile.1 The cardiovascular safety of all marketed newer COX 2 inhibitors requires thorough evaluation in view of the increased cardiovascular4 5 6 and renal risk2 reported for several of these drugs.
Atrial fibrillation is the most common rhythm disorder observed in clinical practice. It more than doubles in prevalence during each advancing decade of life, from 0.5% at age 50-59 years to above 10% at age 80-89 years.7 It is associated with increased mortality and morbidity, mainly due to haemodynamic impairments that exacerbate or even cause heart failure,8 and a threefold to fourfold increased risk of thromboembolic stroke.9
Use of NSAIDs may increase the risk of atrial fibrillation through its adverse renal effects—for example, fluid retention, electrolyte disturbances, and blood pressure destabilisation 2 6—but the evidence for such effects is limited.10 11 Although no original research has been published on COX 2 inhibitors and atrial fibrillation, a meta-analysis summarised data from 114 clinical trials and found that rofecoxib was associated with an increased risk of cardiac arrhythmias (relative risk 2.90, 95% confidence interval 1.07 to 7.88).10 Because the meta-analysis included only 286 incident arrhythmias, precision was low and risk of arrhythmia subtypes such as atrial fibrillation could not be examined.10 Recently, traditional NSAIDs (that is, non-selective NSAIDs and older COX 2 inhibitors) have been associated with increased risk of chronic atrial fibrillation (incidence rate ratio 1.44, 1.08 to 1.91).11
Any confirmed association between use of NSAIDs and atrial fibrillation would have major clinical and public health implications. Older people are of special concern because the prevalence of use of NSAIDs and the incidence of atrial fibrillation increase with age. To address the limitations of the existing literature, we conducted a large population based case-control study examining whether and to what extent use of NSAIDs increases the risk of atrial fibrillation or flutter.
We conducted this population based case-control study in northern Denmark, which has 1.7 million inhabitants (30% of the Danish population). Since 1998 complete computerised prescription records have been available for this population.12 Our study period encompassed 1 January 1999 to 31 December 2008, which yielded at least one year of prescription history for all study participants.
The Danish National Health Service provides universal tax supported healthcare, guaranteeing unfettered access to general practitioners and hospitals and partial reimbursement for prescribed medications, including NSAIDs.13 Most patients with atrial fibrillation or flutter are diagnosed during a hospital admission or at a hospital outpatient clinic.14 Very few cardiologists work outside the public hospital system in Denmark. Linkage among national registries is possible using the unique central personal registry number assigned to each Danish citizen at birth and to residents on immigration.15
Patients with atrial fibrillation or flutter
We used the Danish National Registry of Patients,16 covering all non-psychiatric hospitals since 1977 and emergency room and outpatient clinic visits since 1995, to identify all patients with a first time inpatient or outpatient diagnosis of atrial fibrillation or flutter during the study period. Because atrial fibrillation and flutter share risk factors and to some degree pathophysiology,17 18 we collapsed them into one disease entity.17 18 More than 90% of patients registered with these codes had atrial fibrillation.19 We considered the date of the first diagnosis of atrial fibrillation or flutter to be the index date for cases.
We used the Danish Civil Registration System to select 10 population controls for each case, matched for age and sex.15 This registry has recorded vital statistics for the Danish population since 1968 with daily updates.15 We selected controls using risk set sampling.20 Controls were assigned an index date identical to that of corresponding cases.
Non-steroidal anti-inflammatory drug use
We used the prescription database in the region12 to identify prospectively all prescriptions of NSAIDs redeemed by cases and controls before their index date. Except for ibuprofen in the 200 mg tablet dose, all non-aspirin NSAIDs are available by prescription only.13 Regular users of ibuprofen typically are registered in the database because the cost automatically is partly refunded when the drug is prescribed by a doctor.13
We identified prescriptions for non-aspirin non-selective NSAIDs (ibuprofen, naproxen, ketoprofen, dexibuprofen, piroxicam, and tolfenamic acid), older COX 2 inhibitors (diclofenac, etodolac, nabumeton, and meloxicam), and newer COX 2 inhibitors (celecoxib, rofecoxib, valdecoxib, parecoxib, and etoricoxib).3 21 Because of overlapping COX 2 selectivity, we collapsed the groups of older and newer COX 2 inhibitors into one group.3 Associated ATC (Anatomical Therapeutic Chemical Classification System) codes are provided in the web appendix.
We defined current users of NSAIDs as people who redeemed their most recent prescription within 60 days before their index date. We chose an exposure window of 60 days to capture most current users, as prescriptions of NSAIDs are seldom provided for more than 60 days at a time in Denmark.22 Some side effects may arise shortly after starting treatment2 6 and inclusion of long term users, who are more likely to tolerate the drug, could lead to underestimation of the association with atrial fibrillation or flutter.23 We therefore categorised current users into two groups: new users, defined by having redeemed their first ever prescription within 60 days before the index date, and long term users, defined by having redeemed their first prescription more than 60 days before the index date. We defined people who had redeemed their most recent prescription 61-365 days before the index date as recent users. We defined people with no redeemed prescriptions 365 days before their index date as non-users (reference group).
Because a number of risk factors for atrial fibrillation or flutter can also be associated with use of NSAIDs,24 25 we obtained data from the Danish National Registry of Patients on any previous hospital diagnosis since 1977 of diseases that may increase the risk of atrial fibrillation or flutter (listed in table 1⇓).24 25 To increase the sensitivity of the diagnoses, we used the prescription database12 to obtain data on any use since 1998 of relevant drugs. Furthermore, we identified current use of oral glucocorticoids, because these are associated with increased risk of atrial fibrillation or flutter.26 Associated ICD (International Classification of Diseases) and ATC codes are provided in the web appendix.
Initially, we created contingency tables for the main study variables from which we calculated the frequency of cases and controls in categories of exposure and other variables. We then stratified the contingency tables according to each of the potential confounding factors listed in table 1⇑.27 Next we used conditional logistic regression to compute odds ratios for atrial fibrillation or flutter among current, new, long term, and recent users of non-selective NSAIDs or COX 2 inhibitors.28 Current users of both subclasses of the drugs were treated as a separate group. Because we used risk set sampling of controls, the odds ratios estimated the incidence rate ratios.28 We fitted models controlling for the potential confounding factors listed in table 1. We repeated the analyses in predefined subgroups of sex, age, and presence or absence of cardiovascular disease, chronic kidney disease, osteoarthritis, rheumatoid arthritis, or systemic connective tissue disease. In the stratified analysis, we disregarded the matching and performed unconditional logistic regression with additional adjustments for the matching factors. We repeated the overall analysis for the six most frequently prescribed NSAIDs. To evaluate clinically relevant heterogeneity across drugs, we then compared individual NSAIDs directly using ibuprofen as the referent exposure. Because all patients needed pain relief, this comparison was likely to reduce confounding by indication. We used the tablet dose from the last redeemed prescription as a proxy for the total daily dose and examined the effect associated with low and high tablet dose.
In four secondary analyses we restricted cases to patients with atrial fibrillation or flutter: who had their diagnosis listed as the first diagnosis in the discharge summary, thereby detecting the potential effect of diagnostic surveillance bias among NSAID users;28 who had never redeemed a prescription for digoxin or a vitamin K antagonist before their index date, thereby excluding patients with atrial fibrillation or flutter treated by their general practitioner with no previous hospitalisation; who underwent cardioversion within one year after the index date, thereby relating use of NSAIDs to disease severity; or who had no cancer, chronic obstructive pulmonary disease or asthma, inflammatory bowel disease, rheumatoid or psoriatic arthritis, or systemic connective tissue disease, thereby reducing confounding from systemic inflammation. Finally, using a rule-out approach,29 we estimated how strongly a single unmeasured binary confounder would need to be associated with use of NSAIDs and atrial fibrillation or flutter to fully explain our findings.29
Descriptive data are presented in table 1 for the 32 602 cases and 325 918 population controls (web table 1 divides cases and controls according to their use of NSAIDs). Among the cases, 27 984 (85.8%) were diagnosed with atrial fibrillation or flutter during hospital admission, 4220 (12.9%) at an outpatient clinic, and 398 (1.2%) at an emergency department. The median age was 75 years, and 54% were male. Among cases, 80.1% had been diagnosed previously with cardiovascular disease compared with 58.7% of controls. Cancer, chronic obstructive pulmonary disease or asthma, diabetes mellitus, glucocorticoid use, hyperthyroidism, and osteoarthritis were also more common among cases than controls.
Risk of atrial fibrillation or flutter
As table 2⇓ shows, the age and sex matched incidence rate ratio associating current drug use with atrial fibrillation or flutter was 1.33 (95% confidence interval 1.26 to 1.41) for non-selective NSAIDs and 1.50 (1.42 to 1.59) for COX 2 inhibitors compared with non-users. The crude incidence rate ratios, dissolving the matched sets, were similar to the matched incidence rate ratios, indicating that the matched factors were on balance not associated with the exposure. Adjustment for confounders reduced the incidence rate ratio to 1.17 (1.10 to 1.24) for non-selective NSAIDs and 1.27 (1.20 to 1.34) for COX 2 inhibitors. Older and newer COX 2 inhibitors had similar estimates of effect. The increased risk was driven by new users with an adjusted incidence rate ratio of 1.46 (1.33 to 1.62) for non-selective NSAIDs and 1.71 (1.56 to 1.88) for COX 2 inhibitors.
The stratified analyses showed no observable sign of modified measure of effect by sex, osteoarthritis, or systemic connective tissue disease (data not shown). The data indicated that the risk of atrial fibrillation or flutter associated with use of NSAIDs was highest in the elderly (web table 2). Among patients with chronic kidney disease, the adjusted incidence rate ratio was 2.87 (1.53 to 5.38) for new users of COX 2 inhibitors and 1.75 (1.11 to 2.77) for long term users of non-selective NSAIDs (fig 1⇓). Among patients with rheumatoid arthritis, the adjusted incidence rate ratio was 2.49 (1.40 to 4.42) for new users of COX 2 inhibitors and 1.44 (1.01 to 2.03) for long term users of non-selective NSAIDs). Similar to the overall results, the adjusted incidence rate ratio in the secondary analysis restricted to patients without systemic inflammatory conditions was 1.45 (1.29 to 1.63) for new users of non-selective NSAIDs and 1.64 (1.46 to 1.84) for new users of COX 2 inhibitors.
The results for the individual NSAIDs are shown in table 3⇓. The adjusted incidence rate ratio for atrial fibrillation or flutter among new drug users was 1.43 (1.28 to 1.59) for ibuprofen, 1.44 (0.97 to 2.12) for naproxen, 1.73 (1.53 to 1.97) for diclofenac, 1.51 (1.17 to 1.95) for etodolac, 1.83 (1.44 to 2.34) for celecoxib, and 1.59 (1.24 to 2.02) for rofecoxib. In the direct drug comparison (web table 3), no NSAIDs were associated with a lower risk than ibuprofen, and diclofenac in particular conferred higher risk (1.19, 1.00 to 1.40 for new use). The increased effect estimates associated with use of the individual NSAIDs remained raised for both high dose and low dose tablets. High dose tablets of ibuprofen, naproxen, and diclofenac, however, were associated with higher risks than low dose tablets (data not shown).
Supporting the robustness of our findings, the results of the remaining three secondary analyses were similar to the overall results (web tables 4 and 5 show the results for patients without systemic inflammatory conditions). Finally, we estimated that an unmeasured confounder that was twice as frequent among users of NSAIDs as non-users would need to increase the risk of atrial fibrillation or flutter by a factor of six or more to fully explain the results, if no increased risk actually existed (web figure).
In this large population based case-control study, we found that patients starting treatment with non-aspirin NSAIDs were at increased risk of atrial fibrillation or flutter compared with those not using NSAIDs. The relative risk increase was 40-70%—equivalent to approximately four extra cases per year of atrial fibrillation per 1000 new users of non-selective NSAIDS and seven extra cases per year of atrial fibrillation per 1000 new users of COX 2 inhibitors.7 The risk appeared highest in older people. Patients with chronic kidney disease or rheumatoid arthritis were at particularly increased risk when starting treatment with COX 2 inhibitors.
Several issues should be considered when interpreting our results. The study’s population based design within the setting of a tax supported universal healthcare system largely removed selection biases. The positive predictive value of a diagnosis of atrial fibrillation or flutter has been reported to be as high as 97% in the Danish National Registry of Patients.19 Coding errors were thus unlikely to have had any important influence on our results. We considered cases of atrial fibrillation and flutter together, but our results were driven by atrial fibrillation. Although our findings also related to people treated with cardioversion within one year after first diagnosis, our study was limited by its inability to separate paroxysmal, persistent, and permanent atrial fibrillation.
Data in the prescription database are virtually complete.12 Although we had to use prescription data as a proxy for actual use of NSAIDs, we did not base drug exposure information on written prescriptions,11 but on actual dispensing at pharmacies.12 Requirement of co-payment increased the likelihood of compliance.13 We lacked information on over the counter use of low dose (200 mg/tablet) ibuprofen, which accounted for 30% of total ibuprofen sales and 15% of total NSAID sales during the study period.13 This misclassification of drug exposure would most likely have been non-differential and thus would have biased the effect estimates towards the null. Therefore, to the extent such misclassification occurred, our effect estimates are underestimates.
Our results may be affected by confounding from unmeasured variables, particularly by underlying inflammatory conditions leading to use of NSAIDs. Although our estimates did not change when patients with systemic inflammatory conditions were excluded in a subanalysis, we cannot rule out that new users may have more severe underlying inflammation, which may increase the risk of atrial fibrillation.30 Finally, we lacked data on lifestyle factors, including smoking and body size. Nevertheless, we note that we did adjust partly for lifestyle factors by controlling for history of cancer, chronic obstructive pulmonary disease, and ischaemic heart disease, and that our findings could not be explained by even a strong single unmeasured confounder.
Our study is the first on NSAIDs and atrial fibrillation to include both non-selective NSAIDs and COX 2 inhibitors. A case-control study of patients in the United Kingdom diagnosed in 1996 with chronic atrial fibrillation (n=1035) or paroxysmal atrial fibrillation (n=525) found that contemporary use of traditional NSAIDs was associated with an increased risk of chronic atrial fibrillation (incidence rate ratio 1.44, 95% confidence interval 1.08 to 1.91) and modestly associated with paroxysmal atrial fibrillation (1.18, 0.85 to 1.66)—that is, with magnitude of the association similar to our results.11 By contrast with our findings, however, in the UK study, long term use of NSAIDs was associated with the largest risk increase for atrial fibrillation.
The meta-analysis,10 involving 116 094 patients using newer COX 2 inhibitors, had 6394 composite renal outcome events but only 286 composite arrhythmia outcome events, of which ventricular fibrillation, cardiac arrest, and sudden cardiac death accounted for most.10 Although rofecoxib was associated with an increased relative risk for the composite renal outcome of 1.53 (95% confidence interval 1.33 to 1.76) and the composite arrhythmia outcome (2.90, 1.07 to 7.88),10 the small number and types of arrhythmias available for analysis did not allow for examination of atrial fibrillation as an outcome. In the present study, we found an increased risk of atrial fibrillation or flutter associated with older and newer COX 2 inhibitors. Notably, COX 2 inhibitors, in particular diclofenac, were associated with higher risks than non-selective NSAIDs, indicating the important pharmacological role of COX 2 inhibition.3 5
Use of NSAIDs may increase the risk of atrial fibrillation or flutter through renal and cardiovascular related actions. Five per cent of patients treated with NSAIDs experience nephrotoxic syndromes.2 Both COX enzymes are expressed in distinct anatomic regions of adult kidney tissue.2 Thus, inhibition of synthesis of COX derived prostaglandin impairs inflammation and a variety of physiological processes.2 These changes may induce increases in blood pressure due to expansion of plasma volume, increased peripheral resistance, attenuation of diuretic and antihypertensive drug effects,2 6 and fluctuation of serum potassium as a result of decreased potassium excretion in the distal nephron.2 Thus, the increased risk among new users may be attributable to short term adverse renal effects of NSAIDs, which subsequently trigger atrial fibrillation.24 The finding that patients with chronic kidney disease have a markedly higher risk when starting treatment with COX 2 inhibitors supports this hypothesis.2 6
In conclusion, we found that use of non-aspirin NSAIDs was associated with an increased risk of atrial fibrillation or flutter. Compared with non-users, the association was strongest for new users, with a 40-70% relative risk increase (lowest for non-selective NSAIDs and highest for COX 2 inhibitors). Our study thus adds evidence that atrial fibrillation or flutter need to be added to the cardiovascular risks under consideration when prescribing NSAIDs.
What is already known on this topic
Atrial fibrillation is the most commonly sustained rhythm disorder observed in clinical practice, and NSAIDs are among the most widely used drugs worldwide.
No previous study has examined whether use of COX 2 inhibitors increases the risk of atrial fibrillation.
What this study adds
Use of non-selective NSAIDs or selective COX 2 inhibitors was associated with an increased risk of atrial fibrillation or flutter.
Compared with non-users, the association was strongest for new users, with a 40-70% increase in relative risk (lowest for non-selective NSAIDs and highest for COX 2 inhibitors).
Cite this as: BMJ 2011;343:d3450
Contributors: MS, CFC, and HTS conceived the study idea. All authors designed the study. FM and HTS collected the data. MS, CFC, and HTS reviewed the literature. MS, CFC, FM, and HTS analysed the data. All authors participated in the interpretation of the findings. MS wrote the initial draft. All authors participated in critical revision of the manuscript for important intellectual content and approved the final version. HTS is the guarantor.
Funding: The study was supported by the Danish Medical Research Council (grant 271-05-0511), the Clinical Epidemiological Research Foundation, Denmark, the Danish Heart Association, and an Aarhus University scholarship. Department of Clinical Epidemiology collaborates within the EU Seventh Framework Programme: Arrhythmogenic potential of drugs (ARITMO). None of the funding sources had a role in the study design, conduct, analysis, or reporting.
Competing interests: All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare: no support from any company for the submitted work, although the Department of Clinical Epidemiology is involved in studies with funding from various companies as research grants to (and administered by) Aarhus University, none of which has any relation to the present study; no relation with organisations that might have an interest in the submitted work in the previous three years, except KJR, who received payment from Bayer for a lecture on venous thromboembolism; no non-financial interests that may be relevant to the submitted work.
Ethical approval: This study was approved by the Danish Data Protection Agency (record no 2004-41-4693) and the Aarhus University Hospital registry board. The study does not involve any contact with patients or any intervention, and it is not necessary to procure permission from the Danish Scientific Ethics Committee.
Data sharing: No additional data available.
This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the license. See: http://creativecommons.org/licenses/by-nc/2.0/ and http://creativecommons.org/licenses/by-nc/2.0/legalcode.