Design Population based retrospective cohort study.

Setting Databases of hospital discharge summaries and prescription drug claims in Quebec.

Participants 2256 patients aged 66 or more prescribed celecoxib, rofecoxib, or an NSAID after an index admission for congestive heart failure between April 2000 and March 2002.

Main outcome measures Time to all cause death and recurrent congestive heart failure, combined and separately.

Results The risk of death and recurrent congestive heart failure combined was higher in patients prescribed NSAIDs or rofexocib than in those prescribed celecoxib (hazard ratio 1.26, 95% confidence interval 1.00 to 1.57 and 1.27, 1.09 to 1.49, respectively). The findings were similar when the outcomes were assessed separately. In pairwise analysis, the risks of death and recurrent congestive heart failure, combined and separate, were similar between patients prescribed NSAIDs and rofecoxib.

The role of cyclo-oxygenase-2 inhibition in congestive heart failure is complex. On the one hand, both NSAIDs and cyclo-oxygenase-2 inhibitors have various renovascular side effects, which include increased volume retention, oedema, and blood pressure, all of which can exacerbate heart failure.^19-22 Data, however, suggest that differences exist between the two types of drugs as well as between the cyclo-oxygenase-2 inhibitors themselves, with some,^23-25 but not all,²⁶ studies showing that rofecoxib is associated with more renovascular side effects, including increased oedema and blood pressure, than are NSAIDs and celecoxib. On the other hand, the cyclo-oxygenase-2 enzyme is induced in the myocardium of the failing heart and is associated with myocardial scarring.^27,28 Selective cyclo-oxygenase-2 inhibitors can therefore be cardioprotective at the level of the failing myocardium.

Clinically, NSAIDs have been associated with the onset and exacerbation of congestive heart failure.^29-31 Clinical information on the association between cyclo-oxygenase-2 inhibitors and heart failure is, however, scarce.^32-35 In a recent population based study, users of rofecoxib and NSAIDs, but not celecoxib, had a higher risk of admission for congestive heart failure than controls not taking NSAIDs.³⁶

Patients were eligible for inclusion if they were discharged with a diagnosis of congestive heart failure (code 428.x; international classification of diseases, 9th revision) between 1 April 2000 and 31 March 2002 (the index admission), they had been prescribed celecoxib, rofecoxib, or a NSAID during that time, and they were aged 66 or more at the time of discharge from the index admission. Celecoxib became available on the market in late 1999 and rofecoxib on 1 April 2000. We chose the starting date of 1 April 2000 to ensure that both drugs were available throughout the study. Patients were excluded if they were admitted to hospital for congestive heart failure in the preceding three years. The age cut-off point was determined on the basis of the availability of prescription records, to ensure that all participants had at least one full year of observation in the database before entry into the study.

The most commonly used NSAIDs were diclofenac (4.3%), naproxen (2.6%), indomethacin (1.8%), and ibuprofen (1.0%). Other less commonly used NSAIDs included ketoprofen, sulindac, piroxicam, flurbiprofen, tiaprofenic acid, nabumetone, salsalate, and etodolac.

Significantly more patients prescribed celecoxib survived or remained free of recurrent congestive heart failure than those prescribed rofecoxib or NSAIDs (fig 1). The point of divergence occurred early and persisted over time. The difference in survival was persistently significant when assessed as a single end point (fig 2), and the trend was towards those prescribed celecoxib remaining free of recurrent congestive heart failure compared with those prescribed rofecoxib or NSAIDs (fig 3).

In adjusted analysis, we found a significantly higher risk of death or recurrent congestive heart failure in patients prescribed NSAIDs or rofecoxib than in those prescribed celecoxib (hazard ratio 1.26, 95% confidence interval 1.00 to 1.57 and 1.27, 1.09 to 1.49, respectively; fig 4). These increased risks persisted when death was assessed as a separate outcome (NSAIDs, 1.54, 1.17 to 2.04; rofecoxib, 1.44, 1.17 to 1.78). When recurrent congestive heart failure was assessed separately, the trend was towards an increase in risk (NSAIDs, 1.21, 0.92 to 1.60; rofecoxib, 1.17, 0.96 to 1.42).

An association between cyclo-oxygenase-2 inhibitors and congestive heart failure is emerging, but there seem to be intra-class differences. Case reports have shown worsening heart failure associated with cyclo-oxygenase-2 inhibitors,⁴⁵ and data from a small prospective database for disease management suggests that patients who are prescribed a cyclo-oxygenase-2 inhibitor on discharge are more likely to be readmitted for congestive heart failure within a year compared with those not prescribed a cyclo-oxygenase-2 inhibitor (32.5% versus 22.0%, respectively, P < 0.05).⁴⁶ In that study, the risk of recurrent congestive heart failure differed between rofecoxib and celecoxib (35.5% versus 32.5%, respectively). In a drug safety database, rofecoxib was associated with significantly more reports of cardiac failure than celecoxib.²⁴ Similarly, in a large, population based study of stable hypertensive patients with no history of heart failure, those who were newly treated with rofecoxib were significantly more likely than those newly treated with celecoxib to have a diagnosis of heart failure as an outpatient (rate ratio 1.26, 95% confidence interval 1.06 to 1.48) and as an inpatient (1.52, 1.15 to 2.02).⁴⁷ In that study, the rates of heart failure were similar between the celecoxib and NSAIDs groups. Finally, in a recent population based study of cyclo-oxygenase-2 inhibitors and congestive heart failure, users of rofecoxib and NSAIDs, but not celecoxib, had a higher risk of admission for congestive heart failure than controls receiving non-NSAIDs.³⁶ Findings were similar in a subgroup analysis of patients with a history of admission for congestive heart failure. In pairwise comparisons, they found a significantly higher risk of admission for congestive heart failure in patients prescribed rofecoxib than in those prescribed celecoxib or NSAIDs (adjusted rate ratio 1.8, 95% confidence interval 1.4 to 2.4 and 1.5, 1.1 to 2.1).

Our findings are consistent with these data and are unique because we studied high risk patients with known congestive heart failure and we found a lower risk of death, assessed alone and in combination with recurrent congestive heart failure, in patients prescribed celecoxib compared with those prescribed NSAIDs or rofecoxib. Our data show that celecoxib may be safer than NSAIDs in high risk patients and that there may be important differences between cyclo-oxygenase-2 inhibitors. Differences in selectivity⁴⁸ may, at least in part, explain the different results observed in patients prescribed celecoxib compared with those prescribed rofecoxib. NSAIDs also show differences in selectivity. The number of patients prescribed NSAIDs in our study was, however, small so we grouped the NSAIDs. The relative side effects of individual NSAIDs in congestive heart failure remain unknown.

Our study has several limitations. Firstly, patients who were classified as not taking drugs may have been using over the counter aminosalicylic acid or ibuprofen. In 2000, Santé Québec, a government public health agency, estimated that 17.0% and 2.2% of people aged 65 or more bought over the counter NSAIDs or aminosalicylic acid, respectively.⁴⁹ Non-differential misclassification on exposure would, however, tend to bias the results towards the null. Secondly, by starting the study on the date that rofecoxib became available (1 April 2000), we included only incident users of rofecoxib but incident and prevalent users of celecoxib. We considered whether the results favouring celecoxib may have been the result of attrition and adherencebiases.⁵⁰ We carried out a sensitivity analysis in which we included only new users of celecoxib (and excluded those with a prescription before 1 April 2000). Time to all cause death or recurrent congestive heart failure was similar to that reported in the main analysis (hazard ratio 1.27, 95% confidence interval 1.07 to 1.52 versus 1.27, 1.09 to 1.49).