Risk of congestive heart failure with nonsteroidal antiinflammatory drugs and selective Cyclooxygenase 2 inhibitors: A class effect?

Authors

  • Marie Hudson,

    Corresponding author
    1. The Research Institute of the McGill University Health Centre, Montreal, Quebec, Canada
    • Division of Clinical Epidemiology, The Research Institute of the McGill University Health Centre, 687 Pine Avenue West, Building V, Montreal, Quebec PQ H3A 1A1, Canada
    Search for more papers by this author
  • Elham Rahme,

    1. The Research Institute of the McGill University Health Centre, Montreal, Quebec, Canada
    Search for more papers by this author
    • Dr. Rahme has received consulting fees (more than $10,000 each) from Merck, Pfizer, and Boehringer Ingelheim.

  • Hugues Richard,

    1. The Research Institute of the McGill University Health Centre, Montreal, Quebec, Canada
    Search for more papers by this author
  • Louise Pilote

    1. The Research Institute of the McGill University Health Centre, Montreal, Quebec, Canada
    Search for more papers by this author

Abstract

Objective

Nonsteroidal antiinflammatory drugs (NSAIDs) as a class have been shown to increase the risk of congestive heart failure (CHF) compared with celecoxib. The magnitude of the risk for individual NSAIDs is not known.

Methods

Using administrative databases, we performed a nested case–control study in a population-based cohort of patients ages ≥66 years admitted for CHF between January 1998 and March 2003. Cases were patients readmitted for CHF after cohort entry (index date). Four controls were matched to each case on date of cohort entry and time between cohort entry and index date. Exposure was the current use of an NSAID or a coxib in the 7 days prior to CHF readmission. Using conditional logistic regression, we calculated the odds of readmission for CHF in patients exposed to naproxen, diclofenac, ibuprofen, indomethacin, or rofecoxib compared with celecoxib, after adjusting for possible confounding variables.

Results

We identified 8,512 cases and 34,048 controls. The baseline characteristics between the groups were similar in general. The odds of being readmitted for CHF were higher in patients currently exposed to indomethacin (odds ratio [OR] 2.04, 95% confidence interval [95% CI] 1.16–3.58) or rofecoxib (OR 1.58, 95% CI 1.19–2.11) compared with celecoxib. There was no difference between naproxen, diclofenac, and ibuprofen compared with celecoxib, although the numbers of exposed cases and controls were small.

Conclusion

In elderly patients with known CHF, indomethacin and rofecoxib are associated with a greater risk of recurrent CHF compared with celecoxib. Alternatives should be considered for patients with CHF who require antiinflammatory drugs.

INTRODUCTION

In the recent past, there has been considerable controversy regarding the cardiovascular effects of nonsteroidal antiinflammatory drugs (NSAIDs) and cyclooxygenase 2 (COX-2) inhibitors. Most studies to date have addressed the risks of ischemic heart disease, and evidence of whether individual NSAIDs and COX-2 inhibitors are associated with the same risks or different risks remains contradictory (1–6). There is much less evidence concerning the risks of congestive heart failure (CHF) with NSAIDs and COX-2 inhibitors. In the past, NSAIDs have been associated with the onset and exacerbation of CHF (7, 8) and the burden of illness associated with NSAID-related CHF has been suggested to exceed that of gastrointestinal adverse effects related to NSAIDs (9). More recently, patients exposed to NSAIDs and rofecoxib, but not celecoxib, have been shown to be at an increased risk of hospitalization for CHF compared with nonexposed patients (10), and NSAIDs and rofecoxib have been associated with an increased risk of hospitalization for CHF compared with celecoxib (11). However, in all of these studies, NSAIDs have been studied as a group. Therefore, it is not known whether all NSAIDs are associated with the same risk, in other words, whether there is a class effect of these drugs regarding this side effect (12), or whether the magnitude of the risk varies between individual NSAIDs.

CHF and arthritis are both common conditions in elderly patients. Therefore, elderly patients with CHF may require NSAIDs for the management of their pain and inflammation. However, as mentioned above, whether some NSAIDs are associated with greater or smaller risks of CHF such that some may be more harmful and others preferable in patients with known CHF is not known. We therefore conducted the present study to determine whether there is a class effect among individual NSAIDs on the risk of CHF in a cohort of elderly patients with known CHF.

PATIENTS AND METHODS

Data source and study design.

The hospital discharge summary database of the Province of Québec, Canada (the Maintenance et Exploitation des Données pour l'Étude de la Clientèle Hospitalière) provides information on all hospital admissions for the entire province, including discharge diagnosis and comorbid conditions. The provincial government health insurance agency, the Régie de l'Assurance Maladie du Québec (RAMQ), covers all inpatient and outpatient medical services for the entire Québec population, as well as the costs of outpatient prescription drugs for all individuals ages 65 years and older. Information on vital status is also available in the RAMQ database and is available for virtually all patients (13). NSAIDs, celecoxib, and rofecoxib are approved for therapeutic purposes by the Ministry of Health and are reimbursed by RAMQ, without restrictions concerning their use. Ibuprofen is also available over the counter, although individuals covered by the RAMQ drug insurance plan are likely to seek prescriptions for these medications in order to be reimbursed. The reliability of the coding diagnosis for CHF in administrative databases has been shown to be high (14, 15). Data in the drug claims database have been validated for accuracy of prescription claims (16). The 2 databases can be merged using encrypted health insurance numbers. They have been used in other pharmacoepidemiologic studies (11). Using these administrative data, we designed a nested case–control study. We believe that this design allowed us to assess exposure to NSAIDs, which are often taken intermittently, for short durations, and on an as-needed basis, better than in a cohort study.

Cohort.

Patients were included in the cohort if 1) they were discharged with a primary diagnosis of CHF (as defined in the International Classification of Diseases, Ninth Revision [ICD], code 428.x) between January 1998 and March 2003 (cohort entry) and 2) they were ≥66 years of age at cohort entry. We wished to study patients with new-onset CHF. For this reason, we excluded patients who had a diagnosis of CHF coded during any hospital admission in the 3 years prior to cohort entry. Other new-onset CHF cohorts have been constructed using similar exclusion criteria (15). The age cutoff was used to ensure that all patients had sufficient records in the prescription database to be considered in the various periods of exposure (see below).

Selection of cases and controls.

Using the hospital discharge database, we identified cases as those patients in the cohort who were readmitted with a primary diagnosis of CHF (ICD code 428.x) after cohort entry. The date of hospitalization was identified as the index date. For each case identified, we randomly selected 4 controls using a method of sampling with replacement (17). Because COX-2 inhibitors were not available throughout the study period, controls were matched to cases on cohort entry. In addition, because the risk of recurrent CHF may vary over time after an initial episode of CHF, controls were also matched to cases on time between cohort entry and index date. Controls were selected after ensuring that they were at risk on the day the case was identified, in other words, the controls had to be alive and free of recurrent CHF up until the index date. Controls could be used more than once provided they met the matching criteria mentioned above each time they were selected and they were at risk at the time they were selected as controls (18).

Exposure.

All drugs dispensed to the cases and controls, including NSAIDs, celecoxib, and rofecoxib, were identified. For purposes of comparison, patients exposed to acetaminophen and patients not exposed to NSAIDs, COX-2 inhibitors, and acetaminophen were also identified. Information on the type of prescription, date of filling, and duration of each prescription dispensed was available from the database and used to define periods of exposure.

The main exposure in the present study was the current, exclusive exposure to diclofenac, ibuprofen, indomethacin, naproxen, celecoxib, or rofecoxib in the 7 days prior to and including the index date. We also identified subgroups of exposure by timing of exposure (in addition to those with current exposure, we identified patients who were exposed to a study drug in the 8–180 days prior to the index date [past exposure]).

Covariate information.

Covariate information was defined at the time of cohort entry because the information was equally available for cases and controls at that time and because the time between cohort entry and index date was short (median 134 days, interquartile range 37–401). Covariate information included demographic data, comorbid conditions noted on the discharge summaries (hypertension, diabetes mellitus, chronic obstructive pulmonary disease, myocardial infarction, renal disease, malignancy, rheumatologic disease, and liver disease), concomitant discharge medications that could influence survival in patients with CHF (beta-blockers, angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, diuretics, digoxin, and hydralazine), information on the specialty of the treating physician (3 categories: cardiologist, internist or specialist other than cardiologist, and general practitioner), and length of stay.

Statistical analysis.

The baseline patient characteristics of the cases and controls and of the patients in the various exposure groups were compared using descriptive statistics. Conditional logistic regression models were used to estimate the odds ratios (ORs) of recurrent CHF in patients with current, exclusive exposure to diclofenac, ibuprofen, indomethacin, naproxen, and rofecoxib compared with celecoxib. For purposes of comparison, the odds of recurrent CHF in patients with exclusive exposure to acetaminophen and patients who were currently not exposed to any NSAID, COX-2 inhibitor, or acetaminophen were also compared in the same model with the odds in patients with current, exclusive exposure to celecoxib. A separate model was generated to compare the odds of readmission for CHF in patients exposed to NSAIDs as a group (not including rofecoxib) compared with celecoxib. The models were adjusted for age, sex, comorbidities, concomitant cardiac medications, characteristics of the treating physician, length of stay, and dose. Doses were defined as either at-target doses or above-target doses according to the World Health Organization Anatomical Therapeutic Chemical Classification System with Defined Daily Doses (DDDs) (19), which serves as a tool for drug utilization research. The DDD is the average maintenance daily dose for a drug used for its main indication and may not necessarily reflect the recommended or prescribed daily dose. DDDs have been used in other pharmacoepidemiologic studies of NSAIDs (20). The following were used as at-target or above-target dosages: diclofenac ≤100 mg/day and >100 mg/day, ibuprofen ≤1,200 mg/day and >1,200 mg/day, indomethacin ≤100 mg/day and >100 mg/day, naproxen ≤500 mg/day and >500 mg/day, celecoxib ≤200 mg/day and >200 mg/day, and rofecoxib ≤25 mg/day and >25 mg/day, respectively. The models were adjusted according to whether patients were prescribed above-target doses compared with target doses.

We hypothesized that, given the short half lives of the drugs of interest, the increased odds of CHF would be related to a current exposure and that the odds of CHF would not be increased with past exposures. Therefore, a subgroup analysis according to timing of exposure was undertaken. For this analysis, only patients with at least 180 days of observation prior to the index date were included. Patients were divided into mutually exclusive groups: current exposure (0–7 days prior to index date) and past exposure (8–180 days prior to index date) to diclofenac, ibuprofen, indomethacin, naproxen, celecoxib, or rofecoxib. To facilitate comparison with the main analysis, patients with past, exclusive exposure to celecoxib were compared with patients with current, exclusive exposure to celecoxib. Comparisons were also made for patients with past exposure to acetaminophen and patients not exposed to any NSAID, COX-2 inhibitor, or acetaminophen in the 180 days prior to index date. Models were adjusted with the same covariates as those used in the main analysis. All analyses were performed using the SAS statistical software, version 8.2 (SAS Institute, Cary, NC).

Ethical considerations.

There were minimal risks to the patients observed in this study. All of the data used for study analyses were anonymous and confidential. All personal identifiers were scrambled by the provincial government agencies before we received the data. All results are presented in summary form only. The study was approved by the McGill University Ethics Review Board.

RESULTS

The cohort included 27,259 patients with an admission for CHF between January 1998 and March 2003. Of those, we identified 8,512 cases who were subsequently readmitted for CHF. We selected 34,048 controls, of whom 3,962 later became cases and 7,469 were used as controls more than once. The median age for both cases and controls was 78 years and the majority of patients in both groups were women (Table 1). Cases and controls had similar rates of hypertension (35% and 38%, respectively) and myocardial infarction (23% and 21%, respectively), but cases had more renal disease compared with controls (23% and 16%, respectively). Patients in the various exposure groups were also generally similar (Table 2).

Table 1. Baseline characteristics of cases and controls*
CharacteristicCases (n = 8,512)Controls (n = 34,048)
  • *

    Values are the percentage unless otherwise indicated. IQR = interquartile range; COPD = chronic obstructive pulmonary disease; CHF = congestive heart failure; ACE = angiotensin-converting enzyme; ARBs = angiotensin receptor blockers.

Age, median (IQR) years78 (73–84)78 (72–83)
Male sex4847
Length of stay, median (IQR) days7 (4–12)7 (4–12)
Comorbidities  
 Hypertension3538
 Diabetes3630
 COPD3331
 Myocardial infarction2321
 Renal disease2316
 Malignancy33
 Rheumatologic disease22
 Liver disease11
Medications to treat CHF (before first prescription of a study drug)  
 Diuretics8276
 ACE inhibitors5253
 Digoxin3733
 Beta-blockers3533
 ARBs1111
 Hydralazine42
Treating physicians  
 Cardiologist3133
 Internist or specialist, other than cardiologist1414
 General practitioner5553
Table 2. Baseline characteristics of patients with current exposure to nonsteroidal antiinflammatory drugs, cyclooxygenase 2 inhibitors, or acetaminophen and nonexposed patients*
CharacteristicNaproxen (n = 87)Diclofenac (n = 132)Ibuprofen (n = 35)Indomethacin (n = 63)Celecoxib (n = 631)Rofecoxib (n = 489)Acetaminophen (n = 4,089)Nonexposed (n = 36,453)
  • *

    Values are the percentage unless otherwise indicated. See Table 1 for definitions.

Age, median (IQR) years75 (72–80)77 (74–82)75 (69–82)73 (69–81)78 (73–83)78 (72–83)80 (74, 85)78 (72, 83)
Male sex6146374635383649
Length of stay, median (IQR) days6 (3–12)6 (4–11)5 (4–13)8 (5–11)7 (4–11)7 (4–12)8 (4–14)7 (4–12)
Comorbidities        
 Hypertension4336464441404037
 Diabetes1726341430303131
 COPD3838312733323631
 Myocardial infarction1414201123202022
 Renal disease112592515141917
 Malignancy174303233
 Rheumatologic disease91985631
 Liver disease01001011
Cardiac medications        
 Diuretics8791948787889075
 ACE inhibitors5655666258595852
 Digoxin3245143729263534
 Beta-blockers3030314638443633
 ARBs999616171210
 Hydralazine14000032
Treating physicians        
 Cardiologist2535404426253133
 Internist221623812141514
 General practitioner5349374862615553

The patterns of prescriptions for patients with current exposures are presented in Table 3. Among NSAIDs and COX-2 inhibitors, celecoxib was the most commonly used drug, followed by rofecoxib, diclofenac, naproxen, indomethacin, and ibuprofen. Of the exposed patients, the total duration of current exposure was relatively short and similar numbers of cases and controls had a prescription above target dose.

Table 3. Prescription patterns for cases and controls with current exposures*
 Cases (n = 8,512)Controls (n = 34,048)
  • *

    Values are the number (percentage) unless otherwise indicated. IQR = interquartile range.

Patients with a current exposure  
 Naproxen19 (0.22)68 (0.20)
 Diclofenac26 (0.31)106 (0.31)
 Ibuprofen9 (0.11)26 (0.08)
 Indomethacin21 (0.25)42 (0.12)
 Rofecoxib139 (1.63)350 (1.03)
 Celecoxib132 (1.55)499 (1.47)
 Acetaminophen985 (11.57)3,104 (9.12)
 Nonexposed7,044 (82.75)29,409 (86.38)
Duration of current exposure, median (IQR) days  
 Naproxen13 (7–43)12.5 (6–53.5)
 Diclofenac14 (10–60)17.5 (9–44)
 Ibuprofen10 (7–30)27.5 (7–134)
 Indomethacin6 (4–27)11.5 (5–24)
 Rofecoxib17 (7–60)30 (13–72)
 Celecoxib30 (10–98.5)30 (14–106)
 Acetaminophen15 (7–38)17 (8–46)
Patients with a current exposure whose prescription was above target dose  
 Naproxen15 (79)50 (74)
 Diclofenac15 (58)42 (40)
 Ibuprofen3 (33)11 (42)
 Indomethacin4 (19)14 (33)
 Rofecoxib13 (9)23 (7)
 Celecoxib33 (25)124 (27)

When NSAIDs were examined as a group, the odds of being readmitted for CHF were similar in patients currently exposed to NSAIDs compared with those currently exposed to celecoxib (adjusted OR 1.01, 95% confidence interval [95% CI] 0.74–1.39). However, when NSAIDs were examined individually, the odds of readmission for CHF were significantly higher in patients currently exposed to indomethacin (OR 2.04, 95% CI 1.16–3.58) compared with those exposed to celecoxib, but the odds were similar in patients exposed to naproxen, diclofenac, and ibuprofen compared with celecoxib (OR 1.05, 95% 0.59–1.86; OR 0.82, 95% CI 0.51–1.33; and OR 1.46, 95% CI 0.66–3.21, respectively) (Tables 4 and 5). The odds of recurrent CHF were also greater in patients exposed to rofecoxib compared with celecoxib (OR 1.58, 95% CI 1.19–2.11). There was no difference between patients exposed to acetaminophen and celecoxib and nonexposed patients and celecoxib.

Table 4. Main analysis: odds ratios and 95% confidence intervals for recurrent congestive heart failure in patients exposed to current nonsteroidal antiinflammatory drugs compared with current celecoxib exposure
 Crude odds ratioAdjusted odds ratio*95% confidence interval
  • *

    Adjusted for age, sex, length of stay, dose, comorbidities, physician characteristics, and medications to treat congestive heart failure.

CelecoxibReference  
Naproxen1.051.050.59–1.86
Diclofenac0.920.820.51–1.33
Ibuprofen1.291.460.66–3.21
Indomethacin1.872.041.16–3.58
Rofecoxib1.511.581.19–2.11
Acetaminophen1.191.150.92–1.44
Nonexposed0.900.930.75–1.15
Table 5. Conditional logistic regression model for the main analysis comparing recurrent congestive heart failure in patients exposed to current NSAIDs compared with current celecoxib exposure*
 Crude odds ratioAdjusted odds ratio95% confidence interval
  • *

    NSAIDs = nonsteroidal antiinflammatory drugs; GP = general practitioner; see Table 1 for additional definitions.

CelecoxibReference  
Naproxen1.051.050.59–1.86
Diclofenac0.920.820.51–1.33
Ibuprofen1.291.460.66–3.21
Indomethacin1.872.041.16–3.58
Other NSAIDs0.650.640.34–1.19
Rofecoxib1.511.581.19–2.11
Acetaminophen1.191.150.92–1.44
Nonexposed0.900.930.75–1.15
Other exposure groups1.271.310.97–1.76
Sex (male vs female) 1.040.99–1.09
Age (continuous) 1.011.01–1.02
Dose (high versus low) 1.190.87–1.61
Length of stay 1.081.02–1.14
Myocardial infarction 1.091.02–1.15
COPD 1.040.98–1.09
Renal disease 1.471.39–1.56
Malignancy 0.990.87–1.14
Liver disease 1.020.79–1.33
Diabetes 1.341.27–1.41
Rheumatic disease 0.910.76–1.10
Hypertension 0.900.85–0.94
Cardiologist (versus GP) 0.890.84–0.94
Internist (versus GP) 0.940.87–1.01
Beta-blockers 1.091.03–1.15
ACE inhibitors 0.900.85–0.95
ARBs 0.960.88–1.05
Digoxin 1.141.08–1.21
Hydralazine 2.001.74–2.29
Diuretics 1.361.27–1.46

In subgroup analysis, the odds of recurrent CHF were similar in cases and controls with past exposures to diclofenac, naproxen, ibuprofen, indomethacin, and rofecoxib compared with current celecoxib exposure (Table 6). However, the numbers of exposed cases and controls in the various groups were small and the confidence intervals were wide.

Table 6. Subanalysis by timing of exposure: odds ratios and 95% confidence intervals for recurrent congestive heart failure in patients exposed to nonsteroidal antiinflammatory drugs in the past (8–180 days) compared with current (0–7 days) celecoxib exposure
 Crude odds ratiosAdjusted odds ratios*95% confidence intervals
  • *

    Adjusted for age, sex, length of stay, dose, comorbidities, physician characteristics, and medications to treat congestive heart failure.

  • Defined as never exposed throughout the period of interest, namely, 1–180 days.

Naproxen0.560.480.22–1.06
Diclofenac1.100.970.53–1.78
Ibuprofen0.580.480.15–1.48
Indomethacin0.950.990.50–1.95
Rofecoxib0.700.750.48–1.17
Celecoxib0.610.610.40–0.93
Acetaminophen0.850.890.65–1.23
Nonexposed0.700.770.57–1.06

DISCUSSION

In this population-based study of elderly patients with known CHF, patients exposed to indomethacin and rofecoxib had a higher risk of recurrent CHF compared with those exposed to celecoxib. The risks of recurrent CHF in patients exposed to naproxen, diclofenac, and ibuprofen were not significantly different compared with those of patients exposed to celecoxib. Although these negative findings may suggest that there is no difference between these drugs, the small numbers of exposed cases and controls and the wide confidence intervals suggest that the study may have had too little power to detect small differences. Therefore, although our data do not allow us to rule in or out the presence of a class effect of NSAIDs on the risk of CHF, they do demonstrate, for the first time, that one NSAID in particular (namely, indomethacin) appears to be associated with a higher risk of recurrent CHF and confirm the results of previous studies concerning the increased risk of CHF associated with rofecoxib (10, 11).

We hypothesized that, given the short half lives of the drugs of interest, the increased odds of CHF should be related to a current exposure and that the odds of CHF should not be increased with past exposures (>8 days prior to the index date). The analysis according to timing of exposure (Table 6) demonstrated no increased odds of CHF in patients with past exposures to NSAIDs and confirmed this hypothesis. In so doing, the analysis provided biologic plausibility for our findings in the main analysis, which showed increased odds with current exposure to indomethacin and rofecoxib.

The results of our study are consistent with the limited published data to date. In a study designed to assess the gastrointestinal safety of celecoxib (21), the incidence of cardiac failure was not significantly different between the groups treated with celecoxib, diclofenac, or ibuprofen (22). In a study designed to assess the gastrointestinal safety of rofecoxib (23), the incidence of CHF was higher among patients in the rofecoxib group compared with those in the naproxen group (0.5% versus 0.2%; P = 0.07) (24). However, our study is the first to assess the risks of CHF of several NSAIDs and COX-2 inhibitors in a single cohort of patients at risk for CHF.

Nevertheless, contrary to previous evidence, our study failed to identify an increased risk of recurrent CHF when NSAIDs were assessed as a group. This may be explained by a number of reasons. First, it could be due to the fact that the number of patients exposed to indomethacin was proportionately smaller compared with the number of those exposed to naproxen and diclofenac. Therefore, the overall result for NSAIDs, which is a weighted average of the individual NSAIDs, was driven by the negative results associated with naproxen (OR 1.05) and diclofenac (OR 0.82) rather than the positive result associated with indomethacin (OR 2.04). Second, it is possible that in this cohort of patients with CHF, physicians were very cautious and selected patients to whom they carefully prescribed NSAIDs (for example, perhaps these patients tolerated NSAIDs in the past or perhaps their CHF was particularly well controlled at the time the NSAIDs were prescribed). This careful selection is supported by the rarity of exposure in this large number of patients. Third, exposure to NSAIDs in this study was particularly rare and of short duration (Table 3). Therefore, we may have had low power to detect an association between NSAIDs and CHF. Finally, our result could also be due to misclassification on exposure resulting from the short period chosen as the main exposure. Indeed, patients who were prescribed NSAIDs may have been cautioned about the possibility of exacerbating their symptoms, may have developed symptoms of CHF prior to the main exposure period (0–7 days prior to the index date), and may therefore have decided not to renew their NSAIDs. If they nevertheless were later admitted for CHF, they may have been misclassified as nonexposed. Such misclassification on exposure would bias the results towards the null and in so doing would obscure the harmful effect of NSAIDs.

A possible mechanism for the results observed with indomethacin and rofecoxib may relate to differences in blood pressure control between various NSAIDs and COX-2 inhibitors. NSAIDs have been shown to increase blood pressure, and in this regard the effect of indomethacin appears to be comparable with that of other traditional NSAIDs, including naproxen (25). However, NSAIDs have also been shown to inhibit the blood pressure–lowering activity of many antihypertensive agents (26, 27), which seems to occur to a greater extent with indomethacin than with traditional NSAIDs, including naproxen (28). Therefore, the differential effects on blood pressure control between indomethacin compared with other traditional NSAIDs may explain our results. Rofecoxib has been associated with greater increases in blood pressure compared with celecoxib (29) and naproxen (30). This is also consistent with our findings.

A second possible explanation for our findings concerning indomethacin may relate to the indication for which the various NSAIDs are used. Indomethacin is still considered by some to be the standard treatment for gout (31). However, hyperuricemia is associated with metabolic syndrome and chronic renal failure (32). Patients with CHF and gout, for whom indomethacin is likely to have been prescribed, may have been sicker than those prescribed other traditional NSAIDs who were given the drugs for noninflammatory problems such as osteoarthritis and soft tissue rheumatisms. This possible confounding by indication may, at least in part, explain the increased risk seen with indomethacin. However, the sensitivity of the diagnosis of gout in our administrative data is very poor (33). Therefore, it was not possible to adjust for possible confounding by this indication directly. Nevertheless, we attempted to minimize this possible confounding by indication by adjusting for comorbidities (e.g., hypertension, renal failure) and concomitant medication used for the treatment of CHF (e.g., diuretics) that may also be markers of gout.

A final but less likely explanation for the observed results relates to COX selectivity. The COX-1 and COX-2 selectivities of NSAIDs and COX-2 inhibitors have been shown, both in vitro and ex vivo, to vary between drugs (34, 35). However, these findings do not necessarily predict the efficacy/toxicity ratios of NSAIDs and COX-2 inhibitors in vivo because these are influenced by many other factors including genetic variability in metabolizing enzymes, interactions with other drugs, and individual patient characteristics (36). Indeed, differences in selectivities are not helpful in explaining the results that we obtained in this study because indomethacin and rofecoxib are at the extremes of the range of COX-2 selectivities (35) and yet they were both found to be associated with the greater risk in this study. Other researchers have also recently argued that relative COX-1 and COX-2 selectivities do not explain differences between various NSAIDs and COX-2 inhibitors in the risk of myocardial infarction (37).

Our study has limitations. First, the administrative databases used do not contain detailed clinical information. As with all observational studies, unmeasured confounding might be responsible for the observed differences. To guard against this, we adjusted for multiple potentially confounding variables, including patient and physician characteristics. In addition, the impact of unmeasured confounders in the relationship between COX-2 inhibitors and myocardial infarction using claims databases has been reported to cause only a small underestimation of the association (38). This is likely the case in the present study of the association between NSAIDs and CHF. Second, as in most database pharmacoepidemiologic studies, exposure time to drug was measured by filled prescriptions, which may not faithfully measure pill intake (39). Dose measurement was imprecise and categorized grossly according to target dose. However, the use of a nested case–control design and the use of current exposure as the main exposure of interest increase the chances that patients will have taken the medications at the time and dose assumed. Third, although the reliability of CHF using administrative data has been reported to be high (15), the positive predictive value has been reported to be only 84% (14). This may have contributed to misclassification of outcome. However, because this is likely to have been nondifferential according to exposure status, it would have led to a bias of the results towards the null and made it more difficult to detect an association between CHF and naproxen, diclofenac, and ibuprofen. Fourth, despite an initial cohort of close to 30,000 patients, the numbers of exposed patients were small and some of the negative findings may be due to lack of power to detect a difference. This illustrates the difficulty of studying the question at hand. Indeed, a definitive trial on the question would require considerable time, would be expensive to conduct, and does not appear forthcoming. In the meantime, our observational study based on high-quality administrative data provides valuable information to clinicians faced with the common clinical scenario of a patient with CHF who requires management of his or her arthritis.

NSAIDs as a class have been shown in previous studies to be associated with CHF, but there is little information on the magnitude of the risk of individual drugs. In the absence of head-to-head comparisons in randomized trials, we performed an observational study in a large cohort of patients with CHF. We found that indomethacin and rofecoxib are associated with increased risk of recurrent CHF compared with celecoxib. Therefore, until additional data are available, alternatives such as celecoxib, naproxen, diclofenac, and ibuprofen should be considered for patients with CHF who need to take antiinflammatory drugs.

AUTHOR CONTRIBUTIONS

Dr. Hudson had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Hudson, Rahme, Pilote.

Acquisition of data. Pilote.

Analysis and interpretation of data. Hudson, Rahme, Pilote.

Manuscript preparation. Hudson, Rahme, Pilote.

Statistical analysis. Rahme, Richard.

Obtained funding. Pilote, Hudson.

ROLE OF THE STUDY SPONSOR

The funding sources had no role in the design and conduct of the study, in the analysis and interpretation of the data, in the writing of the report, and in the decision to submit the report to publication.

Ancillary