Relative risk of acute myocardial infarction (AMI). For Figures 1 and 2, the reference group is persons who did not fill any nonsteroidal anti-inflammatory drug (NSAID) prescriptions before the index date. Models were adjusted for all covariates listed in Table 2, including age, sex, ethnicity, Medicaid enrollment, nursing home use, diabetes mellitus, hypertension, congestive heart failure, Charlson Comorbidity Index, number of different drug prescriptions, and number of hospitalizations.
Relationship between cyclooygenase (COX) selectivity and acute myocardial infarction (AMI). This curve was fit using nonlinear assumptions and displays a cubic spline. The solid line represents the log of the odds ratio (OR) of sustaining an MI based on the log of the COX-1:COX-2 ratio for each nonsteroidal anti-inflammatory drug (NSAID) regimen of cases and controls. The dashed lines represent the 95% confidence interval. Subjects using multiple NSAIDs were excluded from these analyses. Zero on the x-axis represents nonselective NSAIDs, whereas positive values represent NSAIDs with COX-1:COX-2 selectivity greater than 1, and negative numbers are ratios less than 1. On the y-axis, the log of 0.0 signifies an OR of 1.0 for acute MI, and values below this signify ORs less than 1.0.
Solomon DH, Glynn RJ, Levin R, Avorn J. Nonsteroidal Anti-inflammatory Drug Use and Acute Myocardial Infarction. Arch Intern Med. 2002;162(10):1099-1104. doi:10.1001/archinte.162.10.1099
Copyright 2002 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2002
Although aspirin has been shown to protect patients from acute myocardial infarction (AMI), the effect of nonaspirin nonsteroidal anti-inflammatory drugs (NSAIDs) is not clear.
To determine whether NSAIDs have a similar effect or whether they differ in their effect on the risk of AMI.
We performed a case-control study of AMI in a large health care database containing information on all filled prescriptions, hospitalizations, diagnoses, and procedures for all patients covered by the New Jersey Medicaid or Medicare and Pharmaceutical Assistance for the Aged and Disabled programs. We identified 4425 patients hospitalized for AMI between January 1, 1991, and December 31, 1995, and 17 700 control subjects. Multivariate models were constructed to control for potential confounders.
A quarter of the cases and controls had filled a prescription for an NSAID in the 6 months before their AMI (cases) or a randomly assigned index date (controls); 9% had filled a prescription for an NSAID that overlapped with their date of AMI or index date. Overall, NSAID users had the same risk of AMI as nonusers, whether such use was measured on the index date (adjusted odds ratio, 1.04; 95% confidence interval, 0.92-1.18; P = .55) or at any time in the prior 6 months (adjusted odds ratio, 1.00; 95% confidence interval, 0.92-1.08; P = .92). However, use of naproxen was associated with a significant reduction in the risk of AMI (adjusted odds ratio, 0.84; 95% confidence interval, 0.72-0.98; P = .03).
Although NSAIDs have anti-inflammatory and antiplatelet effects similar to those of aspirin, we did not find that these drugs confer a protective effect against AMI. However, use of one specific NSAID, naproxen, appeared to be associated with a reduced rate of AMI, an effect recently suggested by a large randomized controlled trial as well.
RECENT PUBLICATION of the study by Bombardier et al1 has rekindled interest in the relationship between nonsteroidal anti-inflammatory drug (NSAID) use and acute myocardial infarction (AMI). Despite the anti-inflammatory and antiplatelet effects of NSAIDs, there has never been a randomized controlled trial that examined whether these drugs confer the protective effects of aspirin against AMI. In fact, in the study by Bombardier et al, patients with rheumatoid arthritis randomized to receive rofecoxib had a significantly higher rate of AMI than patients randomized to receive naproxen.1 Controversy has emerged as to whether this result was the consequence of a protective effect of naproxen, an increase in the risk of AMI caused by rofecoxib, or both.
Traditional NSAIDs inhibit cyclooxygenase-2 (COX-2) and thereby reduce prostaglandin E2 production, resulting in their analgesic, anti-inflammatory, and antipyretic effects.2 Nonsteroidal anti-inflammatory drugs also inhibit COX-1, which in turn lowers the level of the platelet aggregating factor thromboxane B2.3 A similar, but irreversible, inhibition of thromboxane B2 is thought to underlie the protective effect of aspirin against AMI,4- 6 but the relation between traditional NSAIDs and AMI is unclear. One observational study7 examined only women after menopause and found no association between NSAIDs and AMI, but no attempt was made to look at specific NSAIDs.
Understanding the relationship between NSAID use and the risk of AMI is important for several reasons. Quantifying the association between specific NSAIDs and AMI may help to clarify the controversial relationship between selective COX-2 inhibitors and AMI. If 1 or several NSAIDs reduce the risk of AMI, patients taking such agents might not also need to take aspirin, a combination associated with increased risk of gastrointestinal bleeding.8 The goal of the present study was to measure the relation between AMI and ongoing use of NSAIDs in a large, typical population of older adults, and to examine whether specific NSAIDs, particularly naproxen, are associated with a change in the risk of AMI.
Eligible patients were participants between January 1, 1991, and December 31, 1995, in the New Jersey Medicaid or Medicare and Pharmaceutical Assistance for the Aged and Disabled programs. The latter is a state-run drug assistance program for older adults with moderate incomes and provides prescription benefits with no or minimal co-payment. We first identified all patients hospitalized with a main diagnosis of AMI (diagnosis related groups9 121, 122, or 123 or an International Classification of Diseases, Ninth Revision,10 code 410). In addition, as in previous studies,11- 13 we required that the hospitalization be no longer than 180 days, nor could it be less than 3 days if the patient was discharged alive. In a recent study13 of more than 5000 persons, using similar data, these criteria had a positive predictive value of 96.9% for AMI, as validated by primary medical record reviews. Patients with AMI were considered cases, and the date of hospital admission for the AMI was defined as the index date. Potential control subjects were those who did not experience an AMI during the study period. A random clinical encounter was selected as the index date for these subjects. Controls were selected at a ratio of 4:1 to cases, with the groups' frequency matched on the basis of age.
We further required that all potential study subjects demonstrate continuous use of Medicaid or Medicare and Pharmaceutical Assistance for the Aged and Disabled, by having a filled prescription, physician visit, or procedure during the 1 to 180 days, 181 to 365 days, and more than 365 days before their index date. To reduce confounding by other drugs, we excluded all cases and controls who filled a prescription for any aspirin-containing compound or warfarin sodium during the study. We also excluded patients likely to have been regularly using aspirin, including those with evidence of prior AMI, angina, coronary artery bypass graft, percutaneous transluminal coronary angioplasty, cerebrovascular accident, transient ischemic attack, atrial fibrillation, migraine, other chronic headaches, or rheumatoid arthritis.
Based on these criteria, we created a study-specific database that contained information on the following: patient age, sex, race, physician visits, hospital and nursing home use, all recorded diagnoses, all filled prescriptions, and insurance program enrollment. During the study, essentially all prescribed drugs were covered without formulary restrictions. Deductibles and co-payments were absent or minimal, permitting comprehensive capture of medication use for enrolled subjects. All traceable personal identifiers were erased and converted to anonymous study codes to protect patient privacy.
Use of prescribed oral NSAIDs in the 6 months before the index date was the exposure of interest. For each filled prescription, we identified the National Drug Code (product identifier), date filled, dosage, quantity dispensed, and days' supply. Several aspects of NSAID use were characterized.
To determine the proximity of prescribed NSAID use to the index date, we examined which days were covered by any NSAID prescriptions during the prior 180 days. Three categories of proximity were defined: (1) a prescription that extended through the index date; (2) a prescription that ended 1 to 60 days before the index date; and (3) a prescription that ended 61 to 180 days before the index date.
The duration of prescription NSAID coverage in the prior 6 months included the sum of days supplied for all NSAID prescriptions (subtracting duplicate days). We created 3 categories of cumulative duration: 1 to 30 days, 31 to 90 days, and 91 to 180 days.
The dosage of each NSAID prescription was standardized against the 100% maximum anti-inflammatory dosage of each NSAID, based on a standard pharmacology textbook.14 We then calculated the percentage of maximum anti-inflammatory dosage for each day of each NSAID prescription. If multiple NSAID prescriptions covered the same day, we summed the percentage of the maximum anti-inflammatory dosage for each drug. An average was then calculated by summing all daily percentage data and dividing by the total number of days covered. This was divided into 3 categories: 50% or less, 51% to 75%, and 76% or more of maximum anti-inflammatory dosage.
The half-life for each NSAID was determined from a standard pharmacology textbook.14 Half-life was also considered as a continuous variable in 3 categories: shorter than 5 hours, including ibuprofen, indomethacin, ketoprofen, diclofenac, fenoprofen, and mefenamic acid; 5 to 14 hours, including naproxen, sulindac, ketorolac tromethamine, flurbiprofen, tolmetin, and etodolac; and longer than 14 hours, including piroxicam, oxaprozin, and nabumetone.
Each agent's COX-1:COX-2 selectivity ratio was defined based on previously published in vitro assays.15 This ratio was examined as a continuous and categorical variable. Agents with COX-1:COX-2 selectivity ratios greater than 1.0 were considered COX-1 selective. This group consisted of flurbiprofen, ketoprofen, fenoprofen, oxaprozin, tolmetin, indomethacin, and ibuprofen. Preparations with a COX-1:COX-2 selectivity ratio less than 0.25 were considered COX-2 selective, including etodolac and diclofenac. (The newer selective COX-2 inhibitors, celecoxib and rofecoxib, were not in use during the study.) Those NSAIDs with a COX-1:COX-2 ratio of 0.25 to 1.0 were considered nonselective agents; included in this group were naproxen, piroxicam, ketorolac, nabumetone, and sulindac.
Each drug's absolute capacity to inhibit COX-1 and COX-2 was defined separately, based on the concentration necessary to inhibit 50% of COX-1 or COX-2 activity in vitro.15 The daily dosage for each NSAID taken by every subject was converted into a concentration necessary to inhibit 50% of activity in vitro for COX-1, and another for COX-2, based on the peak serum concentration for a typical dosage.16 Separate continuous variables were calculated for COX-1 and COX-2 inhibition for each person taking an NSAID, represented as multiples of the concentration necessary to inhibit 50% of COX-1 or COX-2 activity in vitro; each was then categorized into quintiles.
Variables considered as potential confounders in multivariate models included clinical, sociodemographic, and health care use characteristics. Age, sex, ethnicity, and insurance status were defined from the enrollment files of Medicaid or Medicare. We examined the 180 days before the index date to determine whether patients had evidence of hypertension, diabetes mellitus, or congestive heart failure. In addition, a comorbidity index was calculated using inpatient and outpatient diagnoses, according to the method of Deyo et al.17 Health care resource use was determined for the 180 days before the index date, including the number of hospitalizations, number of different drug prescriptions, and nursing home use. The year of the index date was also considered.
We initially compared the baseline characteristics of patients who experienced an AMI (cases) with those who did not (controls) using χ2 tests. The characteristics of NSAID use were described for all patients and compared for cases and controls. The risk of AMI in patients with and without NSAID exposure was described by calculating odds ratios (ORs) and 95% confidence intervals (CIs) from bivariate logistic models. P values were examined to determine the significance of each OR. Crude ORs were calculated for each NSAID exposure category and for individual preparations. We then used multivariate logistic models that included all covariates to calculate adjusted ORs. No variable selection programs were used. In each of these analyses, nonuse of NSAIDs was the reference exposure. Stratified analyses were performed to compare male vs female naproxen users, and naproxen users aged 65 years or older vs those younger than 65 years.
To further compare NSAIDs with one another, similar analyses were conducted in a subsample of subjects who used NSAIDs. We calculated crude and adjusted ORs, 95% CIs, and P values from logistic models to determine if specific aspects of NSAID exposure or use of specific NSAIDs was associated with AMI. The reference exposure in these calculations was ibuprofen use, which represented the largest group of NSAID users other than naproxen users.
Finally, we used several approaches to examine the relationship between AMI and relative COX selectivity and absolute COX inhibition. First, COX selectivity and COX inhibition were assessed in models that included these (as defined in this section) as continuous variables and quintiles. A follow-up analysis considered only subjects taking a single NSAID throughout the study, plotting the log of the COX-1:COX-2 ratio against the OR of AMI. Because this relationship was nonlinear, curves were fit using cubic splines in Stata.18 All other analyses were performed using SAS statistical software.19
Table 1 presents the baseline characteristics of cases and controls. Because the ages of the 2 groups were frequency matched, the distributions are almost identical. Cases were more likely to be male and white. Slightly more persons with AMI than controls were covered by Medicaid, a marker of low-income status. As expected, patients with an AMI were more likely than controls to have a history of hypertension, diabetes mellitus, or congestive heart failure. They were also more likely to have been diagnosed as having any of the comorbid conditions studied. A larger proportion of cases than controls had been hospitalized or had filled a prescription in the prior 180 days.
We then compared use of any NSAID in the prior 180 days in persons who had an AMI vs controls (Table 2). Twenty-five percent of cases and of controls had filled a prescription for an NSAID during this period, and the characteristics of exposure were similar. An equal percentage had prescriptions covering their index date. Cases and controls also had equal distributions of duration and dosage of NSAIDs. The distribution of persons exposed to most NSAIDs was comparable across cases and controls.
Multivariate logistic models that included all covariates and NSAID exposure in the prior 180 days were examined. No relationship was observed between NSAID use in the prior 180 days and AMI (OR, 1.00; 95% CI, 0.92-1.08; P = .92) (Figure 1), or NSAID use on the index date and AMI (OR, 1.04; 95% CI, 0.92-1.18; P = .55). Likewise, no relationship was found for the proximity of NSAID use to the index date, the duration of NSAID exposure in the prior 180 days, or the dosage of NSAID.
We next focused on comparison of specific NSAID preparations in the 180 days before the index date (Table 3). In multivariate analyses, naproxen use was significantly less common among cases compared with controls; it was associated with a 16% reduction in the risk of AMI (OR, 0.84; 95% CI, 0.72-0.98; P = .03) (Figure 1). By contrast, etodolac (OR, 1.28; 95% CI, 1.00-1.64; P = .05) and fenoprofen (OR, 1.95; 95% CI, 1.16-3.30; P = .01) appeared to be associated with an increased risk of AMI. Ibuprofen use had no association with AMI (OR, 1.02; 95% CI, 0.88-1.18). Similar results were found when the analyses were restricted to NSAID users alone, with ibuprofen as the reference exposure. A reduced risk of AMI in naproxen users was seen in both sexes and all ages. However, there was no relationship between AMI risk and longer duration, higher dosage, or closer proximity to the index date for naproxen exposure (Table 4).
To attempt to understand the relationship between specific NSAID preparations and AMI, we examined whether half-life, COX selectivity, or specific COX-1 or COX-2 inhibitory potency was associated with AMI. No trend was seen between half-life and risk of AMI. Patients taking more COX-1 or COX-2 selective agents appeared to be at a slightly increased risk of AMI, while those taking the nonselective agents appeared to have a slightly lower risk of AMI (Figure 2) (P = .01 for the cubic spline in multivariate analyses). However, these findings were affected by data from groups with the largest sample sizes, naproxen and ibuprofen users. Finally, we assessed the absolute degree of COX-1 and COX-2 inhibition (regardless of selectivity). There was no consistent relationship between COX-1 inhibition and AMI, but patients taking NSAID regimens with higher COX-2 inhibitory capacity appeared to be at a slightly higher risk of AMI. Patients in the highest quintile of COX-2 inhibition had a 25% increased risk of AMI compared with NSAID nonusers (OR, 1.25; 95% CI, 1.08-1.45; P = .01).
After controlling for different potential confounders, NSAID use in general was not associated with an increase or decrease in the risk of AMI. However, naproxen use was associated with an apparent 16% to 20% reduction in AMI risk. This finding persisted in different subgroups of naproxen users and in patients using different dosages and durations of this agent, but there was no relationship between AMI risk and longer duration, higher dosage, or closer proximity to the index date. Overall, patients using agents that were nonselective with respect to COX-1:COX-2 specificity had a somewhat lower risk of AMI, and those taking NSAIDs with the highest COX-2 inhibition appeared to have a small increased risk of AMI.
A potential limitation associated with this study is the possibility of misclassification of NSAID exposure because of unmeasured over-the-counter use of NSAIDs or aspirin, or their use on an as-needed basis. Over-the-counter or as-needed drug use is not captured in the study database of filled prescriptions. However, the study was conducted in the context of prescription benefit programs for low-income and moderate-income patients, through which prescription NSAIDs were available for free or a minimal co-payment. As a result, over-the-counter use of NSAIDs, requiring out-of-pocket payment, was less likely than in the general population. In addition, the study was conducted using data from the first half of the 1990s, when over-the-counter NSAID use was less common. More important, because we found similar results after restricting the analyses to only NSAID users, if over-the-counter NSAID use was substantial, one would have to postulate that there was differential use of such preparations, depending on the prescription NSAID.
We attempted to limit over-the-counter aspirin use as a source of unmeasured confounding by excluding all patients with any diagnoses that might have been managed with aspirin, such as coronary artery disease, cerebrovascular disease, atrial fibrillation, rheumatoid arthritis, and chronic headaches, as well as evidence of prescribed aspirin-containing compounds. If any residual confounding occurred as a result of undetected aspirin use, it is unlikely that it would differentially affect only one NSAID, naproxen. Recently presented data from a detailed patient survey20 suggest that there is not differential use of aspirin between persons taking prescription NSAIDs, further supporting the validity of our conclusions.
Although the database that was used for this study presented some potential limitations, we know of no better sources of data for such a study. Clinical databases that contain information on electrocardiographic results and serum cardiac markers generally do not contain detailed drug exposure information that is based on filled prescriptions for large numbers of patients. The algorithm we used for identification of AMIs has been previously validated.13
The apparent lack of a dose-response, duration-response, or proximity-response relationship between naproxen exposure and AMI raises the possibility that naproxen itself is not involved in the causal pathway reducing the risk of AMI, but may be a marker for some other factor associated with a reduced risk of AMI. Furthermore, because we considered many potential relationships and performed multiple tests of association, there is the possibility of a chance finding.
In the study by Bombardier et al,1 in which patients with rheumatoid arthritis were randomized to receive rofecoxib or naproxen, the rates of AMI were 4-fold higher in patients taking rofecoxib. These findings could have resulted from a protective effect of naproxen, a risk-enhancing effect of rofecoxib, or both. The selective COX-2 inhibitors were not available during the period we studied, and these findings do not clarify the role of rofecoxib. However, the data presented herein are consistent with a possible protective effect of naproxen.
It is not clear from the data why naproxen may lower the risk of AMI. Although a potent COX-1 inhibitor might protect against AMI through inhibiting thromboxane B2 production and thus reducing platelet aggregation, naproxen is not a strong inhibitor of COX-1,15,21,22 and NSAIDs with greater COX-1 inhibition did not demonstrate this effect. Alternatively, an NSAID with strong COX-2 inhibition might reduce the inflammatory component of coronary disease by blocking prostaglandin E2 production23; however, naproxen is also not a potent COX-2 inhibitor.15,21,22 Naproxen is a nonselective inhibitor of COX-1 and COX-2; several other NSAIDs in this category (piroxicam and sulindac) also appeared to be associated with a slightly reduced risk of AMI, although not statistically significant. The effect does not appear to be related to naproxen's long half-life, because other NSAIDs with equally long or longer half-lives did not show this effect.
If confirmed in other populations, the findings of this study suggest that most NSAIDs are not associated with an increased or decreased risk of AMI. This has substantial clinical implications, indicating that patients who regularly use most NSAIDs also require the cardioprotective effect of aspirin. Aspirin may negate the gastrointestinal protective effects of the selective COX-2 inhibitors.24 The data reported herein are also compatible with the possibility that naproxen use is associated with a small reduction in the risk of AMI. To place the effect of naproxen in perspective, in a large randomized trial of daily aspirin use in primary prevention, patients in the intervention arm experienced a 44% reduction in the risk of AMI.5 Therefore, it would be false to equate the more modest effect of naproxen suggested in this study with the cardioprotection afforded by aspirin. Cyclooxygenase selectivity and specific potency may be associated with AMI risk, with nonselective agents possibly related to a reduced risk, and agents with higher COX-2 inhibition potentially conferring a slight risk. Other data suggest that selective COX-2 inhibitors may be associated with an increased risk of thrombosis and AMI.25,26 Although our study does not lend evidence to the mechanisms for this potential relationship, it is known that eicosanoids are important regulators of vascular homeostasis and that the COX-2 gene is preferentially expressed in areas of high endovascular shear stress.27,28 More research is needed to explore the relationship between specific nonselective and selective NSAIDs and the risk of AMI.
Accepted for publication January 31, 2002.
Dr Solomon received an Investigator Award from the Arthritis Foundation, Atlanta, Ga. This work was also supported by grant RO3-AG18395 from the National Institute on Aging, Bethesda, Md. No support from any pharmaceutical company was received for this research.
Corresponding author and reprints: Daniel H. Solomon, MD, MPH, Division of Pharmacoepidemiology and Pharmacoeconomics, Brigham and Women's Hospital, Harvard Medical School, 221 Longwood Ave, Suite 341, Boston, MA 02115 (e-mail: email@example.com).