Does the varied use of NSAIDs explain the differences in the risk of myocardial infarction?

Authors


Dr T.-P. van Staa, General Practice Research Database, Medicines and Healthcare Products Regulatory Agency, 1 Nine Elms Lane, London SW8 5NQ, UK.
(fax: 44 (0)20 7084 2041; e-mail: Tjeerd.vanstaa@GPRD.com).

Abstract.

Objective.  To investigate the risk of myocardial infarction (MI) with diclofenac, ibuprofen and naproxen, taking into account the exposure patterns.

Design.  Retrospective cohort study using the General Practice Research Database.

Setting.  UK primary care.

Subjects.  Patients aged 40+ years prescribed a traditional nonsteroidal anti-inflammatory drug (NSAID). The control cohort was frequency matched by disease risk score.

Intervention.  Observational comparisons of MI rates.

Results.  The study included 729294 NSAID users and 443047 controls. The relative rate (RR) for MI increased with cumulative and daily dose (RR = 1.05 with 0–4 prior prescriptions and RR = 1.49 with 30+; RR = 1.05 with daily dose of <1200 mg ibuprofen and RR = 1.96 with dose of ≥2400 mg per day; for diclofenac, the RR was 1.13 with <150 mg per day and 2.03 with ≥300 mg per day). Diclofenac users had higher risks of MI (RR = 1.21) than ibuprofen (RR = 1.04) or naproxen (RR = 1.03) users, but exposure varied between these drugs. Taking into account these exposure differences, it was found that the risk of MI was comparable in current and past long-term users. The patterns of hazard rates (i.e. absolute risks) of MI were similar in patients using ibuprofen, diclofenac or naproxen with similar history of NSAID use. There was no statistical difference between ibuprofen, diclofenac and ibuprofen in the linear trends for cumulative dose or daily dose.

Conclusions.  Long-term users of traditional NSAIDs have an increased risk of MI that is probably explained by the underlying disease severity. Most of the differences in MI risk between diclofenac, ibuprofen or naproxen may be explained by their varied use.

Introduction

Nonsteroidal anti-inflammatory drugs (NSAIDs) are amongst the most widely used medications in the world. In 2004, rofecoxib, a selective inhibitor of cyclooxygenase 2 (Cox-2), was withdrawn from medical practice following a finding in a large randomized clinical trial of an increased risk of myocardial infarction (MI) [1]. Adverse cardiovascular effects have also been reported in randomized trials for celecoxib, valdecoxib and parecoxib [2–4]. In the last 5 years, there has been growing concern that some of the older traditional NSAIDs such as diclofenac also may have adverse cardiovascular effects similar to those of selective Cox-2 inhibitors. NSAIDs reversibly block both isoforms of Cox, but vary in their degree of selectivity [5]. A recent systematic review of epidemiological studies found that users of the NSAIDs, naproxen, ibuprofen and piroxicam did not have an increased risk of MI [relative rates (RRs) of 0.97, 1.07 and 1.06, respectively], whilst users of the NSAIDs diclofenac and indomethacin showed statistically significant increased risks of MI (RRs of 1.40 and 1.30, respectively) [6]. Whilst there may be biological plausibility for these findings [6], an alternative explanation could be that these drugs are used differently in actual clinical practice. It has been reported that some types of NSAIDs may be channelled to different patients [7–9]. Although a large number of epidemiological studies have evaluated the association between type of NSAID and risk of MI [6], the effects of any differences in NSAID exposure characteristics have not been widely studied. The objective of our study was to evaluate whether the traditional NSAIDs were used differently and whether any variation in exposure may have contributed to differences in MI risk.

Methods

Information for the study was obtained from the General Practice Research Database (GPRD), which comprises of computerized medical records of general practitioners (GPs). GPs play a key role in the UK health care system, as they are responsible for primary health care and specialist referrals. Patients are affiliated to a practice, which centralizes the medical information from the GPs, specialist referrals and hospitalizations. The data recorded in the GPRD include demographic information, prescription details, clinical events, preventive care provided, specialist referrals, hospital admissions and their major outcomes [10]. Data collection started in 1987 and ended for this study in 2006.

The study population consisted of all patients who received a first traditional NSAID prescription during the period of data collection and who were aged 40 years or older at the date of the first prescription (referred to as the index date). They were followed from the first NSAID prescription (i.e. the index date) to the patient’s death, patient’s transfer out of the general practice, the last GPRD data collection available for this study (first quarter of 2006) or the date of the first prescribing of a selective Cox-2 NSAID, whichever date came first. The total follow-up of the NSAID users was divided into periods of current and past exposure, with patients moving between these exposures in accordance to their use of NSAIDs. Current exposure was defined as the period from the date of a NSAID prescription to the end of expected duration plus 3 months. The expected duration of NSAID use was estimated using the daily dose as prescribed by the GP and the number of NSAID tablets. In case of missing data on the prescribed daily dose, the median expected duration of use was used. The period of past exposure was the time period from the end of a current exposure period to the prescribing of a new NSAID or up to end of data collection (whichever date came first). The past exposure period was divided into periods of 6 months.

A control population of patients not using NSAIDs was selected in two steps. First, an age-stratified sample of patients who had not used NSAIDs was taken from the total GPRD (about 1 million patients). Their index date was 1 year after the patient’s start of follow-up. For each gender and 10-year age stratum (40–49, 50–59, 60–60, 70–79 and 80+), 100000 patients were randomly selected. A disease risk score for the risk of MI was developed in the control population using age and gender and the following risk factors: smoking history, use of alcohol, body mass index, the socio-economic class of the practice location, region of the practice, number of visits to the GP in the 6–12 months before, history of diabetes, hypertension, systemic inflammation [rheumatoid arthritis (RA) or systemic lupus erythematosus], atrial fibrillation or heart valve disorders, ischaemic heart disease, cerebrovascular disease, renal failure, bilateral oophorectomy and prescribing 6 months before diuretics, statins, oral glucocorticoids, aspirin, anticoagulants and cardiac glycosides. Forward Poisson regression was conducted using a significance level of 0.05. We also investigated possible statistical interactions between risk factors and age and sex, which were included using a significance level of 0.001 or lower. The disease risk score was then calculated using the beta-coefficients of the final Poisson regression model (the exponential of these predictors are the RRs) both for the control and NSAID cohorts. The second step involved the selection of controls matched by age, gender and disease risk score to the NSAIDs cohort. The selection was carried out using frequency matching of the NSAID cohort by age (10 year), gender and decile of the disease risk score (within each age and gender stratum). The follow-up of the control patients was divided into periods of 6 months.

The rate of MI in periods of current NSAID use was compared to periods of past use as well as the rate found within the disease risk-matched controls. Poisson regression was used to estimate the RRs. We included age (at the date of each NSAID prescription or at the start of each 6-month period), gender and calendar year in the multivariate analyses. The risk factors at the index date that were used to estimate the disease risk score were also included. At each NSAID prescription, the exposure was characterized by type of NSAID, prescribed daily dose, cumulative dose (i.e. number of previous NSAID prescriptions), prior switching of NSAID type (i.e. prescribing of another type of NSAID previously) and number of gaps in NSAID prescribing. A gap was defined to have occurred if there was no repeated prescribing following a NSAID prescription within the period of expected duration of use plus 3 months. We also evaluated the medication possession ratio or compliance for patients who received another NSAID prescription in the 6 months before. The medication possession ratio was the expected duration of NSAID use divided by the time between the two prescriptions. As an example, a medication possession ratio of 0.50 would indicate that the previous NSAID use only covered 50% of the time between prescriptions, suggesting irregular use. Patients who received their last previous NSAID prescription more than 6 months ago were classified as patients with a long gap of use. First time users were patients who received their first NSAID prescription at least 1 year after start of GPRD data collection. At each NSAID prescription, the number of NSAID prescribed in the 1 year before, was also calculated approximating the prior duration of use. For patients with missing information on the expected duration of use, they were classified into a separate category.

The analyses were stratified by aspirin prescribing in the 6 months before each NSAID prescription, indication [Osteoarthritis (OA)/RA or other] and Framingham risk score [11]. The Framingham risk score was based on the weighing and scoring of the following risk factors: gender, age, systolic blood pressure, smoking, total cholesterol divided by high-density lipoprotein cholesterol, diabetes mellitus and left ventricular hypertrophy [11]. The population median was used for patients without values for blood pressure and cholesterol. In accordance with the Framingham equations, we classified ex-smokers within 4 years as current smokers whereas ex-smokers for more than 4 years were classified as nonsmokers.

We also evaluated the patterns of the hazard rates (i.e. absolute risk) for each of the outcomes. The hazard rates were estimated by dividing the follow-up time into 100 periods (over 30 months) and by calculating the absolute rate within each small period (the hazard rate provides the risk of the outcome over a small period of time). These estimates were then smoothed using the methods proposed by Ramlau–Hansen [12]. Two analyses were conducted. One concerned the pattern of hazard rates over duration of NSAID therapy where patients were followed from the start until the end of NSAID therapy or to the date of censoring, whichever date came first. The second analysis concerned an evaluation of the hazard rates after each NSAID prescription. In this analysis, the follow-up time started at the date of a NSAID prescription and ended with the next prescription or at the date of censoring, whichever date came first. This analysis of hazard rates over time after a prescription evaluated whether the (unadjusted) hazard rates varied with changes in exposure. The follow-up time shortly after a NSAID prescription would consist mostly of patients currently exposed to NSAIDs, whilst time distant from a prescription would include mostly past users. The advantage of this analysis is that it provides a graphical representation of the risks with changes in exposure, without the need to use arbitrary cut-offs for current and past exposure.

Results

The study population included 729294 traditional NSAID users. The control cohort included 443047 disease risk score matched patients (Table 1). The different types of NSAID varied as how they were used (Table 2). Compared with ibuprofen, diclofenac was used more by patients with a frequent history of NSAID use and by patients with prior switching between NSAID types. With respect to daily dose, 0.9% of the ibuprofen prescriptions were for a higher daily dose (≥2400 mg daily), 54.3% of diclofenac (≥150 mg) and 65.0% of naproxen (≥1000 mg).

Table 1.   Baseline characteristics of traditional nonsteroidal anti-inflammatory drug (NSAID) and control cohorts
  NSAID cohort (n = 729294)Control cohort (n = 443047)
AgeMean (years)58.058.2
40–4932.5%32.5%
50–5926.1%26.1%
60–6920.3%20.3%
70–7914.0%14.0%
80+7.2%7.2%
GenderWomen54.1%54.1%
Follow-upMean (years)6.15.6
Medical historyDiabetes mellitus4.3%4.5%
Ischaemic heart disease7.4%6.9%
Cerebrovascular disease3.1%3.4%
Table 2.   Characteristics of NSAID exposure stratified by type
 % Of all RxNumber of previous NSAID Rx≥1 Prior gaps in prescribingPrior switching in typeComplianceAspirin Rx, 6 months beforeHistory of OA/RARepeat NSAID Rx in 3 months
First useLong gapNSAID Rx within prior 6 months
0–45–910+Compliance <0.80Compliance 0.80+
  1. NSAID, nonsteroidal anti-inflammatory drug; OA, Osteoarthritis; RA, rheumatoid arthritis.

Ibuprofen31.1%47.9%14.8%37.3%55.8%37.9%21.9%28.2%25.1%24.7%12.3%16.3%56.2%
Diclofenac39.6%38.5%14.9%46.6%58.9%52.5%12.9%19.6%29.1%38.4%11.0%17.2%64.3%
Naproxen9.1%38.7%14.8%46.5%57.6%57.1%10.6%17.1%33.0%39.3%10.9%19.2%62.8%
Meloxicam3.8%22.2%14.4%63.5%64.6%72.7%4.1%9.5%30.9%55.5%16.5%23.2%79.1%
Indomethacin3.6%34.4%15.4%50.1%56.1%52.1%9.6%20.4%29.7%40.3%11.4%20.0%65.3%
Piroxicam2.0%33.0%15.2%51.8%53.2%53.0%7.1%13.5%32.8%46.6%9.6%22.9%68.8%
Mefenamic acid1.9%62.3%13.9%23.8%54.4%42.3%27.7%35.0%23.5%13.9%4.5%8.0%37.7%

It was found that the risk of MI was increased in patients currently using NSAIDs compared with controls (Table 3). Patients who discontinued NSAID therapy had a risk of MI similar to that of controls. The risk of MI was statistically significantly higher in current users of diclofenac (RR = 1.21) compared with controls, and not increased in current users of ibuprofen (RR = 1.04) or naproxen (RR = 1.03). Patients with a history of switching between NSAID types had higher risks of MI. Increasing daily dose was also associated with higher risks for MI for diclofenac and ibuprofen, but not for naproxen (although the numbers were small). When daily and cumulative dose (number of prior prescriptions), extent of intermittent use (i.e. gaps between prescriptions) and extent of switching between types were included into a regression model, it was found that cumulative and daily dose and switching were statistically significantly associated with the risk of MI. There was no statistical difference between ibuprofen, naproxen and diclofenac in the linear trends for cumulative and daily dose and extent of switching. As shown in Table 4, the RR of MI was associated with both NSAID compliance and duration of use. Patients with intermittent NSAID use and frequent short-term use had lower RRs of MI, whilst patients with long-term frequent use had the higher RRs of MI. The highest RR of MI was found in patients with a low Framingham risk score with long-term frequent NSAID use. In patients without recent prescribing of aspirin, the adjusted RR was 1.12 [95% confidence interval (CI): 1.06–1.19] for ibuprofen, 1.35 (95% CI: 1.28–1.42) for diclofenac and 1.11 (95% CI: 1.01–1.23) for naproxen. In patients with recent prescribing of aspirin, the adjusted RRs were 1.28 (95% CI: 1.16–1.40), 1.26 (95% CI: 1.14–1.40) and 1.28 (95% CI: 1.07–1.53), respectively.

Table 3.   RR of MI in current NSAID users compared with control patients or to past NSAID users stratified by NSAID exposure characteristics
NSAID useNumber of CasesNSAID vs. control cohortCurrent vs. past NSAID use
Age-sex-year adjusted RRa (95% CI)Fully adjusted RRb (95% CI)Age-sex-year adjusted RRa (95% CI)Fully adjusted RRb (95% CI)
  1. NSAID, nonsteroidal anti-inflammatory drug; RR, relative rates; MI, myocardial infarction; CI, confidence interval; NA, not applicable; GP, general practitioner.

  2. aAdjusted for age, gender and calendar year. bAdjusted for age, gender, calendar year smoking history, use of alcohol, body mass index, the socio-economic class of the practice location, region of the practice, number of visits to the GP 6–12 months before, history of diabetes, hypertension, systemic inflammation (RA or systemic lupus erythematosus), ischaemic heart disease, cerebrovascular disease and renal failure, and prescribing in the 6 months before use of diuretics, statins, oral glucocorticoids, aspirin, anticoagulants and cardiac glycosides prior to the index date.

Current56901.31 (1.27–1.36)1.12 (1.08–1.17)1.30 (1.26–1.34)1.25 (1.21–1.29)
Type
Ibuprofen (mg per day)19131.22 (1.16–1.28)1.04 (0.98–1.09)1.21 (1.15–1.27)1.16 (1.11–1.22)
 <12001761.29 (1.10–1.50)1.05 (0.91–1.22)1.28 (1.10–1.48)1.18 (1.01–1.36)
 12006001.21 (1.11–1.31)1.02 (0.94–1.11)1.20 (1.11–1.31)1.15 (1.05–1.24)
 1201–23991461.42 (1.20–1.67)1.22 (1.03–1.44)1.42 (1.21–1.67)1.38 (1.17–1.62)
 ≥2400102.28 (1.23–4.24)1.96 (1.05–3.65)2.22 (1.20–4.13)2.16 (1.16–4.01)
Diclofenac (mg per day)20331.40 (1.33–1.47)1.21 (1.15–1.28)1.37 (1.31–1.44)1.34 (1.28–1.40)
 <1506751.30 (1.20–1.40)1.13 (1.04–1.22)1.29 (1.19–1.39)1.26 (1.16–1.36)
 1506501.50 (1.38–1.62)1.28 (1.18–1.39)1.45 (1.34–1.57)1.40 (1.29–1.51)
 151–299351.35 (0.97–1.88)1.18 (0.85–1.65)1.31 (0.94–1.83)1.29 (0.92–1.79)
 ≥300102.28 (1.23–4.24)2.03 (1.09–3.77)2.20 (1.18–4.09)2.19 (1.18–4.07)
Naproxen (mg per day)5261.22 (1.12–1.33)1.03 (0.94–1.13)1.23 (1.12–1.34)1.16 (1.06–1.27)
 <10001551.19 (1.01–1.40)0.99 (0.85–1.17)1.19 (1.02–1.40)1.12 (0.95–1.31)
 10002501.31 (1.15–1.48)1.12 (0.98–1.27)1.30 (1.15–1.48)1.25 (1.10–1.42)
 >1000101.14 (0.61–2.11)0.92 (0.49–1.71)1.11 (0.60–2.06)1.00 (0.54–1.87)
Mefenamic acid951.52 (1.24–1.86)1.18 (0.97–1.45)1.55 (1.27–1.90)1.36 (1.06–1.27)
Indomethacin3091.54 (1.37–1.72)1.27 (1.13–1.43)1.55 (1.38–1.73)1.44 (1.29–1.61)
Meloxicam1371.33 (1.12–1.57)1.12 (0.94–1.32)1.25 (1.06–1.48)1.19 (1.01–1.41)
Piroxicam1221.17 (0.98–1.40)1.01 (0.84–1.21)1.19 (1.00–1.43)1.15 (0.96–1.37)
Number previous NSAID Rx
 0–426411.25 (1.19–1.30)1.05 (1.00–1.10)1.25 (1.20–1.30)1.18 (1.13–1.23)
 5–98051.20 (1.11–1.28)1.03 (0.96–1.11)1.18 (1.10–1.27)1.14 (1.06–1.23)
 10–198561.31 (1.22–1.41)1.14 (1.06–1.22)1.29 (1.21–1.39)1.25 (1.17–1.34)
 20–294501.34 (1.22–1.48)1.18 (1.07–1.30)1.32 (1.20–1.45)1.28 (1.17–1.41)
 ≥309381.67 (1.56–1.79)1.49 (1.39–1.60)1.61 (1.51–1.72)1.60 (1.49–1.71)
Gaps in previous use
 026851.37 (1.31–1.42)1.12 (1.07–1.17)1.37 (1.32–1.43)1.27 (1.21–1.32)
 112031.29 (1.21–1.36)1.10 (1.03–1.17)1.28 (1.21–1.36)1.23 (1.16–1.30)
 ≥218021.26 (1.20–1.32)1.14 (1.08–1.20)1.23 (1.17–1.29)1.24 (1.18–1.30)
Prior switching in type
 031261.27 (1.22–1.32)1.07 (1.02–1.12)1.27 (1.22–1.32)1.20 (1.16–1.25)
 110221.29 (1.21–1.38)1.12 (1.04–1.19)1.28 (1.20–1.36)1.24 (1.16–1.32)
 ≥215421.42 (1.34–1.49)1.24 (1.17–1.31)1.39 (1.32–1.46)1.36 (1.29–1.43)
Past144951.03 (1.00–1.05)0.91 (0.88–0.95)ReferenceReference
No use10834ReferenceReferenceNANA
Table 4.   RR of MI in current compared with past NSAID users stratified by NSAID compliance
 Rate current useaFirst useLong gapNSAID Rx within prior 6 months
Compliance <0.60Compliance 0.60–0.80Compliance 0.80+
Fully adjusted RRb (95% CI) Fully adjusted RRb (95% CI) Fully adjusted RRb (95% CI) Fully adjusted RRb (95% CI)Short-term use
fully adjusted RRb (95% CI)
Medium-term use
fully adjusted RRb (95% CI)
Long-term use
fully adjusted RRb (95% CI)
  1. NSAID, nonsteroidal anti-inflammatory drug; RR, relative rates; MI, myocardial infarction; CI, confidence interval; GP, general practitioner; OA, Osteoarthritis; RA, rheumatoid arthritis.

  2. aNumber of cases per 100 person-years. bAdjusted for age, gender, calendar year smoking history, use of alcohol, body mass index, the socio-economic class of the practice location, region of the practice, number of visits to the GP 6–12 months before, history of diabetes, hypertension, systemic inflammation (RA or systemic lupus erythematosus), ischaemic heart disease, cerebrovascular disease, renal failure and prescribing in the 6 months before of diuretics, statins, oral glucocorticoids, aspirin, anticoagulants and cardiac glycosides prior to the index date.

Overall Framingham risk score0.61.19 (1.10–1.28)1.21 (1.13–1.29)1.16 (1.09–1.25)1.33 (1.18–1.48)1.23 (1.12–1.36)1.38 (1.26–1.50)1.52 (1.38–1.67)
 Low0.11.20 (0.83–1.75)1.24 (0.97–1.59)1.43 (1.05–1.94)0.97 (0.48–1.95)1.03 (0.55–1.93)1.24 (0.73–2.11)2.77 (1.77–4.33)
 Middle0.41.13 (0.95–1.35)1.07 (0.93–1.22)1.16 (1.01–1.34)1.46 (1.16–1.84)1.19 (0.94–1.50)1.40 (1.16–1.69)1.77 (1.45–2.17)
 High0.91.11 (0.96–1.28)1.39 (1.24–1.55)1.06 (0.93–1.20)1.43 (1.19–1.74)1.31 (1.11–1.55)1.41 (1.21–1.63)1.53 (1.29–1.82)
 Very high1.51.23 (1.11–1.37)1.14 (1.02–1.28)1.17 (1.06–1.30)1.15 (0.95–1.38)1.16 (1.01–1.35)1.30 (1.14–1.48)1.30 (1.11–1.51)
Recent aspirin
 No0.51.12 (1.03–1.23)1.22 (1.13–1.31)1.17 (1.08–1.26)1.37 (1.21–1.56)1.21 (1.08–1.36)1.37 (1.25–1.51)1.56 (1.40–1.75)
 Yes1.61.53 (1.32–1.78)1.15 (0.99–1.33)1.12 (0.97–1.30)1.11 (0.86–1.44)1.27 (1.04–1.55)1.30 (1.09–1.54)1.30 (1.07–1.60)
OA/RA
 No0.61.22 (1.13–1.33)1.21 (1.13–1.30)1.18 (1.09–1.28)1.32 (1.16–1.51)1.22 (1.09–1.37)1.36 (1.23–1.51)1.54 (1.37–1.73)
 Yes0.90.97 (0.78–1.21)1.19 (1.00–1.41)1.10 (0.95–1.28)1.35 (1.08–1.68)1.29 (1.05–1.57)1.44 (1.23–1.68)1.51 (1.27–1.81)
Type
 Ibuprofen0.61.11 (0.99–1.24)1.14 (1.03–1.26)1.11 (0.97–1.26)1.28 (1.02–1.61)1.21 (1.00–1.48)1.33 (1.11–1.60)1.46 (1.17–1.83)
 Diclofenac0.61.21 (1.06–1.38)1.31 (1.17–1.46)1.16 (1.03–1.30)1.42 (1.18–1.69)1.26 (1.08–1.48)1.40 (1.23–1.60)1.76 (1.52–2.04)
 Naproxen0.61.22 (0.96–1.56)1.01 (0.81–1.26)1.33 (1.12–1.58)0.95 (0.66–1.38)1.17 (0.88–1.55)1.43 (1.13–1.82)1.06 (0.75–1.49)

Figure 1 shows the RR over duration of nonselective NSAID therapy compared to controls. It was found that the crude RR for MI increased linearly with duration of therapy, similar to the pattern observed for number of previous NSAID prescriptions.

Figure 1.

 Crude relative rates over time of myocardial infarction in current nonsteroidal anti-inflammatory drug (NSAID) users compared with controls (dotted lines represent the 95% CIs). CI, confidence interval.

Figure 2 shows the hazard rate (absolute risk) over time after a NSAID prescription (in case of a repeat prescription, the follow-up time was set to zero). In patients who had not used NSAIDs before, there was a substantially increased risk of MI around the time of the first prescription. After stopping therapy, the hazard rate reduced and stabilized after several months of the prescription. The hazard rate remained elevated in patients with a history of frequent use over several years after stopping NSAID therapy. Full statistical adjustment of these comparisons of current to past users within strata of history of NSAID use prior to a prescription found only statistically significant increases in recent starters: 0, 1.23 (95% CI: 1.15–1.31); 1–4, 1.09 (95% CI: 1.01–1.17); 5–9, RR = 1.05 (95% 0.95–11.5); 20–29, RR = 1.01 (95% CI: 0.87–1.17); 30+ prior prescriptions, RR = 1.11 (95% CI: 0.99–1.25).

Figure 2.

 Hazard rate of myocardial infarction in the time after a nonsteroidal anti-inflammatory drug (NSAID) prescription stratified by number of prior prescriptions (bsl00041 = no previous; bsl00000 = 1–4; bsl00084 = 5–9; bsl00043 = 10–19; ◊ = 20–29; * = 30+ previous NSAID prescriptions).

Figure 3 shows the hazard rate (absolute risk) over time after a NSAID prescription of ibuprofen, diclofenac or naproxen. No difference in the pattern of risk for MI was observed between the three different NSAIDs when stratifying by number of prior prescriptions. The risks were higher in patients with a history of frequent NSAID use, which remained elevated after discontinuation of the NSAID.

Figure 3.

 Hazard rate of myocardial infarction over time after a nonsteroidal anti-inflammatory drug (NSAID) prescription stratified by type and number of prior prescriptions (bsl00041 = diclofenac; bsl00000 = ibuprofen; bsl00084 = naproxen).

Discussion

Increasing duration of NSAID use was associated with larger risks of MI, but these risks remained elevated for several years after discontinuing NSAID exposure. Diclofenac users had higher risk of MI than naproxen or ibuprofen users, but history of use and prior switching differed between these drugs. The patterns of MI risk were similar between diclofenac, ibuprofen and naproxen after taking into account history of use.

We found a substantially increased risk of MI in current NSAID users with a long history of use. This is consistent with other epidemiological studies [6, 13]. However, we also found a similarly increased risk of MI in patients who had stopped taking NSAIDs after a long history of use. There are various possible explanations for an increased risk of MI in patients who stopped NSAIDs after prolonged use. First, patients who stopped NSAIDs may have started on another therapy that may increase the risk of MI. A second explanation for the stable risks after NSAID discontinuation may be that the adverse effects of NSAIDs may last several years after discontinuation. We did not find evidence in literature to support this hypothesis. However, the most plausible explanation for our findings is that long-term NSAIDs users are more severely sick than incidental NSAID users or nonusers. Patients with RA have increased risks of MI [14]. Systemic inflammation accelerates atherosclerosis, increasing the risk of cardiovascular disease. This confounding by indication is rarely fully addressed by including a few risk factors into a regression model. Our findings suggest that the underlying disease severity may contribute substantially to the increased risks of MI in long-term users. A meta-analysis of placebo-controlled randomized trials of at least 6 weeks’ duration found no increased risk of MI in patients using traditional NSAIDs, although naproxen was the most frequent NSAID in this analysis [15].

A meta-analysis of randomized trials could be considered superior to observational evidence, given the blinding and randomization. A substantial part of the evidence of the adverse cardiovascular effects of selective Cox-2 inhibitors has been derived from clinical trials [13, 16]. But randomization and blinding does not guarantee external validity and generalizability to patients in actual clinical practice. The patients in the large trials with selective Cox-2 inhibitors in patients with inflammatory diseases were all selected based on the expectation that they would need anti-inflammatory treatment daily for a prolonged period of time [17–22]. In GPRD, most patients received insufficient medication for long-term daily use of NSAIDs. The RRs for MI were strongly associated with NSAID exposure, lowest in patients with short-term or intermittent use and highest in those with long-term daily use. Given the selection criteria of the larger clinical trials, there are currently few clinical data to support or refute adverse cardiovascular effects of short-term or intermittent NSAID use. Most epidemiological studies, to date, have not addressed the large heterogeneity in the NSAID exposure characteristics in actual clinical practice.

Our finding that diclofenac, ibuprofen and naproxen did not differ substantially in their MI risk, after taking into account the NSAID exposure characteristics, contradicts those of two recent reviews [6, 16]. A review of observational studies reported a RR for MI of 1.40 for diclofenac, 1.07 for ibuprofen and 0.97 for naproxen [6]. A meta-analysis of randomized trials with selective Cox-2 inhibitors reported that diclofenac was associated with a statistically significantly increased risk (RR of 1.63), whilst ibuprofen (RR of 1.51) and naproxen (RR of 0.92) were not. These estimates were based on an indirect statistical estimation of the differences of effects of selective Cox-2 inhibitors in placebo- and active-controlled studies [16]. The main limitation of both meta-analyses is that effect modification by NSAID exposure characteristics was not considered. These meta-analyses would provide biased estimates if the drugs were used differently and the RR was modified by these characteristics. In the meta-analysis of randomized trials, 94% of the information on diclofenac was obtained from trials including patients with osteoarthritis, whilst this was 58% with naproxen (these numbers exclude trials with multiple indications) [23]. Complex clinical problems may not always be solved by simple statistical measures and heterogeneity in terms of the studies or the patients included in a meta-analysis should always be considered [24, 25].

Our association between daily dose and risk of MI is consistent with other studies [26, 27]. Our data suggest that the larger increases occurred at very high daily doses. It would be clinically important to establish whether the excess risks only occur with long-term continuous use or also with intermittent use. The number of patients in GPRD using these high daily doses was too small to evaluate this question. Meta-analysis of individual patient data from multiple large databases may be required for this. The concomitant administration of ibuprofen but not diclofenac has been found to neutralize the platelet stabilizing effects of aspirin [28]. We found no major difference in the risk of MI between ibuprofen and diclofenac in patients recently prescribed aspirin. One limitation of healthcare databases is that the exact timing of drug exposure and the extent of co-administration cannot be determined.

We found that the risk of MI was increased substantially around the time of the first NSAID prescription. This finding may be explained by protopathic bias, which occurs when a drug is inadvertently prescribed for an early manifestation of a disease that has not yet been diagnostically detected. This bias has been described in other settings. A study in the Netherlands found that recent starters of beta-2 agonists have an increased risk of MI especially in the patients with a history of ischaemic heart disease [29].

General Practice Research Database has been used in several other studies to study the risk of MI in patients using traditional NSAIDs [30–36] or selective Cox-2 inhibitors [37]. Despite the use of the same database, the results cannot directly be compared because of varying definitions of the exposure time-windows, different time-periods included and matching variables. Taking this into account, the overall results of the present cohort study were broadly similar to most of the previously published GPRD studies [30–36]. A recent GPRD case–control study concluded that long-term use of diclofenac was associated with substantially increased risks of MI, whilst this conclusion did not apply to ibuprofen or naproxen [35]. This study applied an unusual study design and whilst we were able to replicate these results in a separate analysis, we observed several methodological issues, such as suboptimal matching, lumping of current and past exposure and differential patient inclusion into the study cohorts. A limitation of the previous GPRD studies was that they all concerned case–control studies that lumped duration of exposure into few broad categories. Also, none of the previously published studies took fully into account the NSAID exposure characteristics or compared patients with similar history of NSAID use but different recency of exposure.

There are several limitations of our study. The main limitation is that the comparison groups were not randomized and that there may have been unmeasured confounding. The direction of bias is likely to be an overestimate in the RR with diclofenac compared to ibuprofen, as ibuprofen appeared to be used by less severely ill patients than diclofenac. We also cannot exclude the possibility that there are small differences in cardiovascular toxicity between diclofenac, ibuprofen and naproxen. The RRs for MI risk were generally higher for diclofenac compared with ibuprofen and naproxen, but these values did not reach statistical significance after taking into account the exposure characteristics. Also, the patterns of MI risk over time appeared similar between the different NSAID types. Another limitation is that ibuprofen is available over the counter without prescription which can result in a misclassification of exposure. However, our results comparing current to past users with a similar history of NSAID use did not change when the analysis was restricted to patients aged 65 years or older. For elderly patients, there is no financial charge for dispensing a prescription, whilst there would be a charge for the over the counter medication. Another limitation of the study is that we evaluated a large number of subgroups and that not all subgroup analyses had been prespecified in the protocol. It would be very useful if our results on the association between MI and NSAID exposure characteristics were replicated in another high quality healthcare database.

In conclusion, patients using NSAIDs long-term had a substantially increased risk of MI, but this excess risk is likely to be explained by underlying disease severity. Diclofenac users had increased risks of MI compared with ibuprofen or naproxen, but these differences were mostly explained by heterogeneity in the way these drugs were used. The patterns of MI risk were similar between diclofenac, ibuprofen and naproxen after taking into account exposure characteristics.

Conflict of interest statement

No conflict on interest.

Acknowledgements

We thank Dr Lesley Wise for reviewing the paper. The views expressed in this paper are those of the authors and do not reflect the official policy or position of the Medicines and Healthcare products Regulatory Agency, UK.

Source of funding

This study was funded by the Division of Vigilance and Risk Management of Medicines of the Medicines and Healthcare Products Regulatory Agency, UK. GPRD is owned by the UK Department of Health and operates within the Medicines and Healthcare products Regulatory Agency (MHRA). GPRD is funded by the MHRA, Medical Research Council, various universities, contract research organizations and pharmaceutical companies.

Ancillary