Risk of myocardial infarction in patients taking cyclo-oxygenase-2 inhibitors or conventional non-steroidal anti-inflammatory drugs: population based nested case-control analysis

Study population and data source

We used data from UK general practices contributing to the new QRESEARCH database (http://www.qresearch.org/). This is a new clinical database containing the clinical records of more than 7 million patients ever registered with 468 practices over the past 16 years. The information recorded on the database includes demographics (year of birth, sex, socioeconomic data associated with postcode area), characteristics (height, weight, smoking status), symptoms, clinical diagnosis, consultations, referrals, prescribed drugs, and results of investigations. The database has been validated by comparing birth rates, death rates, consultation rates, prevalence, and mortality with other data sources including the general household survey and the general practice research database.12 The age-sex structure of the population has been compared with that reported in the 2001 census. We found a good correspondence for all of these measures (results available on request), although in some instances our prevalence figures were marginally higher than less recent data.13 We have also compared practices taking part in regional research networks on these and other measures and found a good correspondence.14 Detailed analyses have shown good levels of completeness and consistency.15 Similar databases have been used for studies investigating risk factors for coronary heart disease or effects of conventional NSAIDs.16^–19 In previous studies, the diagnosis of acute myocardial infarction has been confirmed by reviewing hospital discharge notes or comparison with the paper based records and found to be correct in more than 90% of cases.18^–20

The study period ran between 1 August 2000 and 31 July 2004 (the date of the most recent download of QRESEARCH data at the time of the analysis). We used this period as rofecoxib and celecoxib were both available on prescription in the United Kingdom.

Cohort definition

We selected practices that had their current Egton Medical Information Services (EMIS) computer system installed before 1 August 1999. We identified a cohort of patients registered on 1 August 2000 who had been registered for the whole of the preceding 12 months. Patients entered the study period on 1 August 2000 and left the risk period when they developed a myocardial infarction, died, or left the practice or when the study period ended, whichever was earlier.

We identified potential cases of acute myocardial infarction on the basis of a first time diagnosis of acute myocardial infarction recorded with appropriate Read codes during the four year study period. We excluded patients with a myocardial infarction before the start of the study period. We included in the cohort other patients with a diagnosis of coronary heart disease (but without mention of a myocardial infarction). We used this cohort to determine the incidence of myocardial infarction by age and sex.

Case-control analysis

Cases were all patients aged 25 to 100 with a first ever myocardial infarction identified in the cohort analysis. We included patients who had a diagnosis of myocardial infarction recorded as the cause of death. We matched up to 10 controls to each case by age, calendar time, sex, and practice by using incidence density sampling. All controls were alive and registered with the practice at the time their matched case had the myocardial infarction. We derived an index date for each control, which was the date of myocardial infarction of their matched case. We excluded cases and controls who had less than three years of computerised prescribing data available before their index date to ensure that the prescribing data were complete.

Assessment of exposure

We used standardised computerised routines to extract and code data on the medical history and use of prescribed drugs before the index date for each set of cases and controls. We identified all prescriptions for selective and non-selective NSAIDs in the three years before their index date. Twenty seven different NSAIDs were in use during the study period. We grouped the drugs as follows: celecoxib, rofecoxib, ibuprofen, diclofenac (including combination preparations), naproxen, other selective NSAIDS (meloxicam, etoricoxib, etodolac, valdecoxib), and other non-selective NSAIDs. We compared the prescribing rates for each drug per 1000 population with data from the prescribing cost analysis tool (PACT) for 2002 for drugs prescribed by general practice and dispensed in the community and found similar rates and rank order for the preparations.

For each drug group we identified the first and last prescription date and the total number of prescriptions issued in the three years before the index date. We coded the time since last prescription as not prescribed within the past three years, prescribed within 90 days (defined as current use), or prescribed more than 90 days ago. We categorised the total number of prescriptions for each drug group as zero, one to three, and more than three prescriptions. We tested for trend by using the actual number of prescriptions issued within the three year period. General practitioners in the United Kingdom issue patients with sufficient drugs to last one calendar month, so one prescription is approximately equivalent to one month of treatment.

Statistical analysis

We used conditional logistic regression for individually matched case-control studies to derive odds ratios with 95% confidence intervals for myocardial infarction associated with each of our drug groups. We made adjustments for possible confounding effects of smoking (smoker, not smoker, not recorded), comorbidity (diabetes, hypertension, coronary heart disease, osteoarthritis, rheumatoid arthritis, obesity), and deprivation in fifths. We used the Townsend score based on the 2001 census related data associated with the output area of each patient's postcode as a measure of deprivation.21 We categorised obesity as body mass index (kg/m²) less than 30, 30 or more, or not recorded. We also adjusted for the use of other drugs known to affect the risk of myocardial infarction or to be commonly associated with use of NSAIDS—namely, antidepressants16 (selective serotonin uptake inhibitors and tricyclic antidepressants separately) and statins.22 We also adjusted the results for each drug group for the other NSAID groups.

We calculated the numbers needed to harm by applying the incidence of first myocardial infarction per 1000 using the adjusted odds ratio from current usage of drugs. We calculated this separately for all patients aged 25 and over and for those aged 65 and over. We calculated approximate 95% confidence intervals. We examined two way interactions between different NSAIDs and aspirin and coronary heart disease. We fitted a second model restricted to cases and controls with complete data for smoking status and body mass index. We fitted a third model restricted to patients without either coronary heart disease or diabetes in order to reduce possible effects of residual confounding.23 We used Stata (version 8.2) for all the analyses. We selected a P value of 0.01 (two tailed) as significant.

Comparison with other studies

We found evidence to suggest an increased risk of myocardial infarction in patients taking ibuprofen, diclofenac, and rofecoxib and possibly also those taking other selective and non-selective NSAIDs. Patients currently taking ibuprofen had a 24% increased risk of an infarct, and patients who had been on ibuprofen for longer had higher risks. These findings are consistent with those of Ray et al, who reported an increased risk of acute myocardial infarction associated with use of ibuprofen in a high risk population over the age of 50.24 Patients taking diclofenac had a 55% increased risk of myocardial infarction, which is similar to that reported in a much smaller study of non-selective NSAIDs in patients with rheumatoid arthritis conducted on data from the general practice research database.27

Other observational studies conducted before data on COX 2 inhibitors were available reported increased risks of first time myocardial infarction associated with non-selective NSAIDs similar to the risks reported in our study. For example, Schlienger et al reported an adjusted odds ratio of 1.26 (95% confidence interval 1.01 to 1.57) for acute myocardial infarction in patients at low risk taking non-selective NSAIDs.17 However, the analysis did not adjust for the potential confounding effect of socioeconomic status, as this information was not available.

Our study included younger patients and longer follow-up than has been possible before.28 In addition, we had information on obesity, smoking, deprivation, and an important range of comorbidities. Rates of prescription of aspirin in the study population were also sufficient for us to adjust for its potential confounding effect and do analyses in patients taking aspirin compared with those not taking aspirin, which has not been possible before.28 We have reported high levels of comorbidity in both cases and controls, which underlines the need to investigate the risks and benefits of new treatments in elderly populations at high risk, especially given that many trials are not designed to detect adverse effects or exclude high risk patients.9 29

Discussion of methods

This is an observational study and therefore at risk of bias and confounding. For example, some confounding by indication could be present, such as if patients have been prescribed NSAIDs for chest pain that was actually angina. If this had been the case, we would have expected the results to apply equally to all drug groups in our analysis. Similarly, we considered whether channelling might be an explanation for our results.30 Patients who are more at risk of a myocardial infarction may be more likely to be prescribed a COX 2 inhibitor than patients at lower risk. Our analysis included adjustment for many potential confounders, including comorbidity, concurrent drug use, and deprivation, and we expect this to have minimised the impact of any channelling.

Our cases and controls were well matched on age, sex, practice, and calendar time, making this an effective design to assess the effects of different NSAIDs on risk of myocardial infarction. This approach allowed us to examine timing and duration and also to investigate interactions with aspirin and coronary heart disease. Our outcome (whether patients had a myocardial infarction or not) is likely to be well recorded on the general practice clinical databases, especially as the study period coincided with the publication of the National Service Framework for Coronary Heart Disease, which prioritises recording of cardiovascular disease, especially myocardial infarction.31 Apart from an analysis of the number of prescriptions (which can be used as a proxy for cumulative dose), we did not analyse dose. Although data on prescribed dose are available on the QRESEARCH database, we did not think that this would necessarily correlate well with the number of tablets taken by patients, as this can vary day to day according to levels of pain.

No recall bias occurred, as the exposure data were recorded on computer before the date of myocardial infarction. In addition, we included only patients who had been registered with the practice for the entire observation period in order to ensure that the prescribing data were complete. Misclassification of exposure status (that is, use of drugs) is unlikely, as more than 99% of all repeat prescriptions by general practitioners are recorded on computer because these drugs are not currently available in the UK without prescription. Ibuprofen is the only NSAID available without prescription, so some patients obtaining ibuprofen over the counter might have been misclassified as not being on ibuprofen. This is likely to be a small proportion in patients over 65 years, as they are entitled to free prescriptions in the United Kingdom and so tend to have drugs prescribed rather than buy them separately. Our results for ibuprofen were similar in an analysis restricted to patients aged 65 and over. Also, such misclassification, if present and if non-differential, would have had the effect of biasing the odds ratio towards one, making the exposure seem less harmful than it really is.32 Although we have adjusted for several confounding variables, some residual confounding may result from misclassification of those variables and confounding by unmeasured variables.

Conclusions

Since we completed this analysis, Merck has announced the immediate worldwide withdrawal of rofecoxib because of its adverse cardiovascular profile. Since we submitted this manuscript, adverse cardiovascular effects have been reported with both celecoxib and valdecoxib. Our study offers no reassurance that the increased risk of myocardial infarction is specific to rofecoxib alone or specific to COX 2 inhibitors. Indeed, we found similar effects with two commonly used non-selective NSAIDs (diclofenac and ibuprofen). We saw similar odds ratios for naproxen, other selective NSAIDs, and celecoxib, although the results did not reach the 0.01 significance level. This could be because of the relatively low usage of these drugs, which is likely to increase now that rofecoxib has been withdrawn. Lastly, we found no evidence to support a reduction in risk of myocardial infarction associated with naproxen.

This is an observational study and may be subject to residual confounding that we cannot fully correct for. However, we think that enough concerns exist to warrant a reconsideration of the cardiovascular safety of all NSAIDs.