Skip navigation
Link to Back issues listing | Back Issue Listing with content Index | Subject Index


Overview of observational studies
Effect of dose
Duration of treatment
Cardiovascular risk

Bandolier makes no apology for returning to this subject for two reasons. First it is a topic of considerable interest, and second because it is an area in which there has been a very large amount of research, observational and randomised trial, and affords important insights into how we should evaluate evidence.

The first indication that there might be an association between NSAIDs or coxibs and cardiovascular events arose only in 2000, with the publication of a large RCT comparing rofecoxib with naproxen. Subsequent RCTs and observational studies produced mixed results, with some supporting an association, and others not. Complexities included different drugs, doses, duration of use, and arguments about class effects. New analyses from observational studies offer another opportunity to examine the evidence.

Overview of observational studies

A systematic review and meta-analysis of observational studies [1] sought observational studies published between 2000 and 2005, that included case-control and cohort studies of NSAID or coxib use and myocardial infarction, and that provided information to calculate a relative risk comparing users to nonusers.

Sixteen studies were included in the review. They had just over 3.5 million participants with 68,000 cases of myocardial infarction, so 2% of the population investigated had a heart attack. The outcome was typically first myocardial infarction, hospital admission for myocardial infarction, or coronary heart disease death, or combinations. Most studies used computerised pharmacy records or prescriptions, and the definition of exposure was primarily use at the index date or within periods from a few days to months.

Results of the pooled analysis, for all doses, are shown in Figure 1. Combining data for all traditional NSAIDs showed a significantly higher relative risk compared with nonusers. Among individual NSAIDs diclofenac and ibuprofen, but not naproxen, had significantly increased risk. For coxibs, rofecoxib had an increased risk, but not celecoxib.

Figure 1: Relative risk of myocardial infarction, hospital admission for myocardial infarction, or coronary death, according to use of NSAID or coxib (all doses) compared with non-use, in pooled analysis of 3.5 million patients

Effect of dose

The same analysis [1] looked at the effect of dose for coxibs. For rofecoxib, doses of more than 25 mg daily produced a higher relative risk than doses within the licensed range of 12.5 to 25 mg daily (Table 1). For celecoxib there was no dose response.

Table 1: Effect of dose on relative risk of myocardial infarction with rofecoxib, by dose

Rofecoxib dose
(mg per day)
Relative risk compared with nonusers
(95% CI)
25 mg or less
1.2 (1.1-1.3)
more than 25 mg
1.8 (1.4 to 2.3)

A large case-control study of the UK General Practice Research Database [2] confirmed a dose response for rofecoxib, and suggested a possible dose response for celecoxib, but based on a small number of actual events in users of higher doses of celecoxib. There was no evidence of dose response with any NSAID, though this analysis did confirm an elevated relative risk with diclofenac.

Duration of treatment

There is some evidence of a significantly increased risk of myocardial infarction with the first prescription of rofecoxib, but not celecoxib, in older people. A Canadian study [3] compared current users and nonusers. In the period of the first prescription (average duration 28 days), there was a significantly increased rate with rofecoxib (Figure 2), but not at any other time, or at any time with celecoxib.

Figure 2: Relative risk of myocardial infarction by number of prescriptions

By contrast, the UK GPRD study [2] had no increased rate of myocardial infarction with rofecoxib for duration of use less than three months, and showed no consistent duration effect with any drug where there was a sensible number of events.

Cardiovascular risk

The UK GPRD study [2] looked at 3,643 cases of acute myocardial infarction among 490,000 people between 2000 and 2004. It examined the relative risk of myocardial infarction with current coxib use compared with nonusers according to the presence or absence of cardiovascular risk factors.

The results (Figure 3) show that for both rofecoxib and celecoxib there was no increased risk in people with coronary heart disease, and that the risk was highest in those users who did not have coronary heart disease, hypertension, or diabetes. Moreover, relative risks were higher in younger than in older people (Figure 4).

Figure 3: Relative risk of myocardial infarction by levels of preexisting cardiovascular risk

Figure 4: Relative risk of myocardial infarction by age


The evidence in these three studies, in over four million people and with over 70,000 myocardial infarctions means that we need to get our thinking caps on.

The first, and most obvious, thought is that we cannot keep on hiding behind the conventional idea that differences between traditional NSAIDs and coxibs constitute a class effect. Not as far as myocardial infarction is concerned, because at least one commonly used traditional NSAID, diclofenac, has apparently the highest risk.

A second, and much less obvious, thought is that we need to re-think our thinking about risk of myocardial infarction with coxibs or NSAIDs and baseline cardiovascular risk in patients. There is less additional risk with coxib or NSAID in the presence of more coronary heart disease risk factors, as we have seen before (Bandolier 137). This may seem counter-intuitive, but it is supported by the finding of no increased heart attacks in meta-analyses of randomised trials in patients with arthritis (generally higher risk), but finding that even low dose aspirin produces higher rates in people with low cardiovascular risk.

There are other comments, of course, notably about accepting the results of any one study with a pinch of salt, however well it seems to be done. The meta-analysis [1] showed quite clearly the spread of results obtained, both with NSAIDs and coxibs. For almost every analysis some studies had significantly lower risk, and others significantly higher. Only for diclofenac and doses of rofecoxib above 25 mg daily were there consistently higher estimates of risk.

Nor should we regard whole populations as being average. Over 2000 years ago the Roman natural philosopher Titus Lucretius was commenting in individual differences, and 100 years ago Archibald Garrold, the British biochemist, was discussing individual differences in drug metabolism. We continue to be sucked into considering average results from trials (the evidence), when few patients are average. The definition of evidence-based medicine is to use the best available evidence to make decisions about individual patients.

Oh, and perhaps as a final comment, we need to consider the matter of making sweeping changes to policy based on inadequate information. Could it be that a lesson is about to be learned? It may well be that decisions that have been taken seemed to be right at the time, but an alternative view might be that they were made on the basis either of too little information, or possible on an inadequate or inappropriate analysis of the information that was available. That is debatable, of course, but it makes for some fascinating thinking about how we assess evidence about adverse events.


  1. S Hernándex-Díaz et al. Non-steroidal anti-inflammatory drugs and risk of acute myocardial infarction. Basic & Clinical Pharmacology & Toxicology 2006 98: 266-274.
  2. F Andersohn et al. Use of first- and second-generation cyclooxygenase-2-selective nonsteroidal anti-inflammatory drugs and risk of acute myocardial infarction. Circulation 2006 113: 1950-1957.
  3. LE Lévesque et al. Time variations in the risk of myocardial infarction among elderly users of COX-2 inhibitors, Canadian Medical Association Journal 2006 174: Epub ahead of print. DOI: 10.1503/cmaj.051679

previous or next story