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

On supporting belief: NSAIDs and coxibs

Bandolier is occasionally allowed out to attend the odd meeting or two. And some are very odd indeed, especially when it comes to the use of evidence.

There is the wrong way. This is characterised by finding some piece of evidence that backs up whatever opinion is already held. Ideally, something from a randomised trial, or even a meta-analysis does the trick, especially with some incomprehensible output, like an odds ratio portrayed in a forest plot. But of course, it's always then possible to come out with that immortal phrase 'the evidence shows….' It helps, of course, to do this with swaggering confidence.

Then there is the right way, which, is always a bit more tricky. This involves setting some standards as to what constitutes good evidence, and holding up what evidence there is to that standard. Evidence that fails to meet the standard can be discarded, or at least used with caution. Even evidence that does meet the standard may even then not tell the whole story, perhaps being informative about efficacy, but not effectiveness, or being incomplete in some other way.

In a nutshell, then, trying to do things the right way can make anyone look like a hesitant fool compared to the confident assurance of someone who just wants to win the argument. Bandolier has chosen a current controversy, that of cardiovascular problems with NSAIDs and coxibs, as an example of how to win an argument, if not be right.

Selecting some data

If one thing is true, it is that over the last decade research into clinical effects of cyclooxygenase-2 inhibitors has produced a huge amount of information from randomised trials and observational studies. Most comes from patients with pain, but some comes from exploratory work in precancerous conditions or dementia.

For the purposes of this exercise we will choose the following:

The information was obtained from the published papers, or other commentaries that included the data in the form required. Information from the MEDAL programme was provide by Merck Inc.

The raw data

Table 1 shows the raw data used for the analyses, with data for CLASS and TARGET separated by different comparators, which were diclofenac, naproxen, and ibuprofen. Average duration of therapy was derived from information for all patients, and may be slightly different for aspirin or non-aspirin populations, and varied from about a third of a year for CLASS to 1.5 years for MEDAL.



Table 1: Numbers of myocardial infarctions, patients, and patient years of exposure for coxibs and NSAIDs in large randomised trials, for all patients and by low dose aspirin status



Trial Coxib NSAID
Number of MIs
Number of patients
Patient years of exposure
Number of MIs
Number of patients
Patient years of exposure
Average years on treatment
All patients
Coxib
NSAID
Coxib
NSAID
Class Celecoxib Ibuprofen
7
1997
722
7
1985
690
0.36
0.35
Class Celecoxib Diclofenac
11
1990
719
4
1996
694
0.36
0.35
Vigor Rofecoxib Naproxen
20
4047
3035
4
4029
3022
0.75
0.75
Target Lumiracoxib Ibuprofen
5
4376
3326
7
4397
3034
0.76
0.69
Target Lumiracoxib Naproxen
18
4741
3651
10
4730
3595
0.77
0.76
Medal Etoricoxib Diclofenac
111
16833
25858
122
16504
24797
1.54
1.50
Patients not taking low dose aspirin
Coxib
NSAID
Coxib
NSAID
Class Celecoxib Ibuprofen
6
1574
569
2
1567
545
0.36
0.35
Class Celecoxib Diclofenac
7
1580
571
2
1602
557
0.36
0.35
Vigor Rofecoxib Naproxen
20
4047
3035
4
4029
3022
0.75
0.75
Target Lumiracoxib Ibuprofen
4
3401
2585
5
3431
2367
0.76
0.69
Target Lumiracoxib Naproxen
10
3549
2733
4
3537
2688
0.77
0.76
Medal Etoricoxib Diclofenac
65
11104
17057
61
10918
16404
1.54
1.50
Patients who were taking low dose aspirin
Coxib
NSAID
Coxib
NSAID
Class Celecoxib Ibuprofen
1
423
153
5
418
145
0.36
0.35
Class Celecoxib Diclofenac
4
410
148
2
394
137
0.36
0.35
Vigor Rofecoxib Naproxen
0
0
0
0
0
0
0.75
0.75
Target Lumiracoxib Ibuprofen
1
975
741
2
966
667
0.76
0.69
Target Lumiracoxib Naproxen
8
1192
918
6
1193
907
0.77
0.76
Medal Etoricoxib Diclofenac
46
5729
8801
61
5586
8393
1.54
1.50


The number of events was small for most of the trials, a product of low event rates (about 0.5% a year for myocardial infarction in the age group in the trials, and relatively short duration. Only the MEDAL programme, because of its combination of size and duration, had more than 30 MIs in total. Of course, when split by use or non-use of LDA, the numbers of events is smaller still, which increases the possible effects of chance. This is important if the aim of the exercise is to find arguments to use for or against the proposition that coxibs cause heart attacks.

Results

Table 2 has the results for calculations based on aspirin status, by the number of patients, or the number of patient years of exposure, and by comparison with all NSAIDs and non-naproxen NSAIDs. Figures 1-3 show the L'Abbé plots for the comparison with all NSAIDs by number of patients, for all patients (Figure 1), those not taking LDA (Figure 2) and those taking aspirin (Figure 3).



Table 2: Analysis based on the number of patients, or patient years of exposure, by aspirin status for coxibs compared to all NSAIDs and non-naproxen NSAIDs



A: Calculations based on the number of patients
MI/Total (N)
MI/Total (%)
Patient group
Coxib
NSAID
Coxib
NSAID
Relative risk
(95% CI)
NNH
(95% CI)
Comparison with ALL NSAIDs
All patients
172/33984
154/33641
0.51
0.46
1.1 (0.89 to 1.4)
not calculated
Taking LDA
60/8729
76/8557
0.69
0.89
0.77 (0.55 to 1.1)
not calculated
Not taking LDA
112/25255
78/25084
0.44
0.31
1.4 (1.07 to 1.9)
750 (420 to 3900)
Comparison with non-naproxen NSAIDs
All patients
134/25196
140/24882
0.53
0.56
0.94 (0.74 to 1.2)
not calculated
Taking LDA
52/7537
70/7364
0.69
0.95
0.72 (0.50 to 1.03)
not calculated
Not taking LDA
82/17659
70/17518
0.46
0.40
1.2 (0.84 to 1.6)
not calculated
B: Calculations based on the number of patient years of exposure
MI/Total (patient years)
MI/Total (% per year)
Patient group
Coxib
NSAID
Coxib
NSAID
Relative risk
(95% CI)
NNH
(95% CI)
Comparison with ALL NSAIDs
All patients
172/37311
154/35832
0.46
0.43
1.1 (0.86 to 1.3)
not calculated
Taking LDA
60/10761
76/10248
0.56
0.74
0.75 (0.54 to 1.05)
not calculated
Not taking LDA
112/26550
78/25583
0.42
0.30
1.4 (1.03 to 1.8)
850 (450 to 7200)
Comparison with non-naproxen NSAIDs
All patients
134/30625
140/29215
0.44
0.48
0.92 (0.73 to 1.2)
not calculated
Taking LDA
52/9843
70/9342
0.53
0.75
0.71 (0.49 to 1.02)
not calculated
Not taking LDA
82/20782
70/19873
0.39
0.35
1.1 (0.81 to1.54)
not calculated




Figure 1: Individual comparisons of MI rates with coxibs and NSAIDs (solid symbols naproxen) - all patients







Figure 2: MI rates with coxibs and NSAIDs - patients not taking LDA







Figure 3: MI rates with coxibs and NSAIDs - patients taking LDA





What is interesting about the Figures is that there is a tendency for the larger studies to be smack on the line of identity, while the smaller ones, most likely influenced by the random play of chance, are found away from the line of identity.

Patients taking LDA had a higher risk of myocardial infarction than those not taking LDA, by a factor of at least two-fold, as might be expected, so there is an argument that analysis by aspirin status is a legitimate and sensible thing to do.

For almost all the various sub-group analyses undertaken, there was no statistical difference, with the confidence interval of the relative risk including 1. In those taking LDA, the relative risk was invariably below 0.8.

In those patients not taking LDA there was a statistical increase, whether the results were analysed by number of patients or by number of patient years. The relative risk in both cases was 1.4, with the lower confidence interval approaching 1. The NNH was about 800.

Argument for a link between coxibs and MI

It's staring you in the face. In those folk not taking LDA there was a statistically significant increase in MI in people taking coxibs. What more do you want? For every thousand people you give a coxib to, one is going to have a heart attack. And they are being given out like sweeties. It is a potential public health disaster.

Just because people have a little pain? Give me a break! Surely this isn't worth any risk.

Argument against a link between coxibs and MI

All you get is bare statistical significance for one sub-group, of one set of trials, and for one outcome? And not for the comparison with non-naproxen NSAIDs, with overall results influenced by naproxen? Maybe you should wonder whether there is a real effect at all, or perhaps whether naproxen is the only one of your drugs reducing MIs. Why is there no effect in people at greater risk of a heart attack, but only in those at lower risk? Can you explain that? Is there a biologically plausible mechanism for such a situation?

Anyway, overall there is no effect statistically, even with coxib doses generally above those used clinically. Any difference is clinically insignificant and probably unmeasurable even if it were there. It is always useful to have several different therapies available because of the large inter-individual differences between patients, whose genetic reasons we are just beginning to understand. And chronic pain is miserable, with huge negative impact on quality of life. So don't worry.

Comment

Let's reiterate here that this exercise is to demonstrate how opposing views might be generated from the same data, chosen because there was an overall sufficiency of events from large randomised trials in which the MIs were judged by independent blinded endpoint committees.

All of us feel the need to question data, to ask about whether bigger or smaller effects occur in those at higher risk, or on different drugs, or who wear blue or are Welsh. Such salami slicing is disapproved of when used to find a positive result in a trial because of the loss of statistical power that results in turning a randomised trial, in effect, into an observational study with good data collection in a well-defined population.

What's the right answer? Curiously, with the amount of information available, it remains a tad elusive. As more information becomes available, the weight appears to be on the side of no increased risk from coxibs at standard doses, or at least nothing substantial. The trouble is that we know less about NSAIDs than the coxibs; we only know anything sensible about NSAIDs because of their inclusion in coxib trials.

And yet

There's always an 'and yet'. Two simple thoughts stand out.

Firstly, that MI rates seem to be lower with naproxen than other NSAIDs, as the Figures show, at least within the context of regular long term dosing in a clinical trial.

Second, the evidence is clear of higher rates of myocardial harm for coxibs and possibly aspirin compared with placebo in trials in precancerous conditions, though emphatically not in dementia. The former is characterised by low annual event rates with placebo, with rates less than a quarter of those in the latter.

There are some signals where cardiovascular risk is low - but that may just be another case of reading too much into the data. In any event, it is interesting to see the results from these trials and try and relate them to current guidance.

And finally

And finally, this is nothing like the whole story in the arguments about coxibs, NSAIDs, and cardiovascular events. It isn't even a partial story. All it is is an attempt from one set of data to show how "evidence" can be organised to support two sides, or any sides, of an argument.

What we have to do is forget, if we can, our prior beliefs, and judge the evidence in front of us. That includes issues around quality, size, and validity, as well as issues like the weight we give to multiple comparisons without any statistical adjustment for those multiple comparisons. The Figures suggest, once again, that the big battalions might be the ones that don't speak with forked tongue.

previous or next story