General Practice Ltd
Is paracetamol better than nothing for treating OA?
better than nothing for treating OA?
Bandolier editors Dr Andrew Moore and Professor Henry McQuay find the quality of evidence on this simple matter leaves a lot to be desired
Zhang et a · 1
This review chose RCTs of paracetamol versus placebo and/or NSAIDs (which for our purposes loosely means traditional NSAIDs like diclofenac and newer cox-2 selective inhibitors like rofecoxib and celecoxib) published to July 2003. Only studies in osteoarthritis with radiographic or clinical diagnosis, or of pain associated with osteoarthritis were used. Trials had to be randomised.
There were no limits on dose of paracetamol, or of duration of trial, and both abstracts and full publications were accepted.
They found 10 trials, two of which were abstracts. There were four comparisons with placebo and eight with NSAIDs. The dose of paracetamol was 2,600mg to 4,000mg daily, and trial duration was one week to two years, but mostly of six weeks or less. A number of the trials were quite small.
Headline results: paracetamol was barely different from placebo, and was less effective than NSAIDs. Adverse events were no different for paracetamol or placebo, while NSAIDs caused more GI discomfort.
Wegman et a · 2
A different strategy was used, seeking only trials comparing paracetamol and NSAID to the end of 2001. Again osteoarthritis was the target condition, with pain and disability outcomes reported in full publications. It also reviewed guidelines.
This review had five trials with doses of paracetamol between 990 and 6,000mg daily. These had a duration that went from a single dose, to two years. Three of the trials appeared in the Zhang review1.
Headline result: a small benefit in pain relief in favour of NSAIDs.
We know that paracetamol is an analgesic, and experience is that in osteoarthritis it works for some people. Because it is safe, guidelines that make it their first-line option remain sensible. But what exactly
is the evidence?
We examined two recent systematic reviews1,2 detailed in the box below. One looks at trials comparing paracetamol to NSAIDs and placebo, while the other looks at comparisons with NSAIDs alone.
Both share the same problem. There are not that many trials. The problem is confounded because since they were done, newer trials have been published that the reviews did not include.
Let's start by looking at the reviews individually, and then move on to an update with the new trials.
How many trials are there?
A Bandolier search identified 16 trials that examined paracetamol in osteoarthritis and were randomised (not including abstracts). Which of these 16 trials do we want to include in a review? If we accept any size of trial, but specify a minimum quality criterion of being randomised and double blind, then that still leaves us with 16 trials. But five of the 16 trials had fewer than 50 patients and seven had fewer than 100.
What about validity?
Should we accept any dose, for instance? That doesn't make much sense, so we will accept only trials with a paracetamol dose of 3,000 or 4,000mg a day, doses generally used in osteoarthritis treatment. We can remove three trials, with 990mg, 2,000mg, and 6,000mg. (In other places and times different doses to the UK standard 500mg have been used.)
Should we accept any duration?
A single dose trial seems daft, so there should be some minimum. Six to 12 weeks tends now to be the minimum in osteoarthritis trials, so we will accept trials that last at least four weeks, to be generous. There is one single dose trial, four of one week, three of two weeks and one of three weeks.
Another outburst of generosity to equate 2,600mg daily as close to 3,000mg leaves us with seven trials, three with comparisons with placebo, and six with comparisons with NSAIDs or cox-2 inhibitors. The nine we have omitted were all of short duration, and predominantly tiny.
What we have left are trials we think we ought to be able to trust to give sensible and meaningful results. Both reviews found the trials published to the time of their search, five1 and three2 trials respectively.
Paracetamol versus placebo
No simple pooling of results is possible for efficacy in the three reports (2,000 patients) comparing paracetamol 4,000mg daily with placebo over six to 12 weeks. One report had four trials, and only one of those four trials showed any hint of difference between paracetamol and placebo.
The largest and best tria · 3 used a series of cut-off points, none of which showed a difference. It did suggest there might be a benefit of paracetamol over placebo in a subgroup of patients who had no sudden increase in pain, and without night pain, but this was a small proportion of the whole.
There were no differences in discontinuations overall or because of lack of efficacy, or in adverse events (table 1).
Paracetamol v NSAIDs and cox-2s
Again, simple pooling of efficacy results was not possible in the six studies (2,100 patients) that reported them. Paracetamol doses were 4,000mg daily in five trials and 2,600mg in one. One report had two trials, and in seven trials overall five found a significant benefit for NSAID over paracetamol for at least one efficacy measure.
Paracetamol had more discontinuations both overall and because of lack of efficacy. Over six to 12 weeks, compared with NSAIDs, one more patient would discontinue overall for every 10 patients treated with paracetamol, and one more would discontinue for lack of efficacy for every 15 patients treated. Paracetamol was associated with fewer patients discontinuing because of adverse events, with one fewer patient discontinuing because of adverse events for every 31 treated (though with very wide confidence intervals see
table 1). There were no differences between paracetamol and NSAIDs for patients reporting any adverse event, or for gastrointestinal adverse events or serious adverse events (table 1).
A 'heap' of evidence?
The words of the ninth centry historian Nennius I have made a heap of all that I could find when writing his Historia Brittanorum have resonance for us today. So precisely where does this 'heap' of evidence take us? We know that NSAIDs and
cox-2s are more effective than paracetamol. More patients with osteoarthritis will have a good effect, but more of them will have some adverse event that will make them want to stop taking it.
Overall, in six to 12 weeks after starting on paracetamol or NSAIDs something like 30 per cent of people will stop taking their medicine for one reason or another. About one in three or four will have some adverse event, even with placebo; about one in 10 will have a gastrointestinal adverse event, even with placebo; and one in 100 or so will have a serious adverse event, even with placebo.
Pulling these trials together is not an exercise in intellectual nihilism. We have deliberately chosen the best trials, in terms of their quality, their validity, and their size. Yet we are disappointed at their ability to discriminate between paracetamol, a known analgesic, and placebo, and between NSAIDs and paracetamol, when we know NSAIDs are likely to be better. It would be difficult, though, if paracetamol was a new drug. However cheap it was, it would be hard to convince purchasers to pay for something that did no more than doing nothing.
When you're in a hole, stop digging. This is possibly one of the most sensible bits of advice for the situation we find ourselves in.
Could the key message be that the trials are lacking in sensitivity, and that while they are the best we have, they are just not good enough? Reviews like this may not change clinical practice, but they may change research practice, and that in turn should lead to us being better informed.
This makes us think about how we should be wary of health economic analyses that fail to take account of all the evidence. If you sought cost comparisons between paracetamol, NSAIDs and cox-2s in osteoarthritis, you would find at least one apparently useful paper4.
It would tell you that paracetamol was quite a good buy, based on limited information. It may be, but if we can't tell the difference between paracetamol or nothing, how do we know?
More pertinent, how do we know we are not doing the wrong thing?
Andrew Moore is honorary professor of health sciences at University College Swansea and editor-in-chief of Bandolier
Henry McQuay is professor of pain relief
at the Oxford Pain Relief Unit and
co-editor of Bandolier
1 W Zhang et al. Does paracetamol (acetaminophen) reduce the pain of osteoarthritis?: a meta-analysis of randomised controlled trials.
Annals of the Rheumatic Diseases 2004 63: 901-907
2 A Wegman et al. Nonsteroidal anti-inflammatory drugs or acetaminophen for osteoarthritis of hip or knee? A systematic review of evidence and guidelines. Journal of Rheumatology 2004 31: 344-354
3 C Miceli-Richard et al. Paracetamol in osteoarthritis of the knee.
Ann Rheum Dis 2004 63: 923-930
4 CC Kamath et al. The cost-effectiveness of acetaminophen, NSAIDs, and selective COX-2 inhibitors in the treatment of symptomatic knee osteoarthritis. Value in Health 2003 6: 144-157