My attention fell this week on the reporting of an article accepted for publication in one of the main journals in my field. According to the ABC, placebo treatments for pain can be effective even when patients know that it’s a placebo.
The source for this report is the following article which is about to be published in the premier pain research journal Pain (as an interesting aside, note how the more authoritative a scientific journal is, the shorter its name is. I give you Nature, Science, Cell, Circulation, Lungto name just a few).
On closer examination, this publication becomes somewhat less newsworthy. It has all the classic hallmarks of a fluke result that is partly the result of the statistics used, and partly over-interpretation of fairly modest results. Given the authors claim they:
turn our understanding of the placebo effect on its head
I thought I had better take a look since this would be big news if true.
The study took place in Portugal, in a hospital pain clinic enrolling chronic back pain patients. The methodology was a small randomised controlled trial. Although double-blinding could have been possible and would have added substantially to the strength of the study design, for some reason there was a clear difference between the way the two groups were treated.
The group who received the open-label placebo treatment were given large orange capsules while the “treatment as usual” (TAU) group received nothing. Double- blinding could have been achieved by giving all the subjects placebo pills but having a separate researcher deliver the information about treatment allocation.
Instead, the placebo group were given a special briefing that was not given to the other group. To further compound the difference in treatment, the TAU group were told they would be given an opportunity to take the pills after the three weeks of the study period had been completed. Such a design is often used as a recruitment tool, but it inevitably enhances expectations and is part of a separate, distinct bias in research known as the Hawthorne effect.
There are a couple of other, smaller more technical flaws in the design but even if we decide not to quibble about those details, the results are far from impressive.
The authors report that despite their effective randomisation procedures, there was a difference in baseline pain score of 1.25 on a 10 point scale. This is not a small difference. The claimed effect of the placebo treatment was 1.5 points. Allowing that the treatment as usual group was already suffering slightly less pain on average, one would anticipate the gains from any treatment to be slightly smaller. That is exactly what was shown.
In assessing a new treatment for consideration in clinical use, the minimum effect size that will attract interest is at least 2.0 points on a 10 point scale.