What's wrong with Shang et al.?

Shang et al. recently published a meta-analysis comparing homeopathy with 'conventional' treatments in the Lancet (Lancet 366; 726-732). This is the most recent of the meta-analyses cited by Ben Goldacre in his excellent Guardian piece on homeopathy. Homeopaths have not been amused by this paper, and have tried very hard to discredit it. This is perhaps predictable since the paper concludes that homeopathic treatments have no effect beyond placebo, once the poorly-designed and biased studies have been stripped out. The consensus amongst homeopaths seems to be that the study is outrageously flawed, and scientists show themselves up by referring to it. Some recent discussions of it involving homeopaths can be found here and here. Despite the large amount of noise surrounding the discussion of this paper, I haven't seen a lot of signal, and so far I haven't seen a critique that stands up to scrutiny.

What did the paper do? The authors set out to test the hypothesis that homeopathic treatment effects can be attributed to the placebo effect. If that were the case, then any positive trial results for homeopathy would have to result from poor study design and/or bias. The authors tested this proposition by identifying 105 papers reporting 110 trials of homeopathy, and matching them with 110 trials of 'allopathy', or conventional treatments, on the basis of disorder treated and type of outcome measured. The authors then assessed the methodological quality of the papers, based on factors such as whether the trial was adequately blinded and whether it was adquately randomised. The paper found that for all the homeopathic trials, there was an effect beyond placebo. However, when the trials that were of low methodological quality and/or had sample sizes that were small were stripped out of the analysis, the remaining 8 trials showed no effect beyond placebo. On the other hand, when the same procedure was followed for the conventional medicine trials, the six remaining trials did show an effect beyond placebo.

So far, so good. What have been the criticisms of the paper?

One criticism has been that the trials deemed to be large and of higher quality were not identified, and that the reporting of the meta-analysis was inadequate. This criticism does carry some weight, and the reporting in the original paper was not good enough. However, the authors recognised the problem, and rectified it by identifying the trials in a reply to published criticisms that appeared in the Lancet (Lancet 366: 2083). You can find all the details of the study via apgaylard's blog here. So, this criticism is no longer valid.

Another criticism has been that the meta-analysis only uses 8 papers out of 105 to conclude that homeopathic remedies are no better than placebo. This seems to totally miss the point of the study. For one thing, it's a meta-analysis, so it pools studies in order to get more statistically significant results than single studies. The eight studies of homeopathy have a total n of 1,923, which is quite respectable. Also, Shang et al. have not employed some sort of sleight of hand to dismiss the other 97 papers. They have filtered them out because they are of inadequate methodological quality and/or size, based on clearly stated criteria. This allows the authors to compare the results from all the studies with the results from the best studies. When you use only the best studies, there is no longer any benefit for homeopathy beyond placebo. In contrast, using the best studies of conventional treatments, there is an effect beyond placebo. Again, this is the whole point of the study, and criticising it on the basis that it seeks to use the best-quality studies seems somewhat misguided.

Another common criticism from homeopaths is that the study doesn't test 'real' homeopathy. Shang et al. split studies of homeopathy into four types:

1. Classical homeopathy: individualised treatment based on homeopathic history-taking
2. Clinical homeopathy: no history-taking involved, each patient gets the same remedy
3. Complex homeopathy: patients take a mixture of several different remedies
4. Isopathy: the agent judged to be the cause of the disorder was used

For example, here's a website where they state flat out that there is no such thing as clinical homeopathy. This would be news to anyone who has wandered into Boots and seen the homeopathic remedies on sale there. More commonly, the criticism is that only 'classical homeopathy' is really homeopathy, and the other types don't count. Even if we allow this criticism, the fact is that 18 of the included trials were of 'classical homeopathy', as defined by the authors, and two of those made it into the group of eight large, high quality trials. The statistical analysis also showed that there was little evidence that effects differed between different types of homeopathy. So, not only did the study include trials of individualised homeopathy, it showed that these were no more effective than the other forms of homeopathy.

So, on the whole, it seems to me that the methodology of Shang et al. is reasonable, and the conclusions justified. I think it's probably true that no study is entirely without flaws, and I'm willing to be corrected on this. But so far I've seen no good criticism of the Shang et al. study that invalidates its conclusions.

Edit: Just as an aside it's interesting to read the second last paragraph of Shang et al., where they discuss the place of homeopathy in treatment systems. I take the liberty of reproducing the paragraph below:

"We emphasise that our study, and the trials we examined, exclusively addressed the narrow question of whether homoeopathic remedies have specific effects. Context effects can influence the effects of interventions, and the relationship between patient and carer might be an important pathway mediating such effects. Practitioners of homoeopathy can form powerful alliances with their patients, because patients and carers commonly share strong beliefs about the treatment’s effectiveness, and other cultural beliefs, which might be both empowering and restorative. For some people, therefore, homoeopathy could be another tool that complements conventional medicine, whereas others might see it as purposeful and antiscientific deception of patients, which has no place in modern health care. Clearly, rather than doing further placebo-controlled trials of homoeopathy, future research efforts should focus on the nature of context effects and on the place of homoeopathy in health-care systems."

This seems to be entirely reasonable, and suggests that Shang et al. have no particular bias against homeopathy.