There are days when I wish I could say, “Look for a meta-analysis that answers your question and trust it blindly.” Do vitamin supplements work? Look for a meta-analysis! Are sausages bad for your health? Meta-analysis! Will an aspirin a day keep the doctor away? Meta-analysis!
Meta-analyses are often held up as the best form of scientific evidence. Forget relying on doctors reporting on a single case or a string of cases; forget depending on scientists looking backwards in time at a group of people who had an intervention versus another that did not; forget counting on prospective studies; forget even clinging onto a single randomized clinical trial. A meta-analysis has the power to put all of the studies together, synthesize their results, and produce a single magical number. This number can tell you if the intervention works or doesn’t work according to the best evidence we have. Why couldn’t we trust this magical number?
Although meta-analyses are wonderful, they can suffer from one problem: garbage in, garbage out. As an example, if a student takes four tests meant to assess the same aptitude but the tests were badly designed, the fact that the student’s mean score ends up being 92% is irrelevant. If the tests that go into the averaging machine are bad, the final result is just as bad.
We were recently sent
a meta-analysis to look at of 10 trials on the use of craniosacral therapy for pain. The conclusion of this meta-analysis is that craniosacral therapy works. But does it?