Detailed guides to painful problems, treatments & more

Recommendations are made in the absence of any good treatments

PainSci » bibliography » Colquhoun 2017
Tags: scientific medicine, stats

One article on PainSci cites Colquhoun 2017: The “Impress Me” Test

PainSci commentary on Colquhoun 2017: ?This page is one of thousands in the bibliography. It is not a general article: it is focused on a single scientific paper, and it may provide only just enough context for the summary to make sense. Links to other papers and more general information are provided wherever possible.

Dr. David Colquhoun briefly but persuasively argues that clinical guidelines and scientific reviews routinely make recommendations based on inadequate evidence, substantially due to a common failure to appreciate the risk of false positives in positive studies of treatments with low prior plausibility: “every false positive not only harms patients (and budgets) but also provides ammunition for the antiscience brigade, who are now so evident.“

~ Paul Ingraham

original abstract Abstracts here may not perfectly match originals, for a variety of technical and practical reasons. Some abstacts are truncated for my purposes here, if they are particularly long-winded and unhelpful. I occasionally add clarifying notes. And I make some minor corrections.

Heneghan and colleagues discuss the need for better clinical guidelines. One problem with guidelines is that in very many cases no good treatment exists, yet recommendations are still made.

A good example is non-specific low back pain—the 2009 guidelines from the National Institute for Health and Care Excellence recommended acupuncture despite little evidence that it works to any useful extent. The guidelines were revised in 2016 to say “do not offer acupuncture.” This left few other recommended treatments, but the new guidelines fail to say explicitly that it’s an unsolved problem.

Even Cochrane reviews don’t seem to appreciate the distinction between P values and the risk of false positives. If you observe a P value close to 0.05, then to achieve a false positive rate of 5% you must assume that you are 87% certain of a real effect before the trial is done. That’s clearly preposterous. Even if you observe P=0.001, you would still have a false positive rate of 8% if the hypothesis was implausible (prior probability 0.1). And these numbers apply to perfect unbiased experiments, before you add all the other problems of P hacking, multiple comparisons, and so on.

These misunderstandings must be responsible for many false positive results. And every false positive not only harms patients (and budgets) but also provides ammunition for the antiscience brigade, who are now so evident.

related content

This page is part of the PainScience BIBLIOGRAPHY, which contains plain language summaries of thousands of scientific papers & others sources. It’s like a highly specialized blog. A few highlights: