An immense new review of exercise for pain

And it sounds like good news, but I’m not so sure. The salamander is not easily impressed.

Singh et al. is a sweeping new super-review of many kinds of exercise for many kinds of pain: a mighty review of 157 other reviews, based on almost 2800 trials, and over 220,000 study subjects. So it’s a big boy. The authors conclude that exercise modestly reduces pain for most conditions — with the counterintuitive twist that gentler, shorter programs shone brightest in the data.

I think that conclusion is a bit misleading. 🤏🏻

The paper is certainly useful as a map of the literature, but the confident punchline reaches beyond what data from a lot of unblinded trials can support. It’s not quite “garbage” in-and-out this time, because some of the exercise-for-pain literature is decent, and there’s quite a lot of it. But even the good studies have predictable limitations.

The data come from mostly unblinded trials, so expectation and attention effects are baked right into their standardized mean difference (SMD) of −0.59 — it measures the size of the effect but says nothing about what caused it. It’s also not a number people can really relate to — statistical units (standard deviations), not points on a pain scale — so a respectable-looking −0.59 doesn’t actually tell you the effect clears the “clinically meaningful” bar. It might. But we can’t know. There are other kinds of numbers for that (minimal clinically important difference), but there’s nothing about MCID anywhere in this paper. So “clinically meaningful pain relief” is not a fair summary of “there’s a real, non-zero difference … but we can’t say it’s big enough for a patient to notice, and we can’t rule out that it’s an artifact of unblinding.”

After a cynical adjustment, there might not be much left of the result.

And another thing: that odd inverse dose-response … bigger effects from shorter, easier programs? Singh et al. explain it away (people not sticking with it after the trial ends), and even spin it into a case for low-dose programs (“more realistic”). Fine, maybe so. But they also don’t suggest the possibility that it’s an artifact of low-quality data that their own analysis flags elsewhere. Bit of an oopsy there, and another sign that they were spinning things in a wishful-thinking-about-exercise way.

So exercise might be somewhat effective as a relatively short-term treatment for many kinds of pain, but it remains surprisingly uncertain, and this paper doesn’t settle it. See also Citation check for “exercise is good for chronic pain” — also rather damning.

The value of a long-term investment in fitness is a separate question, more of an unknown — and that’s where a lot of my own hope lives.