551 back pain trials, 71,126 patients, and almost none of it is good evidence

The largest synthesis yet assembled of non-surgical treatments for chronic low back pain found that six of them beat doing nothing by a clinically meaningful margin in the short term, that almost none of that advantage survived to twelve months, and that 98.6% of the underlying evidence was rated very low certainty on the standard grading scale [s1]. Both halves of that sentence are the finding. The rankings are real; the confidence you can place in them is not.

The time-course network meta-analysis, published in BMJ Medicine, pulled together 581 reports of 551 studies covering 71,126 patients, drawn from six databases and 302 earlier systematic reviews [s1]. It compared fourteen treatment categories — acupuncture, education or advice, electrotherapy, exercise training, manual treatment or manipulation, massage, the McKenzie method, pharmacotherapy, psychological treatments, traction, physical therapy, multidisciplinary pain management, usual care and placebo — against no treatment at all, and it did so at four separate time points rather than pooling everything into a single average [s1]. Of the included trials, 510 enrolled people with non-specific chronic low back pain and 41 enrolled people with radicular pain [s1].

What cleared the bar, and for how long

The threshold the analysis used for a clinically important difference was half a standard deviation [s1]. On a 0–100 back pain scale, six treatments exceeded it in the short term, defined as one day to three months: massage (mean difference −25.61, 95% credible interval −30.42 to −10.91), acupuncture (−20.91, −24.00 to −11.95), manual treatment (−19.48, −22.17 to −11.74), electrotherapy (−18.98, −21.84 to −10.95), multidisciplinary pain management (−18.96, −22.26 to −9.58) and exercise (−15.59, −17.51 to −10.05) [s1].

For disability, a narrower set cleared the threshold in the short term: acupuncture (−10.52, −11.84 to −6.59), massage (−9.95, −11.45 to −5.50) and multidisciplinary pain management (−12.56, −13.91 to −8.55) [s1]. Immediately after treatment, only the McKenzie method exceeded it; at intermediate follow-up, only massage [s1].

Then the effects go. In the long term — twelve months or more — two of the fourteen treatments for back pain and nine of fourteen for disability were statistically significantly better than no treatment, but none of those differences reached clinical significance [s1]. A statistically detectable effect that is too small to matter to a patient is the standard result at one year.

The certainty grade is the story

Across interventions and time points, the GRADE assessment rated 1.4% of the evidence low certainty and 98.6% very low certainty [s1]. Nothing was rated moderate or high.

"Very low certainty" in the GRADE framework means the true effect is likely to be substantially different from the estimate. Applied to a table of six treatments and their credible intervals, it means the ordering — massage ahead of acupuncture ahead of manual therapy ahead of exercise — should not be read as a ranking anyone should act on. Two of the results, massage and the McKenzie method, were explicitly unstable in the analysis's own sensitivity checks [s1].

It also means the comparison being made matters enormously. These are effects against no treatment: a true control arm, not a sham or a credible placebo. Any treatment involving attention, touch, expectation and time will beat no treatment. Separating out how much of a 25-point massage effect is massage requires a sham comparison, and this analysis was not built to supply one.

Radicular pain has almost no evidence at all

Only 41 of 551 trials enrolled people with radicular chronic low back pain — pain radiating along a nerve root [s1]. The authors report that treatment effects for radicular pain did not appear to differ from those for non-specific pain, and are careful to add that the sensitivity analyses could not establish this positively, so an evidence gap remains for that subpopulation [s1]. For a presentation this common, that is a striking absence.

How this sits against guidance

The American College of Physicians guideline reaches recommendations that are compatible with the network analysis and arrive with the same soft grading underneath. For chronic low back pain it recommends starting with non-pharmacological treatment — exercise, multidisciplinary rehabilitation, acupuncture and mindfulness-based stress reduction at moderate-quality evidence, and tai chi, yoga, motor control exercise, progressive relaxation, biofeedback, low-level laser therapy, operant therapy, cognitive behavioural therapy or spinal manipulation at low-quality evidence [s2]. Drug treatment sits behind that, as a weak recommendation on moderate-quality evidence [s2].

For acute and subacute back pain the guideline's framing is more important than its list: most patients improve over time regardless of treatment [s2]. Much of what looks like a treatment effect in an uncontrolled setting is the natural course of the condition.

The broader clinical literature has moved in the same direction. A Lancet review of chronic pain describes clinical trials and guidelines as typically recommending a personalised, multimodal, interdisciplinary approach, which may combine pharmacotherapy, psychotherapy, integrative treatments and invasive procedures [s3]. That is a description of a field with many partially effective options and no dominant one — which is exactly what a network meta-analysis with 551 trials and no clear winner looks like from the inside.

What a reader can take from this

Three things are reasonably solid. Short-term relief from several hands-on and active treatments is real relative to doing nothing [s1]. Nothing tested here reliably changes where someone is at twelve months [s1]. And the confidence attached to every specific number in that table is very low [s1].

The unglamorous implication is that treatment choice for chronic back pain is not, on current evidence, a technical question with a right answer waiting to be looked up. What the data support is trying something active, at a time horizon of weeks to months, and not expecting the choice among the active options to be the decisive variable.

This article is informational and is not medical advice. Treatment decisions for back pain belong with a reader and their clinician.

Sources

• Conservative treatments for chronic non-specific low back pain: time course network meta-analysis — BMJ Medicine, 2026-06-30
• Noninvasive Treatments for Acute, Subacute, and Chronic Low Back Pain: A Clinical Practice Guideline From the American College of Physicians — Annals of Internal Medicine, 2017-02-13
• Chronic pain: an update on burden, best practices, and new advances — The Lancet, 2021-05-29