Attempting to separate placebo effects from exercise in chronic pain: a systematic review and meta-analysis

Key points

1. 01Placebo and non-specific effects (natural history, regression to the mean, the Hawthorne effect) can account for a large share of pain improvement, so isolating exercise's own effect matters.
2. 02Only 4 of 79 trials compared exercise to a placebo, and none used a true exercise-based placebo; their placebos were sham electrotherapy or oral supplements.
3. 03Against placebo, exercise was not statistically more effective for pain (Hedges g 0.94, but the confidence interval crossed zero).
4. 04Against no-treatment and usual-care controls, exercise was clearly more effective, reflecting the combined value of training plus the clinical encounter.
5. 05Every comparison was graded very low certainty, and all studies were at high risk of bias, largely from unblinded, self-reported pain.

How it was conducted

Design

Systematic review with pairwise random-effects meta-analysis, GRADE certainty, PROSPERO registered

Search

MEDLINE, CINAHL, SPORTDiscus, EMBASE, CENTRAL from inception to February 2021; 79 RCTs included

Participants

4843 adults with chronic primary musculoskeletal pain (over 3 months), including fibromyalgia, knee/hip osteoarthritis, and chronic low back pain

Intervention

Exercise training alone versus placebo, true (no-treatment or waitlist) control, or usual care

Outcomes

Pain intensity (VAS, numeric scales, or disease-specific pain subscales), pooled as Hedges g

Analysis

Pairwise random-effects meta-analysis with subgroup analyses by comparator type and funnel-plot and Egger's tests for publication bias

What they found

• Exercise vs placebo: g 0.94 (95% CI -0.17 to 2.06, P=0.098, I2=92.46%, 4 studies, n=253), not statistically significant.
• Exercise vs all controls combined: g 0.84 (95% CI 0.64 to 1.04, P<0.001, I2=90.02%, 79 studies, n=4843), with strong publication bias (Egger P<0.001).
• Exercise vs true (no-treatment) control: g 0.99 (95% CI 0.66 to 1.32, P<0.001, 42 studies, n=2361).
• Exercise vs usual care: g 0.64 (95% CI 0.44 to 0.83, P<0.001, 33 studies, n=2229).
• Every comparison was graded very low certainty of evidence under GRADE; all studies were at high risk of bias.

Limitations

• Only 4 trials had a placebo comparator and none used a true exercise placebo, so the specific effect of exercise on pain could not be isolated.
• All 79 studies were at high risk of bias, mainly from unblinded participants and self-reported pain (100% high detection-bias risk).
• Strong evidence of publication bias was found in the control comparisons (Egger P<0.001).
• Usual care was poorly defined and never standardized across trials, confounding the exercise-versus-usual-care comparison.

Why it matters

For patients

Exercise reliably reduces chronic pain compared with doing nothing, but part of that benefit may come from the care and attention around it rather than the exercise alone.

For clinicians

Keep prescribing exercise, since it beats no care and usual care, but recognize the specific drug-like effect of exercise on pain is not yet proven and contextual factors matter.

For readers

Exercise clearly outperforms no treatment for chronic pain, yet whether it beats a true placebo remains unanswered.