Aller au contenu
Santé mentale et cognitionMéta-analyse d’essais10 min read

Exercise eased depression in 218 trials, but confidence in the evidence is low

Collectif éditorial MoveVitals · Study published · Reading updated

Original paper:Effect of exercise for depression: systematic review and network meta-analysis of randomised controlled trials · BMJ · 2024 · Article original ↗

À retenir

In a network meta-analysis of 218 trials with 14,170 people who met criteria for major depression, walking or jogging, yoga, strength training, mixed aerobic exercise and tai chi or qigong each gave moderate reductions in depression compared with active controls (walking or jogging: Hedges' g −0.62, 95% credible interval −0.80 to −0.45). Confidence in the evidence was low for walking or jogging and very low for the other treatments, mainly because only one study met the low-risk-of-bias criteria.

Study design

Méta-analyse d’essais

Reading basis

Full text

Reading focus

Design & applicability

Cette explication éditoriale existe en anglais, chinois simplifié et chinois traditionnel. Cette interface affiche la lecture en anglais.

Question de recherche

Major depressive disorder is a leading cause of disability worldwide, many people respond poorly to drugs or psychotherapy, and access to treatment is limited: coverage is only 51% in high-income countries and 20% in low and lower-middle income countries. Guidelines in the US, UK and Australia already recommend physical activity as part of treatment, but they disagree about how much to do and which type to choose.

Earlier pairwise meta-analyses pooled very different exercises, or left out yoga, tai chi and qigong, and two earlier network meta-analyses (15 and 71 studies) may have been too small to examine dose and type. The authors set out to identify the best dose and type of exercise for people who meet clinical cut-offs for major depression, compared with psychotherapy, antidepressants and control conditions. They also tested whether sex, age, baseline severity, intensity, behaviour change techniques and autonomy support changed the effect.

Population et méthode

This is a systematic review and Bayesian network meta-analysis of randomized trials, published in the BMJ in 2024 and registered on PROSPERO. A network meta-analysis compares many treatments at once by combining head-to-head and indirect evidence. The authors searched the Cochrane Library, Medline, Embase, SPORTDiscus and PsycINFO with no filters, date limits or language restrictions (searches on 17 December 2018 and 7 August 2020, last updated 3 June 2023). From 18,658 records they included 218 unique studies described in 246 reports, with 495 arms and 14,170 participants. Trials had to include an exercise arm and participants who met criteria for major depressive disorder, either by clinician diagnosis or by scoring above an established threshold on a self-report scale (for example above 13 on the Beck Depression Inventory-II). People with physical conditions and postpartum depression were allowed. Exercise meant planned, structured, repetitive movement to improve or maintain fitness, which let yoga, tai chi and qigong in. Any comparator counted: SSRI antidepressants, cognitive behavioural therapy, active controls (usual care, placebo tablet, stretching, educational control, social support) or a waitlist.

The outcome was change in depressive symptoms from baseline on scales such as the Beck Depression Inventory and the Hamilton Depression Rating Scale, converted to one scale as Hedges' g, where larger negative numbers mean a bigger fall in symptoms. Because scores improve even in control groups, results are reported as improvement beyond active controls. Those controls were pooled because they gave similar results (g between −0.93 and −1.00, no significant differences), while waitlist was kept separate. The authors set the minimum clinically important difference at 0.5 standard deviations, which equals g −0.20 against active controls. Outcomes were measured directly after treatment or up to six months later. Models were Bayesian, arm-based and multilevel, run in R with preregistered informative priors, and results are given as 95% credible intervals. The authors tested moderators including sex, age, baseline severity, comorbidities, intensity, weekly dose, programme length, autonomy support, behaviour change techniques, group delivery and funding, ran sensitivity analyses for risk of bias and publication bias, rated confidence in each comparison with the CINeMA tool, and assessed acceptability through dropout rates.

Résultats

  1. 1
    Five types of exercise gave moderate reductions in depression compared with active controls, all larger than the minimum clinically important difference of g −0.20: walking or jogging −0.62 (95% credible interval −0.80 to −0.45; 1210 participants), yoga −0.55 (−0.73 to −0.36; 1047), strength training −0.49 (−0.69 to −0.29; 643), mixed aerobic exercise −0.43 (−0.61 to −0.24; 1286) and tai chi or qigong −0.42 (−0.65 to −0.21; 343). The intervals overlap, so the ranking among them is not firm.
  2. 2
    Dance had the largest effect (g −0.96, 95% credible interval −1.36 to −0.56) but rests on only 107 participants, mostly young women (88% female, mean age 31 years). Exercise added to SSRIs (−0.55, −0.86 to −0.23; 268 participants) and aerobic exercise added to psychotherapy (−0.54, −0.76 to −0.32; 404) also helped. For reference, cognitive behavioural therapy alone scored −0.55 (−0.75 to −0.37; 712) and SSRIs alone −0.26 (−0.50 to −0.01; 432), though the authors warn their search was not designed to find all trials of those treatments.
  3. 3
    Harder exercise worked better. Across types, light activity such as walking or hatha yoga gave g −0.58 (−0.82 to −0.33) and vigorous exercise such as running or interval training gave −0.74 (−1.10 to −0.38). Weekly energy expenditure did not explain this, and the dose–response curve for it was unclear. Weak evidence suggested shorter programmes did somewhat better (10 weeks −0.53, −0.71 to −0.35; 30 weeks −0.37, −0.79 to 0.03). Effects held up when measured right after treatment (−0.59, −0.80 to −0.39) or up to six months later (−0.63, −0.87 to −0.40).
  4. 4
    Strength training and yoga were the easiest to stick with. The odds of dropping out were lower than with active control for strength training (odds ratio 0.55, 95% credible interval 0.31 to 0.99; 247 participants) and yoga (0.57, 0.35 to 0.94; 264), and not significantly different for any other arm.
  5. 5
    Exercise appeared equally effective for people with and without other health conditions and across different baseline levels of depression. Sex and age shifted which type looked best: strength training seemed more effective for women and younger people, yoga and tai chi for men, and yoga more so for older people. The share of variance explained rose from 0.65 with exercise type alone to 0.71 with sex and 0.69 with age, but not with baseline depression (0.67) or comorbidities (0.66).
  6. 6
    Several expected effects did not appear, and some surprises did. No behaviour change technique significantly changed the results. Studies that gave participants more choice over frequency, intensity, type or time showed weaker effects (g −0.28, −0.78 to 0.23) than those that did not (−0.75, −1.17 to −0.33). Funded studies showed stronger effects (−1.01, −1.19 to −0.82) than unfunded ones (−0.77, −1.09 to −0.46). A test for small-study bias was significant (P<0.001), but publication bias alone could not shrink the effects to zero; it would take significant results being reported 58 times more often than non-significant ones to push them below the clinical threshold.

Limites et incertitude

  1. 1
    The authors note that only one study met the Cochrane criteria for low risk of bias and few blinded participants or staff, so the effects may include expectancy (placebo-type) effects. This is why confidence was rated low for walking or jogging and very low for every other treatment. They also note that the older risk-of-bias tool they used can be conservative for unblinded trials with self-reported outcomes.
  2. 2
    The authors note that the evidence for many types of exercise comes from small trials, and that most dance research is in young women (88% female, mean age 31 years), so they stop short of recommending dance more strongly. Many studies focused on one sex, older adults or people with other conditions, which raised concerns about how directly the results apply to a typical adult with major depression.
  3. 3
    The authors note that the age and sex patterns are between-study differences, not individual-level ones, and may be confounded because programmes were often tailored to a group; mean age was available for only 71% of studies. Our observation: the male and female and older and younger patterns should be treated as hypotheses, and the surprising autonomy result rests on the authors' own coding of trial methods, not on what participants felt.
  4. 4
    The authors note that the search was not built to cover unpublished studies or trial registries, and that the multilevel Egger's test pointed to some publication bias. They also changed the analysis plan after registration, replacing the GRADE approach with CINeMA and adding the dropout analysis at a reviewer's request.
  5. 5
    Our observation: the comparisons with cognitive behavioural therapy and SSRIs are indirect and rough, and participants were people meeting depression cut-offs, often by self-report scales, so the results do not speak to mild low mood. Small wording differences between the abstract and the results sections (for example walking or jogging g −0.62 versus −0.63) show the estimates are approximate.

Mise en perspective

If you are living with depression, this analysis supports raising exercise with your clinician as one serious option alongside psychotherapy or antidepressants. Walking or jogging, yoga, strength training, mixed aerobic exercise and tai chi or qigong all lowered symptoms by a moderate amount, none clearly beat the others, and the evidence leaned toward pushing the effort up when you are able. Because strength training and yoga had lower dropout than the controls, they may be easier to keep doing. A practical reading is to pick something you will actually repeat.

It does not show that exercise works as well as therapy or medication, because confidence in the evidence was low or very low. It does not tell you which type suits you, since the age and sex patterns are tentative, and the intensity advice comes from group averages and ignores individual health conditions. Decisions about starting, changing or stopping treatment belong with your clinician. The overall direction is consistent across many trials; the exact size of benefit is uncertain.

Vos données

Log each session with its type, minutes and effort using the same watch or app, ideally with heart-rate zones or a simple effort rating, since the trials separated light activity such as walking from vigorous exercise such as running. Wrist heart rate can under-read during strength work and yoga, so rely on effort ratings there. Track mood weekly with the same validated questionnaire your clinician uses, for example the Beck Depression Inventory, and judge change over weeks to months rather than days: the programmes in the trials were often around 10 weeks. Wearable HRV, sleep and readiness scores are not measures of depression, and illness or medication changes can shift them for other reasons.

Comprendre ce type de preuve

Pooling trials can improve precision, but compare populations, controls, bias and uncertainty. Network rankings include indirect evidence and do not identify a universal best treatment.

Sources et attribution

Effect of exercise for depression: systematic review and network meta-analysis of randomised controlled trials

BMJ · 2024-02-14

DOI: 10.1136/bmj-2023-075847 ↗PubMed · 38355154 ↗

Interprétation éditoriale MoveVitals · contenu éducatif · aucune évaluation clinique individuelle enregistrée

Sources vérifiées 2026-11-01

Guides pratiques

À lire ensemble

·Synthèse de revues

Exercise gave small-to-moderate cognitive gains across 133 reviews of trials

An umbrella review of 133 systematic reviews (2724 trials, 258,279 participants) found that exercise improved general cognition (SMD 0.42, 95% CI 0.37 to 0.47), memory (0.26) and executive function (0.24), with the largest memory gains in children and adolescents and the largest executive function gains in people with ADHD. Most of the reviews (94) were rated critically low in quality and the effects were smaller in the best-rated ones (0.23 for general cognition), and the outcomes were test scores, not dementia.

British Journal of Sports Medicine

Explications approfondiesLire l’étude →
·Méta-analyse observationnelle

More daily steps linked to fewer depressive symptoms across 33 studies of adults

Pooling 33 observational studies of 96,173 adults, higher daily step counts went with fewer depressive symptoms, and in the two prospective cohorts that could be pooled, 7000 or more steps a day went with a lower risk of depression than fewer (RR 0.69, 95% CI 0.62-0.77). Most studies were cross-sectional, so low mood may be reducing steps as much as steps affect mood, and the authors say the prediction intervals call for caution.

JAMA Network Open

Explications approfondiesLire l’étude →
·Synthèse de revues

Physical activity eased depression and anxiety across 97 reviews of trials

Across 97 systematic reviews that pooled 1039 randomized trials with 128,119 participants, physical activity had medium effects on depression (median effect size −0.43) and anxiety (−0.42) compared with usual care, and higher-intensity activity went with larger gains. Most of those reviews (77) were rated critically low in quality, and the evidence mainly concerns mild-to-moderate symptoms.

British Journal of Sports Medicine

Explications approfondiesLire l’étude →
·Méta-analyse d’essais

Strength training eased depressive symptoms in 33 trials, but blinded trials found less

Pooling 33 randomized trials with 1877 participants, resistance exercise training was followed by a moderate reduction in depressive symptoms compared with non-active control conditions (mean effect size 0.66, 95% CI 0.48 to 0.83; P < .001). Trials that blinded allocation and/or assessment found smaller reductions, and results differed a great deal between trials (I² 76.0%), so the true effect is likely smaller and less certain than the headline number.

JAMA Psychiatry

Explications approfondiesLire l’étude →
Santé mentale et cognition →