Exercise as Medicine: What the Largest Meta-Analysis on Depression Actually Found

This is the largest synthesis of its kind — 41 randomised controlled trials, over 2,200 participants, all diagnosed with major depressive disorder or scoring above clinical thresholds on validated depression scales. Crucially, every trial compared exercise against a genuinely inactive control group (usual care, wait-list, or health education) — not against stretching, other therapies, or medication, which is where a lot of earlier reviews got muddled.

The overall effect was large: a standardised mean difference of −0.946. In plain terms, that means for every two people put through an exercise intervention, roughly one experiences a clinically meaningful drop in depressive symptoms compared to someone left untreated. That's a number needed to treat of 2 — a figure that holds up remarkably well against psychotherapy (NNT ≈ 2.5) and antidepressant medication (NNT ≈ 4.3) from comparable large-scale analyses.

Even when the analysis was restricted to only the highest-quality, lowest-bias trials, the effect remained moderate-to-large (SMD = −0.666), which is the kind of result that survives scrutiny rather than evaporating under it.

Not all exercise types performed equally: Aerobic and resistance training on their own produced large effects. Mixed aerobic + resistance training produced smaller effects — a genuinely counterintuitive finding. Supervised exercise clearly outperformed unsupervised exercise. Moderate and vigorous intensity both worked; the small pool of light-intensity trials didn't reach significance, likely from low study numbers rather than true ineffectiveness. Group and individual settings performed similarly.

Shorter trials showed larger effects than longer ones — the opposite of what's often assumed about "needing time" for exercise to work. The likely explanation isn't that exercise stops working over time, but that longer trials see more dropout, which dilutes the measured effect. It doesn't mean quick fixes; it means adherence over time is the real variable to manage.

For anyone building programmes around clients dealing with low mood, the takeaway isn't vague ("exercise helps stress") — it's specific: supervised, structured aerobic or resistance sessions at moderate-to-vigorous intensity, delivered consistently, carry an effect size comparable to first-line clinical treatments. That's a genuinely different conversation to have with a client than "movement is good for you."

This is not a replacement for psychotherapy or medication where those are needed — the authors are explicit that individual clinical judgement should guide treatment choice. But it does mean exercise deserves to be treated as a real, dosed intervention rather than a lifestyle suggestion tacked onto the end of a session.