REF-1580 · Ciria et al., 2023 · Nature Human Behaviour · Paper · PubMed

Does regular exercise sharpen my mind?

In healthy people, trials give d = 0.22, 0.13 after moderators and 0.05 (-0.09 to 0.14) after bias correction: benefit not shown, not disproved, and experts dispute it.

Main caveat: Abstract only for the review; the author correction and the previews of the Matters Arising and the authors' reply were also read. No full text is held.
fairly strong evidenceMeasured on a mixed group

Most of the 24 meta-analyses reported a positive overall effect. The authors' audit found low statistical power in primary trials, selective inclusion of studies, publication bias and much variation from analytic and pre-processing choices.

They advise caution in claims and recommendations linking regular exercise to cognitive benefit in healthy people until more reliable causal evidence accumulates. Note what they are not saying: their conclusion is about the weakness of evidence, and small benefits survive the first adjustment.

Protocol

Included: meta-analyses limited to randomised trials in the healthy population.
New analysis: all primary trials in those reviews, with moderators (active control, baseline differences) and publication-bias correction.
Correction of 2 September 2026 (10.1038/s41562-026-02532-2): the article said 109 primary studies met inclusion. Seven gave too little information for an effect size and two pairs reported the same samples, so analyses rest on 100 non-redundant studies. The correction says these clarifications leave estimates and conclusions unchanged, that one wrong citation moved results only beyond the reported decimals, and that a report of preregistration deviations was added.
Not read: ages, doses, durations, cognitive domains, the bias-correction method.
Funding and interests: Comunidad de Madrid and the Spanish science ministry per the Europe PMC record; the publisher's page declares no competing interests.

Key Results

Estimate of exercise vs control on cognitiond (95% CI)
All primary trials0.22 (0.16 to 0.28)
After key moderators (active control, baseline differences)0.13 (0.07 to 0.20)
After publication-bias correction0.05 (-0.09 to 0.14)
Read the last row with care. An interval around zero does not establish zero, and no equivalence test is in the abstract.
Dispute. Dupuy and 20 co-authors (Matters Arising, 2024, 10.1038/s41562-024-01949-x) reject the reading on methodological and theoretical grounds and call the authors' caution about WHO recommendations unwarranted.The authors replied in the same issue (8:1464-1466, 10.1038/s41562-024-01950-4), holding that underpowered designs, selective reporting and publication bias preclude a conclusive benefit. Only the opening paragraph of each was read, so the arguments cannot be weighed here.

Mechanisms

Measured: cognitive outcomes as pooled; no mechanism in what was read.
Authors' interpretation: low power, selective inclusion, publication bias and varied analytic choices undermine the evidence, and the estimate becomes negligible after bias correction.
Inference: short trials in healthy people may be poor tests of a slow, long-term effect, which would limit both this review's null-leaning end and the positive reviews before it.

Practical Application

Does regular exercise sharpen my mind? In healthy people, trials give d = 0.22, 0.13 after moderators and 0.05 (-0.09 to 0.14) after bias correction: benefit not shown, not disproved, and experts dispute it.
Good for: keeping brain claims modest for someone training for health. Train for outcomes with better evidence and do not promise a sharper mind (Inference).
Not good for: saying that exercise does nothing for cognition. The first two estimates are positive, and the dispute is open.
Not good for: dementia or impairment, which this review of healthy people does not address.
Not good for: a verdict on any one domain, age or exercise type, since none is in the abstract.

Limitations

Abstract only for the review; the author correction and the previews of the Matters Arising and the authors' reply were also read. No full text is held.
Bias correction is model-dependent; the method and sensitivity to it were not read.
Primary trials underpowered by the authors' own finding, so a true small effect may be hidden or a false one inflated.
Healthy samples, varied tests. No athletes; no clinical outcomes.
Contested, with only the opening of the critique and of the authors' reply read.
How this entry was checked and changed
Why this level

(from the abstract) — an umbrella review earns A+ by design, but the rubric's A- row fits the evidence it carries: the authors themselves report low power in the primary trials, selective inclusion and publication bias, its bottom-line number depends on a bias-correction model, the reading is formally contested, and only the abstract was read