Ask almost anyone whether exercise is good for the brain and you will get a confident yes. Ask what kind of evidence sits behind that yes, and the answer gets more interesting. Most of it comes from observational cohorts, where people who move more also tend to be people with better mental health, more income, less pain and more years of schooling — and untangling those threads is the hard part.
A systematic review published on 10 November in npj Exercise Medicine and Health takes a different route. It collects every study that has used Mendelian randomisation to ask whether physical activity, sedentary behaviour or physical fitness causes changes in brain health, and reads the results together. The headline is not the one the fitness industry would choose: across 61 studies, the authors found the evidence “remains limited and mixed”, with 30 of the included studies reporting only null or directionally opposite effects.
What Mendelian randomisation is actually doing
The method borrows the logic of a randomised trial from genetics. Because gene variants are dealt out at conception more or less at random, people who carry variants associated with being more active can be compared with people who do not, without the usual puzzle of whether they chose to be that way.
That is the theory. In practice the method leans on three assumptions — that the variants really do shift the behaviour, that they are not tangled up with anything else that affects health, and that they reach the outcome only through the behaviour in question. For a trait like alcohol use, where a gene variant sits in a metabolising enzyme, the biology is fairly clear. For something as diffuse as “time spent watching television”, it is not, and the paper is unusually candid about that.
The review screened 1,114 unique records, searched PubMed and medRxiv with no date restriction, and registered its protocol in advance. Of the 61 studies that met the criteria, 27 looked at cognitive outcomes, 30 at psychiatric ones and four at both.
Being active: mostly null, one consistent exception
Self-reported physical activity showed no consistent protective effect on depression, schizophrenia, anxiety or bipolar disorder. Some studies found a protective signal for depression; most did not. Accelerometer-measured activity told a slightly different story — several studies reported weak evidence that it lowers depression risk — but those studies scored lower on quality than the ones that found nothing, and their results were less stable across sensitivity analyses.
Cognitive outcomes were more consistent, and still mixed. Multiple studies found no causal effect of measured activity on Alzheimer’s disease; one that did found the result vanished in its validation sample, and another got it only with a lenient threshold for selecting genetic instruments. One study reported that moderate-to-vigorous activity could lower cognitive performance scores.
The clearest exception was aerobic exercise, which came out protective for depression risk in all three studies that examined it specifically. The authors note that other intensities — strenuous sport, heavy do-it-yourself work, light activity — did not show the same pattern, and they are careful not to over-read three studies.
Fitness held up better than activity
Where the results did line up, they lined up around fitness rather than movement. Handgrip strength, appendicular lean mass and walking pace showed the most consistent protective associations, both for mental health and for cognitive measures — with the strongest support on the cognitive side, where nine of twelve studies pointed the same way.
That shift in what is being measured matters. Grip strength and lean mass are objective, cheap to measure, and pick up the fitness of people who cannot or do not run or swim. They also partly capture frailty and ageing, and the authors say so directly: lower muscle mass and slower walking are entangled with the conditions that also affect thinking, so the arrow may not run only one way. At least one analysis found that the walking-pace and grip-strength effects on depression did not survive adjustment for body mass index, while the lean-mass effect did.
Television, sitting, and a body-mass asterisk
Sedentary behaviour was equally uneven. Time spent watching television was the most consistent sedentary exposure linked to higher depression risk, and was also reported as detrimental to cognitive performance and dementia risk. Computer use either had no effect or, in one study, appeared protective. In several analyses the sedentary effects were mediated or explained away by body mass index, and the authors point out that sedentary behaviour is genetically entangled with socioeconomic status, physical illness and family environment — precisely the kind of confounding that weakens a genetic instrument.
The quality problem inside the pile
Perhaps the most useful part of the review is its assessment of the studies themselves. Thirty-two of the 61 scored as insufficient on the authors’ risk-of-bias tool, 28 as sufficient, and exactly one as good. Only 11% had demonstrably strong genetic instruments. Just 26 of the 61 — 43% — used independent summary statistics for a given exposure and outcome pairing; the rest re-used the same underlying datasets, so apparent agreement between papers is often the same data analysed twice rather than replication.
The authors put it plainly: this redundancy “risks overstating the consistency of the evidence”. They also note that every included study drew on UK Biobank for at least one trait, a cohort that skews older, wealthier, healthier and more educated than the population it stands for.
Why trials and genes disagree
None of this erases the trial literature, and the review does not claim it does. Randomised exercise studies generally find more consistent benefits, and the paper offers two explanations for the gap. Trials measure a defined, short-term intervention in a specific group; Mendelian randomisation estimates the effect of lifelong genetic liability to move more. And trials control confounding by design, whereas the genetic approach assumes its instruments are clean — a strong assumption for behaviour as complicated as exercise.
A separate meta-analysis published on 27 October in Frontiers in Psychiatry shows what the trial side looks like at its best. Pooling 22 randomised trials covering 1,599 adolescents, it found exercise improved executive function with a pooled effect of 0.55 standard deviations — a moderate number — with the lower bound of the confidence interval still comfortably above zero. But heterogeneity was severe, and the authors graded certainty as low for the overall result: the trials varied widely in activity type, duration and control condition.
Read together, the two papers are less a contradiction than a lesson in what each design can see. The genetic method is good at stripping out confounding and bad at capturing the thing a person actually does on a Tuesday evening.
What public guidance still rests on
Health agencies have not built their recommendations on Mendelian randomisation, and nothing here changes the arithmetic they do cite. The World Health Organization’s fact sheet notes that nearly a third of adults — 1.8 billion people — fall short of 150 minutes of moderate-intensity activity a week, up five percentage points between 2010 and 2022, and that insufficiently active people carry a 20–30% higher risk of death. Its benefits list includes reduced symptoms of depression and anxiety and enhanced brain health. The CDC’s adult guidance is 150 minutes of moderate activity weekly plus two days of muscle-strengthening work.
MedlinePlus frames the cognitive claim in the register the evidence supports: exercise helps “keep your thinking, learning, and judgment skills sharp as you age”, without promising a specific amount of protection.
What we did not find
The review found no genetic study able to test dose, frequency or timing. The underlying datasets measure activity in coarse blocks — active versus inactive, moderate-to-vigorous minutes, hours of television — so questions like how much, how often or whether short bouts count are outside what this design can answer. Nor does any of it test exercise as a treatment for people already diagnosed with a cognitive disorder: these studies estimate lifelong genetic liability across a population, not the effect of starting to move after a diagnosis, and the instruments rarely separate activity types beyond broad categories.
The authors close with the sentence that keeps the whole exercise honest: the absence of a genetic association “does not preclude a substantial role for environmental influences”. Genes proxy a tendency to be active; they do not proxy the park you walk in, the sleep you get afterwards, or the mood that made you go.
For anyone deciding how to spend half an hour, the practical picture has not moved much: the trials support moving, the genetic studies fail to confirm a lifelong causal chain, and the two are answering different questions. If you have a condition that affects how much you can exercise, or you are weighing exercise against a symptom that worries you, that is a conversation for a healthcare professional rather than a news article — our medical disclaimer sets out the limits of what this site can do.
For the adjacent research, we have written about what short movement breaks can and cannot do and about a walking-intensity trial in frail older adults.