Creatine and the aging brain

Pooled memory effects land near a third of a standard deviation. The larger number comes from older adults alone, is highly heterogeneous, and a published letter disputes the statistics behind it.

mixedStudies disagree, or the mechanism is contested.Revised 03 Oct 2026

Plain white powder in a small glass dish on a pale wooden surface

Eight studies pooled. The result is a fraction of a standard deviation, and the largest number in the field is the least stable.

The claim

Creatine is a molecule the brain holds in small quantities and spends on buffering the energy currency of neurons. Supplementing it raises those stores, and brain creatine is measurable after a period of supplementation. The interesting question for an ageing brain is whether that changes anything you would notice.

The literature is small, mostly short, and split between people who think it works and people who have tried to measure it. Here is what the measurements say.

The pooled number

The most-cited synthesis is Prokopidis and colleagues, 2023. Ten randomized placebo-controlled trials met the criteria for the review and eight went into the meta-analysis.

Across the whole pool, memory performance improved against placebo by a standardized mean difference of 0.29 (95% CI 0.04 to 0.53; I² = 66%; P = 0.02). The subgroup split is where the paper becomes interesting for this audience, and it is also where the caveats stack up.

Older adults, 66 to 76 years: SMD 0.88 (95% CI 0.22 to 1.55; I² = 83%; P = 0.009).
Younger adults, 11 to 31 years: SMD 0.03 (95% CI -0.14 to 0.20; I² = 0%; P = 0.72).

A near-nil effect in the young and a large one in the old is the shape one would hope for, and it is the shape the abstract leads with. Three things deserve to sit next to it. I² of 83% means the studies in that subgroup disagreed with each other substantially. The confidence interval runs from 0.22 to 1.55, which is wide enough to contain both a modest and a very large effect. And the subgroup rests on far fewer trials and participants than the eight in the pool overall.

A published letter in the same journal later challenged the statistical handling of that meta-analysis specifically, arguing that double-counting and inadequate statistics led to false-positive findings. The authors replied. The dispute is unresolved in print, and it attaches to the headline result.

The broader 2024 pool

A separate 2024 meta-analysis widened the net: 16 randomized trials, 492 participants, ages 20.8 to 76.4, including both healthy people and patients with specific conditions.

Memory: SMD 0.31 (95% CI 0.18 to 0.44).
Attention on timed tests: SMD -0.31 (95% CI -0.58 to -0.03), faster.
Processing speed: SMD -0.51 (95% CI -1.01 to -0.01), faster.
Overall cognitive function: no significant improvement.
Executive function: no significant improvement.

The pattern is consistent with the 2023 result on the one domain both pools agree on, memory, at almost the same magnitude. Where they diverge is everything else. Timed attention and processing speed improved here and not there; that is what you expect from a small literature where different trials test different things with different tools.

The authors graded the evidence themselves, which is more useful than the headline. Memory is moderate certainty. Attention, processing speed, overall cognitive function and executive function are all low certainty. Subgroup analysis found the effects larger in people with disease, in the 18-to-60 band, and in women, none of which replicated cleanly in the 2023 pool.

One housekeeping note, because a reader deserves it: this review carries a published corrigendum from February 2025. I could not read what it changes — the notice is a citation stub with no text attached — so the figures quoted here are the ones printed in the original abstract, and anyone leaning on them should check the corrected version first.

The 2018 review, and who responds

An earlier systematic review from 2018 included six studies and 281 individuals. Its conclusion was narrower than the later ones: there was evidence that short-term memory and intelligence or reasoning might improve, while long-term memory, spatial memory, memory scanning, attention, executive function, response inhibition, word fluency, reaction time and mental fatigue were all conflicting. Performance in young individuals was unchanged.

That review also reported that vegetarians responded better than meat-eaters on memory tasks, which is the origin of the low-intake hypothesis. The reasoning is sound: dietary creatine comes almost entirely from red meat and fish, so people who eat neither start from a lower baseline and have more room to move.

The strongest direct test of that hypothesis is a preregistered crossover trial published in 2023, with 123 participants, roughly half of them vegetarians, on a six-week supplementation period. It is the largest cognitive trial of its kind and the results are worth sitting with.

Bayesian evidence supported a small beneficial effect. On backward digit span the effect bordered significance (P = 0.064, partial eta-squared 0.029). On Raven’s Advanced Progressive Matrices it did not (P = 0.327, partial eta-squared 0.008). The eight exploratory tests showed no indication of improvement. Vegetarians did not benefit more than omnivores. Side effects were reported significantly more often on creatine than on placebo (P = 0.002).

So the largest trial says: small effect, at best, and no special advantage for the group the mechanism would most favour.

The observational signal, and why it is not evidence

There is a small cross-sectional study of 42 adults over 60 that examined habitual dietary creatine intake against memory. Forward Corsi block correlation was r = 0.703 (P < 0.001), reverse Corsi r = 0.715 (P < 0.001), and the memory component of the mini-mental r = 0.406 (P = 0.004). Median dietary intake across the group was just under four tenths of a gram a day, and people above the median scored better on both Corsi versions.

Those correlations are striking and they are the kind of number that gets quoted without its design attached. Forty-two people. No supplement, no randomization, no measurement of stores, no follow-up — and in a cross-sectional study like this one, every test in the battery is post hoc by construction. People who eat more red meat and fish differ from people who eat less in a great many ways, and this study cannot separate any of them. It is a hypothesis about where to look, not a finding.

Where that leaves the claim

Small, measurable, and thin. That is the honest summary, and it is a different sentence from both “creatine does nothing for the brain” and “creatine boosts cognition”.

The strongest case for taking it seriously is the convergence: two independent meta-analyses landing on nearly the same memory effect, from different literatures and different study sets, with a plausible mechanism and a measurable change in brain stores. The weakest case is everything around it. The subgroup effect in older adults is the number people quote and it is the least stable, with the highest heterogeneity in the analysis and a live methodological dispute attached.

One more thing this literature does not resolve. Almost every trial here studies people who are not deficient by any measure, and almost none stratifies by baseline brain creatine. The low-intake hypothesis has not actually been tested properly in the group that would benefit from it most. The 2023 trial is the closest thing, and its answer was no difference between vegetarians and omnivores.

What the data do not show

They do not show a large cognitive effect in older adults. The 0.88 figure comes from a small subgroup with I² of 83% and a confidence interval from 0.22 to 1.55.
They do not show the 2023 headline is safe from criticism. A letter in the same journal disputes its statistics; the authors replied; neither has been retracted.
They do not show that vegetarians or vegans benefit more. The largest trial tested it directly and found no difference.
They do not show an effect on overall cognitive function or executive function. Both 2024 pools came out null on those, and the graded certainty is low.
They do not show anything about dementia. No trial in these reviews enrolled people with cognitive impairment or a dementia diagnosis. The mechanism is a hypothesis about healthy ageing brains.
They do not show that the observational correlations are causal. Forty-two people, measured once.

Practical reading

If the question is whether creatine deserves a place on an ageing brain reading list, the answer is that it belongs there with a confidence label attached, and the label is mixed. If the question is what to do about it, that is a clinical decision rather than an editorial one, and it belongs with a clinician who knows the rest of the picture. For the wider context on how decline is measured and misreported, cognitive decline covers what the population numbers say. For the muscle side of the same molecule in the same age group, strength training and muscle after fifty is where the stronger evidence sits.