Creatine has an unusually good evidence base for a supplement, and an unusually clear pattern of what that evidence does and does not cover. Two independent meta-analyses of randomized trials, published one year apart, agree on memory and disagree about almost everything else. That disagreement is the interesting part.
The 2024 analysis: sixteen trials, 492 people
Frontiers in Nutrition, 2024, following PRISMA 2020, with the protocol registered on PROSPERO (CRD42024533557). Randomized controlled trials published between 1993 and 2024, retrieved from PubMed, Scopus and Web of Science. Sixteen RCTs involving 492 participants aged 20.8 to 76.4 years, healthy people and patients with specific diseases. Creatine monohydrate was the form used in all included studies.
What improved, with standardized mean differences:
- Memory: SMD 0.31 (95 % CI 0.18 to 0.44), Hedges’ g 0.3003 (0.1778 to 0.4228).
- Attention time: SMD −0.31 (−0.58 to −0.03).
- Processing speed time: SMD −0.51 (−1.01 to −0.01).
What did not: overall cognitive function and executive function. No significant improvement on either.
The subgroup analysis is where it gets interesting. More benefit in individuals with diseases, in those aged 18 to 60, and in females. And no significant difference between short interventions (under 4 weeks) and long-term ones (4 weeks or more) — which is itself a finding, because it means the trials that ran for a day and the trials that ran for months are being pooled.
On quality: low-to-moderate risk of bias, no significant publication bias, and a GRADE assessment that gives moderate certainty for memory and low certainty for processing speed, overall cognitive function, executive function and attention. The authors are explicit that larger and more robust trials are warranted.
The 2023 analysis: memory only, and a ten-fold age difference
Nutrition Reviews, 2023, searched PubMed, Web of Science, Cochrane Library and Scopus from inception to September 2021. Twenty-three eligible RCTs identified; ten compared creatine against placebo on memory measures in healthy individuals and went into the systematic review, eight into the meta-analysis.
Overall: memory improved against placebo, SMD 0.29 (95 % CI 0.04 to 0.53), I² 66 %, P 0.02. That is the same answer as the 2024 paper, from a partly different set of trials, one year earlier.
Then the subgroup analysis, which is the number worth remembering: in adults aged 66 to 76 years, SMD 0.88 (0.22 to 1.55), I² 83 %, P 0.009. In adults aged 11 to 31 years, SMD 0.03 (−0.14 to 0.20), I² 0 %, P 0.72. The confidence interval in the young group sits squarely on zero.
The authors also checked the obvious confounders and found they did not explain the result: amount (roughly 2.2 to 20 grams a day), intervention duration (5 days to 24 weeks), sex and geographical origin had no influence on the findings.
Reading the two together
The honest summary of the whole literature is: a small, reliable effect on one specific domain — memory — concentrated almost entirely in older adults.
Three things follow, and they matter more than the headline number.
First, the age gradient is the finding. If you are 25, the estimate for you is 0.03, with a confidence interval that includes no effect. The 0.88 belongs to somebody else.
Second, the two analyses disagree about executive function and overall cognition only in the sense that one of them pooled healthy people and patients with disease while the other restricted to healthy people. When a subgroup analysis in one paper finds the benefit concentrated in “individuals with diseases” and the other finds it concentrated in “adults aged 66 to 76”, the sensible reading is that the benefit tracks the people with the most to lose, not the youngest users.
Third, heterogeneity is high where the effect is largest: I² 83 % in the older adult subgroup, I² 66 % overall. These are not interchangeable numbers.
How to judge the next claim you read
The transferable lesson is not about creatine. It is about reading a supplement claim: find the subgroup that the headline hides, check the confidence interval for the age group and the condition you are actually in, and check whether the paper registered its protocol and rated its own certainty with GRADE. A supplement with a registered protocol and a published GRADE rating is telling you more about its own limits than most marketing copy ever will. That is the same habit that separates a report you can audit from one you can only take on trust, applied to a trial instead of a vial.
What this evidence does not show
It does not show that creatine improves overall cognitive performance or executive function: both were null in the analysis that specifically measured them. It does not show a dementia-prevention effect — no trial here studied progression of any disease, and the “individuals with diseases” subgroup is a catch-all, not a diagnosis. It says nothing about people under 18, where the estimate in the one age band measured is indistinguishable from no effect. The GRADE authors themselves rate four of the five domains as low certainty. And a meta-analysis of 492 people across 20 to 76 years is a small corpus: it is a well-run summary of a thin literature, not a wide one.
This article summarises published meta-analyses of randomized trials. It is not medical advice and not a recommendation for any individual.
