- Design
- Network meta-analysis of randomised trials
- Population
- 24 trials, 2,082 people with mild cognitive impairment or dementia
- Primary outcome
- MMSE and MoCA scores
- Effect
- Cognitive training on MoCA SMD 2.96 (95% CI 1.57–4.35); exercise SMD 2.26 (0.95–3.57)
This network meta-analysis combined 24 randomised trials with 2,082 people with mild cognitive impairment or dementia, comparing multidomain interventions, music therapy, cognitive rehabilitation, physical exercise, cognitive training and health education against control.
On the MMSE, cognitive training and health education improved scores against control. On the MoCA, physical exercise, health education and cognitive training all improved scores, with cognitive training ranked highest. Music therapy, cognitive rehabilitation and multidomain interventions did not show significant benefit.
The standardised effect sizes (up to about 3) are far larger than usually seen in dementia trials, and the confidence intervals are very wide, which suggests small, heterogeneous or biased trials. The ranking of health education so high is surprising. The practical message that structured cognitive training and exercise may help cognition is consistent with other evidence, but the size of benefit should not be quoted to patients. It was published in September 2026.
- Encourage regular physical exercise for people with mild cognitive impairment.
- Structured cognitive training may help cognitive test scores.
- Do not quote these effect sizes to patients or families; they are likely overestimates.
- Set realistic expectations: benefits in trials are usually modest.
Why it matters
Families ask what they can do beyond medicines, and exercise and cognitive training are reasonable answers.
Don't overread it
Very large, imprecise effect sizes suggest small or biased trials, not dramatic real-world benefit.
The statistics, in plain English
A standardised mean difference of 1.43 to 2.96 would be enormous for any dementia intervention; most effective treatments achieve well under 0.5. Very wide confidence intervals, such as 0.05 to 4.00, show great uncertainty. Rankings from a network meta-analysis can look decisive when the underlying evidence is weak.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for geriatrics, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free