Design: systematic review and random-effects meta-analysis of sham-controlled randomised trials, GRADE-rated
Population: 357 patients with dementia across nine randomised trials
Primary outcome: global cognition immediately after stimulation and at follow-up
Effect: immediate global cognition SMD 0.28 (95% CI 0.01-0.54, I2 = 39%); not significant at follow-up

Nine randomised trials of transcranial alternating current stimulation against sham, in 357 patients with dementia, were pooled across seven databases with GRADE rating.

Immediate global cognition improved by a standardised mean difference of 0.28 (95% CI 0.01 to 0.54, I2 = 39%) - a small effect whose confidence interval nearly touches zero. Immediate recall, delayed recall and associative memory also improved immediately. At follow-up, the global cognition benefit was no longer significant. Neuropsychiatric symptoms did not improve, executive function evidence was inconclusive, and no serious adverse events occurred. A physiological marker moved in parallel: short-latency afferent inhibition, an index of cholinergic transmission, improved by 0.34 with no heterogeneity.

The cholinergic signal is the interesting part and also the easiest to over-read. A measurable change in a neurophysiological index is not a clinical benefit, and a cognitive gain that has gone by follow-up is not disease modification. Nine small trials with heterogeneous stimulation parameters cannot tell you which protocol, for how long, or in whom. This is a research-stage intervention, and nothing here supports offering it to a family asking what else can be done.

Effect on global cognition was small and did not persist to follow-up.
No benefit for neuropsychiatric symptoms, which is what most families ask about.
No serious adverse events across nine trials, so the safety question is not the limiting one.
Stimulation parameters varied between trials; no protocol can be recommended from this.
Do not offer this outside a trial - tell families it is research-stage.

Why it matters

Families increasingly ask about brain stimulation by name, and this puts a number on what it currently offers.

The statistics, in plain English

A standardised mean difference of 0.28 is conventionally small, and a confidence interval running from 0.01 to 0.54 means the result only just clears significance - a slightly different set of nine trials could easily have produced nothing. The loss of effect at follow-up is the more important finding, because a cognitive intervention that does not hold is measuring the testing session rather than the disease.