Cannabinoids for agitation in dementia: no benefit, twice the somnolence

Design

Systematic review and meta-analysis of randomised trials, random-effects and Bayesian synthesis, GRADE-rated

Population

334 participants with Alzheimer's disease or other dementia across nine randomised trials

Primary outcome

Agitation on the Cohen-Mansfield Agitation Inventory, and neuropsychiatric symptoms on the NPI nursing home version

Effect

CMAI SMD −0.58 (95% CI −1.71 to 0.55, I² 84%); NPI-NH total SMD −0.02 (−1.00 to 0.96); somnolence RR 2.03 (1.29 to 3.20)

Nine randomised trials with 334 participants in total were pooled, covering cannabinoid-based treatments in Alzheimer's disease and other dementias, with GRADE certainty assessment and both frequentist and Bayesian synthesis. Agitation on the Cohen-Mansfield Agitation Inventory did not improve, standardised mean difference −0.58 (95% CI −1.71 to 0.55) with heterogeneity of 84%. Neuropsychiatric Inventory total was flat, SMD −0.02 (−1.00 to 0.96), and its agitation subscale −0.44 (−1.45 to 0.57). Mini-Mental State Examination scores gave SMD 0.86 with an interval from −16.33 to 18.06 and heterogeneity of 96% — a number that carries no information at all. Bayesian posteriors sat close to zero. Somnolence was twice as common as with placebo, RR 2.03 (1.29 to 3.20).

Certainty was rated moderate for the behavioural outcomes and low for cognition, so the behavioural conclusion is reasonably secure: on the current evidence these drugs do not treat agitation in dementia. The one consistent effect is sedation, and sedation in a patient with dementia is not a therapeutic response — it raises falls, aspiration and confusion, and it makes the behaviour harder to reassess.

This matters because families ask. Cannabis-derived products are marketed directly to carers as a gentler alternative to antipsychotics, and the honest answer is that the evidence does not support them and the sedation is real. That does not make antipsychotics the fallback either. The first move in agitation remains looking for the cause — pain, constipation, urinary retention, infection, a change of surroundings — before any drug is considered.

• Do not offer cannabinoid products for agitation in dementia; the behavioural evidence is moderate-certainty negative.
• Warn about somnolence specifically if a family has already started one — it is twice as common as with placebo.
• Sedation is not improvement; record what the behaviour was, not just whether the patient settled.
• Screen for pain, constipation, retention and infection before any pharmacological step.
• The trials are small and mostly in nursing home populations — the finding does not stretch to other indications for cannabinoids.

The statistics, in plain English

Heterogeneity of 84% on the primary outcome means the individual trials disagree far more than chance explains, so the pooled figure is an average over studies that were measuring different things in different populations. That is why the interval, −1.71 to 0.55, is wide enough to include a large benefit — yet the Bayesian analysis, which handles small studies more honestly, put the posterior near zero. The MMSE result with 96% heterogeneity and an interval from −16 to +18 should simply be disregarded. Note that a wide interval usually means 'we do not know'; here the convergence of the frequentist point estimates near null, the Bayesian posteriors and the GRADE rating of moderate is what turns it into a usable negative.

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.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.