Agitation is the part of Alzheimer's that wrecks caregivers: the pacing, the shouting, the swing at the aide who came to help with a shirt. It is also where cannabis research in dementia has its only real results. In 2019 a Toronto team gave nabilone, a prescription synthetic cannabinoid, to 39 people with moderate to severe Alzheimer's and watched agitation scores fall against placebo. In late 2025 a Johns Hopkins-led team reported that prescription dronabinol (synthetic THC) beat placebo on one of its two agitation scales over three weeks.
Here is where this page lands. If cannabis has a place in Alzheimer's agitation, it is a prescription nabilone or dronabinol question for the person's prescriber and care team, shaped like the trials, with a measured dose, a consent process and someone rating the behavior. A dispensary gummy stirred into the pudding cup is a different act entirely, with no trial behind it and a person on the receiving end who often cannot say how it feels. The broader picture of cannabis in dementia care (the appetite trials, the consent rules for nursing homes, the emergency-visit data) is on our appetite loss in older adults page, and none of it is repeated here.
What counts as agitation in Alzheimer's?
The International Psychogeriatric Association's 2015 consensus definition, which the Johns Hopkins dronabinol trial used to screen its participants, has four parts: a dementia or cognitive impairment, behavior consistent with emotional distress, excessive motor activity or verbal or physical aggression, and behavior that causes excess disability and is not solely attributable to another disorder.
That last clause does more work than it looks. A woman who started shouting on Tuesday and turns out to have a urinary tract infection on Wednesday does not meet the definition, because something else explains her behavior. The dronabinol trial went further and excluded anyone with delirium outright.
What should happen first when agitation starts or suddenly changes?
A clinician should look for a medical cause. The 2015 BMJ review by Kales, Gitlin and Lyketsos calls it crucial to investigate urinary tract and other infections, constipation, dehydration and pain, and to review the current medications, before treating the behavior itself. Non-drug approaches come first, it says, except in emergencies where someone is in danger.
The numbers behind that advice are not subtle. In one study the review cites, 36% of community-dwelling people with dementia had an undetected illness linked to their behavioral symptoms, agitation and crying out among them. Pain is associated with aggression in dementia, and treating the pain can reduce it.
So a sudden change in agitation goes to a clinician before anyone reaches for a drug. THC can itself cause confusion, and a dose given to someone quietly developing delirium is a good way to miss the delirium.
Does nabilone help agitation in Alzheimer's?
In one small crossover trial, yes. Herrmann's 2019 study gave 39 people with Alzheimer's six weeks of nabilone (a mean dose of 1.6 mg reached, against a target of 1 to 2 mg) and six weeks of placebo. Agitation on the Cohen-Mansfield inventory improved by 4.0 points more on nabilone, and caregiver distress fell too.
The participants were about as frail as trial populations get: mean age 87, mean standardized Mini-Mental score 6.5 out of 30, 77% men, recruited from a long-term care facility and geriatric psychiatry clinics. Sedation showed up in 45% of patients during the nabilone phase against 16% on placebo, though sedation bad enough to limit treatment was not significantly different. The cognition results pointed both ways: the brief Mini-Mental test favored nabilone, while a longer battery for severe impairment, completed by 25 patients, favored placebo. The clinicians' global impression of change improved in 47% on nabilone and 23% on placebo, a gap that missed significance.
A 2026 reanalysis of the same 39 patients found five traits that went with a bigger response, and the first on the list was pain (a score of 3 or more on a pain scale built for advanced dementia). Patients in the top third of predicted responders responded 82% of the time, against 40% in the bottom third. Pain at the top of that list fits the BMJ review's point above, and it raises an awkward question for the whole field: how much of the agitation that responded was pain that a painkiller would also have reached.
The trial everyone is waiting for is NAB-IT, the same Sunnybrook group's phase 3 study: 112 participants planned, eight weeks of nabilone or placebo, agitation on the same inventory. Its registry record says it is still recruiting, with completion estimated for April 2027.
What did the Johns Hopkins dronabinol trial find?
A partial win. Rosenberg's THC-AD trial randomized 75 people with Alzheimer's agitation at five Eastern US centers to three weeks of dronabinol or placebo, on top of their existing medications. Dronabinol beat placebo on the Pittsburgh Agitation Scale and missed on the second primary scale, and every secondary outcome, the Cohen-Mansfield inventory included, came out level.
Per the trial's registry entry, the capsules held 2.5 mg twice a day in week one, then 5 mg twice a day for weeks two and three. Most participants were already on antidepressants or antipsychotics. The published abstract names somnolence as the one adverse event tied to dronabinol, and the posted results add detail: two serious adverse events, both in the dronabinol arm (one delirium, one broken upper arm), and none on placebo, with falls at 3 of 37 on dronabinol and 5 of 38 on placebo. Three weeks is short, and a fracture in a 75-person trial is one event too many to wave away.
Why did the low-dose THC trial come up empty?
Probably the dose, though nobody can be sure. The Dutch trial led by van den Elsen gave 1.5 mg of THC three times a day, 4.5 mg in total, for three weeks to 50 people with dementia. Symptoms improved in both groups, and THC did no better than placebo on overall neuropsychiatric symptoms or on agitation.
A 2026 meta-analysis from the Toronto group behind the nabilone trial found the agitation benefit concentrated in studies giving more than 10 mg of THC-equivalent, and in severe dementia. That cuts uncomfortably for anyone hoping a microdose gummy is the gentle version of the trials. Each 1.5 mg dose in the Dutch trial sits in what our dosing guide calls the microdose tier (1 to 2.5 mg), and it did nothing measurable, while sedation was the price attached to the doses that did.
| Trial | Drug and dose | Who and how long | Agitation result | Side effects reported |
|---|---|---|---|---|
| van den Elsen 2015 | THC 1.5 mg three times daily | 50 people with dementia, 3 weeks | No better than placebo; both arms improved | Mild or moderate events similar (16 vs 14 patients) |
| Herrmann 2019 | Nabilone, mean 1.6 mg | 39 with Alzheimer's, 6-week crossover | 4.0 points better than placebo on the Cohen-Mansfield inventory | Sedation 45% vs 16% |
| Albertyn 2025 | Sativex spray (1:1 THC and CBD), titrated | 29 UK care home residents, 4 weeks | Feasibility study, not built to answer | No adverse reactions, no withdrawals |
| Rosenberg 2026 | Dronabinol up to 5 mg twice daily | 75 with Alzheimer's, 3 weeks | Better on one of two primary scales | Somnolence; 2 serious events, both on dronabinol |
The UK study set out to randomize 60 care home residents within a year, ran into the pandemic, and managed 29, which tells you the trial can be run and nothing yet about whether the drug works.
What do the reviews conclude?
Cautiously positive on agitation, with sedation as the cost. Two 2026 meta-analyses found lower agitation scores on cannabinoids than placebo, and both found sedation or somnolence roughly doubled. The 2021 Cochrane review, working from four trials and 126 people, could not say whether cannabinoids help or harm people with dementia.
The details are where the caution lives. The Brazilian-led review in the American Journal of Geriatric Psychiatry pooled six randomized trials and one open-label cohort, 221 participants in all, and put the somnolence risk ratio at 2.25, calling falls and fatigue imprecisely estimated. The Toronto review in CNS Drugs pooled ten studies and 328 participants, and its agitation result stopped being statistically significant once the studies at high risk of bias were removed. Its ten studies, combined, still enroll fewer people than the single brexpiprazole trial below.
How does brexpiprazole compare?
It has the approval and the bigger trial. The FDA approved brexpiprazole (Rexulti) on May 11, 2023 as the first treatment for agitation associated with dementia due to Alzheimer's. In the 345-patient confirmatory trial, agitation fell 5.32 points more than placebo over 12 weeks on the same Cohen-Mansfield inventory the nabilone trial used.
That trial also shows how strong placebo is in this field. Patients on placebo improved by 17.3 points over 12 weeks, against 22.6 on the drug, which is why an open-label report of a grandmother who calmed down on a tincture proves so little.
The label is blunt about the cost. Its boxed warning says elderly patients with dementia-related psychosis treated with antipsychotics are at increased risk of death; in 17 placebo-controlled trials across the class, drug-treated patients died at 1.6 to 1.7 times the placebo rate, about 4.5% against 2.6% over a typical 10-week trial, mostly from heart problems or infections such as pneumonia. It also warns of falls, and it says brexpiprazole is not for as-needed use. The label's agitation schedule starts at 0.5 mg once daily and steps up over two weeks to a 2 mg target.
None of the cannabinoid trials ran long enough or large enough to say whether nabilone or dronabinol carries a mortality risk of its own. The fair comparison is an approved drug with a known, labeled danger against two unapproved ones whose long-term risks in this population have not been measured.
Why is a dispensary edible the wrong tool for this?
Because it strips out everything that made the trials interpretable. The trials used capsules of known strength, given on a schedule, with clinicians rating behavior and recording side effects. A gummy carries a label claim, starts working 30 minutes to 2 hours after eating and lasts 4 to 8 hours, per our guide to how long edibles last.
Put that timing against a person whose agitation comes in episodes. A gummy given at the first sign of trouble lands after the episode is over, or in the middle of dinner, and the person cannot tell anyone whether they feel dizzy, frightened or simply stoned. A caregiver who sees no effect at the one-hour mark and gives a second piece has doubled the dose before the first has peaked. Sedation, the side effect both 2026 reviews agree on, is a fall risk in a person already prone to falls, and THC's effect on standing blood pressure is set out on our Parkinson's tremor page.
A drug given to quiet someone who cannot object works as a restraint unless it treats something the person is suffering from, which is why the work-up comes first. THC can also cause paranoia, a risk covered on our psychosis risk page, and a person with Alzheimer's has fewer ways to make sense of it.
Who decides when the person cannot?
The prescriber, the care team and the person's legal surrogate, through a documented consent process. The long-term care medicine society AMDA asks for informed consent from the surrogate, agent or guardian, a written facility policy and monitoring for adverse effects.
That process is set out in full on the appetite loss page, alongside AMDA's caution against widespread routine use of marijuana in long-term care, and our news piece on cannabis in dementia care covers how aging experts talk families through it.