A Cannabis Trial in Severe Dementia Missed Its Main Target. Read Past the Headline.
THC/CBD oil didn't beat placebo on agitation. It did cut as-needed sedative use — a secondary result.
A team in Geneva ran a serious trial of cannabis oil in people with severe dementia, and published the results in Age and Ageing this month. The trial asked one main question: does THC/CBD oil reduce agitation compared with placebo? The answer was no. On a secondary measure, how often staff reached for as-needed sedatives and antipsychotics, the cannabis periods came out significantly lower. Trade headlines have led with the second finding. The practical upshot: this is a well-run negative trial with one interesting loose thread, and it should be read that way.
Before you send this to a family member, the four things that matter most:
- The main outcome was negative. No significant difference from placebo on agitation.
- The secondary outcome was positive. Fewer as-needed psychotropic doses during cannabis periods.
- It was small. 25 people randomised, 19 finished.
- It was safe. No serious adverse events attributed to the study drug.
This article is for education and is not medical advice. A qualified clinician can help you evaluate individual risks, medications, and treatment decisions. Never change a medication for yourself or someone in your care based on a news article.
What the trial actually did
The design is the strong part, so it's worth walking through.
It was a multicentre, randomised, double-blind, placebo-controlled crossover trial across five long-term care facilities in Geneva, Switzerland. Every one of those words is doing work:
- Randomised, who got what was decided by chance, not by a clinician's judgment.
- Double-blind, neither staff nor assessors knew which period was which.
- Placebo-controlled, the comparison was an inactive oil, not "no treatment."
- Crossover, each patient received both the drug and the placebo, in sequence. Each person is their own comparison.
Participants had severe dementia and received oral THC and CBD oil in a 1:2 ratio, up to 20 mg THC and 40 mg CBD daily. Active and placebo periods each ran eight weeks, separated by a one-week washout.
Twenty-five patients were randomised, with an average age of 83. Nineteen completed both periods.
Reporting on the trial puts the average achieved dose after titration at 17.85 mg THC and 35.7 mg CBD per day, close to the ceiling, which suggests the researchers pushed the dose rather than under-treating.
Primary outcome: agitation, measured with the Cohen-Mansfield Agitation Inventory, a standard rating scale for agitated behaviours in dementia.
Secondary outcomes: other behavioural scales, pain, as-needed psychotropic medication use, and safety.
What it found, and what it didn't
The trial's own results paragraph is unambiguous:
No significant difference was observed between active treatment and placebo for the CMAI and other main behavioural outcomes.
That's the headline finding. THC/CBD oil did not outperform placebo on agitation, and it did not outperform placebo on the other main behavioural measures either.
Then the loose thread. Use of pro re nata psychotropic medications, the "as-needed" doses of antipsychotics and sedatives that care staff administer when someone is distressed, was significantly lower during the active treatment period. Reported figures put it at 64 administrations during cannabis treatment versus 153 during placebo.
And the safety result: the treatment was well tolerated, with no serious adverse events related to the study drug.
So: negative on the thing it set out to measure, positive on a thing it also measured, and safe.
Why "primary endpoint" is the only word that matters here
This is the part worth understanding, because it applies to every trial you'll ever read about.
Before a trial starts, researchers declare one main outcome, the primary endpoint. That declaration is a commitment device. It stops a team from measuring twenty things, finding that two came out well by chance, and writing the paper about those two.
The reason it works is statistics. If you test enough outcomes, some will look significant purely by luck. Declaring the primary endpoint in advance means there's exactly one result the trial is designed to answer, powered to answer, and honest about.
Everything else is secondary, real data, worth reporting, worth following up. But secondary findings are generated in an environment where chance has more room to operate, and they are not what the trial was sized to detect.
That's the honest status of the as-needed medication finding. It's genuinely interesting. It is not established.
How the headline inverted the finding
Here's what actually happened in the coverage.
The trial's own conclusion says THC/CBD oil did not show superiority over placebo in reducing agitation, while noting it was safe and associated with a significant reduction in as-needed medication.
The trade headline reads: "Clinical Trial: THC/CBD Oil Significantly Reduced Use of Psychotropic Drugs in Patients With Severe Dementia."
Every word of that headline is factually accurate. It's still the wrong story. A reader who sees only the headline learns that a cannabis trial in dementia succeeded. The trial's authors reported that it didn't.
We're not accusing anyone of dishonesty. This is what happens when a negative result contains one positive number, the positive number is the interesting-looking sentence, and it travels. But the effect on a caregiver reading a feed is real, and it's the kind of thing that gets someone asking a nursing home to try cannabis oil on the strength of a study that specifically failed to show it helps agitation.
Grading this evidence honestly
We'd grade this as limited but well-conducted. Both halves of that matter.
What's strong: the design. Randomised, double-blind, placebo-controlled, crossover, multicentre, with a pre-declared primary endpoint and a published negative result. Publishing a negative trial is a genuine contribution, the literature is distorted by trials that quietly disappear when they don't work.
What's limiting:
- Size. 25 randomised, 19 completers. Small trials can miss real effects, and they can also produce spurious ones.
- Population. Severe dementia in long-term care in one city. This tells you little about earlier-stage dementia or home care.
- The secondary finding's status. The abstract doesn't describe correction for multiple comparisons on the PRN outcome. Without that, its statistical significance should be read cautiously.
- Abstract, not full text. We read the published abstract. The full paper will contain detail on effect sizes, dropouts, and analysis choices that would sharpen all of this.
What's unknown: mechanism. If cannabis oil didn't move measured agitation but staff reached for sedatives less often, something is going on, perhaps in sleep, pain, or how distress presents rather than how much of it there is. The trial doesn't tell us, and speculation isn't evidence.
What a family should take from this
If you are caring for someone with severe dementia, here's the fair reading.
Don't take this as permission. The trial did not show cannabis reduces agitation. That was the question, and the answer was no.
Don't take it as a closed door either. It was safe at meaningful doses in a frail, elderly population, 83 years old on average, with severe dementia. That's not nothing. Safety data in this population is scarce and hard to collect.
Do take the as-needed medication finding to a clinician, if the person you care for is receiving frequent PRN antipsychotics. Not as a request for cannabis, but as a prompt for a conversation about PRN use generally, which is a live issue in dementia care independent of anything cannabis-related.
Do ask what the goal is. "Less agitation" and "fewer sedative doses" are different targets. This trial suggests they can move independently.
Did the trial show cannabis helps agitation in dementia?
No. The abstract states no significant difference was observed between active treatment and placebo on the Cohen-Mansfield Agitation Inventory or other main behavioural outcomes.
What did the trial find?
Use of as-needed (pro re nata) psychotropic medication was significantly lower during cannabis treatment than placebo, a secondary outcome. Reported figures are 64 administrations versus 153.
How large was the study?
Twenty-five patients were randomised, average age 83. Nineteen completed both treatment periods.
What dose was used?
Oral THC and CBD oil in a 1:2 ratio, up to 20 mg THC and 40 mg CBD daily. Reporting puts the average achieved dose at 17.85 mg THC and 35.7 mg CBD.
Was it safe?
The treatment was well tolerated and no serious adverse events related to the study drug were observed. That's a meaningful result in a frail elderly population, though the sample was small.
Why does the primary endpoint matter so much?
It's declared before the trial begins and the study is sized to answer it. Secondary outcomes are more vulnerable to chance findings, so a positive secondary alongside a negative primary points toward the next study rather than a change in care.
Should I ask about cannabis for a relative with dementia?
That's a conversation for their treating clinician, who knows their medications and risks. This trial does not support cannabis as a treatment for agitation.
Key takeaways
- Primary outcome: negative. No significant difference from placebo on agitation.
- Secondary outcome: as-needed psychotropic use significantly lower during cannabis periods (64 vs 153 administrations).
- Design: randomised, double-blind, placebo-controlled crossover; five Geneva care facilities; two 8-week periods with a 1-week washout.
- Size: 25 randomised, 19 completers, average age 83.
- Safety: well tolerated; no serious adverse events attributed to the study drug.
- Status of the evidence: limited but well-conducted; hypothesis-generating, not practice-changing.
The Cannible Newsroom's take
What we'd tell a friend: this is a good study that didn't work, and that's more useful than it sounds.
We want to be careful here, because it would be easy to write this piece as "media gets cannabis story wrong" and feel clever. That's not quite it. The as-needed medication result is genuinely worth someone's attention. If a safe intervention reduces how often frail, distressed elderly people get dosed with antipsychotics, that is a real thing to chase, antipsychotic use in dementia care carries known risks and has been a target for reduction for years.
But it has to be chased properly, in a trial designed to answer that question, with enough people to answer it. That's what a hypothesis-generating finding earns you: a next study, not a conclusion.
What concerns us is the reading environment. Caregivers of people with dementia are often exhausted, under-supported, and looking hard for something that helps. That's exactly the audience least served by a headline that reports a trial's most encouraging number without its main one. We'd rather hand that reader the disappointing truth and the honest loose thread than the flattering summary.
What we'd flag for the field: publish the negatives, loudly. This team ran a rigorous trial, got an unwelcome answer, and published it. That's how the evidence base gets trustworthy.
This article will be updated as the full paper is examined and as follow-up research appears.
Authoritative sources for further reading
- A randomised, double-blind, placebo-controlled crossover trial on cannabinoid-based medication for behavioural disorders in old patients with severe dementia: the study results: Bianchi F, Broers B, Desmeules JA, Langlois A, Wampfler J, Curtin F, Pautex S. Age and Ageing 55(8), Oxford University Press on behalf of the British Geriatrics Society, August 2026. PMID 42613673.
- Clinical Trial: THC/CBD Oil Significantly Reduced Use of Psychotropic Drugs in Patients With Severe Dementia: The Marijuana Herald, August 19, 2026. Included as the coverage discussed above.
Talk to a qualified clinician about any treatment decision for a person in your care. This article is education, not advice.
Cannabinoid-based medication for behavioural disorders in severe dementia, trial results — Age and Ageing, August 2026, the primary source for every figure in this article