In January 2020 the Parkinson’s Foundation sent an anonymous survey to 7,607 people with Parkinson’s disease who had previously engaged with the Foundation. It drew 1,339 responses, 1,064 of them complete. About a quarter of respondents, 24.5 percent (261 people), had used cannabis in the previous six months. That is the number that gets quoted.
The other 803 had not. Each could pick up to three reasons, and the one chosen most often was a lack of evidence that cannabis works: 481 people, which is 59.9 percent of non-users and 45 percent of all respondents. The question for the randomized trials is what they have shown, and whether it answers those patients.
What clinicians are being asked
The best picture of what Parkinson’s specialists face is a 73-item survey of neurologists at National Parkinson Foundation Centers of Excellence, published by Bega and colleagues in 2017. Fifty-six people responded. Of those, 80 percent had patients who used cannabis and 95 percent had been asked to prescribe it. Twenty-three percent reported any formal education on the subject. Asked how they approached patients, 52 percent took a neutral position, 39 percent encouraged use and 9 percent discouraged it. Fifty-six responders cannot measure how often the question arises, but they show that it arrives in clinic.
The randomized record, by compound
The trials do not test one intervention. Sorted by what patients were given:
Nabilone, a synthetic THC analogue, was tested in two randomized trials. Sieradzan and colleagues reported in 2001 that it reduced levodopa-induced dyskinesia in seven patients in a crossover design. Peball and colleagues, in 2020, reported improved non-motor symptoms, discussed in the next section.
A whole-plant cannabis extract was tested by Carroll and colleagues in 2004: 19 patients randomized in a crossover design, four weeks per phase, with no effect on levodopa-induced dyskinesia or parkinsonism.
Cannabidiol (CBD) alone has been tested most often. Chagas and colleagues, 2014: 21 patients in three groups of seven, given placebo, 75 mg or 300 mg a day, with no difference in UPDRS scores; the 300 mg group differed from placebo on the PDQ-39 quality-of-life total at p=0.05. A phase 2/3 trial of 75 to 300 mg for REM sleep behavior disorder in 33 patients, followed for 14 weeks, found no difference on its primary outcomes. A crossover trial of a single 300 mg dose in 24 patients found lower anxiety and smaller tremor amplitude during a simulated public-speaking test, a stressor created for the experiment. A 12-week trial of a CBD-dominant sublingual product in Thailand, which averaged 26 mg of CBD and 1.2 mg of THC a day, randomized 60 patients and analyzed 51; it found no difference in its primary measure, delayed recall, and a small gain in a naming score, with no serious side effects reported.
CBD combined with THC has produced the two largest trials. Liu and colleagues at the University of Colorado randomized 61 patients to an extract delivering about 192 mg of CBD and 6.4 mg of THC a day, for two weeks including dose escalation. The motor difference was −1.80 points (95% confidence interval −5.88 to 2.27); the authors reported that sleep, cognition and activities of daily living favored placebo. Kubota and colleagues randomized 101 patients with Parkinson’s pain to an oil delivering 43.88 mg of CBD and 0.96 mg of THC a day for nine weeks; 87 completed. The primary pain score did not differ (p=0.757), nor did the other non-motor scales. The trial stopped after a planned interim analysis fell below its futility threshold.
Why the nabilone result answers a different question
Peball’s trial was an enriched-enrollment randomized withdrawal study at the Medical University of Innsbruck. Forty-eight people were screened and 47 entered an open-label phase in which everyone received nabilone, titrated to a maximum of 2 mg a day. Nine did not continue: five were non-responders, defined as patients who did not rate their non-motor symptoms “much improved” or “very much improved,” and four stopped, one by dropping out and three because of adverse events. The other 38, 81 percent of those who started, were randomized: 19 stayed on their nabilone dose and 19 switched to matching placebo for four weeks. All 38 finished.
On the primary measure, the MDS-UPDRS Part I non-motor score, the placebo group worsened by 2.63 points and the nabilone group by 1.00. The difference was 1.63 points (95% confidence interval 0.09 to 3.18, p=0.030). A secondary non-motor scale, the NMSS, favored nabilone by 6.95 points, but its interval crossed zero (−0.66 to 14.55, p=0.147).
What that comparison tests is whether, among people who had already improved on nabilone in an uncontrolled phase, switching to placebo makes things worse over four weeks. It does not test whether starting nabilone helps a patient who has not been selected for response, and the open-label phase, where everyone was treated, had no control group. The authors acknowledge that a withdrawal design builds in negative expectations for those switched to placebo and that restricting to responders can overestimate an effect; they argue the second concern is small because most enrolled patients responded.
The trial was sponsored by the university. Nabilone’s manufacturer, AOP Orphan Pharmaceuticals, supplied the drug and placebo and paid for in-person study visits; one author is its employee and four others report travel grants or personal fees from it. Kubota’s oils were supplied by GreenCare Pharma, which the authors say had no role in the design, analysis or publication. Liu’s trial was funded by Colorado’s medical marijuana research grant program, with extract from the National Institute on Drug Abuse.
What the pooled analyses add
Messak and colleagues, in Acta Neurologica Belgica in May, identified 11 randomized trials and pooled six. Cannabinoids did not significantly improve disease severity, the motor examination, motor experiences of daily living or non-motor symptoms. Heterogeneity in the non-motor result was 64.3 percent and fell to zero in a sensitivity analysis that excluded Peball. The authors read that as a nabilone-specific signal for non-motor symptoms and call for dedicated trials. A sensitivity analysis shows that Peball differs from the other trials. The design is one possible reason and the compound is another, and the analysis cannot separate them.
Fernandes and colleagues, published online in Revue Neurologique on October 31, pooled seven randomized trials against placebo, with searches through December 2025, so Kubota’s August 2026 report is not included. They found no clear differences on efficacy outcomes. Their Bayesian analysis put the probability of any benefit on the UPDRS motor examination at 74.0 percent, and the probability that the true standardized effect exceeds their prespecified threshold of 0.20 at 42.8 percent. They concluded that the randomized evidence is insufficient to establish clinically meaningful benefit, limited by few trials and participants, short follow-up, differing interventions and imprecise estimates.
Safety, with denominators
Fernandes reported a pooled risk ratio of 4.44 for confusion (95% confidence interval 1.17 to 16.87) and 2.78 for dizziness (1.65 to 4.68), across products. The intervals are wide, and both exclude one.
The Colorado trial’s registry record gives counts. With 31 patients on the CBD and THC extract and 30 on placebo: dizziness in 18 against six, decreased concentration in 10 against four, feeling abnormal in eight against two, feeling drunk in seven against two, confusion in five against one. The paper reports that 13 patients, all on the extract, did not reach the protocol dose because of intolerance. One serious adverse event, a pneumonia, occurred on the extract. The extract contained both compounds, so the trial cannot say which drove these effects.
The other trials differ. In Peball’s open-label phase with nabilone, the paper’s table lists fatigue (17), dizziness (9), daytime sleepiness (5), dry mouth (4) and confusion or disorientation (3) among the most common events; in the randomized phase, 32 percent on nabilone and 42 percent on placebo had any adverse event, and there were no serious events. The Thai CBD trial reported no serious side effects. Kubota’s oil produced more irritability than placebo (p=0.005); two severe events, Guillain-Barré syndrome and cryptococcal meningitis, were judged unrelated.
Dizziness in 58 percent of patients on the Colorado extract against 20 percent on placebo matters in a condition that already affects balance. What these trials have not shown is more falls. Liu’s registry lists four falls among 31 patients on the extract and three among 30 on placebo, and Peball’s randomized phase had one in each arm. Dizziness and confusion are plausible routes to falls, but that inference is not a measured result, which is why this article does not make it.
What patients are treating, and what has been tested
Among the survey’s users, the most common targets were anxiety (45.5 percent), pain (44.0 percent) and sleep (44.0 percent). Sixty of the 261, 23.0 percent, had stopped in the previous six months. Among those who had stopped, or who had a bad earlier experience, lack of symptom improvement was the most common reason (35.5 percent, 33 of 93). Users were more likely than non-users to say prescription medicines were not controlling their non-motor symptoms (p=0.03), and 610 of the 803 non-users (76.0 percent) said evidence supporting cannabis for Parkinson’s would be most influential in any decision to start.
The survey recorded what people reported, from people already connected to the Foundation, with a 17.6 percent response rate. It cannot show whether cannabis helps or does not. It shows demand and stated reasons, and those can be set against the trials by symptom. For pain, the one trial identified with pain as its primary endpoint, Kubota’s, found no difference. For sleep, the REM sleep behavior disorder trial found no difference on its primary outcomes, while the Peball authors attribute their result partly to anxious mood and night-time sleep. For anxiety, the evidence is a 24-patient trial of a single dose under a staged stressor, and a 27-patient Warsaw trial of CBD oil for anxiety and depression that finished in November 2024 and has posted no results.
So patient experience and randomized evidence do not line up neatly. The symptoms users treat have been tested, in trials that were small, short or varied in product, with null or mixed results.
What the evidence still cannot answer
Compounds, doses and durations differ enough that a null pooled estimate is not a verdict on any single product, in either direction. Peball’s nabilone is a synthetic THC analogue. Chagas, the sleep trial and the Thai trial used CBD alone or CBD with a little THC. Liu’s extract carried about 6.4 mg of THC and Kubota’s less than 1 mg, and Kubota’s authors suggest the low THC dose may help explain the contrast with earlier positive reports. CBD doses ran from an average of 26 mg a day in the Thai trial to 300 mg in Chagas’s higher-dose arm, and the two largest nulls, Kubota’s 43.88 mg and Liu’s 192 mg, do not describe the same exposure.
The longest randomized follow-up among these trials is 14 weeks. The longest nabilone exposure reported is an uncontrolled, open-label extension in which 22 earlier responders took the drug for six months. Neither says much about a progressive disease. The trial counts here come from randomized trials indexed in PubMed.
Professional guidance has not caught up. The Movement Disorder Society’s evidence-based review of non-motor treatments searched the literature through December 31, 2016, before Peball, Liu and Kubota. The Parkinson’s Foundation’s 2020 statement, drawn from a March 2019 convening, does not endorse cannabis use for Parkinson’s symptoms and lists dizziness and problems with balance among side effects to watch for. A PubMed search for guidelines and consensus statements found nothing more recent addressing cannabis in Parkinson’s specifically.
More trials are not assured. A 100-patient study of medical cannabis for non-motor symptoms at Sheba Medical Center has carried the status “unknown” since November 2021, with an estimated completion in August 2023. An 88-patient, four-arm CBD trial from Argentine company Laboratorio Elea Phoenix is listed as recruiting, with a primary completion date of November 2025 that has passed; its primary endpoint is adverse events, not efficacy.
The evidence has not shown that cannabinoids fail in Parkinson’s disease. Pooled randomized trials have found no clear benefit, the one positive non-motor trial asked a continuation question in selected responders, the most consistent safety signal is dizziness and confusion, and the trials are too few, short and varied to settle what any one product does. The evidence gap the survey’s non-users described is still open.