Cannabis and Epilepsy: What Cannabidiol Actually Does, and What It Does Not
Epilepsy is the only indication where a cannabis-derived medicine has cleared the full regulatory bar anywhere in the world — and that approval is far narrower than the headlines suggest.
1. Why Epilepsy Became the One Approved Cannabinoid Indication
Cannabis has been proposed for pain, anxiety, nausea and much else, on observational evidence. Epilepsy is different for procedural reasons: seizures per 28 days is a hard number, and because three severe childhood syndromes are drug-resistant by definition, randomised placebo-controlled add-on trials were ethically acceptable, with children keeping their existing medicines while cannabidiol or matched placebo went on top.
2. The Approved Purified Cannabidiol Medicine
Which regulators approved it, and when
The medicine is a purified cannabidiol oral solution from GW Pharmaceuticals, now part of Jazz Pharmaceuticals, sold as Epidiolex in the United States and Epidyolex in Europe and Australia: pharmaceutical grade, defined concentration, negligible tetrahydrocannabinol.
The United States Food and Drug Administration approved it in 2018 for Lennox-Gastaut syndrome and Dravet syndrome, adding tuberous sclerosis complex in August 2020; the label now covers all three from one year of age. The European Commission authorised it on 19 September 2019, with a wording difference that matters: in Europe those two syndromes are licensed only in conjunction with clobazam, from two years. The tuberous sclerosis indication followed in April 2021 without that condition; Australia registered it in September 2020, Canada in December 2023. No approval covers cannabis, or cannabidiol generally, for epilepsy.
The three syndromes on the label
| Feature | Dravet syndrome | Lennox-Gastaut syndrome | Tuberous sclerosis complex |
|---|---|---|---|
| Nature | Genetic encephalopathy, often SCN1A-linked | Multiple seizure types from early childhood | Multisystem disorder with benign tumours |
| Endpoint counted | Convulsive seizures | Drop seizures (tonic or atonic falls) | All associated seizures |
| Approvals | 2018 United States, 2019 Europe | 2018 United States, 2019 Europe | 2020 United States, 2021 Europe |
3. What the Pivotal Trials Measured and Found
Each study ran a four-week baseline count, then randomised patients to cannabidiol or placebo added to an existing regimen of typically three or four anti-seizure medicines. Both arms are shown, because a reduction figure without its placebo comparator misleads.
| Trial | Syndrome, seizures counted | Dose | Cannabidiol | Placebo | Result |
|---|---|---|---|---|---|
| Devinsky 2017 | Dravet, convulsive | 20 mg/kg/day | 38.9% | 13.3% | Difference −22.8 points, P=0.01 |
| Miller 2020 | Dravet, convulsive | 10 and 20 mg/kg/day | ~49% and 46% | ~27% | P=0.0095, P=0.0299 |
| Devinsky 2018 | Lennox-Gastaut, drop | 10 and 20 mg/kg/day | 37.2% and 41.9% | 17.2% | P=0.002, P=0.005 |
| Thiele 2018 | Lennox-Gastaut, drop | 20 mg/kg/day | 43.9% | 21.8% | Difference −17.21, P=0.0135 |
| Thiele 2021 | Tuberous sclerosis, all | 25 and 50 mg/kg/day | 48.6% and 47.5% | 26.5% | P<0.001, P=0.002 |
Reading the numbers without being misled
Placebo arms fell 13 to 27 per cent with no active drug: fluctuation in severe epilepsy, closer diary-keeping, regression to the mean, expectation. The drug effect is the gap between the columns, about 20 to 25 points.
Responder rates say the same. In the first Dravet trial, 43 per cent achieved at least a 50 per cent reduction against 27 per cent on placebo, which alone did not reach significance (P=0.08). In tuberous sclerosis, responder rates were 36 per cent at 25 mg/kg/day and 40 per cent at 50 mg/kg/day against 22 per cent on placebo. Seizure freedom was rare: 5 per cent against zero. About half of treated children were not responders, and a phase 3 trial in Japan missed its endpoint in 2024.
4. Doses Used in the Trials, Reported as Study Facts
These figures record what the studies gave under supervision with laboratory monitoring. They are not instructions: nothing here should be used to start, stop or calculate a dose.
The Dravet and Lennox-Gastaut trials studied 10 and 20 mg/kg/day, reached by gradual titration; the tuberous sclerosis trial studied 25 and 50 mg/kg/day. The prescribing information sets maintenance at 10 mg/kg/day for Lennox-Gastaut and Dravet syndromes with a maximum of 20, and 25 mg/kg/day for tuberous sclerosis complex, where the higher dose was no more effective but less tolerable.
Titration, monitoring and any regimen change belong to the treating neurologist. Families should never reduce or replace an existing anticonvulsant on their own initiative; abrupt changes to established anti-seizure medication can cause serious harm.
5. Adverse Effects Recorded in the Randomised Trials
From the label’s Lennox-Gastaut and Dravet trial data, with placebo alongside: somnolence 23 per cent at 10 mg/kg/day and 25 per cent at 20 mg/kg/day against 8 per cent; decreased appetite 16 and 22 per cent against 5 per cent; diarrhoea 9 and 20 per cent against 9 per cent, so at the lower dose diarrhoea was no commoner than with placebo. In the tuberous sclerosis trial at 25 mg/kg/day: diarrhoea 31 against 25 per cent, appetite loss 20 against 12 per cent, somnolence 13 against 9 per cent.
Liver enzyme elevations
Alanine aminotransferase above three times the upper limit of normal occurred in 13 per cent of cannabidiol-treated patients in the Lennox-Gastaut and Dravet trials against 1 per cent on placebo, and 12 per cent in the tuberous sclerosis trial. Transaminase elevation as a reported adverse reaction: 8 per cent at 10 mg/kg/day, 16 per cent at 20, against 3 per cent.
The risk is uneven: at therapeutic doses the label reports such elevations in 30 per cent of patients on both valproate and clobazam, 21 per cent on valproate alone, 4 per cent on clobazam alone and 3 per cent on neither. In the three-year Dravet extension, 21 per cent of 315 patients had them, 93 per cent of those on valproate.
6. Drug Interactions That Change the Picture
Clobazam, valproate and the enzyme picture
Clobazam is a benzodiazepine widely used in these syndromes. Cannabidiol moderately inhibits CYP2C19, which clears N-desmethylclobazam (norclobazam), clobazam’s active metabolite, so norclobazam rises substantially — Rogawski describes roughly 2.5 to 3-fold increases, meaning some sedation attributed to cannabidiol is a benzodiazepine effect. The label advises prescribers to consider a clobazam dose reduction if adverse reactions appear: a physician’s decision with drug-level monitoring, never a family’s. Valproate barely changes cannabidiol levels but clearly raises liver enzyme elevations.
Cannabidiol is metabolised by CYP2C19 and CYP3A4; it inhibits CYP2C19 moderately, CYP1A2 weakly and UGT1A9, and may interact with CYP2B6 and CYP2C8 substrates.
| Co-medication | What happens | Why it matters |
|---|---|---|
| Clobazam | Active metabolite N-desmethylclobazam rises markedly | Added sedation; clouds attribution of benefit |
| Valproate | Higher incidence of liver enzyme elevations | Largest single driver of transaminase risk |
| Stiripentol | Stiripentol exposure increases | Common in Dravet regimens |
| Everolimus | About 2.5-fold increase in exposure | Directly relevant in tuberous sclerosis |
7. How Much of the Benefit Was the Clobazam Interaction?
Rogawski, in Epilepsy and Behavior in 2019, highlighted open-label data in which 51 per cent of patients taking clobazam were responders against 27 per cent of those not taking it, and argued controlled trials were needed to show whether cannabidiol has independent anti-seizure activity at all. The European licence requiring clobazam is a regulator declining to answer. Against that, Devinsky and colleagues, in Acta Neurologica Scandinavica in 2020, pooled four randomised trials covering 714 patients and found significant reduction against placebo both with and without clobazam. Subgroup analyses cannot settle it.
8. Purified Isolate Against Full-Spectrum Extract
The “whole plant is better” claim usually traces to one paper: Pamplona, da Silva and Coan, in Frontiers in Neurology in 2018, a meta-analysis of observational data. On the loosest measure — improvement of any kind — 71 per cent on cannabidiol-rich extracts improved against 46 per cent on purified cannabidiol. At the threshold of at least a 50 per cent reduction the difference vanished: 37 against 42 per cent. Mean dose was far lower in the extract group, about 6 against 25 to 27 mg/kg/day.
Why the comparison is not settled
Every included study was observational and unblinded with no placebo arm; as the authors wrote, it is impossible to quantify how much of the response is placebo effect. Doses were not comparable, so formulation and dose are confounded. On the loose endpoint most vulnerable to expectation bias extracts looked better; on the rigorous endpoint they did not differ. No randomised head-to-head trial has settled it.
9. Unregulated Cannabidiol Products Are a Different Category
Bonn-Miller and colleagues, in the Journal of the American Medical Association in 2017, tested 84 cannabidiol products from 31 online companies. Around 30 per cent contained cannabidiol within 10 per cent of the labelled amount; roughly 42 per cent contained more than claimed and about 26 per cent less. Tetrahydrocannabinol was detected in a meaningful minority — serious when the user is a child.
Contamination and counterfeits
Between October 2017 and January 2018, Utah public health authorities investigated 52 acute poisonings from products sold as cannabidiol that in fact contained a synthetic cannabinoid; around 60 per cent were hospitalised, with altered mental status, vomiting, and seizures or shaking. A product marketed to control seizures caused them. Pesticide and solvent residues, heavy metals and microbial contamination are further risks. A wellness oil is not the approved medicine at a friendlier price: the epilepsy evidence attaches to the regulated product, not to the molecule.
10. The Charlotte Figi Story, Told Accurately
Charlotte Figi was born in Colorado on 18 October 2006, had her first seizure at around three months, and was later diagnosed with Dravet syndrome. By age five she was reported to have up to 300 convulsive seizures a week, after many failed medications. In 2012 her family obtained a cannabis extract low in tetrahydrocannabinol and high in cannabidiol, from a variety renamed Charlotte’s Web, and reported a dramatic fall in seizures; in 2013 she appeared in Sanjay Gupta’s CNN documentary Weed. She died on 7 April 2020, aged 13, of respiratory failure and cardiac arrest after pneumonia triggered seizures.
What it demonstrated, and what it did not
The policy consequences were fast: families relocated to Colorado, jurisdictions passed cannabidiol-specific access laws, and the resulting pressure made the formal trials above fundable.
Her case demonstrated that one child improved dramatically after starting a cannabidiol-rich extract, and that a single case can change law faster than a decade of committee work. It did not demonstrate the average effect of cannabidiol, because a single case cannot; it did not show that a full-spectrum extract beats a purified one, because no comparison was made; and it established no dose or formulation. That is no criticism of the Figi family, who never claimed otherwise, but of how the story was later used.
11. What Is Still Not Established
Adult focal epilepsy and the role of tetrahydrocannabinol
The commonest form of adult drug-resistant epilepsy has no comparable evidence base. A randomised trial of adjunctive transdermal cannabidiol in adults with focal epilepsy, in JAMA Network Open in 2022, found no difference against placebo: 2.51 seizures per 28 days at the lower dose and 2.59 at the higher, against 2.49 on placebo (P=0.89 and P=0.32). Its open-label extension looked encouraging, 60.8 per cent reaching at least a 50 per cent reduction by month six — showing why extensions cannot replace a placebo arm. No randomised evidence shows tetrahydrocannabinol reduces seizures, and it has its own risks in a developing brain.
Long-term outcomes
Open-label extensions are the longest view available, with no placebo group. In the three-year Dravet extension of 315 patients, retention was 72 per cent at one year, 53 per cent at two and 45 per cent at three; adverse events were reported by 97 per cent, most commonly diarrhoea 43 per cent, pyrexia 39 per cent, decreased appetite 31 per cent. Effects on cognition, development and behaviour over a childhood remain unknown.
12. Practical Realities for Families in Southeast Asia
Thailand’s position changed sharply in 2025. The Notification of the Ministry of Public Health Re: Controlled Herbs (Cannabis), B.E. 2568 signed 23 June 2025 and effective from 26 June 2025, reclassified cannabis flower as a controlled herb and made supply prescription-only: an official form, an authorised practitioner, 30 days maximum, no refill. Advertising cannabis is banned on all channels, which is why no legitimate Thai source markets a cannabis product as a treatment for a child’s epilepsy. Non-compliant supply carries up to a year’s imprisonment, a fine of up to 20,000 baht, or both. The route is therefore the ordinary medical one: a paediatric neurologist, a documented diagnosis, a prescription inside the licensed system.
Thai clinical experience exists. Two centres under the Department of Medical Services — the Neurological Institute of Thailand and Queen Sirikit National Institute of Child Health — treated 14 children with drug-resistant epilepsy using a product with at least 20 parts cannabidiol to one part tetrahydrocannabinol, median 5.6 mg/kg/day. Among the nine continuing for a median 18 months, the 50 per cent responder rate was 50 per cent for convulsive and 43 per cent for total seizures. Adverse effects were common — somnolence 64 per cent, worsening seizures 43 per cent, irritability 42 per cent, reduced appetite 28 per cent — and five of the 14 stopped for serious adverse effects, one with hepatitis. It was uncontrolled, with no placebo arm.
Cost, supply and the cross-border trap
The approved medicine is expensive and is not registered everywhere in the region. Where it is unavailable, families may be offered domestic cannabidiol-enriched preparations, which are not the product studied in the pivotal trials even when the molecule is nominally the same; concentration, contaminants, batch consistency and tetrahydrocannabinol content all differ, so the trial evidence does not transfer. Nor should anyone carry cannabis across borders here: several Southeast Asian jurisdictions impose severe criminal penalties for possession and importation.
If your family has lived through this — the diagnosis, the waiting lists, the seizure diaries, the decision about whether to try one more medicine — your experience is worth more to other families here than any summary of trial data. Tell us what the access pathway looked like where you are and what the monitoring involved, and please keep to lived experience rather than product recommendations. One point is non-negotiable: every decision about starting, stopping, titrating or replacing an anti-seizure medicine belongs with your child’s treating neurologist, never with a forum thread.
This article is part of the Asiannabis Community educational series on medical cannabis. It is general information, not medical advice — treatment decisions belong with a licensed practitioner. Content is for educational purposes within jurisdictions where medical cannabis is legally permitted.