Edibles and Seizure Medications: What the Epidiolex Label Measured, at What Dose, and Why the Neurologist Owns the Decision

Most cannabinoid interaction warnings are inference: an enzyme, a drug that uses it, a shrug. Seizure medications are the exception, for a reason nobody planned. CBD is one of them. Every Epidiolex seizure trial added CBD to existing drugs, so its maker ran interaction studies in healthy volunteers, then double-blind trials in patients, and the FDA label prints the answers. No other drug class holds all three layers, by a PubMed search on November 7, 2026.

That label is the story, and the generic CBD warning skips it, handing one paragraph to someone on lamotrigine, where nothing moved, and someone on clobazam, where a number tripled. Then comes the part a dispensary counter has no reason to raise: those numbers came from 10 to 20 mg/kg a day of pharmaceutical CBD, and THC carries a seizure warning on its own label with no human threshold measurement behind it.

Our position: the dose gap is a reason for calm and no reason for silence, because nobody has measured the floor. The neurologist, who can order a level or a liver panel, owns the decision. The budtender can tell you what the gummy tastes like. (Efficacy lives on our epilepsy page, the wider enzyme map on our CBD interactions hub.)

Which seizure medications does CBD actually change?

Clobazam's active metabolite is the one large, repeated finding, about threefold by the Epidiolex label. Stiripentol rises modestly, topiramate rises inside its normal range, and valproate's level holds while liver risk climbs. Phenytoin has two published cases. Lamotrigine showed no trend across 63 patients, and rufinamide, zonisamide and eslicarbazepine faded in the final Alabama analysis.

The table lists each drug, what CBD did to its blood level, the CBD dose at which that was measured, and the source.

Two kinds of evidence sit there. Morrison, VanLandingham and Ben-Menachem were built to test interactions at fixed doses, and all came from the manufacturer, whose own summary (Patsalos 2020) finds low interaction potential except for clobazam. The University of Alabama numbers come from an open-label state program whose analysts skipped multiple-comparison corrections on purpose. Its 2017 analysis of 81 patients produced the list that still circulates. Its 2023 final analysis of 169 lost rufinamide, zonisamide and eslicarbazepine and rated the clinical significance of its findings likely minor.

Why is clobazam the interaction everyone cites?

Because it is large, repeated and explained by one enzyme. CBD moderately inhibits CYP2C19, which clears N-desmethylclobazam, clobazam's active metabolite. The Epidiolex label reports about threefold at 750mg twice daily, Morrison measured 3.4-fold in volunteers, and a double-blind trial in patients found 2.6-fold at 20 mg/kg a day.

Geffrey's 13 children at Massachusetts General showed the clinical version: a mean 500 percent rise at four weeks, clobazam cut in 10 of the 13, side effects easing with the cuts. The studies split on the parent drug, which rose for Geffrey, fell in Gaston's 2017 analysis, rose in 2023 and barely moved in the dedicated trials. The label prices the metabolite at somnolence and sedation in 46 percent of Epidiolex patients on clobazam against 16 percent of those not on it, and our benzodiazepines guide carries that arithmetic. Gaston and Szaflarski's 2018 review pins the cause on CYP2C19; the label's CYP3A4 and CYP2C9 probes showed no change, and CYP2D6 is not named.

The Onfi label never mentions CBD, and it already has a page for this. It warns that moderate CYP2C19 inhibitors can raise the metabolite, and it reports plasma levels 5-fold higher in genetic CYP2C19 poor metabolizers, a group it starts at 5mg a day and titrates to half the usual dose. Trial-dose CBD pushes an ordinary patient part of the way there.

Stiripentol makes it a stack. Diacomit is indicated only for Dravet patients taking clobazam and by its own label raises the metabolite 5-fold, and in the Epidiolex Dravet trial 65 percent of patients were on clobazam, 57 percent on valproate and 43 percent on stiripentol. When Wirrell and Sullivan polled 26 US Dravet experts in 2026 (Biocodex, which makes Diacomit, gave operational support), 92 percent agreed on periodic liver tests with valproate. Asked whether clobazam needs no further cut when CBD joins stiripentol and clobazam, 42 percent agreed, short of consensus. Specialists split on that question. A dispensary counter will not settle it.

What does CBD do to valproate and the liver?

Little to valproate's total level and a good deal to the liver. Morrison saw no clinically relevant change and patients saw small falls, but ALT rose past three times the upper limit of normal in 21 percent of Epidiolex patients on valproate without clobazam, against 3 percent of those on neither drug.

Those are the label's Lennox-Gastaut and Dravet figures at 10 and 20 mg/kg a day: 30 percent with valproate and clobazam together, 21 with valproate alone, 4 with clobazam alone, 3 with neither. The dose slope is steep, with ALT crossing that line in 17 percent of patients at 20 mg/kg a day against 1 percent at 10. The label asks for a panel before starting and at one, three and six months, and suggests more often on valproate.

One case report sits on the retail side. Singh and colleagues described a 2-year-old boy on clobazam and zonisamide, found to have severe liver dysfunction during a visit for a respiratory illness. His parents had been giving him nonprescription hemp extract, his enzymes fell after it was held, and the reporters rated the link probable. The abstract gives no dose, and one child proves little.

Do carbamazepine and phenytoin change how CBD or THC works?

Probably, by label logic, though nobody has measured either directly. Both are potent enzyme inducers. Rifampin, the inducer the Epidiolex label tested, cut CBD exposure 32 percent and its active metabolite 63 percent, and the label allows up to a doubling of Epidiolex alongside a strong inducer. Phenytoin also runs the other way: CBD raised it in two cases.

The Tegretol label calls carbamazepine a potent CYP3A4 inducer, and the Diacomit label names phenytoin and carbamazepine alongside rifampin. The Epidiolex label names only rifampin, tested at 600mg a day against single 750mg doses of CBD, and a PubMed search on November 7, 2026 found no human study of what either drug does to CBD levels. In the other direction carbamazepine looks quiet, which fits: its label routes it through CYP3A4, whose probe did not move at 750mg of CBD twice daily.

Phenytoin is the one to watch, on the evidence of two people. Blond and colleagues described two Lennox-Gastaut patients whose phenytoin rose after CBD was added. The first went a year free of tonic-clonic seizures, then relapsed when phenytoin was lowered over liver concerns. The second developed ataxia and falls that resolved on a lower dose. The abstract gives no CBD dose, and the lead author has consulted for Jazz. An average across 11 Alabama patients cannot rule that out, and the Dilantin label says why individuals matter: phenytoin runs partly on CYP2C19, its metabolism saturates, and a dose increase of 10 percent or more can raise levels disproportionately, into intoxication.

THC sits on the same hook. The Marinol label says inducers can lower dronabinol exposure, and rifampicin cut peak 11-hydroxy-THC 87 percent in Stott's Sativex study; the same search session found no human pharmacokinetic study of THC with carbamazepine or phenytoin. Our trigeminal neuralgia page covers the trap: a dose built up around an inducer lands harder if the neurologist lowers or switches it.

A patient.info thread from May 2017 shows how this plays out. One poster described years without seizures on Tegretol and phenobarbital, then a move to Keppra because both interacted with Harvoni, the hepatitis C drug. The same poster was taking CBD capsules for neuropathy pain, bought at a dispensary with no prescription, and could not see a doctor for another two months. Across 21 posts nobody gave a milligram figure. One interaction earned a change of drugs, and the other earned a shopping trip.

How does a 25mg gummy compare with the doses in these trials?

For an adult, it is a small fraction. The trials ran 10 to 20 mg/kg a day, which for a 70kg adult is 700 to 1,400mg, or 28 to 56 gummies at 25mg. The label's volunteer dose of 1,500mg a day is 60 gummies. A 20kg child reaches the trial range at 8 to 16.

The pivotal trials gave 20 mg/kg a day in Dravet (Devinsky 2017) and 10 or 20 in Lennox-Gastaut (Devinsky 2018, Thiele 2018). Green Roads' Relax Berries, thirty 25mg gummies for $49.99 on our CBD gummy ranking, put the volunteer dose at two jars and $99.98 a day, against 15 millilitres of Epidiolex. Through a 1:1 gummy like Wyld's Pomegranate, 700mg of CBD arrives with 700mg of THC. Herdegen and Cascorbi's 2023 review estimates that clinically relevant interactions start above 300mg of CBD a day. Food widens the uncertainty: the label's high-fat meal raised peak CBD five-fold and exposure four-fold.

The floor is unmeasured. A PubMed search on November 7, 2026 pairing CBD with these drugs and terms for gummies, edibles, hemp and artisanal products found no study that measured a gummy. The nearest data point is uncomfortable. Cohen's chart review at Children's National compared 11 children on artisanal CBD, at a median 4.3 mg/kg a day, with 31 on pharmaceutical CBD at 25. Serum CBD was lower in the artisanal group, and mean N-desmethylclobazam was not: 8,054 ng/mL against 6,780, with a reference range of 300 to 3,000. Six artisanal children were on clobazam, no pre-CBD level is reported and caregivers set the dose, so this proves nothing about cause. It does mean the low end cannot be assumed quiet. For a 20kg child, 4.3 mg/kg is 86mg a day, under four gummies.

Does THC lower the seizure threshold?

Nobody has measured that in people. The Marinol label for synthetic THC warns of seizures and asks prescribers to weigh the risk in patients on antiepileptic drugs. A PubMed search on November 7, 2026 found case reports, surveys, registry data and animal studies, and no human study of seizure threshold.

That search, for THC or cannabis with seizure threshold or proconvulsant effects in humans, returned nine records and none measured a threshold. A second, for trials of THC, dronabinol or nabilone with seizures or epilepsy, returned nine and none tested THC alone against seizures.

The human record points both ways. Kaczor's 2022 scoping review found 10 of 11 acute-exposure studies reporting more seizures than background rates predict, proposed THC as the cause and said its data cannot establish one; it excluded epilepsy populations. French drug-safety systems logged 130 convulsive reports among 4,296 cannabinoid notifications through July 2023, mostly recreational and often with other drugs or alcohol. One was a 3-year-old with Dravet syndrome who died of status epilepticus while taking a CBD and THC product bought online against medical advice; spontaneous reports cannot assign cause. Navarro's open-label Colombian cohort reported fewer seizures on a CBD-rich oil with a median 4mg of THC a day, with no control arm and too much CBD to say what the THC did.

Mice do not settle it. Injected THC at 10 mg/kg triggered electrographic seizures in Malyshevskaya's animals. In Anderson's Dravet model, very low doses protected against heat-induced seizures; neither cannabinoid alone changed spontaneous seizures, and the two together in chow worsened them and raised mortality. A mouse is no dosing guide, but that pairing is the one printed on a ratio gummy. The American Epilepsy Society calls the evidence for cannabis itself highly limited.

What the neurologist can use

Whether to order a level is itself a judgment call. Geffrey's authors called clobazam monitoring necessary, and the 2026 Dravet consensus rated routine plasma monitoring unnecessary while allowing it to clarify an interaction. The American Epilepsy Society urges people to make these decisions with an epilepsy specialist. Four details give that conversation numbers.

1. Every seizure medication and its dose, flagged if it is clobazam, valproate, stiripentol or phenytoin.
2. CBD in milligrams per day, by product, gummy or oil, and whether it goes down with food.
3. THC in milligrams per day, including the small amount in a ratio gummy.
4. The start date, any change in seizure count or sleepiness since, and any medication dose skipped or cut.

How to open that conversation is covered in our guide to using edibles for medical purposes.