The CBD Drug Interactions That Are Actually Documented, and the Doses at Which They Happen

Four interactions have real human evidence behind them: clobazam, where cannabidiol raises the active metabolite N-desmethylclobazam severalfold; valproate, where the combination raises transaminases without a plasma-level interaction; warfarin, documented in a case report with rising INR; and tacrolimus, documented in a trial participant with roughly threefold higher dose-normalized concentrations.

Every one of those observations came from pharmaceutical-grade cannabidiol at doses between roughly 700 and 2,900 mg per day. Ordinary retail cannabidiol products deliver 10 to 50 mg. Those are not the same exposure, and the interaction literature does not transfer cleanly from one to the other in either direction.

Patients do not report cannabidiol as a medication. They report it as a supplement, if at all, and often not at all, because it is sold next to vitamins. The result is that a genuine pharmacokinetic perpetrator drug enters a regimen without anyone adjusting monitoring.

The consequences are not theoretical. Sedation on clobazam, transaminase elevation on valproate, an unstable INR on warfarin and a supratherapeutic tacrolimus level are all documented, and all four are manageable if someone knows to look.

Rongrong Jiang and colleagues at Hokuriku University characterized cannabidiol metabolism in human liver microsomes in Life Sciences in 2011. Seven of fourteen recombinant cytochrome P450 enzymes tested could metabolize it, and correlation and inhibition work identified CYP3A4 and CYP2C19 as the principal routes, with formation of the active metabolite 7-hydroxy-cannabidiol driven predominantly by CYP2C19.

So cannabidiol is a substrate of the same enzymes that handle a large share of the prescription formulary. It is also an inhibitor of them. Joshua Brown and Almut Winterstein at the University of Florida summarized this in the Journal of Clinical Medicine in 2019 with a phrase worth keeping: cannabidiol is implicated as both a victim and a perpetrator of drug interactions.

Sumit Bansal, Mary Paine and Jashvant Unadkat at the University of Washington quantified the inhibition and modeled what it should produce. Publishing in Drug Metabolism and Disposition in 2022, they predicted maximum interactions for oral cannabidiol at 700 mg per day against probe drugs, expressed as area under the curve ratios: midazolam 14.8 for CYP3A, diclofenac 9.6 for CYP2C9, omeprazole 7.3 for CYP2C19, theophylline 4.0 for CYP1A2, and smaller ratios for CYP2B6, CYP2D6 and CYP2C8. They also found that the metabolite 7-hydroxy-cannabidiol inhibits most of these enzymes and inactivates CYP2C19 and CYP3A. The authors stated plainly that in vivo studies are needed to verify these predictions, and that caveat should travel with the numbers.

Alexandra Geffrey, Elizabeth Thiele and colleagues at Massachusetts General Hospital reported the clinical version first, in Epilepsia in 2015. Thirteen children with refractory epilepsy were taking clobazam while receiving cannabidiol under an expanded access protocol. At four weeks, mean clobazam levels had risen 60 percent, and mean N-desmethylclobazam levels, the active metabolite, had risen 500 percent.

The clinical texture of that report is worth repeating. Nine of the thirteen had greater than 50 percent seizure reduction. Ten of thirteen reported side effects. Those side effects were relieved by reducing the clobazam dose, and clobazam doses were in fact reduced in ten of the thirteen patients over the course of treatment. In other words, a substantial part of both the benefit and the toxicity attributed to cannabidiol in that setting was a clobazam interaction.

A controlled healthy-volunteer study confirmed the direction and refined the magnitude. Gilmour Morrison and colleagues ran an open-label fixed-sequence trial published in Clinical Pharmacology in Drug Development in 2019. Cannabidiol had little effect on clobazam itself, about 1.2-fold on both peak concentration and area under the curve, while N-desmethylclobazam exposure rose 3.4-fold. Stiripentol exposure rose modestly. Clobazam in turn increased 7-hydroxy-cannabidiol exposure by roughly 1.5 to 1.7-fold, making the interaction bidirectional.

This one is easy to get wrong, because the mechanism is not what most people assume. Tyler Gaston, Jerzy Szaflarski and colleagues at the University of Alabama at Birmingham monitored serum levels of 19 antiepileptic drugs in 39 adults and 42 children receiving pharmaceutical cannabidiol titrated from 5 up to 50 mg per kilogram per day, publishing in Epilepsia in 2017.

They found significant changes in topiramate, rufinamide and N-desmethylclobazam levels, a decrease in clobazam, and in adults increases in zonisamide and eslicarbazepine. Apart from clobazam and its metabolite, the changes stayed within accepted therapeutic ranges. Separately, aspartate aminotransferase and alanine aminotransferase were significantly higher in participants taking concomitant valproate.

The healthy-volunteer study found no clinically relevant pharmacokinetic interaction between cannabidiol and valproate in either direction. Put those two findings together and the picture is specific: the valproate problem is a hepatic signal, not a plasma-level interaction, which means checking a valproate level will not detect it. Liver function tests will. Orrin Devinsky’s Dravet syndrome trial, published in the New England Journal of Medicine in 2017 with 120 participants at 20 mg per kilogram per day, listed abnormal liver-function results among the adverse events more common with cannabidiol than placebo.

Leslie Grayson, Jerzy Szaflarski and colleagues published a case report in Epilepsy and Behavior Case Reports in 2017 describing an interaction between warfarin and cannabidiol, with the practical recommendation that INR be monitored when cannabinoids are introduced. Warfarin’s S-enantiomer is cleared substantially by CYP2C9, which cannabidiol inhibits, so the mechanism is coherent. The evidence is one patient.

The tacrolimus report is similarly small and similarly important. Abbie Leino and colleagues at the University of Cincinnati described in the American Journal of Transplantation in 2019 a participant in a cannabidiol epilepsy trial who was also taking tacrolimus and showed an approximately threefold increase in dose-normalized tacrolimus concentrations while receiving 2,000 to 2,900 mg of cannabidiol per day.

A lower-dose counterpoint exists and should be read alongside it. A Uruguayan group led by L. Cunetti gave cannabidiol to seven kidney transplant recipients for chronic pain, escalating from 100 to 300 mg per day, and reported in Transplantation Proceedings in 2018 that tacrolimus levels were variable, requiring dose reduction in some patients, while patients on cyclosporine remained stable. Seven patients, three weeks, no control group. The honest summary across both reports is that cannabidiol can raise calcineurin inhibitor concentrations and that nobody has characterized the dose threshold at which it starts to matter.

Here are the doses at which the documented interactions were observed. The Dravet trial used 20 mg per kilogram per day, which is 1,400 mg daily for a 70 kilogram adult. The Alabama safety study titrated to as much as 50 mg per kilogram per day. The healthy-volunteer liver study used 1,500 mg per day. The tacrolimus case involved 2,000 to 2,900 mg per day. The interaction modeling was run at 700 mg per day.

A typical retail cannabidiol product delivers 10 to 50 mg per serving. That is one to two orders of magnitude below every figure in the previous paragraph. Treating the trial interaction data as directly applicable to a 25 mg gummy overstates the risk considerably.

The opposite conclusion is equally unsupported. No published human study has characterized interaction risk at consumer doses. The correct statement is that the risk at 25 mg per day is uncharacterized, not that it is zero, and there is a specific reason to withhold reassurance: Marcel Bonn-Miller and colleagues published a research letter in JAMA in 2017 comparing label claims against laboratory analysis for cannabidiol extracts sold online, and label accuracy in that market cannot be assumed. A patient taking what they believe is 25 mg may not be.

Paul Watkins and colleagues at the University of North Carolina reported something that deserves more attention than it gets, in Clinical Pharmacology and Therapeutics in 2021. Sixteen healthy adults with no epilepsy and no concomitant antiepileptic drugs received 1,500 mg of cannabidiol daily for about three and a half weeks as part of a phase 1 interaction study.

Seven of the sixteen, 44 percent, had a peak alanine aminotransferase above the upper limit of normal. In five of them, 31 percent, the value exceeded five times the upper limit, meeting international consensus criteria for drug-induced liver injury. Elevations began within two to four weeks of starting cannabidiol. Six participants were discontinued from the protocol, and some had symptoms consistent with hepatitis or hypersensitivity.

Two features of that result matter clinically. There was no correlation between transaminase elevation and baseline characteristics, CYP2C19 genotype, or cannabidiol plasma concentration, meaning susceptibility could not be predicted in advance. And these were healthy adults, so the liver effect is not solely a consequence of combining cannabidiol with valproate. At pharmaceutical doses, cannabidiol is a drug with a hepatic safety profile, and it should be monitored like one.

The clobazam interaction is the strongest item here, supported by a clinical cohort and a controlled healthy-volunteer pharmacokinetic study that agree on direction and roughly on magnitude. The liver findings are also strong, coming from a randomized trial program plus a dedicated phase 1 study in healthy adults. Treat both as established at pharmaceutical doses.

The warfarin and tacrolimus evidence is weaker in study design and stronger in consequence. A single case report and a single trial participant are thin, and both involve narrow therapeutic index drugs where a threefold concentration change causes real harm. The correct clinical response to thin evidence about a dangerous interaction is monitoring, not dismissal.

The interaction magnitudes most often quoted in patient-facing material come from static pharmacokinetic modeling rather than from dosing people. The University of Washington group predicted a nearly fifteenfold increase in midazolam exposure with 700 mg of cannabidiol and explicitly stated that in vivo studies are warranted to verify the predictions. Reproducing that number without the caveat misrepresents the source.

The clinical reports are also small and uncontrolled. Thirteen children on clobazam. One patient on warfarin. One trial participant on tacrolimus. Seven transplant patients over three weeks. These are the right studies to change monitoring practice and the wrong studies to quantify risk for an individual patient.

This literature does not show that low-dose retail cannabidiol causes clinically meaningful interactions. It also does not show that it is safe. No published human study has measured interaction magnitude at 10 to 50 mg per day, which is where most consumer use sits, so both the alarming and the reassuring versions of that claim are unsupported.

The evidence does not identify who is susceptible. In the healthy-adult liver study, transaminase elevation showed no correlation with baseline characteristics, CYP2C19 genotype or cannabidiol plasma concentration. There is currently no way to screen a patient in advance and predict who will have a problem.

The regulatory situation created this confusion. The same molecule exists as a prescription drug with a full interaction section in its labeling and as a retail supplement with none, at doses differing by a factor of fifty. Patients encounter both under the same three letters and reasonably assume the safety information transfers.

It mostly does not, and the direction of the error runs both ways. People taking 25 mg are frightened by trial-dose warnings, while people taking 1,500 mg of a high-strength oil from an unregulated supplier receive no monitoring at all. The second group is the one clinicians should be looking for.

The clobazam data changed how I think about this whole subject. When a child on clobazam improves after starting cannabidiol, some meaningful part of that improvement may be the fivefold rise in an active benzodiazepine metabolite, and some meaningful part of the sedation is the same thing. That is not an argument against cannabidiol. It is an argument for knowing what you are actually changing.

In practice my rules are unglamorous. I ask about cannabidiol by name, because patients do not volunteer it. If someone is on warfarin, tacrolimus, cyclosporine, clobazam or valproate, I want to know the milligram dose and the product, and I want baseline labs. And I tell patients on high-dose oils that this is a drug with a liver signal in healthy volunteers, which is usually the first time anyone has framed it that way for them.

Cannabidiol inhibits CYP3A4, CYP2C19 and CYP2C9 and has documented interactions with clobazam, warfarin and tacrolimus, plus a transaminase signal with valproate that plasma levels will not detect. Those observations come from doses of roughly 700 to 2,900 mg per day. At retail doses of 10 to 50 mg the risk is uncharacterized rather than absent. Patients on narrow therapeutic index drugs should disclose cannabidiol use and have appropriate monitoring, including liver function tests and INR where relevant.

Carry two things forward. First, the named interactions are real and have specific monitoring attached: clobazam levels and sedation, liver function tests with valproate or high-dose cannabidiol, INR on warfarin, drug levels on calcineurin inhibitors. Second, dose is the variable that determines whether any of this applies, and the doses in the literature are far above what most people are taking. A source that omits the dose is not giving you enough information to act on.