Does CBD Reduce Adult Epilepsy Seizures While Causing Drowsiness or Liver-Test Changes? A 37-Patient Study

Short answer: In a 2026 retrospective study of 37 adults with severe epilepsy who received prescription cannabidiol (CBD) alongside other antiseizure medications, 17 of 37 patients (45.9%) had at least a 50% reduction in recorded seizure frequency at 12 months. Yet 24 of 37 (64.9%) experienced a reported adverse event, including drowsiness in 10 (27.0%) and abnormal liver tests in seven (18.9%). Caregivers also reported better alertness, behavior, or functioning in 11 (29.7%). These observations are not contradictory: seizure response, perceived daily function, sedation, and liver laboratory findings are different endpoints, observed in an uncontrolled clinical cohort. They do not establish CBD’s independent causal contribution or any particular patient’s forensic impairment.

The primary research is Sánchez-Gamino and colleagues’ 2026 multicenter study of adult cannabidiol treatment in Epilepsia Open. The most defensible conclusion is descriptive: these outcomes occurred during treatment of complex epilepsy in three Spanish centers, not in a randomized CBD-versus-placebo trial. The study did not directly quantify CBD or THC in any blood, urine, or oral-fluid specimen and did not test driving, psychomotor fitness, or case-specific causation.

What was the actual clinical study?

The investigators reviewed records from three medical centers in Spain for adults diagnosed with Lennox–Gastaut syndrome, Dravet syndrome, or epilepsy associated with tuberous sclerosis complex. Eligibility required starting CBD before August 2024 and having sufficient medical-chart follow-up. The cohort comprised 37 people, 25 men and 12 women, with a median age of 25 years (interquartile range 22–39). Thirty-four had Lennox–Gastaut syndrome, one Dravet syndrome, and two tuberous-sclerosis-associated epilepsy according to the demographic table.

This population is not representative of all adults who take CBD or cannabis products. Participants had severe, treatment-resistant developmental or epileptic disorders and typically received several other antiseizure medications. The median number of previous medications in the source table was four (interquartile range three to four). The clinicians made ordinary treatment decisions rather than assigning participants randomly to CBD or an otherwise identical placebo group. No untreated control cohort was followed concurrently.

Medical charts contained seizure diaries completed by patients or caregivers, treatment information, caregiver observations, and adverse-event reports. The investigators compared recorded total seizure burden during the six months before CBD with follow-up at six and twelve months. They did not reliably separate each seizure type for outcome analysis. That distinction is important when a person has several kinds of seizures: a change in a combined count need not mean the same proportional change for each seizure type or clinical circumstance.

What CBD exposure was documented—and what was not measured?

The study recorded prescribed oral CBD dosing in milligrams per kilogram of body weight per day. The reported median starting dose was 4.41 mg/kg/day (IQR 3.33–4.86); median maintenance dose was 10.45 mg/kg/day (IQR approximately 7.455–13.975); and the median maximum prescribed dose was 11 mg/kg/day (IQR 9–14.3). Those are administered or prescribed regimen descriptors, not chemically measured concentrations in a patient’s circulation.

The authors did not present a CBD, THC, or metabolite concentration from venous whole blood, serum, plasma, urine, oral fluid, or breath. They also did not estimate an individual concentration-time curve from timed samples. Thus, the paper cannot establish what blood CBD concentration corresponded to drowsiness, a caregiver observation, a seizure improvement, or an abnormal liver result. The absence of cannabinoid assays in this design should not be filled in using a generalized pharmacokinetic table or a purported CBD impairment cutoff.

For a separate explanation of why prescribed dose, specimen collection and measured concentration answer different questions, see therapeutic drug monitoring and specimen timing. That reference offers general method context; it does not add measurements to this particular study.

How many patients met the seizure-response thresholds?

The study reported that 24 of 37 (64.9%) reached at least a 25% seizure-frequency reduction at six months, declining to 22 of 37 (59.5%) at twelve months. A stricter threshold, at least a 50% reduction, was met by 18 of 37 (48.6%) at six months and 17 of 37 (45.9%) at twelve months. These are responder counts using recorded total seizure frequency, not a controlled estimate of drug-attributable efficacy. The authors also reported shorter seizure duration in two of 37 (5.4%) at six months and four of 37 (10.8%) at twelve months.

All percentages shown are source counts divided by 37, not estimates from a random sample of general CBD users. The records do not provide a patient-specific probability of response or a validated causal prediction rule. Outcome ascertainment depended on the quality of caregiver diaries and retrospective chart abstraction.

Did higher CBD maintenance doses predict better seizure response?

Not significantly in the reported comparisons. The authors compared median maintenance doses across seizure-response categories and reported Kruskal–Wallis p values of 0.138 and 0.36. The paper also reported no significant relationship between concomitant clobazam use and achieving the 50% response criterion (Fisher exact p = 0.645). These findings do not prove dose is irrelevant, that clobazam has no clinical interaction, or that equal doses produce equal patient outcomes.

The exposure categories were not randomized, groups were small, and clinicians could adjust treatment in response to effects. That creates confounding by clinical indication and tolerability: patients may be maintained at a lower dose precisely because of adverse symptoms or because physicians judged a lower regimen appropriate. A nonsignificant comparison does not exclude a dose relationship of a size the study was underpowered to detect.

What adverse events were actually reported?

At least one adverse event was recorded in 24 of 37 patients (64.9%). The authors tabulated drowsiness in 10 (27.0%), abnormal liver function in seven (18.9%), and irritability in four (10.8%). Gastrointestinal symptoms, insomnia and paradoxical worsening each appeared in two (5.4%). Increased appetite, decreased appetite and urinary retention each appeared in one patient (2.7%). Categories may overlap in the same person and should not be summed into an independent number of affected patients.

Six patients discontinued because of adverse events; separately, the study states seven patients stopped overall—one for ineffectiveness alone, one for adverse events alone, and five for both. These categories describe discontinuation reasons and should not be represented as seven independent drug toxicities. For an individual case, the event timing and all contemporaneous medical decisions would have to be reviewed.

Do seven abnormal liver results establish CBD-caused liver injury?

No. The study reports seven abnormal liver-function events, each below twice the local normal value. Six resolved without a treatment adjustment; one resolved after a reduction in concomitant valproic acid. The authors reported no correlation of the abnormal laboratory findings with CBD maintenance or maximum dose. They did not present an adjudicated causal liver-injury diagnosis for each case or an independently controlled estimate of how many abnormalities CBD itself caused.

A change in a liver laboratory test is evidence of a laboratory value changing; it is not automatically proof of hepatocellular necrosis, liver failure, or attribution to one particular drug. The relevant original values, chronology, other medical conditions, co-prescribed therapies and subsequent tests are critical to any specific causation question. The presence of valproic acid in the one resolved case is an additional reminder that combination treatment complicates single-agent attribution.

Nor should this study be used to infer a therapeutic or toxic blood concentration: no CBD concentration was determined and no direct matrix-specific exposure–response calibration was presented. General background on the difference between a sampled specimen and an inference is available in the biological-matrices guide.

Can patients be reported as more alert while others become drowsy?

Yes, because the paper measured the two outcomes differently and at the cohort rather than universal individual level. Caregivers said 11 of 37 patients improved in alertness, social interaction, behavior or everyday functioning after treatment began. Those reports were not generated from a standardized neuropsychological instrument, validated alertness scale, reaction-time protocol, or a blinded independent performance assessment. By contrast, drowsiness was an adverse-event entry in clinical records for 10 people.

The study did not present evidence that the 11 people described as improved and the 10 with drowsiness were disjoint sets. Even if they were, an improved caregiver impression in some and sedation in others would be entirely compatible with heterogeneous experiences during complex therapy. An individual might also experience fewer seizures but still have medication-related sleepiness. These are not mutually exclusive outcomes.

The investigators found no statistically significant difference in maintenance dose (p = 0.699), seizure response (p = 0.67), or concomitant medicine reduction (p = 0.43) between patients with and without caregiver-reported cognitive-behavioral improvement. Those negative tests do not validate the caregiver rating as a cognitive test or show that the behavior change was independent of treatment changes.

Why does a caregiver’s observation differ from a measured functional task?

A caregiver observation can be clinically important: it captures experiences at home over longer periods. But it is a different form of evidence from a timed reaction task, standardized neuropsychological score, laboratory vigilance assessment, or actual operation of a motor vehicle. The source did not record participant response times, divided-attention failures, simulator lane deviations, or on-road performance. Therefore it did not establish a quantitative psychomotor impairment rate.

In evidence interpretation, a description such as “more interactive” can reflect improvement in seizure burden, better sleep, adjustment of co-medications, contextual changes, subjective expectation, or several of these simultaneously. Without a blinded control and predefined validated outcome, one cannot assign that change uniquely to CBD. Similarly, “drowsy” in a chart is an important symptom, not a measured blood level or a statutory impairment finding.

What did the study show about changing other antiseizure medicines?

The median number of concomitant antiseizure medicines decreased from four (IQR three to four) before CBD initiation to three (IQR three to four) afterward, with a reported paired Wilcoxon p = 0.022. A median difference is not a count of every prescription that was stopped, and the IQR overlapping between time points does not negate the reported paired statistical test. The study did not randomly assign which other drug should be decreased.

These contemporaneous medication changes matter particularly for assessing sedation or cognition. A patient’s apparent alertness might improve if a sedating co-medication is reduced, while seizure frequency and other symptoms also change. Conversely, adding CBD may have effects independent of other drugs that an observational chart review cannot isolate. The paper supplies useful descriptive context but not an instrument to allocate fractions of benefit or harm among several medications.

Does the high continuation rate mean CBD was harmless?

No. Thirty of 37 patients (81.1%) were retained at twelve months in the authors’ classification. Continuation is influenced by perceived seizure benefit, caregiver preference, access to treatment, clinician judgment, competing therapies and tolerability. It is not equivalent to “no adverse effects,” especially when 24 patients had reported adverse events and six stopped because of adverse events with or without ineffectiveness.

Follow-up length varied considerably: the Results describe a median follow-up of 28 months for all patients with IQR 18.5–55, and 34 months among those continuing CBD with IQR 22.75–58.25. Some discontinued before a year, even though the eligibility language refers to at least twelve months of follow-up data. These design nuances caution against converting retention into a time-to-event safety curve that the original paper did not estimate.

Can the findings be applied to over-the-counter CBD or smoked cannabis?

Not directly. The study concerns prescribed, highly purified cannabidiol as adjunctive epilepsy therapy in adults with specific severe syndromes. Retail CBD products may differ in formulation, actual delivered mass, impurities, labeling and co-exposures. Smoked cannabis and cannabis containing substantial THC involve different compounds and exposure pathways. The study did not evaluate recreational products, inhaled THC, oral-food dosing, or blood THC thresholds.

Even the mg/kg/day dose in this paper cannot be converted into a measured human blood cannabinoid concentration without additional individual pharmacokinetic evidence. Those concepts are developed separately in the pharmacokinetics guide. The most important practical distinction is between a documented prescription, documented adherence, analytically measured exposure and an actual clinical outcome.

What would be necessary for a specific toxicology or injury assessment?

An evidence-bound assessment would start with the precise medication and label, administration timeline, dose changes, concomitant medications, baseline seizure and cognitive status, contemporaneous medical records, and documented onset and resolution of symptoms. For a liver-test dispute, it would additionally require actual laboratory results, reference ranges, serial trends and alternative potential causes. For a claimed driving or workplace problem, actual functional observations and task evidence would be needed; the retrospective seizure cohort provides no validated threshold for such a conclusion.

Specimen chemistry matters as well: a laboratory report for CBD in whole blood cannot be substituted for plasma, a metabolite screen, or prescribed milligrams per day. This particular primary paper reported none of those cannabinoid assay results. It is therefore inappropriate to claim that it established a specific circulating CBD concentration associated with seizure benefit, drowsiness, or abnormal liver tests. The same caution applies in reverse: absence of a measured concentration is not proof of absence of CBD exposure.

What are the most consequential study limitations?

The study had only 37 patients, nearly all with Lennox–Gastaut syndrome. There was no randomized placebo or untreated comparison, no systematic drug concentration monitoring, no standardized objective cognition or driving test, and no reliable seizure-type-specific outcome analysis. Medication changes during follow-up were part of routine care and could affect efficacy and adverse-event interpretation. Caregiver impressions were subjective and potentially affected by expectations; seizure diaries can miss or misclassify events.

The maintenance-dose and adverse-event relationship is especially vulnerable to reverse causation because a patient experiencing an adverse event may have their CBD dose lowered. The paper’s relatively long but unequal periods of follow-up also complicate naive comparisons between retained patients and those who stopped. Statistically nonsignificant results in this modest sample should not be presented as proof that an interaction or dose effect is absent.

The original manuscript notes industry-funded medical writing support and author relationships with pharmaceutical companies. These are relevant source disclosures, not evidence that its numerical results are incorrect. The appropriate scientific response is to interpret the design, denominators, methods and reported uncertainty as written rather than either assuming bias or overlooking it.

Frequently asked questions

Did almost half the adults improve their seizure counts on CBD?

Seventeen of 37 (45.9%) had at least a 50% reduction in recorded seizure frequency at twelve months. That describes a threshold reached during treatment; without a concurrent control group it cannot quantify how much CBD alone caused the improvement.

Did 27% experience drowsiness?

Drowsiness was recorded in 10 of 37 (27.0%) patients. No standardized alertness test or on-road driving task determined the functional consequence for those ten people.

Does the liver-test finding mean 19% developed liver failure?

No. Seven of 37 (18.9%) had reported abnormal liver tests, all less than twice the source’s normal threshold. The paper did not diagnose liver failure or establish CBD-only causation.

Was a higher CBD dose associated with more adverse events?

The opposite cross-sectional pattern was reported: patients with adverse events had a lower achieved median maintenance dose (7 versus 12.5 mg/kg/day, p < 0.001). That does not prove higher doses were protective because dosing may have been modified after symptoms.

Can caregiver-reported better cognition offset drowsiness?

The two observations have different measurement methods. The caregiver improvement reports in 11 patients were not validated objective cognitive results and were not established as a causal counterweight to the ten recorded drowsiness events.

Did investigators test actual blood CBD, THC or impairment thresholds?

No. The source reports medication doses and clinical chart outcomes. It provides no measured cannabinoid concentration in blood, plasma, urine or oral fluid, and no driving or psychomotor cutoff.

Can the results establish causation in a particular patient’s medication case?

No. A patient-specific assessment needs the full medication and laboratory chronology, baseline condition, adverse-event documentation and competing explanations. Group-level uncontrolled observations cannot substitute for that evidence.

Primary source and scope

Sánchez-Gamino S, Chamorro Muñoz MI, Serrano-Castro PJ, et al. Effectiveness and safety of cannabidiol in adult patients with epilepsy: A multicenter, retrospective study. Epilepsia Open (first published September 21, 2026). Original primary research: doi:10.1002/epi4.70331.

This article explains what the specified observational study measured and what its evidence cannot resolve. It does not offer individual medical advice, establish clinical medication safety, interpret an actual person’s impairment, or express an opinion on a particular legal case.