Cannabidiol and High-CBD Cannabis Extracts in Pediatric Autism Spectrum Disorder

Key Findings

• A 2026 systematic review and meta-analysis of 54 randomized trials reported a reduction in autistic traits with cannabinoids (SMD -0.36, 95% CI -0.66 to -0.07), one of only four indications showing any signal in a review that found nothing for anxiety, PTSD, psychotic disorders, or anorexia; 24 of the 54 trials carried high risk of bias and certainty was graded low for most outcomes [1].
• The largest placebo-controlled trial in this population (n = 150) split across its two registered primary endpoints: no difference from placebo on a parent-rated behavior questionnaire, and much or very much improved disruptive behavior in 49% on whole-plant extract versus 21% on placebo (p = 0.005) on the clinician-rated co-primary, an endpoint specified in advance rather than found afterward [2].
• The purified cannabidiol literature splits by design: two double-blind trials were null on their primary outcomes, one reporting a strong placebo effect as its central finding, while an open-label trial in 23 children without intellectual disability reported 44% response and a large effect on social responsiveness (d = 1.36) with no control arm to separate it from placebo [3, 4, 5].
• A 20:1 extract titrated to 10 mg/kg/day of cannabidiol delivers roughly 0.5 mg/kg/day of THC, about 15 mg daily for a 30 kg child, and no trial in this literature ran long enough to characterize chronic THC exposure during neurodevelopment [2, 8].
• Somnolence and decreased appetite occurred about three times more often on active treatment than placebo in the largest trial (28% and 25% versus 8% and 15%), giving caregivers a plausible route to infer allocation in trials whose outcomes they rate [2].
• Risperidone and aripiprazole may reduce irritability relative to placebo (mean difference -7.89 and -6.26, low certainty), while the randomized-trial estimates for weight gain and extrapyramidal effects rest on very low certainty evidence limited by risk of bias and imprecision [6].
• Reported response rates of 44 to 62 percent imply a substantial responder subgroup, but no trial has identified a validated predictor of who belongs to it, so a clinician cannot determine in advance which child is likely to benefit [5].

---

Introduction

Autism spectrum disorder has no approved pharmacologic treatment for its core features. Risperidone and aripiprazole carry indications for irritability associated with autism, not for social communication differences or restricted and repetitive behavior, and both carry metabolic and extrapyramidal effects that make long-term pediatric use a recurring concern. An agent that moved social responsiveness without those liabilities would be the first treatment to touch the core features of autism at all, which is why investigators keep returning to a literature that has so far refused to settle, and why a great deal of use has run ahead of the evidence. The biological case for looking is thinner than that use suggests without being absent: animal models show endocannabinoid signaling abnormalities, the rationale the largest trial cites for its own existence [2], and the endocannabinoid system modulates the excitatory and inhibitory balance implicated across neurodevelopmental conditions [11], the same receptor biology that underlies interest in cannabinoids for dementia. Neither observation predicts a clinical effect in autism. Together they establish that the hypothesis is not arbitrary, a lower bar than it is usually made to carry.

Over the past five years the pediatric literature has grown from case series into a small body of randomized trials and, as of 2026, into pooled analysis. The results are usually called promising but mixed, which is accurate and unhelpful, because it obscures the more interesting point: the trials do not disagree at random. Whether a trial had a placebo arm, which instrument it registered as primary, and who did the rating predict the reported result better than whether the bottle held a whole-plant extract or a purified molecule.

This article organizes the evidence around those design features, taking control, endpoint selection, blinding, product composition, and population definition in turn, with a section on safety and exposure alongside them and a final one asking whether the extract versus isolate question is one the current evidence can answer at all. Reading the literature this way has a payoff: design problems have design solutions, and the reasons these trials disagree are largely fixable at modest cost. The scope is pediatric and adolescent autism, drawing on randomized trials, systematic reviews, and meta-analyses through mid-2026. Cannabidivarin, which has its own pediatric development program, falls outside it.

---

Why the Comparator Matters

The case for investigating cannabinoids in autism usually opens with the shortcomings of atypical antipsychotics. That framing deserves scrutiny, because the comparator turns out to be less firmly characterized than the argument assumes.

A 2025 Cochrane network meta-analysis of 17 randomized trials, 16 in children, found risperidone and aripiprazole may reduce irritability against placebo short term, mean differences -7.89 (95% CI -9.37 to -6.42) and -6.26 (95% CI -7.62 to -4.91), both low certainty [6]. Lurasidone probably produces little or no difference.

The harms data are weaker than the efficacy data. Weight gain above predefined thresholds carried a risk ratio of 2.40 (95% CI 1.25 to 4.60) across 434 participants, and extrapyramidal effects 2.36 (95% CI 1.22 to 4.59) across 511. The review graded both very low certainty, citing risk of bias and serious imprecision [6]. These are short-term trial data, so they speak to the first weeks and months of exposure, not the multi-year use that generates most clinical concern. The certainty grade describes what randomized trials have established. It is not a verdict on the metabolic risks documented in large observational cohorts and pharmacovigilance reporting, which is where most clinical knowledge of long-term antipsychotic harm comes from.

This matters for how cannabinoid trials should be read. If the argument is comparable benefit at lower risk, both halves of that comparison rest on low or very low certainty trial evidence. A field cannot establish superiority against a comparator whose trial-level harm profile is itself poorly quantified.

The Control Problem

The single largest source of divergence in this literature is whether a trial had a placebo arm.

An open-label phase 2 trial published in 2026 enrolled 23 children and adolescents with autism, fluent verbal language, and IQ at or above 80, using a Bayesian optimal interval design to test pharmaceutical-grade cannabidiol at 3, 6, or 9 mg/kg/day [5]. Ten of 23, 44%, met the responder definition on an individually designated target domain, rising to 62% at the highest dose. The Social Responsiveness Scale improved with an effect size of 1.36, shifting the group mean from severe into moderate.

The trial had no control arm, and its authors present it as preliminary. Set against the controlled data, the reason for caution becomes concrete instead of formal. In the double-blind crossover of purified cannabidiol at up to 20 mg/kg/day in autistic boys aged 7 to 14 with severe problem behaviors, active and placebo phases both produced improvement on repetitive behavior and child behavior checklists, with no significant difference between them [3]. Observational assessment improved on placebo only, and that improvement disappeared once concomitant medications entered the analysis. The authors identified a strong placebo effect as a principal finding and drew the explicit lesson that placebo control is indispensable here.

A retrospective analysis of the 88 placebo recipients in the largest controlled trial gives the effect a partial mechanism. Placebo response was associated with the child's comprehension of the treatment's purpose (p = 0.037), with a trend for children whose symptoms had recently worsened before enrollment (p = 0.053) [7]. Parental expectations, prior experience, locus of control, and adherence showed no association, cutting against the simplest placebo-by-proxy hypothesis while confirming the response is substantial and structured.

The implication for reading this literature is direct. An uncontrolled response rate of 44% and a controlled null are the expected outputs of two designs applied to a condition with a large placebo response and fluctuating symptoms.

The Endpoint Problem

Within a single trial, the reported answer often depends on which instrument was designated primary.

The proof-of-concept trial by Aran and colleagues randomized 150 participants aged 5 to 21 to a whole-plant 20:1 cannabidiol to THC extract, a purified preparation at the same ratio, or placebo, for 12 weeks with crossover [2]. The registered primary outcome, a parent-rated home behavior questionnaire, showed no difference between groups. The co-primary, Clinical Global Impression-Improvement with disruptive behavior anchors, showed much or very much improved status in 49% on extract versus 21% on placebo (p = 0.005). Both endpoints were registered in advance, so this is a trial that split across its primaries, not one that missed and recovered a secondary. The Social Responsiveness Scale, a secondary, improved by a median 14.9 points versus 3.6 (p = 0.009). Parenting stress showed nothing. The authors call the efficacy evidence mixed and insufficient.

A 2025 crossover trial in 29 autistic children shows the same pattern from the opposite direction [4]. Cannabidiol oil with terpenes at 10 mg/kg/day produced no significant effect on the Social Responsiveness Scale, its registered primary outcome (p = 0.125), while secondaries improved: social relating (p = 0.024), anxiety (p = 0.002), and parental stress (p = 0.044). Adaptive functioning showed no change.

Anxiety shows the difficulty from another angle. It was the strongest secondary in the crossover trial at p = 0.002, stronger than the social outcome that trial was built around [4], while the same meta-analysis found no anxiety effect across any indication [1]. A domain-specific effect in autism is possible, and so is a secondary outcome behaving as secondary outcomes do.

The Social Responsiveness Scale carries the positive secondary finding in the first trial and the null primary finding in the second. What differs is its designated role, and with it the multiplicity correction, the powering assumption, and the interpretive weight the result can bear.

This is a familiar problem in psychiatry, where constructs are measured by questionnaire rather than assay, and the number of scales aggravates it. A companion analysis of the largest trial found cannabinoid treatment no better than placebo on any sleep domain [12]. A study measuring seven behavioral domains has many chances to find something, so registered primary outcomes deserve heavy weight and secondary findings stay hypothesis-generating.

The Blinding Problem

Nearly every outcome in pediatric autism trials is rated by a parent or caregiver, and nearly every cannabinoid preparation produces noticeable side effects. Those two facts interact.

In the 150-participant trial, somnolence was reported by 28% on whole-plant extract, 23% on purified cannabinoids, and 8% on placebo; decreased appetite by 25%, 21%, and 15% [2]. Sedation and appetite change are exactly the effects a parent notices and interprets. The 2026 open-label trial, with no blinding to preserve, reported increased sleep duration in 39%, dream activity in 35%, salivation in 30%, and sedation in 26% [5]. A caregiver observing that cluster in a blinded trial has a reasonable basis for guessing allocation, and one who guesses is no longer blinded.

None of the trials discussed here reports a formal test of blinding integrity, a notable omission given that the primary outcomes depend on it. Asking raters to guess their assignment at study end is inexpensive and would materially change how these results can be read.

The problem compounds in crossover designs, which both purified-cannabidiol trials used [3, 4]. A crossover improves statistical efficiency in a small sample while giving caregivers a within-family comparison that makes allocation easier to infer. The washouts, four weeks and eight, address pharmacologic carryover but not the memory of which period felt different.

Blinded clinical impression offers a partial check. In the purified-cannabidiol crossover, roughly two thirds of participants were judged by blinded clinicians to have improved on cannabidiol, a third showing no change or improving on placebo, even as the quantitative primaries came out null [3]. That divergence is a reasonable prompt to ask whether the instruments capture what clinicians observe. Imperfect blinding would produce it too, as would accurate instruments paired with mistaken impressions.

The Product Problem

Much of this literature describes its intervention by ratio. "20:1 CBD-rich extract" appears as though it named a compound, when it names the proportion of two constituents among hundreds.

The 2026 systematic review of 12 pediatric studies found most used full-spectrum extracts at high cannabidiol to THC ratios, commonly 20:1, titrated to roughly 10 mg/kg/day [8]. Nothing in that description fixes terpene content, minor cannabinoid content, or batch consistency, a problem examined in our review of cannabis product standardization and quality. Two trials using 20:1 extracts from different producers may have tested materially different products under one label.

Dosing compounds the problem, and it runs backwards. Doses span 3 to 20 mg/kg/day, nearly sevenfold [3, 5]. The open-label trial found its highest response at its highest dose, 9 mg/kg/day, with dose correlating with global improvement [5], while the trial dosing to 20 mg/kg/day found nothing [3]. A compound acting dose-dependently should have performed best where the dose was highest.

Trials of "a Mediterranean diet" long produced inconsistent results because the label described a pattern, not a defined exposure. Nutrition research answered that by specifying components and quantities. The 20:1 extract sits roughly where the diet label sat before that shift.

Safety and Exposure

The ratio fixes THC exposure as well as cannabidiol exposure, which trial reports rarely state directly. A 20:1 preparation titrated to 10 mg/kg/day of cannabidiol delivers 0.5 mg/kg/day of THC, roughly 15 mg daily for a 30 kg child, continuously for 12 weeks [2, 8]. Chronic THC exposure during neurodevelopment is the safety question this literature is least equipped to answer, since no trial ran long enough to address it.

Against that, the short-term tolerability record is consistent. No treatment-related serious adverse events occurred in the 150-participant trial [2], the purified crossover reported an acceptable safety profile [3], and the open-label trial found vital signs, liver function tests, and blood counts unaffected [5]. Across 54 randomized trials, cannabinoids raised the odds of any adverse event but not of serious events or study withdrawal [1]. The gap in the safety picture is duration, not severity.

Pharmacokinetics are largely absent. The 150-participant trial names their lack as a limitation alongside its wide age and functional range [2]. Where levels were measured they complicate interpretation: plasma cannabidiol did not correlate with response in the open-label trial [5], and the purified crossover found concomitant behavioral medications may reduce cannabidiol blood levels [3]. Where polypharmacy is common and the study drug interacts with co-medications, nominal dose and delivered exposure diverge.

Those interactions are characterized in the epilepsy literature. Cannabidiol interacts bidirectionally with clobazam, raising 7-hydroxy-cannabidiol and norclobazam concentrations with attendant sedation, and elevates liver transaminases particularly alongside valproate [9]. Neither is exotic where antiseizure medication is common.

The Population Problem

Response rates of 44 to 62 percent across these trials imply that a substantial subgroup improves [5]. No validated predictor identifies which children those are, so a clinician asked who is likely to benefit has no evidence-based answer.

The candidate moderators are visible but untested, and the trials differ sharply in whom they enrolled. The open-label trial required fluent verbal language and IQ at or above 80 [5]; the purified-cannabidiol crossover enrolled boys aged 7 to 14 with severe problem behaviors [3]; the 150-participant trial spanned ages 5 to 21 with a wide range of functional levels, flagged by its authors as a limitation [2]. These populations overlap only partially, and comparing across them compares different clinical questions.

It is tempting to read the contrast between the open-label trial in children without intellectual disability and the null crossover in children with higher support needs as evidence that cognitive profile moderates response. That reading is unsupported: the two differ in control design as well as population, and the control difference alone predicts the divergence.

Sex is a second gap. One trial enrolled boys exclusively [3], another was 18 of 29 male [4], and the largest reports sex without powering subgroup analysis [2]. Since girls are diagnosed later and differently, a literature under-enrolling them yields findings of unknown generalizability.

Baseline severity is a third. The retrospective placebo analysis found a trend toward placebo response among children whose symptoms had worsened shortly before enrollment [7], pointing to regression to the mean operating on criteria that select for current distress. Trials recruiting families at their most difficult moment build in a tendency toward apparent improvement regardless of assignment.

The emerging biomarker literature offers a route forward. A 2026 secondary analysis of electroencephalography from the purified-cannabidiol crossover found aperiodic EEG measures varying with cannabidiol metabolite levels in blood, and those levels associating positively with receptive vocabulary, nonverbal intelligence, and visuomotor coordination [10]. That is a small exploratory finding in a 24-participant subsample, and the first candidate stratification variable here grounded in something other than symptom questionnaires.

Does the Extract Versus Isolate Distinction Survive?

The question most often asked here is whether whole-plant extracts outperform purified cannabinoids, and whether the entourage effect explains the difference. In autism, the evidence supports a narrower answer than either advocates or skeptics offer.

One trial was positioned to test it. The 150-participant study included a whole-plant extract at 20:1, a purified preparation of the same two molecules at the same ratio, and placebo [2]. That design isolates the botanical matrix, the terpenes and minor cannabinoids present in the extract and absent from the purified arm, as close to a controlled test of the entourage hypothesis as this field has produced.

It was not designed or powered to answer it. The registered primary comparison was extract against placebo, no powered head-to-head appears, and the adverse event profiles of the two active arms were similar [2]. The trial establishes tolerability for both and efficacy for neither with confidence, which is what its authors concluded.

Two claims need separating. Whether the botanical matrix contributes anything in autism is open, and the trial best placed to address it was underpowered for that contrast. Whether full-spectrum extracts have been shown superior to purified cannabinoids is not open, and the answer is no. A 2026 review of 17 studies across autism, Fragile X, and Rett syndrome reports exploratory analyses suggesting differential effects of cannabidiol versus cannabidiol plus THC, while concluding current evidence cannot support routine clinical use [11].

A chemical explanation runs alongside the design one, and the two cannot be separated. Both active arms in the 150-participant trial contained THC, while the three THC-free studies were null on their primaries or uncontrolled [3, 4, 5]. The only trial producing a positive signal on any endpoint is therefore also the only one containing THC, and simultaneously the largest and best powered. Whether the signal tracks THC, the botanical matrix, or sample size is not recoverable from the published data.

Neither preparation has controlled evidence of efficacy on a registered primary outcome in this population, and the comparison between them remains untested. Product marketing on both sides runs well ahead of the trial evidence.

Future Research Directions

The design problems above suggest specific remedies, several of which are inexpensive.

Register blinding integrity as a reported outcome. Asking caregivers and raters to guess allocation at trial end costs almost nothing and lets readers calibrate parent-rated outcomes against functional unblinding. Its absence here is the most easily corrected deficiency identified.

Adopt a core outcome set. A single scale serves as a positive secondary in one trial and a null primary in another, which invites selective emphasis. A consensus set developed with autistic people and their families would constrain multiplicity and make trials comparable.

Characterize the product fully and report it. Trials should publish full cannabinoid and terpene profiles with batch-level certificates of analysis, not ratios, and state absolute THC dose in mg/kg/day. Without this, replication is impossible in principle.

Design the head-to-head comparison properly. A trial powered for the extract versus purified contrast at matched cannabinoid content, not against placebo alone, would answer the entourage question here. The 150-participant trial shows the arms are feasible; it lacked the sample size and a prespecified plan for that comparison. A THC-free extract arm would separate matrix effects from THC.

Collect pharmacokinetics routinely, and record concomitant medications as exposure variables. Behavioral medications may lower cannabidiol levels [3], and plasma levels did not track response in an open-label setting [5], leaving the dose-exposure-effect relationship unmapped.

Pursue stratification biomarkers. The aperiodic EEG signal associated with cannabidiol metabolite levels [10] should be tested prospectively instead of analyzed post hoc. An enrichment design, using a candidate marker to identify likely responders before randomization, would address the responder-prediction gap directly.

Extend follow-up. These trials ran 6 to 12 weeks per condition. Families considering cannabinoids for a lifelong condition need data measured in years, particularly since the comparator concern motivating this line of inquiry is long-term risk, and since chronic pediatric THC exposure is itself uncharacterized.

Challenges and Ethical Considerations

Research here carries obligations beyond methodological rigor.

Consent and assent are complicated when participants have communication differences and intellectual disability. That placebo response tracked the child's comprehension of the treatment purpose [7] is both a methodological and an ethical result: comprehension varied, so some participants understood little about their own participation.

Outcome selection embeds a value judgment. Instruments measuring social responsiveness treat movement toward non-autistic social behavior as improvement, a framing many autistic people reject. Measures of disruptive behavior and caregiver stress partly capture burden on families, a legitimate target and a different one from benefit to the child. Trials reporting reduced parental stress alongside null child-centered outcomes [4] should be described precisely, not summarized as showing benefit.

The comparator asymmetry noted earlier has an ethical dimension, and it invites misreading. The observational record on antipsychotic metabolic harm in youth is substantial, so thin randomized-trial evidence should not be taken to make that harm speculative, and nothing here is a reason to stop a prescribed medication. The narrower point is that trial-level evidence is limited on both sides of the comparison, which argues for caution about cannabinoids [6].

Access runs ahead of evidence. High-CBD products are widely available, and families use them with or without clinical supervision. The 2026 systematic reviews converge on cautious, supervised, adjunctive use at most [8, 11]. The reasonable posture is to ask about use, know the clobazam and valproate interactions [9], and be candid that trial evidence does not establish efficacy.

Publication and interpretation bias deserve attention in a field where commercial interests attach to particular answers. Both the botanical extract industry and pharmaceutical developers of purified cannabinoids have stakes in how these comparisons are reported.

Frequently Asked Questions

Does cannabidiol improve core autism symptoms? Not on the evidence available. Two double-blind trials of purified cannabidiol were null on their primary outcomes [3, 4], and the largest extract trial was null on its registered primary while positive on a clinician-rated co-primary [2]. The pooled 2026 estimate for autistic traits was small and graded low certainty [1].

Is the 44% response rate in the 2026 open-label trial meaningful? It describes what happened in that trial but cannot be attributed to cannabidiol, since there was no control arm [5]. Placebo response in the controlled trials was large [3, 7].

Are CBD-rich extracts better than purified cannabidiol for autism? No randomized trial has shown that. The one trial containing both arms at a matched 20:1 ratio was not powered for that comparison [2]. The question is open; the claim of superiority is unsupported.

Is cannabidiol safe in children with autism? Trials reported acceptable short-term tolerability with mostly mild adverse events, commonly somnolence and appetite change [2, 5]. In the 2026 meta-analysis pooling 54 randomized trials across mental health indications, cannabinoids raised the odds of any adverse event (OR 1.75, number needed to harm 7) without raising serious adverse events [1]. Cannabidiol interacts with clobazam and elevates transaminases alongside valproate [9]. A 20:1 extract also delivers meaningful daily THC alongside the cannabidiol. Long-term pediatric data are absent.

What should a clinician tell a family already using a CBD product? Ask what is being used and at what dose, review concomitant medications for interactions [9], and be clear that current evidence supports neither efficacy claims nor a particular product choice.

The Takeaway

The pediatric autism cannabinoid literature is small, and it is more coherent than its reputation for mixed results suggests. Trials with placebo arms have been largely null on the parent-rated questionnaires they registered as primary, with the notable exception of a clinician-rated co-primary met at p = 0.005 in the largest of them. Trials without placebo arms have reported substantial response. Positive findings cluster in clinician-rated global impressions, in a population with a documented, structured placebo response and caregiver-rated endpoints vulnerable to functional unblinding.

That pattern does not establish that cannabinoids are ineffective in autism. It establishes that no trial yet run was built to detect an effect if one is there, and it identifies with unusual precision what a trial that could would look like: adequately powered, blinding verified, product fully characterized including absolute THC dose, primary outcome agreed in advance, and follow-up measured in years. None of that is exotic and most of it is inexpensive. A field that knows which study to run next is in better condition than one that has run the right studies and found nothing.

For the extract versus isolate question specifically, the evidence supports no preference. Families and clinicians are being offered confident answers by parties on both sides of a comparison that has never been properly run.

---

References

• [1]: Wilson J, Dobson O, Langcake A, Mishra P, Bryant Z, Leung J, Dawson D, Graham M, Teesson M, Freeman TP, Hall W, Chan GCK, Stockings E. The efficacy and safety of cannabinoids for the treatment of mental disorders and substance use disorders: a systematic review and meta-analysis. Lancet Psychiatry. 2026;13(4):304-315. https://pubmed.ncbi.nlm.nih.gov/41856154/
• [2]: Aran A, Harel M, Cassuto H, Polyansky L, Schnapp A, Wattad N, Shmueli D, Golan D, Castellanos FX. Cannabinoid treatment for autism: a proof-of-concept randomized trial. Mol Autism. 2021;12(1):6. https://pubmed.ncbi.nlm.nih.gov/33536055/
• [3]: Trauner D, Umlauf A, Grelotti DJ, Fitzgerald R, Hannawi A, Marcotte TD, Knight C, Smith L, Paez G, Crowhurst J, Brown A, Suhandynata RT, Lund K, Menlyadiev M, Grant I. Cannabidiol (CBD) Treatment for Severe Problem Behaviors in Autistic Boys: A Randomized Clinical Trial. J Autism Dev Disord. 2025. https://pubmed.ncbi.nlm.nih.gov/40410546/
• [4]: Parrella NF, Hill AT, Enticott PG, Botha T, Catchlove S, Downey L, Ford TC. Effects of Cannabidiol on Social Relating, Anxiety, and Parental Stress in Autistic Children: A Randomized Controlled Crossover Trial. Autism Res. 2025;19(2):e70159. https://pubmed.ncbi.nlm.nih.gov/41452412/
• [5]: Lawson J, Robinson L, Conlon GR, Shalev RA, Cervantes PE, Yoncheva Y, Hirsch GS, Troxel AB, Friedman D, Devinsky O, Castellanos FX. A Phase-2 Open-Label Trial of Cannabidiol to Treat Core and Associated Symptoms of Autism in Children and Adolescents Without Intellectual Disability. J Child Adolesc Psychopharmacol. 2026;36(6):350-358. https://pubmed.ncbi.nlm.nih.gov/42204954/
• [6]: Meza N, Franco JV, Sguassero Y, Núñez V, Escobar Liquitay CM, Rees R, Williams K, Rojas V, Rojas F, Pringsheim T, Madrid E. Atypical antipsychotics for autism spectrum disorder: a network meta-analysis. Cochrane Database Syst Rev. 2025;5(5):CD014965. https://pubmed.ncbi.nlm.nih.gov/40396498/
• [7]: Aran A, Harel M, Ovadia A, Shalgy S, Cayam-Rand D. Mediators of Placebo Response to Cannabinoid Treatment in Children with Autism Spectrum Disorder. J Clin Med. 2023;12(9):3098. https://pubmed.ncbi.nlm.nih.gov/37176538/
• [8]: Vasconcelos QDJS, de Freitas RF, Freitas MC, de Andrade IP, Brandão CB, Nascimento CE, Neves KR, Magalhães PS, Filho VB, Oliveira GL, de Moura MAM, Aragão GF. Cannabinoid-based interventions for behavioral outcomes in children and adolescents with autism spectrum disorder: A systematic review of safety and effectiveness. Prog Neuropsychopharmacol Biol Psychiatry. 2026;146:111697. https://pubmed.ncbi.nlm.nih.gov/41974303/
• [9]: Strzelczyk A, Schubert-Bast S. A Practical Guide to the Treatment of Dravet Syndrome with Anti-Seizure Medication. CNS Drugs. 2022;36(3):217-237. https://pubmed.ncbi.nlm.nih.gov/35156171/
• [10]: Cazares C, Hutton A, Paez G, Trauner D, Voytek B. Cannabidiol blood metabolite levels after cannabidiol treatment are associated with broadband EEG changes and improvements in visuomotor and non-verbal cognitive abilities in boys with autism requiring higher levels of support. Transl Psychiatry. 2026;16(1). https://pubmed.ncbi.nlm.nih.gov/41611664/
• [11]: Boeri S, Bosisio L, Prato G, Martinez Popple M, De Grandis E, Striano P, Benke TA, Nobili L. Efficacy and Safety of Cannabinoid-Based Products in Children and Adolescents with Autism Spectrum Disorder, Fragile X Syndrome and Rett Syndrome: A Systematic Review. J Child Adolesc Psychopharmacol. 2026;36(7):393-404. https://pubmed.ncbi.nlm.nih.gov/42339654/
• [12]: Schnapp A, Harel M, Cayam-Rand D, Cassuto H, Polyansky L, Aran A. A Placebo-Controlled Trial of Cannabinoid Treatment for Disruptive Behavior in Children and Adolescents with Autism Spectrum Disorder: Effects on Sleep Parameters as Measured by the CSHQ. Biomedicines. 2022;10(7):1685. https://pubmed.ncbi.nlm.nih.gov/35884990/