A meta-analysis of 149 randomized trials found behavioral interventions far outpaced drugs on core autism symptoms, but study design inflated the largest reported gains.
Key Takeaways
- The comparison has long been murky: Clinicians, payers and families have lacked a clear head-to-head read on whether medications or behavioral approaches do more for the core features of autism. A new Nature Mental Health meta-analysis pooling 149 randomized trials sets out to close that gap.
- The timing raises the stakes: The findings arrive as federal officials reshape the national autism research agenda and as Medicaid programs and the HHS Office of Inspector General sharpen scrutiny of autism services, including ABA billing. What works for core symptoms now carries direct reimbursement and policy weight.
- The margin is wide: Nonpharmacological interventions showed a large pooled effect on core symptoms (Hedges’ g = 0.70), more than triple the small effect measured for drugs and supplements (g = 0.20). The difference was statistically significant, and medications were never designed to treat autism’s core social and behavioral features.
- Design drives the numbers: Larger reported effects clustered in smaller, shorter, older and non-Western studies and in trials relying on unblinded clinician ratings. The authors call for more rigorous, well-controlled trials before the size of the behavioral advantage can be taken at face value.
A new meta-analysis offers one of the broadest head-to-head looks yet at how autism treatments compare, and its headline is likely to resonate across the behavioral health sector: behavioral and psychosocial interventions substantially outperformed medications and dietary supplements at improving the core symptoms of autism spectrum disorder. Published May 25, 2026 in Nature Mental Health, the analysis was led by researchers at Shanghai Jiao Tong University and Indiana University Bloomington, who pooled 149 randomized controlled trials enrolling 9,011 participants.
The same paper, however, complicates any clean victory lap. Its authors report that the largest treatment effects tended to surface in the weakest study designs, a pattern that should give pause to anyone, on either side of the pharmacological divide, inclined to treat the top-line numbers as settled fact.
For a field built largely on behavioral services, the context matters. The U.S. Centers for Disease Control and Prevention now estimates that 1 in 31 eight-year-olds were identified with autism as of its 2022 surveillance, up from 1 in 36 two years earlier, and demand for services continues to climb after a decade of private-equity expansion across autism therapy. At the same time, payers and regulators are pressing harder on outcomes and integrity: the HHS Office of Inspector General is auditing Medicaid ABA billing and pursuing recoupments, and federal health leadership has reoriented the national autism agenda. Against that backdrop, a rigorous comparison of what actually moves core symptoms is more than an academic exercise.
How the Autism Treatment Meta-Analysis Was Conducted
The team searched PubMed, Embase, PsycINFO and the Cochrane databases through April 2025, and prospectively registered the protocol. Eligible studies were randomized trials, with active or inactive controls, that measured social communication, restricted or repetitive behaviors, or overall symptom severity. Effects were standardized as Hedges’ g and pooled using random-effects models in R; the data and analysis code were posted publicly.
Of the 149 trials, 69 (enrolling 2,889 participants) tested nonpharmacological approaches, while the remainder (enrolling 6,122 participants) tested medications or dietary supplements. One important nuance for sector readers: the nonpharmacological category is broad. It spans behavioral, developmental and psychosocial programs, social-skills training, cognitive behavioral therapy and technology-delivered interventions. ABA sits within that grouping rather than being isolated as a single comparator, so the results speak to behavioral and psychosocial care in aggregate, not to any one model.
Behavioral Interventions Showed a Large Effect on Core Autism Symptoms
The separation between the two arms was stark. Nonpharmacological interventions produced a pooled effect of g = 0.70 on core symptoms, against g = 0.20 for pharmacological and dietary-supplement approaches, a between-group difference of 0.43 that was highly statistically significant (P < 0.0001). On the conventional Cohen benchmarks (0.2 small, 0.5 medium, 0.8 large), that places behavioral and psychosocial care in the medium-to-large range and drugs near the bottom of the small range.
One figure deserves a caution flag of its own: the behavioral arm was far more heterogeneous than the drug arm (I-squared of 75.6% versus 44.1%). High heterogeneity means the individual trial results scattered widely, so the 0.70 average masks a meaningful spread between studies that found large benefits and those that found little. The pooled number is a center of gravity, not a guarantee of what any single program delivers.
Why Autism Medications Scored Low on Core Symptoms
A small effect for drugs on core symptoms is, in one sense, exactly what the pharmacology would predict. No medication is approved to treat autism’s core social-communication and repetitive-behavior features. The two agents the FDA has cleared in autism, risperidone and aripiprazole, are indicated for irritability and aggression associated with the condition, not for the core phenotype the meta-analysis measured. Read that way, the g = 0.20 result is less an indictment of medication than a reminder that drugs and behavioral programs are aimed at different targets. Families and prescribers who use medication to manage irritability, anxiety, sleep or attention are addressing co-occurring symptoms the study did not set out to score.
Study Design and Bias: Why the Effect Sizes May Be Inflated
The moderator analysis is where the paper turns genuinely instructive for the field. Larger reported effects were associated with smaller sample sizes, shorter trial durations, non-Western settings, clinician-rated (rather than blinded or objective) outcomes, and earlier publication dates. In the meta-regression, sample size and publication year stood out as significant predictors: the smaller and older the study, the bigger the apparent benefit.
That cluster of red flags points to a familiar measurement problem, and behavioral research is especially exposed to it. Drug trials can be double-blinded against an identical-looking placebo; a parent-training program or a therapist-delivered intervention generally cannot. When the people rating improvement know which children received the treatment, expectations can color the scores. The pattern echoes earlier work such as Project AIM, a 2020 meta-analysis that found behavioral intervention effects shrank sharply once analyses accounted for detection bias and study quality. The new paper does not say behavioral care fails to work; it says the field’s most eye-catching numbers may be partly an artifact of how the studies were built and judged.
What the Findings Mean for ABA Providers and Payers
For behavioral health operators, the finding is double-edged. It reinforces the clinical consensus that behavioral and psychosocial approaches are the front-line response to autism’s core features, a position that supports the value proposition of ABA and allied services. At the same time, it hands payers and auditors a sharpened argument that headline efficacy claims need to be backed by rigorous, well-measured outcomes rather than legacy effect sizes drawn from small or unblinded trials.
That tension is likely to push the conversation toward standardized, validated outcome measures, longer follow-up, larger samples and, where feasible, blinded or objective assessment. Providers that can demonstrate durable gains using defensible measurement are positioned to fare better as reimbursement scrutiny intensifies. The study, in effect, raises the evidentiary bar for everyone in the autism services market, even as it affirms the broad direction of behavioral care.
Autism Policy and Medicaid Scrutiny Raise the Stakes
The analysis lands in an unusually politicized environment for autism science. Federal health leadership has overhauled the Interagency Autism Coordinating Committee, redirected research priorities toward identifying causes, and floated data initiatives that have drawn objections from researchers and the autistic community. With the OIG continuing to examine Medicaid ABA payments and state programs revisiting rates and prior-authorization rules, questions about which interventions deliver measurable benefit are no longer confined to journals. They feed directly into coverage decisions, oversight and the public debate over how autism care is funded. A peer-reviewed comparison that both validates behavioral approaches and warns against overstating them is, for that reason, a useful corrective on multiple fronts.
The bottom line for the sector: the weight of randomized evidence still favors behavioral and psychosocial interventions for core autism symptoms by a wide margin, but the credibility of that margin now depends on the field’s willingness to fund and run the kind of rigorous, well-controlled trials the authors say are still missing.
