Published on 25 September 2026
Atypical antipsychotics for autistic children: reading a network ranking against the clinical threshold
The essentials
This Cochrane review with network meta-analysis pools 17 randomised trials and 1,027 participants, of whom 996 were children and 31 came from a single adult trial. For short-term irritability in autistic children, measured with the ABC-I (0 to 45), risperidone may reduce the score by 7.89 points against placebo (95% CI -9.37 to -6.42) and aripiprazole by 6.26 points (-7.62 to -4.91), with low certainty; lurasidone probably makes little or no difference (-1.30; -5.46 to 2.86; moderate certainty). The confidence intervals of the first two cross the 6.6-point minimal important difference chosen by the authors, and the aripiprazole estimate lies below that threshold. The ranking puts risperidone first (SUCRA 99.6%), yet it is only 1.63 points ahead of aripiprazole. Weight gain and extrapyramidal signs seem more frequent on treatment, with very low certainty. There are no data beyond six months.
Context
Atypical antipsychotics are widely prescribed for behavioural symptoms of autism, including irritability and aggression. According to the review, only aripiprazole and risperidone have FDA approval for irritability in autistic children and adolescents. Direct comparisons between drugs are scarce: two trials compare risperidone with aripiprazole.
A network meta-analysis can rank treatments that have not been compared head to head. The ranking itself has to be read with caution: a very high rank can hide a clinically minimal difference.
The study at a glance
| Item | |
|---|---|
| Population | |
| Children (16 trials, 996 participants) and adults (1 trial, 31 participants) with a clinical diagnosis of autism spectrum disorder; trials published from 1998 to 2019, in the United States, Canada, Japan, Iran and India; 11 to 105 participants randomised per trial | |
| Intervention | |
| Oral atypical antipsychotics: risperidone, aripiprazole, lurasidone, olanzapine | |
| Comparator | |
| Placebo, or another atypical antipsychotic (two trials compare risperidone with aripiprazole) | |
| Outcomes | |
| Irritability (ABC-I, 0 to 45), aggression, weight gain, extrapyramidal effects, obsessive-compulsive behaviours, inappropriate speech; primary outcome in the short term (under three months), limited medium-term data (three to six months), none in the long term (over six months) | |
| Design | |
| Cochrane systematic review with network meta-analysis for irritability and pairwise meta-analyses for the other outcomes; protocol published in 2022 | |
Quality check
| Item | Verdict |
|---|---|
| Published protocol | Published in 2022 |
| FindingProtocol published in 2022; some methods have since been modified, with the detail given in supplementary material not consulted here. | |
| Literature search | 14 sources |
| FindingSearch of 14 sources (12 databases, 2 trial registers), last searched on 3 January 2024. | |
| Risk of bias of the trials | Reservations |
| FindingRoB 2 outcome by outcome; only one trial at low overall risk (Owen 2009); four trials at high risk for outcome measurement, and one trial (DeVane 2019) at high risk for missing data. | |
| Transitivity and consistency | Assessed |
| FindingAssessed; partially connected network, with no intransitivity or global inconsistency reported; prediction intervals provided. | |
| Certainty of evidence | GRADE and CINeMA |
| FindingA minimal important difference of 6.6 points was used to judge imprecision; this threshold comes from an estimate in people with intellectual disability and aggressive behaviour, not specific to autism. | |
| Funding | No dedicated funding |
| FindingNo funding dedicated to the review; internal support (salary, in-kind support) from three institutions; conflicts of interest declared by author in the publication. | |
Results
| Item | Value |
|---|---|
| Risperidone versus placebo | -7.89 |
| Finding95% CI -9.37 to -6.42; prediction interval -9.63 to -6.15; low certainty. | |
| Aripiprazole versus placebo | -6.26 |
| Finding95% CI -7.62 to -4.91; prediction interval -7.86 to -4.67; low certainty. | |
| Lurasidone versus placebo | -1.30 |
| Finding95% CI -5.46 to 2.86; moderate certainty: probably little or no difference. | |
| Ranking (SUCRA) | 99.6; 66.4; 24.4 |
| FindingRisperidone, aripiprazole, lurasidone. Gap between risperidone and aripiprazole: 1.63 points (95% CI 0.34 to 2.93). | |
| Weight gain (predefined thresholds) | RR 2.40 |
| Finding95% CI 1.25 to 4.60; 7 trials, 434 participants; 123 versus 51 per 1,000; very low certainty. In kilograms: +1.22 kg (95% CI 0.55 to 1.88; 3 trials, 297 participants), very low certainty. | |
| Extrapyramidal effects | RR 2.36 |
| Finding95% CI 1.22 to 4.59; 6 trials, 511 participants; 121 versus 51 per 1,000; very low certainty. | |
Critical appraisal
| Item | Verdict |
|---|---|
| Cochrane methodology | Complete |
| FindingPublished protocol, 14 sources, RoB 2 by outcome, transitivity and consistency assessed, prediction intervals, CINeMA and GRADE, clinical threshold of 6.6 points. | |
| Authors’ conclusions | In line with the certainty |
| FindingCautious wording (may, probably, very uncertain); conclusions drawn drug by drug against placebo, with no recommendation on choosing between drugs. | |
| Body of evidence | Weak |
| FindingSmall trials (11 to 105 participants), only one at low overall risk of bias, partially connected network with two direct head-to-head trials. | |
| Adults and long term | Almost absent |
| FindingA single adult trial, 31 people, medium-term follow-up; no data beyond six months. | |
| Wording in the text | Authors should correct labels |
| FindingAt least two qualifiers in the body of the text do not match the numerical interval (stereotypy, risperidone versus aripiprazole: “favourable” for -0.80, CI -3.09 to 1.49; social-communication adaptive functioning in adults: “null” for -0.47, CI -0.81 to -0.13). The figures used here are those of the estimates, not the qualifiers. | |
Level of evidence
The level of evidence is that of a systematic review of randomised trials with network meta-analysis (Oxford CEBM level 1a). Confidence is high in the quality of the synthesis and in the transparency of the Cochrane process.
It is more limited for clinical decisions: small trials, a partially connected network, a single adult trial, no long-term data, low certainty for the two main results and very low certainty for weight gain and extrapyramidal effects. The ranking measures a probability, not the size of an advantage.
The colleague test
What an experienced colleague would say if you presented this study in two minutes, between two consultations.
“In autistic children, risperidone and aripiprazole may reduce short-term irritability by 6 to 8 points out of 45, around the clinical threshold, with low certainty; the trials also report more sedation and, with very low certainty, more weight gain. Risperidone’s first place rests on 1.63 points.”
Translation for practice: the ranking is not enough to choose a drug; you weigh a possible benefit, close to the clinical threshold, against frequent adverse effects, over the duration of the trials only.
What you can do with this
- What you can understand: a SUCRA of 99.6% does not say that risperidone is clearly superior, only that it is the most likely to be the best; the gap with aripiprazole is 1.63 points on a 45-point scale.
- What you can tell parents: risperidone and aripiprazole may reduce short-term irritability, by roughly 6 to 8 points out of 45, with low certainty; weight gain and motor effects seem more frequent, with very low certainty, and there are no data beyond six months.
- What you can monitor: weight, sedation, extrapyramidal signs and whether the benefit persists, since the trials say nothing about the long term.
- Prescribing framework in France, from the summaries of product characteristics consulted in the public medicines database of the ANSM (the French medicines agency) on 25 September 2026: risperidone is indicated in children from 5 years of age for persistent aggression in conduct disorder with below-average intellectual functioning or intellectual disability, for six weeks at most; aripiprazole has no indication in autism spectrum disorder. Irritability in autism is therefore not, as such, the labelled indication. In the United States, the review reports FDA approval for irritability associated with autism (aripiprazole from 6 to 17 years, risperidone from 5 to 16 years); labelling varies between countries and should be checked in the local summary of product characteristics.
- In adults: a single trial of 31 people, with no network analysis; the data are too sparse to guide practice.
Frequently asked questions
Which antipsychotic should be chosen for irritability in autistic children?
The review does not conclude in favour of one drug. Risperidone and aripiprazole may reduce irritability by roughly 6 to 8 points out of 45, with low certainty; the gap between them is 1.63 points (95% CI 0.34 to 2.93), below the 6.6-point threshold.
What does a SUCRA of 99.6% mean?
It is the relative probability of being the best treatment in the network, not a measure of advantage. It has to be read alongside the score difference, here 1.63 points on a 45-point scale.
Is lurasidone effective for irritability in autism?
In this review it probably makes little or no difference compared with placebo (-1.30; 95% CI -5.46 to 2.86), with moderate certainty.
Are these drugs safe?
The data do not allow a conclusion of safety. Weight gain (123 versus 51 per 1,000) and extrapyramidal effects (121 versus 51 per 1,000) are more frequent than on placebo in the trials, but the authors judge these effects very uncertain (very low certainty). Sedation (RR 3.55; 95% CI 2.09 to 6.02; 8 trials, 673 participants) and somnolence (RR 4.68; 2.65 to 8.24) are also more frequent, for outcomes absent from the GRADE summary tables. No data are available beyond six months.
Do these results apply to autistic adults?
The review does not allow a conclusion for adults: a single trial, 31 participants, medium-term follow-up, and no network analysis possible.
Annotated bibliography
Source study. Meza N, Franco JVA, 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 of Systematic Reviews 2025; (5): CD014965. DOI 10.1002/14651858.CD014965.pub2. PMID: 40396498. Funding: the review received no dedicated funding; internal sources: the Instituto Universitario Hospital Italiano de Buenos Aires (salary of C. M. Escobar Liquitay for conducting Cochrane reviews), the Universidad de Valparaíso (in-kind support to N. Meza and E. Madrid), Heinrich Heine University Düsseldorf (in-kind support to J. V. A. Franco); no external sources. Conflicts of interest as declared: Meza, Núñez, Escobar Liquitay, Rees and Rojas F.: none. Franco: Contact Editor of Cochrane Urology and Managing Editor of Cochrane Metabolic and Endocrine Disorders, not involved in the editorial process of this review. Sguassero: works part time for Cochrane Response, editor of Cochrane Developmental, Psychosocial and Learning Problems (DPLP) and of Cochrane Clinical Answers, not involved in the editorial process. Williams: ongoing grant (since 1 January 2020) from Epsilon Healthcare (formerly THC Global Group Ltd) to develop an experimental product and a placebo for children in a multisite phase III trial funded by the MRFF, paid to the Murdoch Children’s Research Institute; grant from Tilray (28 November 2018 to 27 November 2019) for a pilot trial of cannabidiol in severe behaviour disorders of children with intellectual disability, with or without autism; NHMRC grant (1 June 2020 to 31 May 2023), of which she is principal investigator, for a phase III cannabidiol trial, paid to institutions; participation in an ongoing NHMRC-funded study on predictors of autism outcomes, whose results might be eligible for this review; editor of DPLP; unpaid member of a data monitoring committee for a selective serotonin reuptake inhibitor trial; principal investigator of a funded randomised trial at the Monash Children’s Clinical Trials Team (ACADIA) and of a funded trial (Axial Therapeutics); consulting fees as a member of the Scientific Committee (Human Health); co-author of a book that generates royalties; a grant from the Australian Medical (wording as published). Rojas V.: paediatric neurologist at Hospital Dr Gustavo Fricke and Universidad de Valparaíso. Pringsheim: neurologist at Alberta Children’s Hospital and Foothills Medical Centre; declares having expressed opinions on the subject in academic work; grants from the Azrieli Accelerator (University of Calgary) and the Canadian Institutes of Health Research. Madrid: contributor to the Cochrane Sustainable Healthcare Group, has published opinions on the subject, including a 2021 Cochrane editorial on low-value care during the COVID-19 pandemic. Supplementary material: of the nine supplementary materials of the publication, those partly consulted were the characteristics of included, excluded, awaiting classification and ongoing studies (materials 2 to 5, without exhaustive reading) and the data package (material 8, CSV and JSON files). Supplementary materials 1 (search strategies), 6 (risk of bias), 7 (analyses) and 9 (differences from the protocol) were not consulted, and no data appearing only there are reused here.
Context reference. Hassiotis A, Melville C, Jahoda A, Strydom A, Cooper SA, Taggart L et al. Estimation of the minimal clinically important difference on the Aberrant Behaviour Checklist-Irritability (ABC-I) for people with intellectual disabilities who display aggressive challenging behaviour: a triangulated approach. Research in Developmental Disabilities 2022; 124: 104202. DOI 10.1016/j.ridd.2022.104202. This reference, cited by the review, is the source of the 6.6-point ABC-I threshold used to judge imprecision; it concerns a population with intellectual disability and aggressive behaviour, not specific to autism.
Editorial collections
Tags
Verified on 25 September 2026 against the full text of the publication and its supplementary material where available. This analysis underwent an independent double reading. The English version was checked for conformity on 25 September 2026, against the figures of the French version and against the source. How we verify what we publish
This analysis is intended for healthcare professionals. It does not constitute a prescribing recommendation and does not replace individual clinical judgment.
Analysis from Psychiatry Evidence Base, evidence-based psychiatry, explained with rigor.