Atypical antipsychotics for autism spectrum disorder: a network meta‐analysis

Objectives

This is a protocol for a Cochrane Review (intervention). The objectives are as follows:

Primary

To assess the comparative efficacy of atypical antipsychotics for irritability through network meta‐analysis in children and adults with autism spectrum disorder (ASD).

Secondary

To assess the safety and efficacy of atypical antipsychotics as medications for core and non‐core symptoms in children and adults with ASD.

Background

Description of the condition

Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM‐5; APA 2013), identifies autism spectrum disorder (ASD) as a lifelong condition based on 1) social and communication deficits, and 2) restricted and repetitive behaviours, as manifested by at least two of the following elements: insistence on sameness, fixed and unusual interests, and abnormal sensory processing. The current criterion for ASD as a single diagnosis is the result of merging a broad variety of nosological entities, as defined by both Diagnostic and Statistical Manual of Mental Disorders, 4th Edition‐Text Revision (DSM‐IV‐TR; APA 2000), and the International Statistical Classification of Diseases and Related Health Problems, 11th Revision (ICD‐11; WHO 2021). ASD is considered a neurodevelopmental disorder, but for most patients, the exact genetic and environmental factors contributing to its cause remain unknown. 

People with ASD have considerable symptom variability that may affect patients throughout life, which generates special concerns in social inclusion issues in adult life (Magiati 2014). Both DSM‐5 (APA 2013) and ICD‐11 (WHO 2021) have stated severity degrees according to functionality relative to core symptoms (such as poor language, stereotyped behaviours, social impairment, among others) and the presence of intellectual disability, respectively. Secondary characteristics such as emotional and behavioural problems are extremely common; examples include irritability, aggression, emotional dysregulation, self‐injury and injury to others (Lecavalier 2006). As no biomarkers have been well established yet (Frye 2019; Hiremath 2021), the diagnosis of ASD is based on behavioural features. 

Estimates of the prevalence of ASD vary by the diagnostic criteria used, the age of children screened, and the study location (Elsabbagh 2012; Persico 2021). More recent prevalence studies have yielded higher estimates: up to 1/54 in children (Maenner 2020) and 1/45 in adults in the USA (Dietz 2020); and to 1/85 in children in Italy (Narzisi 2018), and 1/102 in adults in the UK (Brugha 2011). A 2016 report by the US Centers for Disease Control and Prevention (CDC) found the point prevalence of ASD to be 18.5 per 1000 eight‐year‐olds (Maenner 2020). A systematic review found that the median worldwide prevalence of autistic disorder was 17 per 100,000, with a range of 2.8 to 94 per 100,000, and the median worldwide prevalence of pervasive developmental disorder was 62 per 100,000, with a range of 1 to 189 per 100,000 (Elsabbagh 2012). Recently, the current global prevalence of ASD has been reported to be as high as 1% (Lyall 2017).

Individuals with ASD are also frequently affected by other neuropsychiatric conditions, the most common of which are: attention deficit hyperactivity disorder (ADHD); anxiety disorders; sleep‐wake disorders; oppositional defiant disorder (ODD); depressive disorders, obsessive‐compulsive disorder (OCD); bipolar disorders; and schizophrenia spectrum disorders (Lyall 2017). 

Behavioural symptoms of ASD during adult life represent a specific challenge (Im 2021); indeed, neurobiological findings have shown age‐related structural brain network changes in ASD (Baxter 2019; Koolschijn 2017; Powell 2017; Walsh 2019). Clinical trials regarding pharmacological interventions for behavioural symptoms (e.g. challenging behaviours) in adults with ASD have been conducted (Im 2021; Sawyer 2014), but these are scarce in comparison with studies exploring the efficacy of these treatments in children (Im 2021). Furthermore, the difference (in terms of effect measures) of pharmacological interventions among different age groups of patients with ASD are still unclear.

Both pharmacological and non‐pharmacological interventions are available for children and adults with ASD (LeClerc 2015; Meza 2022). Non‐pharmacological interventions include educational, behavioural, and social communication strategies used alone or in combination as part of an individual plan to enhance learning and community participation (Meza 2022). These interventions strive to improve communication, social skills, daily living skills, play and leisure skills, academic achievement, and maladaptive behaviours (Meza 2022; Myers 2007). A range of drugs has been assessed for behavioural symptoms in ASD, both in children and adults. However, only two atypical (second‐generation) antipsychotics (i.e. aripiprazole (FDA 2009) and risperidone (FDA 2005)) have received approval from the US Food and Drug Administration (FDA) for their use for irritability in children and youth with ASD (Fieiras 2022). 

Description of the intervention

Antipsychotics, also known as neuroleptics, are a class of psychotropic agents traditionally used to treat psychotic phenomena in schizophrenia but nowadays are used in the management of other conditions (Brunton 2018). They are classified as typical (first‐generation) or atypical (second‐generation). In comparison with typical antipsychotics (which have been associated with extrapyramidal side effects (Campbell 1997) such as acute dystonic reactions, withdrawal dyskinesias, and tardive dyskinesia), atypical (or second‐generation) antipsychotics may be associated with a lower risk of drug‐induced movement disorders except for risperidone in high doses (Brunton 2018), but it remains unclear. Nevertheless, atypical antipsychotics have shown a greater risk of weight gain (Brunton 2018; Pierre 2005), but more evidence is needed, considering potential differences among age groups.

Typical and atypical antipsychotics have been widely evaluated and prescribed to treat core and behavioural symptoms (e.g. irritability, aggression, obsessions, repetitive behaviours, etc.) in children and adults with ASD (Accordino 2016; Posey 2008). A systematic review of some atypical antipsychotics (i.e. risperidone, aripiprazole, and lurasidone) for ASD in children indicated that risperidone and aripiprazole may be effective and equally safe in treating symptoms of irritability (Fallah 2019). Nevertheless, head‐to‐head comparisons show risperidone as no superior (Ghanizadeh 2014) or slightly more effective than aripiprazole, with more side effects (DeVane 2019). Although short‐term randomised controlled trials (RCTs) suggested the efficacy of antipsychotics for improving some symptoms of ASD, side effects may limit their use (D’Alò 2020). 

How the intervention might work

Alterations in dopaminergic and serotonergic neurotransmission have been implicated in ASD and its symptoms, as it has been elucidated through neuroimaging and metabolic studies (Posey 2008). Typical antipsychotic medications, such as haloperidol, are potent antagonists at D₂‐dopamine receptors. Second‐generation antipsychotics, in general, are weak antagonists at D₂‐dopamine receptors, and stronger antagonists at D₄‐dopamine receptors and 5‐HT_2A‐serotonin receptors (Aringhieri 2018; Kapur 2001; Seeman 2002). Atypical antipsychotics are also antagonists of histamine receptors and noradrenaline receptors. Thus, both typical and atypical antipsychotics would affect the manifestations of ASD through the interactions with dopamine and serotonin receptors, but this might vary according to their affinity to such receptors. Aripiprazole, an atypical antipsychotic agent, is considered a partial dopamine agonist due to its dual agonist/antagonist action depending on synaptic dopamine levels (Goodnick 2002). Additionally, the 5‐HT₂‐serotonin receptor antagonism of aripiprazole is higher than most atypical antipsychotics but lower than ziprasidone or risperidone. However, its degree of antagonist activity, along with partial D₂‐dopamine receptor agonist activity, is at an optimal level, which would reduce the risk of extrapyramidal signs, from a theoretical perspective (Goodnick 2002). Moreover, there might be a differential effect of these drugs in their interaction with other receptors. For example, aripiprazole has a moderate affinity to the H₁‐histamine receptor, which would result in decreased sedation and risk of weight gain compared with clozapine and olanzapine (Goodnick 2002).

Why it is important to do this review

This protocol, broader in scope than previous Cochrane Reviews on the safety and efficacy of aripiprazole (Hirsch 2016) and risperidone (Jesner 2007) for children and adults with ASD, aims to perform a comprehensive and updated analysis of all studied atypical antipsychotics. Moreover, it will include head‐to‐head comparisons following the latest innovations in the production of systematic reviews. Of note, other atypical antipsychotics—not yet included in a Cochrane Review—such as olanzapine or lurasidone, have shown variable results on behavioural symptoms of ASD in either children or adults, as reported in other systematic reviews and clinical trials (D’Alò 2021; Fallah 2019; Fieiras 2022; Persico 2021; Pillay 2017). The FDA has approved only aripiprazole and risperidone for children and adolescents to treat irritability in patients with ASD aged from 6 to 17 years (FDA 2009) and 5 to 16 years (FDA 2005), respectively. It is important to assess both the efficacy and the side effects of the different atypical antipsychotics in individuals with ASD. It is also relevant to determine whether the effects of atypical antipsychotics depend on factors such as age, posology, duration of treatment, and pharmacological co‐interventions (e.g. non‐pharmacological interventions, antidepressants, melatonin, among others), considering a broad range of critical outcomes.

Objectives

Primary

To assess the comparative efficacy of atypical antipsychotics for irritability through network meta‐analysis in children and adults with autism spectrum disorder (ASD).

Secondary

To assess the safety and efficacy of atypical antipsychotics as medications for core and non‐core symptoms in children and adults with ASD.

Methods

Criteria for considering studies for this review

Types of studies

All randomised controlled trials (RCTs) comparing any atypical antipsychotic drug with placebo or another atypical antipsychotic drug for autism spectrum disorder (ASD). Given the longitudinal nature of the intervention and the expected fluctuation in symptoms, we will exclude cross‐over trials. 

Types of participants

Adults (aged 18 years or older) and children (up to 17 years old) with a clinical diagnosis of ASD. The diagnosis could be reached according to any version of the Diagnostic and Statistical Manual of Mental Disorders (DSM) (APA 1980; APA 1994; APA 2000; APA 2013) or the International Statistical Classification of Diseases and Related Health Problems (ICD) (WHO 2011; WHO 2021), Autism Diagnostic Interview‐Revised (Lord 1994), Autism Diagnostic Observation Schedule (Lord 2000), or any other ‐ current or past ‐ standardised diagnostic criteria or instrument. We will include participants with diagnoses using terms relevant to ASD such as Asperger’s disorder, pervasive developmental disorder‐not otherwise specified, and autistic disorder, regardless of comorbidity with other neuropsychiatric disorders. We will place no restrictions based on age or severity. In addition, we will not exclude RCTs that include participants receiving concomitant pharmacological and non‐pharmacological therapies in both study arms that are considered balanced co‐interventions.

We will exclude studies focused exclusively on a genetic or chromosomal condition associated with ASD (e.g. Rett syndrome, fragile X syndrome, tuberous sclerosis, Phelan‐McDermid syndrome, among others), but we will include studies that recruit ASD that may or may not have included some children with Rett and childhood disintegrative disorder (CDD) will be included.

We will not include studies conducted on patients with unconfirmed cases of ASD.

Types of interventions

Atypical antipsychotics (i.e. risperidone, aripiprazole, lurasidone, clozapine, olanzapine, quetiapine, zotepine, amisulpride, ziprasidone, cariprazine, brexpiprazole, asenapine, sertindole, and iloperidone) for ASD; administered orally, at any dosage and period of prescription.

Comparisons for network meta‐analysis

We will perform network meta‐analyses for children and adults separately. We will include trials comparing atypical antipsychotics to other drugs of the same class or placebo (see the representation of the network in Figure 1). We will group different posologies for the same drugs in the same. Participants in the networks could in principle be randomised to any of the nodes, and we will verify this by comparing characteristics of study design, participants, interventions, and comparisons (Salanti 2012) while considering potential sources of clinical heterogeneity and effect modification (see Subgroup analysis and investigation of heterogeneity). These network meta‐analyses will focus on the first primary outcome (irritability ‐ short term).

1.

Representation of possible comparisons between atypical antipsychotics and placebo. Considering that there are numerous atypical antipsychotics and we do not expect all of them to be evaluated in clinical trials, we illustrated the framework for grouping with some examples.

Comparisons for pairwise meta‐analysis

We will perform pairwise meta‐analyses for children and adults separately. We will include studies that compare the intervention to a placebo or to another atypical antipsychotic drug. We will group different posologies by the same antipsychotic agent.

Types of outcome measures

We will not use the measurement or non‐measurement of the outcomes assessed in this review as an eligibility criterion.

Primary outcomes

• Irritability, as rated by the Aberrant Behaviour Checklist (ABC) Irritability subscale (Aman 1986).

• Aggression, as rated by the Overt Aggression Scale (OAS; Yudofsky 1986), or Behavior Problems Inventory (BPI) Aggressive/Destructive Behavior subscale (Rojahn 2001).

• Weight gain and metabolic side effects, as well as other side effects (e.g. somnolence, sedation, insomnia, headache, constipation, suicidal ideation, suicidal attempt, hyperprolactinaemia, among others).

Secondary outcomes

• Extrapyramidal side effects, as measured by scales such as the modified Webster Scale (mWS; Webster 1968), the Abnormal Involuntary Movements Scale (AIMS; Guy 1976), the Extrapyramidal Symptom Rating Scale (ESRS; Chouinard 2005), and the Barnes Akathisia Scale (BARS; Barnes 1989).

• Obsessive‐compulsive behaviours, as rated by the Children’s Yale‐Brown Obsessive‐Compulsive Scale (CY‐BOCS; Goodman 1989).

• Social‐communication adaptive functioning (social functioning), as rated by the Vineland Adaptive Behavior Scale (VABS) Socialisation subscale (Sparrow 1993).

• Inappropriate speech, as rated by the ABC Inappropriate Speech subscale (Aman 1986).

• Lethargy/withdrawal, as rated by the ABC Lethargy/Withdrawal subscale (Aman 1986).

• Hyperactivity, as rated by the ABC Hyperactivity subscale (Aman 1986).

• Stereotypy, as rated by the ABC Stereotypy subscale (Aman 1986) or BPI‐Stereotyped Behavior subscale (Rojahn 2001).

• Clinical improvement, as rated by the Clinical Global Impression (CGI) ‐ Improvement scale (Guy 1976).

For both primary and secondary outcomes, we will include results measured by validated clinician‐ or parent‐reported instruments. In case of any aforementioned scale is not reported in a specific outcome, we will include another reliable or commonly used scale if considered in a primary study.

Timing of outcome measures

We plan to synthesise results for the following time points: short term (less than 3 months), medium term (3 to 6 months), and long term (over 6 months). When multiple results are reported for each outcome, we will include the longest follow‐up in each category.

Search methods for identification of studies

Electronic searches

We will search the following sources from the beginning of each database to the date of the search without restrictions on the date, language, or publication status. We will use search terms suggested by our clinical experts taking into consideration the recently published search criteria and classification of pharmacological treatments, including atypical antipsychotics (Shokraneh 2021). 

• Cochrane Central Register of Controlled Trials (CENTRAL; current issue) in the Cochrane Library, and which includes the specialised register of Cochrane Developmental Psychosocial and Learning Problems.

• MEDLINE Ovid SP and Epub Ahead of Print, In‐Process, In‐Data‐Review & Other Non‐Indexed Citations, Daily and Versions (1946 onwards).

• Embase Elsevier (1974 onwards).

• PsycINFO Ovid SP (1967 onwards).

• Cumulative Index to Nursing and Allied Health Literature (CINAHL) EBSCOhost (1937 onwards).

• LILACS (lilacs.bvsalud.org/es/; all available years).

• Science Citation Index Expanded (SCI‐E) Web of Science Clarivate (1970 to present).

• Social Sciences Citation Index (SSCI) Web of Science Clarivate (1970 to present).

• Conference Proceedings Citation Index ‐ Science (CPCI‐S) Web of Science Clarivate (1990 to present).

• Conference Proceedings Citation Index ‐ Social Science & Humanities (CPCI‐SSH) Web of Science (1990 to present).

• ProQuest Dissertations & Theses Global (all available years).

• Epistemonikos (www.epistemonikos.org).

• US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov).

• World Health Organization International Clinical Trials Registry Platform (trialsearch.who.int/).

We will search MEDLINE using the strategy in Appendix 1. This strategy will be adapted for other databases using appropriate syntax and indexing terms. 

Searching other resources

We will check the reference lists of all primary included studies and relevant review articles for additional references. We will contact experts in the field to identify additional unpublished data. We will search relevant manufacturers’ websites for trial information, including the following specialised sources of adverse events reports: Australian Adverse Drug Reactions Bulletin (www.tga.gov.au/adr/aadrb.htm); Current Problems in Pharmacovigilance (www.mhra.gov.uk); European Medicines Evaluation Agency (www.emea.eu); and the US Food and Drug Administration (FDA) Medwatch (www.fda.gov/medwatch). Close to the publication of the review, we will search MEDLINE and Embase to identify retractions, errors, or corrections of included studies. We will also search RetractionWatch (retractionwatch.com). 

Data collection and analysis

Selection of studies

Two review authors (Eva Madrid (EM) and Nicolás Meza (NM)) will independently screen titles and abstracts obtained from the searches and select those potentially relevant.

We will retrieve the full‐text of study reports deemed potentially relevant, and two review authors (EM, NM) will independently screen them to identify studies that meet the criteria for inclusion. We will identify ineligible records and record the reasons for exclusion. We will resolve any disagreements through discussion, and, if required, a third review author will be consulted (Juan VA Franco (JVAF)/Yanina Sguassero (YS)).

We will also identify and collate multiple reports of the same study so that each study rather than each report is the unit of interest in the review. We will use Covidence software for study selection (Covidence 2021). We will not impose restrictions on the date, language, or publication status.

We will present our study selection process in detail as a PRISMA flowchart (Page 2021).

Data extraction and management

Two review authors (NM, EM) will independently extract data from the included studies and enter them onto a pre‐designed data extraction form, which has been piloted on at least two studies. We will extract the following data.

• Participants: age (i.e. mean and standard deviation, or median and interquartile range), gender, diagnostic criteria, intelligence quotient (IQ), inclusion or exclusion criteria (or both), and baseline characteristics of study groups.

• Interventions: antipsychotic drug, dosage, frequency of administration, duration of treatment, comparator, co‐interventions, and excluded medications.

• Outcomes: primary and secondary outcomes, and time points.

• Methods: study dates and procedures (e.g. recruitment of participants), number of participating centres and locations, study setting, details of any 'run‐in' period, study design, and risk of bias assessment (e.g. method of randomisation and allocation concealment).

We will resolve all discrepancies concerning the data extraction through discussion. An additional review author (Valeria Rojas (VR)) will enter data into Review Manager (RevMan Web 2020), and another (JVAF) will double‐check against the study report to ensure the accuracy of the data extracted. We will provide a summary of the characteristics of included studies as additional tables that will be presented separately for adults and children.

Assessment of risk of bias in included studies

Two review authors (NM, EM) will independently assess the risk of bias for the following results:

• irritability (short term);

• aggression (short term);

• weight gain (short term);

• extrapyramidal side effects (short term);

• obsessive‐compulsive behaviours (short term); and

• inappropriate speech (short term).

For this purpose, we will use RoB 2, the current Cochrane risk of bias tool to assess the risk of bias in randomised trials (Higgins 2022a; Sterne 2019). Any disagreements will be resolved by discussing or involving a third review author as a referee (JVAF/YS). We will assess the following risk of bias domains: 1) randomisation, 2) deviations from intended interventions, 3) missing outcome data, 4) measurement of the outcome, and 5) selection of the reported results.

Answers to signalling questions and supporting information will collectively lead to a domain‐level judgement in the form of 'low risk', 'some concerns', or 'high risk' of bias. These domain‐level judgements will inform an overall risk of bias judgement for each result. We will consider the algorithm‐proposed judgements and provide a quote from the study report together with a justification for our judgement in the risk of bias table. We will summarise the risk of bias judgements across different studies for each of the domains listed. When judging the 'Bias due to deviations from intended interventions', we will focus the analyses on the effect of assignment to intervention (Higgins 2022a). We will aim to source study protocols for the assessment of selective reporting. Where information on the risk of bias relates to unpublished data or correspondence with a trialist, we will note this in the risk of bias table.

We will use the RoB 2 Excel tool to manage the data supporting the answers to the signalling questions and risk of bias judgements (available at www.riskofbias.info/). All these data will be publicly available as supplementary material in the Open Science Framework platform (osf.io/).

We will conduct the review according to this published protocol and report any deviations from it in the 'Differences between protocol and review' section of the systematic review, in order to address any potential bias in conducting this systematic review.

Measures of treatment effect

We will enter the outcome data for each study into the data tables in Review Manager to calculate the treatment effects (RevMan Web 2020).

Dichotomous data

We will analyse dichotomous data as risk ratios (RRs). We will report the corresponding 95% confidence intervals (CIs). We will aim to calculate these from the data reported in the trial or following the transformations suggested in Section 6.4 of the Cochrane Handbook for Systematic Reviews of Interventions (hereafter referred to as the Cochrane Handbook; Higgins 2022b).

Continuous data

We will analyse continuous data as mean differences (MD) or standardised MD (SMD) when studies use different scales to measure the same outcome. We will report the corresponding 95% CIs. We will enter data presented as a scale's post‐intervention ('final value') with a consistent direction of effect.

Unit of analysis issues

Cluster‐RCTs 

We do not anticipate finding cluster‐RCTs. However, if we do, we expect that authors will have controlled for clustering in their analyses. If this is not clear, we will contact the study authors and if they have not controlled for clustering, we will request that they supply the individual patient data (IPD), so we may calculate the intracluster correlation coefficient (ICC). If IPD data are not available, we will obtain an external estimate of the ICC from similar studies, or if this is not possible, seek statistical advice to obtain an estimate of the ICC. We will use the ICC to reanalyse the data using the generic inverse variance approach in RevMan Web, to obtain approximately correct analyses (Higgins 2022c).

Studies with multiple treatment groups

The unit of analysis will be the individual participant. Where multiple trial arms are reported in a single trial, we will include only the treatment arms relevant to the review clinical question. If two comparisons (e.g. drug A versus placebo and drug B versus placebo) are combined in the same study, we will follow the guidance in Section 6.2.9 of the Cochrane Handbook (Higgins 2022b), to avoid double‐counting in pairwise meta‐analysis. Our preferred approach will be to combine groups to create a single pairwise comparison.

Dealing with missing data

We will contact study authors or study sponsors to verify key study characteristics and obtain missing numerical outcome data, where possible (e.g. when a study is identified as abstract only), or to explain why they are missing and not available in the study manuscript. If we are unable to obtain a response, we will report our attempts to obtain the missing outcome data and its potential impact on the findings of the review in the 'Discussion' section.

If numerical outcome data are missing, such as standard deviations or correlation coefficients, and they cannot be obtained from the authors, we will calculate them from other available statistics, such as P values, according to the methods described in the Cochrane Handbook (Page 2022).

We will record loss to follow‐up data for risk of bias purposes and analyse beneficial data according to an intention‐to‐treat (ITT) analysis whenever the data are available. We will not impute data, and in case of substantial missing outcome data, this will affect the risk of bias rating and the overall certainty of the evidence.

Assessment of heterogeneity

Network meta‐analysis

Assessment of the transitivity assumption

Before conducting a network meta‐analysis, we will assess the transitivity assumption. Network meta‐analysis rests on the assumption of transitivity, that is, that effect modifiers have a comparable distribution across treatment comparisons in a network (Cipriani 2013; Jansen 2013). To assess the plausibility of this assumption, we will visually inspect the comparability of distributions of age, diagnostic criteria, posology, co‐interventions, intelligence quotient, and overall risk of bias as potential treatment effect modifiers across comparisons (Salanti 2014). We will assess the similarity of inclusion and exclusion criteria of all studies, including participants, treatments, and outcomes, to evaluate whether they impact treatment effects.

Assessment of statistical consistency

Lack of transitivity in a network can threaten the validity of the consistency assumption, that is, the statistical agreement between direct and indirect evidence (Caldwell 2005; Lu 2004). Results can be misleading in the presence of inconsistency in the network. We will evaluate the presence of inconsistency both locally and globally. We will evaluate each network locally using the loop‐specific method by generating an inconsistency factor along with a 95% CI for each closed‐loop (Veroniki 2013). This way, we will identify which piece of evidence will be responsible for inconsistency, and we will explore this further. We will also apply a global assessment for consistency in each network by applying the design‐by‐treatment interaction model (White 2012). It has been shown that inconsistency tests have low power to detect true inconsistency (Song 2012; Veroniki 2014). Hence, we will assess transitivity even in the absence of evidence for inconsistency. If an inconsistency is found, we will follow the guidance provided in the Cochrane Handbook (Section 11.4.4.4; Chaimani 2022).

Pairwise meta‐analysis

For clinical heterogeneity, we will compare the distribution of participant characteristics between studies (e.g. age, gender), whereas, for methodological heterogeneity, we will compare trial characteristics such as randomisation, allocation concealment, blinding, and loss to follow‐up. We will visually inspect the horizontal lines representing each trial on the forest plot for overlapping CI for statistical heterogeneity. We will also use the Chi² test for homogeneity (interpreting a P value less than 0.10 as heterogeneity) and the I² statistic to measure heterogeneity amongst the trials in each analysis. If we identify substantial heterogeneity, we will report it and explore possible causes by pre‐specified subgroup analyses (Subgroup analysis and investigation of heterogeneity).

We will use the rough guide to the interpretation of the I², as outlined in Chapter 10 of the Cochrane Handbook (Deeks 2022), as follows:

• 0% to 40%: might not be important;

• 30% to 60%: may represent moderate heterogeneity;

• 50% to 90%: may represent substantial heterogeneity; and

• 75% to 100%: considerable heterogeneity.

We will avoid using absolute cut‐off values but interpret I² in relation to the size and direction of effects and strength of evidence for heterogeneity (see Data synthesis). If we consider that the studies are too clinically heterogeneous, we will not conduct a primary meta‐analysis and will provide a narrative synthesis of the results instead (see Data synthesis).

Assessment of reporting biases

If the searches identify trial protocols, clinical trial registrations, or abstracts indicating the existence of unpublished studies, we will attempt to determine the status of any unpublished studies by contacting the authors. For the pairwise meta‐analyses, if we are able to pool more than 10 trials, we will create and examine a funnel plot to explore possible small‐study and publication biases. For the network meta‐analyses, if we are able to pool more than 10 trials, will use comparison‐adjusted funnel plots to assess small‐study effects (Chaimani 2013). Several explanations can be offered for the asymmetry of a funnel plot, including true heterogeneity of effect with respect to trial size, poor methodological design (and hence bias of small trials), and publication bias. We will, therefore, interpret results carefully. There is also the possibility of novelty bias, which refers to the mere appearance that a new treatment is better when it is new, which can become apparent in these comparison‐adjusted funnel plots (Salanti 2010).

Data synthesis

We will perform a meta‐analysis if we identify two or more studies with no important differences in methods, addressing the same outcome and similar participants, interventions, and comparators in their underlying clinical questions.

Methods for indirect and mixed comparisons

We will fit a random‐effects network meta‐analysis model because we anticipate methodological and clinical heterogeneity across studies. We will assume a common within‐network heterogeneity estimate across comparisons, and we will estimate this using the restricted maximum likelihood (REML) method (Veroniki 2016). This is a reasonable assumption, given that all treatments included in the network are of the same nature. An advantage of this approach is that treatment comparisons informed by a single study can borrow strength from the rest of the studies in the network (Higgins 1996; Salanti 2008). Each network meta‐analysis treatment effect estimate will be presented along with a 95% CI and a 95% predictive interval (PrI). A PrI is an interval within which the treatment effect estimate of a future study is expected to lie, accounting for both uncertainties of the treatment effect and between‐study variance estimates (Higgins 2009; Riley 2011). We will conduct network meta‐analysis by using the network suite of commands in Stata (White 2012; White 2015).

Methods for pairwise treatment comparisons

We will pool data from studies we judge to be clinically homogeneous using Review Manager software (RevMan Web 2020). If more than one study provides usable data in any single comparison, we will perform a meta‐analysis. We will use a random‐effects model as we assume that clinical heterogeneity will be high among study populations even when children and adults are assessed separately. For continuous outcomes, we will use the inverse variance method. For dichotomous outcomes, we will use the Mantel‐Haenszel method. We will include all studies in the primary analysis, and we will explore the effect of bias in a sensitivity analysis (see Sensitivity analysis). If a meta‐analysis is not possible due to clinical heterogeneity, we will follow the guidance provided in Chapter 12 of the Cochrane Handbook for Synthesis Without Meta‐analysis (SWiM) (McKenzie 2022).

Subgroup analysis and investigation of heterogeneity

We will undertake a subgroup analysis to explore the possibility of differential responses to atypical antipsychotics compared with placebo on our primary outcomes based on:

• age range: early childhood (up to 5 years old); middle childhood (6 to 11 years old); and adolescence (12 to 17 years old);

• posology (fixed‐dose scheme versus flexible‐dose scheme);

• intelligence quotient (low, normal, superior); and

• pharmacological co‐interventions (i.e. atypical antipsychotics as add‐on pharmacological therapy versus atypical antipsychotics as single pharmacological therapy).

We will use the Chi² test to test for subgroup interactions in Review Manager (RevMan Web 2020).

Sensitivity analysis

We will perform sensitivity analyses by repeating the meta‐analyses after excluding:

• unpublished studies (if there were any); and

• studies with an overall high risk of bias.

Summary of findings and assessment of the certainty of the evidence

Two review authors (NM, EM) will use the five GRADE domains (overall RoB 2 judgement, consistency of effect, imprecision, indirectness, and publication bias) to assess the certainty of the body of evidence from studies that contribute data to pre‐specified outcomes. We will resolve any disagreements by discussion or by involving another author (JVAF/YS). Based on how the evidence meets these criteria, we will downgrade the evidence up to two levels for each domain and assign an overall certainty rating of high, moderate, low, or very low (Schünemann 2022). These ratings, shown below, will reflect the overall certainty we have in the estimate of effects per outcome and comparison.

• High certainty: review authors are very confident that the true effect lies close to that of the estimate of the effect.

• Moderate certainty: review authors are moderately confident in the effect estimate; the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.

• Low certainty: confidence in the effect estimate is limited; the true effect may be substantially different from the estimate of the effect.

• Very low certainty: study authors have very little confidence in the effect estimate; the true effect is likely to be substantially different from the effect estimate.

We will create two summary of findings tables for network meta‐analyses, one for children and one for adults, of the comparison of each atypical antipsychotic versus placebo for the outcome irritability (short term) using the Confidence in Network Meta‐analysis (CINeMA) framework and software (Chaimani 2022; CINeMA; Salanti 2014). We will create the summary of findings tables for each outcome using the approach presented by Yepes‐Nuñez 2019 (Additional Table 1).

1. Summary of findings table: network meta‐analyisis of antipsychotic drugs versus placebo.

Patient or population: children and adults with autism spectrum disorder Interventions: atypical antipsychotics Comparator (reference): placebo Setting:
Outcome: irritability Measured by:
X studies X participants	Anticipated absolute effect (95% CI) *		Certainty of the evidence
	With placebo	With atypical antipsychotics
Atypical antipsychotic 1 (direct or mixed estimate)
Atypical antipsychotic 2 (direct or mixed estimate)
Atypical antipsychotic 3 (direct or mixed estimate)
Atypical antipsychotic 4 (direct or mixed estimate)
Atypical antipsychotic 5 (direct or mixed estimate)
CI: confidence interval, MD: mean difference. Network meta‐analysis summary of findings table definitions: * Estimates are reported as mean difference and confidence interval (CI).
GRADE Working Group grades of evidence (or certainty of the evidence). High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

In addition, we will create two summary of findings tables for pairwise comparisons, one for children and one for adults, of atypical antipsychotic versus placebo for the following outcomes:

• aggression (short term);

• weight gain (short term);

• extrapyramidal side effects (short term);

• obsessive‐compulsive behaviours (short term); and

• inappropriate speech (short term).

For pairwise comparisons, we will use methods and recommendations described in Chapter 14 of the Cochrane Handbook (Schünemann 2022) using GRADEproGDT software (GRADEpro GDT 2020).

Appendices

Appendix 1. Search strategy for MEDLINE Ovid

Ovid MEDLINE(R) and Epub Ahead of Print, In‐Process, In‐Data‐Review & Other Non‐Indexed Citations, Daily and Versions(R) 

1. exp Autism Spectrum Disorder/

2. exp Child Development Disorders, Pervasive/

3. (pervasive development* disorder* OR PDD OR PDDs).kf,tw.

4. Asperger*.kf,tw.

5. (autis* OR ASD OR ASDs).kf,tw.

6. Kanner*.kf,tw.

7. or/1‐6

8. Antipsychotic Agents/

9. (atypical adj2 (anti‐psychotic* or antipsychotic*)).kf,tw.

10. "second generation antipsychotic*".kf,tw.

11. Amisulpride/

12. (Amisulpride OR Sultopride OR solian OR Barnetil).kf,tw,rn.

13. Aripiprazole/

14. (Aripiprazol* or Abilify).kf,tw,rn.

15. (asenapine or Secuado or Saphris).kf,tw,rn.

16. (blonanserin or lonasen).kf,tw,rn.

17. Brilaroxazine.kf,tw,rn.

18. (brexpiprazole or Rexulti).kf,tw,rn.

19. (Carpipramine or Defekton or Prazinil).kf,tw,rn.

20. (cariprazine or Vraylar).kf,tw,rn.

21. Clozapine/

22. (Clozaril or Leponex or Clozapine).kf,tw,rn.

23. (chlorocarpipramine or clocapramine).kf,tw,rn.

24. (clot?iapine or Entumine).kf,tw,rn.

25. fluperlapine.kf,tw,rn.

26. (iloperidone or Zomaril or Fanapt).kf,tw,rn.

27. Lurasidone Hydrochloride/

28. (Lurasidone or Latuda).kf,tw,rn.

29. (levosulpiride or levopraid).kf,tw,rn.

30. Olanzapine/

31. olanzapine.kf,tw,rn.

32. Paliperidone Palmitate/

33. (Paliperidone or Invega).kf,tw,rn.

34. (pimavanserin or Nuplazid).kf,tw,rn.

35. perospirone.kf,tw,rn.

36. Perlapine.kf,tw,rn.

37. (Melperone or methylperon).kf,tw,rn.

38. Molindone/

39. Molindone.kf,tw,rn.

40. Moban.kf,tw,rn.

41. mosapramine.kf,tw,rn.

42. (nemonapride or emonapride).kf,tw,rn.

43. Quetiapine Fumarate/

44. (Quetiapine or Seroquel).kf,tw,rn.

45. Remoxipride/

46. Remoxipride.kf,tw,rn.

47. Risperidone/

48. risperid*.kf,tw,rn.

49. (sertindole or Serlect).kf,tw,rn.

50. Sultopriderisperidone.kf,tw,rn.

51. Tiapride Hydrochloride/

52. Tiapride.kf,tw,rn.

53. Velotab.kf,tw,rn.

54. (ziprasidone or Geodon).kf,tw,rn.

55. (zotepine or Zoleptil or Nipolept).kf,tw,rn.

56. (ziprasidone or Geodon).kf,tw,rn.

57. (Zyprexa or Zolafren).kf,tw,rn.

58. or/8‐57

59. 7 and 58 

Contributions of authors

Nicolás Meza: NM
Juan VA Franco: JVAR
Reginald Rees: RR
Camila Micaela Escobar Liquitay: CMEL
Yanina Sguassero: YS
Katrina Williams: KW
Tamara Pringsheim: TP
Valeria Rojas: VR
Eva Madrid: EM

Roles and responsibilities (protocol)
Guarantor of the review	NM
Co‐ordination of the writing of the protocol	NM, JVAF
Draft the protocol (providing input across the manuscript)	NM, EM, JVAF, RR, VR, CMEL, YS, KW, TP
‐ Writing the background section	NM, RR, VR, KW, TP
‐ Writing the methods section	NM, EM, JVAF, CMEL, YS
Developing the search strategy	NM, CMEL, EM, JVAF

Sources of support

Internal sources

• Instituto Universitario Hospital Italiano de Buenos Aires, Argentina

  Provides a salary for Camila Micaela Escobar Liquitay and Juan VA Franco for the conduct of Cochrane Reviews. 

• Universidad de Valparaíso, Chile

  Provides in‐kind support for Nicolás Meza and Eva Madrid.

External sources

• No sources of support provided

Declarations of interest

Nicolás Meza: has declared that he has no conflicts of interest.

Juan VA Franco is a Contact Editor with Cochrane Urology and Managing Editor for Cochrane Metabolic and Endocrine Disorders. 

Reginald Rees: has declared that he has no conflicts of interest.

Camila Micaela Escobar Liquitay: has declared that she has no conflicts of interest.

Yanina Sguassero works for Cochrane Response part‐time as a Systematic Reviewer. She is also an Editor with Cochrane Developmental, Psychosocial and Learning Problems (DPLP) and with Cochrane Clinical Answers. 

Katrina Williams (KW) reports an ongoing grant (commencing 1 January 2020) from Epsilon Healthcare (formerly THC Global Group Ltd), to develop an interventional product and placebo for children for an MRFF‐funded phase III multisite trial, and for which the Victorian Government are providing the investigational product; paid to Murdoch Children's Research Institute. KW reports a grant (28 November 2018 to 27 November 2019) from Tilray, for a pilot trial of cannabidiol for severe behaviour problems in children with intellectual disability, with or without autism, and for which Tilray provided the investigation drug and placebo; paid to Murdoch Children's Institute. KW also reports a grant (1 June 2020 to 31 May 2023) from the National Health and Medical Research Council (NHMRC), on which she is named Chief Investigator, for a phase III trial of cannabidiol for severe behaviour disorder in children with an intellectual disability, with or without autism; paid to institutions (Monash University, Murdoch Children's Research Institute and Sydney Children's Hospital Network). KW reports being involved in an ongoing study about predictors of autism outcome that will also publish diagnostic stability outcomes, which could be eligible for inclusion in this systematic review; the study is funded by the NHMRC. KW is an Editor with DPLP. Lastly, KW reports being a member of the data monitoring committee for a trial for SSRIs for restricted and repetitive behaviours; unpaid position.

Tamara Pringsheim works as a Neurologist at the Alberta Children's Hospital and Foothills Medical Centre, Canada. She reports declaring opinions on the topic in her academic work at the Department of Clinical Neurosciences, University of Calgary, Canada. 

Valeria Rojas works as a Child Neurologist at the Dr Gustavo Fricke Hospital and the Universidad de Valparaíso, Chile.

Eva Madrid is a contributor with the Cochrane Sustainable Healthcare Group and has published opinions on the topic, most recently in the following Cochrane Editorial: Clarke M, Born K, Johansson M, Jørgensen KJ, Levinson W, Madrid E, et al. Making wise choices about low‐value health care in the COVID‐19 pandemic. Cochrane Database of Systematic Reviews 2021, Issue 9. Art. No.: ED000153. DOI: 10.1002/14651858.ED000153.

New