Efficacy of very early interventions on neurodevelopmental outcomes for infants and toddlers at increased likelihood of or diagnosed with autism: A systematic review and meta‐analysis

Abstract

The aim of this systematic review was to determine the efficacy of very early interventions for infants and toddlers at increased likelihood of or diagnosed with autism for autism symptomatology, developmental outcomes and/or neurocognitive markers. Eight databases were searched (14 April 2022) with inclusion criteria: (i) RCTs with care as usual (CAU) comparison group, (ii) participants at increased likelihood of or diagnosed with autism and aged <24 months corrected age (CA), (iii) parent‐mediated and/or clinician directed interventions, and (iv) outcome measures were autism symptomatology, cognition, language, adaptive skills, or neurocognitive assessments (EEG and eye tracking). Quality was assessed using Risk of Bias 2 and GRADE. Nineteen publications from 12 studies reported on 715 infants and toddlers. There was low to moderate certainty evidence that clinician‐assessed outcomes did not show significant treatment effects for: autism symptomatology (ADOS CSS: MD −0.08, 95% CI −0.61, 0.44, p = 0.75), cognitive outcome (Mullen Scales of Early Learning‐Early Learning Composite (MSEL‐ELC): SMD 0.05, 95% CI −0.19, 0.29, p = 0.67), receptive language (MSEL—Receptive Language: SMD 0.04, 95% CI −0.21, 0.3, p = 0.74) or expressive language (MSEL‐Expressive Language: SMD 0.06, 95% CI −0.1, 0.23, p = 0.45). Neurocognitive outcomes (EEG and eye tracking) were heterogeneous, with inconsistent findings. There is low to moderate certainty evidence that very early interventions have limited impact on neurodevelopmental outcomes by age 3 years.

Lay summary

Earliest possible diagnosis and intervention for infants and toddlers with autism are widely recommended to harness early brain plasticity and improve developmental outcomes. We have low to moderate certainty evidence that very early interventions for infants and toddlers at increased likelihood or diagnosed with autism, commencing within the first 2 years of life, have little‐to‐no impact on autism symptomatology, cognition or language outcomes by age 3 years.

INTRODUCTION

The past decade has seen significant progress in early autism research. There has been a dramatic rise in the number and rigor of randomized controlled trials (RCTs) (Green & Garg, 2018), research to facilitate earlier detection and accurate diagnosis (Baranek et al., 2022; Barbaro et al., 2022; McPartland et al., 2020; Wolff et al., 2018), and a proliferation of prospective studies exploring early developmental and neurocognitive trajectories of infants at increased likelihood of autism (Jones et al., 2019; Szatmari et al., 2016). For these young infants at increased likelihood of autism, there has also been an increase in “pre‐emptive,” or very early interventions beginning prior to diagnosis. The paradox is that as intervention science and methodology have become more rigorous, the efficacy of earlier support is less certain (Lord et al., 2022; Sandbank, Bottema‐Beutel, & Woynaroski, 2021).

Several large, rigorous systematic reviews with meta‐analyses of early interventions for children with autism aged 0–8 years have demonstrated limited impacts of early interventions on child developmental outcomes (Bottema‐Beutel, 2020; Crank et al., 2021; Rodgers et al., 2020; Sandbank, Bottema‐Beutel, Crowley, Cassidy, Dunham, et al., 2020; Sandbank, Bottema‐Beutel, Crowley, Cassidy, Feldman, et al., 2020; Sandbank, Chow, et al., 2021). A Project AIM meta‐analysis found that for RCTs using measures without detection bias, no intervention type was associated with significant effects on any child developmental outcome (Sandbank, Bottema‐Beutel, Crowley, Cassidy, Dunham, et al., 2020). A further meta‐analysis of language outcomes by the Project AIM group demonstrated small but significant effects on expressive and composite language outcomes, but quality concerns mitigated these findings (Sandbank, Bottema‐Beutel, Crowley, Cassidy, Feldman, et al., 2020). Significant concerns regarding methodological rigor were also highlighted in their meta‐analysis of Naturalistic Developmental and Behavioral Interventions (NDBI), particularly in the reliance on outcome measures with high detection bias (Crank et al., 2021). A subsequent individual participant data meta‐analysis (IPD‐MA) of all early intensive applied behavior analysis‐based interventions for young autistic children compared with treatment‐as‐usual or eclectic interventions found small benefits for adaptive skills and cognition, but all included studies were rated as having serious risk of bias (Rodgers et al., 2020). A recent systematic review and meta‐analysis of parent mediated interventions for infants at increased likelihood of autism rated most studies as low risk of bias and found no significant differences on summary effect sizes for any child developmental outcomes (Hampton & Rodriguez, 2021). To date, no systematic reviews have used standardized measures of certainty in their published findings, as recommended by the World Health Organization (WHO), National Institute for Health Care Excellence (NICE) and National Health and Medical Research Council (NHMRC). This is considered critical to the interpretation of results of systematic reviews and the quality, or certainty, of the evidence, as well as for the translation of evidence into clinical practice guidelines.

The Lancet Commission on the future of care and clinical research in autism recommended that intervention research should focus on ‘what intervention and support strategies are effective for whom and when, and which interventions lead to changes beyond their proximal outcomes’; in other words mediation, moderation and distal outcomes (Lord et al., 2022). Hampton and Rodriguez (2021) found that while the studies fell under two different theoretical frameworks (Developmental and Naturalistic Developmental and Behavioral Interventions (NDBI)), many of the targets and strategies were similar. In their meta‐analysis, effect sizes were pooled for four specific domains: receptive language (Hedge's g − 0.021, p = 0.91), expressive language (Hedge's g 0.032, p = 0.82), social communication (Hedge's g 0.075, p = 0.55) and autism symptomatology (Hedge's g 0.114, p = 0.31). They found that very early interventions were associated with increased parental responsiveness and implementation of support strategies, however these proximal changes were not associated with subsequent distal outcomes in any of the four domains. Project AIM meta‐analyses found that summary effect sizes were larger for proximal than distal outcomes (Sandbank, Bottema‐Beutel, Crowley, Cassidy, Dunham, et al., 2020), though not when limited to language outcomes (Sandbank, Bottema‐Beutel, Crowley, Cassidy, Feldman, et al., 2020). In further moderation analyses, no association was found between cumulative intervention intensity (Bottema‐Beutel, 2020; Sandbank, Bottema‐Beutel, Crowley, Cassidy, Feldman, et al., 2020), intervention type, autism symptomatology and age of the child and outcomes (Sandbank, Bottema‐Beutel, Crowley, Cassidy, Feldman, et al., 2020).

The use of neuroimaging and neurocognitive biomarkers, such as magnetic resonance imaging (MRI), electro‐encephalography (EEG), eye tracking and functional near infra‐red spectroscopy (fNIRS), has grown within autism research (Jones et al., 2019; Webb et al., 2019). These have also been incorporated into RCTs of early intervention as a means of elucidating underlying neural mechanisms that may precede or elude observable developmental changes, particularly given the limitations of behavioral observations and lack of sensitivity to changes arising as a result of treatment effects (Anagnostou et al., 2015; Provenzani et al., 2020). Treatment responsive biomarkers may also improve understanding of neural‐to‐behavioral mechanisms (Green & Garg, 2018). There have been no systematic reviews to date that have included neurocognitive markers in their findings.

The aim of this study was to systematically review the literature to determine whether interventions delivered in the first 2 years of life, for infants at high likelihood of or diagnosed with autism, impact upon child neurodevelopmental outcomes. This was in order to address the questions as to (i) what interventions are effective for (ii) which infants and toddlers, (iii) within the earliest window of neuroplasticity, and (iv) lead to changes in the distal outcome measures of autism symptomatology, language, cognition, adaptive skills and neurocognitive markers of EEG and eye tracking, compared to community care‐as‐usual. A further secondary aim was to assess the certainty of our outcomes using GRADE.

METHODS

The Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) (Liberati et al., 2009) were followed. The protocol was registered in the Prospective Register of Systematic Reviews (CRD42020158688).

Search strategy

A systematic literature search was performed by searching the following databases by AM until 14 April 2022: Cochrane Central Register of Controlled Trials (CENTRAL), PubMed, PsycInfo, CINAHL and EMBASE. The search terms included: autis* OR asperger* OR pervasive development* AND parent* OR intervention OR therapy OR therapies OR treatment OR behavior* OR behavior* AND randomized OR randomized OR trial AND infant OR toddler OR child* OR baby OR babies OR newborn. As previous reviews have reported that the majority of intervention studies were conducted in the USA, the Clinical Trials register ClinicalTrials.gov was also searched. Protocols that matched the search terms were explored, including US National Institute of Health (NIH) grant number/s and compared to the publication records. A search of the International Standard Randomized Controlled Trial Number (ISRCTN) register was performed. References of all included studies and relevant systematic reviews were examined.

Selection criteria

Studies that met the following selection criteria were included: (1) randomized controlled trial (RCT) published in a peer review journal or as a thesis, in order to capture all published and unpublished trial data, up until April 14, 2022 with no limits on language, (2) study participants were infants and toddlers recruited at an average age less than or equal to 24 months, with either diagnostic ascertainment of autism, increased likelihood of autism based on the presence of a first degree relative with autism, or showing early signs of autism on validated screening tool, (3) with a care‐as‐usual comparison group, (4) utilizing any psychological or psychosocial intervention including programs delivered by the clinician or via the parent, (5) infant outcomes included either autism diagnosis/symptomatology, language, cognition, adaptive skills or neurocognitive markers of EEG and eye tracking. Studies were limited to <24 months corrected age (CA) to focus on studies optimizing the earliest provision of intervention and period of presumed greatest neuroplasticity (Webb et al., 2014; Zwaigenbaum et al., 2015). Both infants with increased likelihood of autism and toddlers diagnosed with autism were included, as the method of diagnostic ascertainment of autism was similar enough to the ascertainment of likelihood status across the literature that they did not form two distinct categories. For example, the Autism Diagnostic Observation Schedule (ADOS) was used to identify infants and toddlers in both groups. Neurocognitive biomarkers were limited to EEG and eye tracking based on an initial scoping review.

Study selection was performed independently by two reviewers (AM, LT) using Covidence (Covidence systematic review software, n.d.). Reasons for exclusion were reached by consensus between reviewers and documented (See Supplement 1 for PRISMA diagram; see Supplement 2 for list of excluded studies). Where consensus was not reached a third independent reviewer was consulted (KW).

Data extraction

Data was extracted by two reviewers (AM, LT). Information extracted from each included study on (1) characteristics of study (year, country, and number of participants), (2) characteristics of trial participants (including age, sex, autism status and method of classification), (3) type of intervention (including type, dose, duration and method of delivery of program) and (4) type of outcome measure; autism symptomatology, autism outcome, cognition, language and adaptive skills outcomes, using validated tools such as ADOS (Lord et al., 2000), Mullen Scales of Early Learning (MSEL) receptive and expressive language scales (Mullen, 1995) and Vineland Adaptive Behavior Scales (VABS) (Sparrow et al., 2005). Two authors (AM, LT) independently extracted unadjusted post‐test outcome data for all behavioral outcome domains; autism symptomatology, cognition, receptive and expressive language and adaptive skills. Where data was incomplete, corresponding authors were contacted for the additional data. One author (AM) extracted neurocognitive outcome data, which was checked by a second author (KW).

Data synthesis

Autism symptomatology, cognition, language and adaptive skills outcomes were assessed using continuous variables and were summarized using means, standard deviations and 95% confidence intervals for the intervention and control groups.

RevMan 5.4 (Review Manager (RevMan), 2020) was used for data analysis. A synthesis of neurocognitive outcomes was performed by one author (AM) and checked by a second author (KW).

Risk of bias and GRADE certainty of evidence

Cochrane Risk of Bias 2 tool (Sterne et al., 2019) was used by three reviewers (AM, KW and JB) to assess methodological quality of outcome measures included in the meta‐analysis. GRADE (GRADE handbook for grading quality of evidence and strength of recommendations, 2013) was used by three reviewers (AM, KW and JB) to assess the certainty of evidence and entered into GRADEpro (Evidence Prime, 2020) to produce Summary of Findings tables for child development outcomes and neurocognitive biomarkers. Supplement 7 contains additional information for the reader on RoB2 and GRADE.

RESULTS

A total of 4654 articles were identified from our searches and screened by title and abstract by two reviewers independently (AM and LT). One hundred and twenty‐seven papers were retrieved for full text review independently by two reviewers (AM and LT), of which 19 met full inclusion criteria (see PRISMA diagram, Supplement 1; excluded studies, Supplement 2). Seventeen studies were published in peer reviewed journals (Aaronson et al., 2021; Baranek et al., 2015; Carter et al., 2011; Dawson et al., 2010; Dawson et al., 2012; Drew et al., 2002; Estes, Munson, et al., 2015; Green et al., 2015; Green et al., 2017; Jones et al., 2017; Kasari et al., 2014; Rogers et al., 2012; Rogers et al., 2019; Watson et al., 2017; Whitehouse et al., 2019; Whitehouse et al., 2021; Yoder et al., 2020) while two were published as PhD theses (Hartford, 2010; Sullivan, 2012). One study was published as a conference proceeding only (Gulsrud et al., 2020); the corresponding author was contacted but post‐intervention data was unavailable and was therefore excluded from the review. These 19 articles were derived from 12 unique studies (Baranek et al., 2015; Carter et al., 2011; Dawson et al., 2010; Drew et al., 2002; Green et al., 2015; Hartford, 2010; Jones et al., 2017; Kasari et al., 2014; Rogers et al., 2019; Sullivan, 2012; Watson et al., 2017; Whitehouse et al., 2019; Yoder et al., 2020). The RCT by Dawson et al., 2010 had three additional follow up articles, with one at a later time point (Estes, Munson, et al., 2015) and two reporting neurocognitive outcome measures (Aaronson et al., 2021; Dawson et al., 2012), as well as a PhD thesis that performed a mediation analysis of the results of the original study (Sullivan, 2012). Studies led by Green et al. (2015) and Whitehouse et al. (2019) both had an additional paper for a later follow up on the same group of participants (Green et al., 2017; Whitehouse et al., 2021) and one RCT (Rogers et al., 2019) was a multi‐site collaboration and expansion of the initial Parent‐Early Start Denver Model (P‐ESDM) study (Rogers et al., 2012). Only two studies reported outcomes after the age of 39 months (Estes, Munson, et al., 2015; Rogers et al., 2019), with oldest age at follow up 73 months (Estes, Munson, et al., 2015). Post‐intervention participant data was extracted from 17 studies; see Table 1.

TABLE 1.

Study characteristics

Study	N	Child characteristics					Family characteristics		Intervention characteristics
		Age in months
		Mean	SD	Range	% male	ASD status (method of determination)	% non‐Caucasian	% some college education	Study location	Intervention	Type	Duration (months)
Drew et al. (2002)	24	22.5	3.4	–	79	ASD (ADI‐R, 2 expert clinicians)	Not reported	Not reported	UK	Soc‐prag joint attention	NDBI	12
Hartford (2010)	13	–	–	12–14	–	EST (FYI)	33	100	US	RT	NDBI	6
Dawson et al. (2010)	48	23.9	3.9	18–30	–	ASD (ADI‐R, ADOS)	Not reported	Not reported	US	ESDM	NDBI	24
Dawson et al. (2012) a	29
Sullivan (2012) a	48
Estes, Munson, et al. (2015) a	39
Aaronson et al. (2021)ª	20
Carter et al. (2011)	62	20.25	2.6	15–25	82	EST (STAT and/or met criteria for ASD based on expert clinician)	53	84	US	HMTW	Dev	3.5
Rogers et al. (2012)	98	21	3.5		78	EST (ADOS‐T and 2 expert clinicians)	27	57	US	P‐ESDM	NDBI	27
Rogers et al. (2019) b	118	20.6	3.37		78	ASD (ADOS, 2 expert clinicians, DSM‐IV)	33	Av mat education 15 yrs	US	P‐ESDM then ESDM	NDBI	27
Kasari et al. (2014)	66	22.56	3.48		79	EST (CHAT, CSBS‐DP)	62	93	US	FPI	Dev	3
Baranek et al. (2015)	16	14		13–17	88	EST (FYI) ± HLS	19	88	US	ART	NDBI	6–8
Green et al. (2015)	54	9.1	0.8		52	HLS	26	46% with at least college degree	UK	iBASIS‐VIPP	Dev	5
Green et al. (2017) c	53
Watson et al. (2017)	87	13.8	0.71	13–15	69	EST (FYI)	31	84	US	ART	NDBI	6–8
Jones et al. (2017)	33	9 d			64	HLS	12	Not reported	US	PFR	Attachment	2.5
Whitehouse et al. (2019)	97	12.4	1.93		68	EST (SACS‐R)	Not reported	60% with college degree	AUS	iBASIS‐VIPP	Dev	5
Whitehouse et al. (2021) e	89
Yoder et al. (2020)	97	14	2		56	HLS	23	93	US	ImPACT	NDBI	3
Total	715

^a Follow on studies from Dawson et al., 2010.

^b Extension of Rogers et al. (2012).

^c Follow on from Green et al. (2015).

^d Initial assessment at 6.4 m but intervention commenced at 9 months.

^e Follow on from Whitehouse et al. (2019).

Abbreviations: ADI‐R, Autism Diagnostic Intervention‐Revised; ADOS, Autism Diagnostic Observation Scale; AOSI, Autism Observation Scale in Infants; ASD, autism spectrum disorder; CSBS‐DP, Communication and Symbolic Behavior Scales‐Developmental Profile; DSM‐IV, Diagnostic and Statistical Manual, fourth edition; ESDM, Early Start Denver Model; EST, early symptomatic toddler; FYI, First Years Inventory; HLS, high likelihood sibling; HMTW, Hanen's More Than Words; iBASIS‐VIPP, Intervention in British Autism Study of Infant Siblings‐Video Interaction to Promote Positive Parenting; M‐CHAT, Modified CHecklist for Autism in Toddlers; P‐ESDM, Parent‐Early Start Denver Model; SACS‐R, Social Attention and Communication Surveillance‐Revised.

The study sample sizes varied widely, from n = 13 (Hartford, 2010) to n = 118 infants (Rogers et al., 2019), with mean age at recruitment ranging from 6 months (Jones et al., 2017) to 23.9 months (Dawson et al., 2010). Most studies were undertaken in the USA, with the exception of two studies in the UK (Drew et al., 2002; Green et al., 2015) and one in Australia (Whitehouse et al., 2019). All studies were undertaken within the university or hospital setting. Sociodemographic variables were inconsistently reported. Most studies recruited families where the majority identified as White ethnicity and mothers had post‐secondary school levels of education. Nine studies recruited infants at increased likelihood of autism due to showing early signs on standardized screening tools such as the First Years Inventory (FYI) (Reznick et al., 2007), Modified Checklist for Autism in Toddlers (M‐CHAT) (Robins et al., 2001), ADOS (Lord et al., 2000), and Social Attention and Communication Surveillance‐Revised (SACS‐R, (Barbaro et al., 2022)) (Baranek et al., 2015; Carter et al., 2011; Hartford, 2010; Kasari et al., 2014; Watson et al., 2017; Whitehouse et al., 2019) or by recruiting prospective baby siblings of older children with autism (Green et al., 2015; Jones et al., 2017; Yoder et al., 2020). Three studies recruited infants with an early diagnosis of autism at mean ages 20.6 months (Rogers et al., 2019), 22.5 months (Drew et al., 2002) and 23.9 months (Dawson et al., 2010). There was substantial overlap between measures, such as use of ADOS and expert clinician consensus, to screen toddlers as included for at‐risk studies due to showing signs of autism (Carter et al., 2011; Rogers et al., 2012), as well as those that recruited toddlers diagnosed with autism (Dawson et al., 2010; Drew et al., 2002; Rogers et al., 2019). There was also a substantial overlap in toddler ages for the five above‐mentioned studies, with ages between 20 and 24 months. Characteristics of the included studies are reported in Table 1. For additional characteristics of interventions as well as therapies received by Community Care‐As‐Usual, see Table 1, Supplement 3.

A number of different types of autism specific interventions were trialed, with the majority being NDBI: Early Start Denver Model [ESDM] (Dawson et al., 2010; Rogers et al., 2019), Improving Parents as Communication Teachers [ImPACT] (Yoder et al., 2020), variations on Adapted Responsive Teaching [ART] (Baranek et al., 2015; Hartford, 2010; Watson et al., 2017) and a social‐pragmatic joint attention focused parent training program (Drew et al., 2002). Three were Developmental interventions; Intervention in British Autism Study of Infant Siblings—Video Interaction to Promote Positive Parenting [iBASIS‐VIPP] (Green et al., 2015; Whitehouse et al., 2019), Focused Playtime Intervention (Kasari et al., 2014) and Hanen's More Than Words (Carter et al., 2011). Promoting First Relationships [PFR] (Jones et al., 2017) is an attachment‐based intervention to promote social–emotional development that had previously been trialed in vulnerable families in contact with statutory child protection agencies. A number of the studies included the developers of the interventions as lead researchers in the RCT of the interventions (Dawson et al., 2010; Green et al., 2015; Kasari et al., 2014) or in the replication of the initial RCT across different sites (Rogers et al., 2019; Whitehouse et al., 2019), reflecting the emerging nature of this field of research.

Most interventions were parent‐mediated, with only the two ESDM studies using clinicians to deliver most of the program content. The duration and intensity of programs varied widely, ranging from 3 months (Carter et al., 2011; Yoder et al., 2020) to 2 years (Dawson et al., 2010; Rogers et al., 2019). All of the parent mediated programs were of low intensity (0.5–2 sessions per week with therapist supporting the parent) and dose of parent practice with infant varying from 15 mins per day (Whitehouse et al., 2019) to 7 h per week (Drew et al., 2002), with only four studies reporting the amount of parent practice time (Dawson et al., 2010; Drew et al., 2002; Whitehouse et al., 2019; Yoder et al., 2020). In comparison, the two ESDM studies used an intensive approach, with infants receiving 15–20 h per week of clinician delivered program content in addition to parent practice over 2 years (Dawson et al., 2010; Rogers et al., 2019) (see supplement 3, Table 1).

Child development outcomes measures

Autism symptomatology, cognition, receptive and expressive language outcomes were measured via clinician assessed and parent‐reported measures (see Tables 2–7, Supplement 3). Additional data received from four authors of seven trials was included (Rogers et al., 2019; Watson et al., 2017; Whitehouse et al., 2019; Yoder et al., 2020). Contact details were not found for one thesis (Hartford, 2010); therefore that data was unavailable. Measures included in the meta‐analysis were taken from the 30 to 39 month timepoints to improve comparison of measures across time, unless study follow up concluded prior to this age (Watson et al., 2017; Yoder et al., 2020). For studies with infants at increased likelihood of autism, later autism diagnostic outcomes were extracted and reported (Table 4, Supplement 3).

Autism symptomatology

Six trials including 451 participants contributed to the meta‐analysis for clinician assessed autism symptomatology (ADOS Calibrated Severity Scale). Interventions were associated with a small but not significant reduction in autism symptomatology (mean difference [MD] = −0.08, 95% CI −0.61 to 0.44, p = 0.75; Figure 1, Supplement 4). There was limited statistical heterogeneity between studies (chi² = 5.14, df = 4, p = 0.27, I ² = 23%). Subgroup analyses were performed by age at enrolment, age at outcome assessment and by autism likelihood status however numbers of studies contributing to each subgroup were small and standardized mean differences did not reach statistical significance. Cochrane Risk of Bias 2 was conducted for studies contributing to the clinician assessed measures of autism symptomatology (Supplement 5). A sensitivity analysis was performed by RoB2 status (high vs. low risk of bias), with moderate statistical difference (chi² = 1.46, df = 1, p = 0.14, I ² = 54.8%) between groups (Figure 1, Supplement 4).

Cognitive outcomes

Eight trials including 530 participants contributed to the meta‐analysis for clinician assessed cognitive outcomes, with all trials using the Mullen Scales of Early Learning‐Early Learning Composite (MSEL‐ELC) or MSEL‐Developmental Quotient (MSEL‐DQ). Interventions were not associated with an improvement in cognitive outcomes (SMD = 0.05, 95% CI −0.19 to 0.29, p = 0.68; Figure 2, Supplement 4). There was moderate statistical heterogeneity between studies (chi² = 12.76, df = 7, p = 0.08, I ² = 45%). Subgroup analyses were performed by age at enrolment, age at outcome assessment and by autism likelihood status, however the number of studies contributing to each subgroup in these meta‐analyses were small and standardized mean differences did not reach statistical significance. Cochrane Risk of Bias 2 was conducted for studies contributing to the clinician assessed measures of cognition (Supplement 5). A sensitivity analysis was performed by RoB2 status (high vs. low risk of bias), with no statistical difference between status groups (chi² = 0.14, df = 1, p = 0.71, I ² = 0%).

Receptive language

Nine trials including 582 participants contributed to the meta‐analysis for clinician assessed receptive language using MSEL‐Receptive Language (MSEL‐RL). Interventions were not associated with an improvement in receptive language (SMD = 0.04, 95% CI −0.21 to 0.30, p = 0.74; Figure 3, Supplement 4). There was moderate statistical heterogeneity between studies (chi² = 18.07, df = 8, p = 0.02, I ² = 56%). Subgroup analyses were performed by age at enrolment, age at outcome assessment and by autism likelihood status, however the numbers of studies contributing to each subgroup in these meta‐analyses were small and standardized mean differences did not reach statistical significance. Cochrane Risk of Bias 2 was conducted for studies contributing to clinician rated measures of receptive language (Supplement 5). A sensitivity analysis was performed by RoB2 status (high vs. low risk of bias), with no statistical difference (chi² = 0.02, df = 1, p = 0.88, I ² = 0%) between groups.

Expressive language

Nine trials including 582 participants contributed to the meta‐analysis for clinician assessed expressive language using MSEL‐Expressive Language (MSEL‐EL). Interventions were not associated with an improvement in expressive language (SMD = 0.06, 95% CI −0.10 to 0.23, p = 0.45; Figure 4, Supplement 4). There was no statistical heterogeneity between studies (chi² = 7.78, df = 8, p = 0.46, I ² = 0%). Subgroup analyses were performed by age at enrolment, age at outcome assessment and by autism likelihood status however numbers of studies contributing to each subgroup were small and standardized mean differences did not reach statistical significance. Cochrane Risk of Bias 2 was conducted for studies contributing to clinician rated measures of expressive language (Supplement 5). A sensitivity analysis was performed by RoB2 status (high vs. low risk of bias), with no statistical difference (chi² = 1.10, df = 1, p = 0.29, I ² = 9.2%) between groups.

Autism diagnostic outcome

Five studies including 314 participants at high likelihood of autism contributed to the meta‐analysis for autism diagnostic outcome. Support programs did not lead to a difference in autism outcome (OR 1.05, 95% CI 0.49–2.26, p = 0.22). There was low statistical heterogeneity between studies (chi² = 5.70, df = 4, p = 0.22, I ² = 30%) (Figure 5, Supplement 4).

Adaptive skills

Seven studies including 413 participants contributed to some aspects of parent rated adaptive skills outcomes, however not every study reported all five possible domains from the VABS‐2 (Adaptive Behavior Composite [ABC], Communication, Daily Living Skills, Socialization and Motor).

Four studies with 224 participants contributed to the VABS‐2 ABC meta‐analysis. Interventions were associated with small improvements in adaptive skills as measured by VABS‐2 ABC (SMD 0.36, 95% CI 0.09–0.63, p = 0.009; SMD was calculated as one study (Rogers et al., 2019) reported Age Equivalents rather than Standard Scores, see Figure 6, Supplement 4) and daily living skills (4 studies, 195 participants, MD 3.64, 95% CI 0.17–7.12, p = 0.04). Communication and Socialization domains were more commonly reported (6 studies, 330 to 333 participants) however they did not lead to statistically significant improvements; Communication (MD 1.9, 95% CI −1.22 to 5.02, p = 0.23, see Figure 7, Supplement 4) or in Socialization (MD 2.06, 95% CI −0.6 to 4.72, p = 0.13, see Figure 8, Supplement 4). All studies were rated as high risk of bias due to detection bias (Supplement 5).

Neurocognitive biomarkers

Four studies (Aaronson et al., 2021; Dawson et al., 2012;Green et al., 2015; Jones et al., 2017) with 94 participants contributed to EEG biomarkers, including four different EEG tasks: Event Related Potentials (ERP) to social versus non‐social stimuli (Dawson et al., 2012; Jones et al., 2017), ERP auditory oddball task (Green et al., 2015), mu rhythm attenuation (Aaronson et al., 2021) and spectral analysis to social versus non‐social stimuli (Dawson et al., 2012; Jones et al., 2017). Two studies (Green et al., 2015; Jones et al., 2017) with 79 participants contributed to the eye tracking biomarkers, with two different eye tracking tasks: Gap‐Overlap task (Green et al., 2015) and habituation times to social versus non‐social stimuli (Jones et al., 2017). One study (Whitehouse et al., 2019) performed an eye tracking assessment (Gap‐Overlap task) that was not included in the protocol registration and subsequent publication, as this was supplementary to the original study. Due to heterogeneity in neurocognitive measures used, a meta‐analysis was unable to be performed. A synthesis of the extracted data was therefore undertaken according to Cochrane guidelines (Tables 8 and 9, Supplement 3). Assessment of the technical aspects of the included neurocognitive biomarkers was outside the scope of this systematic review.

Our biomarker findings were inconsistent, with some measures showing between group differences in favor of support programs, for example at one time point but not at another, and other measures showing no differences between groups (Tables 8 and 9, Supplement 3).

Cochrane RoB2 was performed for all EEG and eye tracking outcome measures (Supplement 6). All results were assessed as high risk of bias across multiple domains, apart from the Gap‐Overlap task, which was rated as low risk of bias as there was a prespecified analysis plan and few infants with missing data. The moderate effect size for change in gap‐effect in the Gap‐Overlap task, as hypothesized, was very close to statistical significance. Missing data for study participants was otherwise common, particularly for EEG outcomes, however only two studies reported that infants not contributing data (particularly those unable to tolerate or complete the assessments) differed from those who did contribute (Aaronson et al., 2021; Dawson et al., 2012).

There were significant discrepancies between prespecified protocols and publications. Only one study (Green et al., 2015) published a detailed pre‐specified analysis plan. This updated analysis plan differed significantly from the original pre‐specified analysis protocol (iBASIS Team, 2011), without explanation as to why the theoretical assumptions underlying the original measures in the analysis plan had changed. The original plan theorized that social attention biomarkers in comparison to non‐social biomarkers would be preferentially impacted by the intervention; the published analysis theorized that non‐social attention biomarkers only would be affected. Three studies did not provide pre‐specified analysis plans for the neurocognitive markers (Aaronson et al., 2021; Dawson et al., 2012; Jones et al., 2017); one of these study protocols also included the use of fMRI that was unpublished (Dawson et al., 2010). We found one study where eye tracking assessments were undertaken but not reported in the final publications, as the eye tracking measures were not part of the pre‐specified primary outcome measures, and results remain unpublished. In addition, there was one pre‐registered study in ClinicalTrials.gov where the neurocognitive measures of EEG and eye tracking were unregistered but published, while the pre‐registered behavioral observations of child development and autism symptomatology remain as yet unpublished.

GRADE

GRADE is an approach for rating the quality of evidence from a systematic review and is the most widely adopted tool used for assessing the quality or certainty in the reviewed evidence (Alonso‐Coello et al., 2016; GRADE handbook for grading quality of evidence and strength of recommendations, 2013; Guyatt et al., 2008). When reporting the results of any systematic review, GRADE posits that there are two key considerations; the effect of the intervention (reported numerically in some way) and the certainty (or confidence) we can have in the evidence for that effect (GRADE handbook for grading quality of evidence and strength of recommendations, 2013). Quality, or certainty of a body of evidence, is defined as the extent to which there can be confidence that an estimate of effect is close to the quantity of interest. The approach considers within‐study risk of bias, directness of evidence, heterogeneity, precision of effect estimates and publication bias. There are four possible levels of quality: high, moderate, low and very low certainty evidence. High certainty means “the authors have a lot of confidence that the true effect is similar to the estimated effect,” moderate certainty means that “the authors believe that the true effect is probably close to the estimated effect,” low certainty means that “the true effect might be markedly different from the estimated effect,” and very low certainty evidence that “the true effect is probably markedly different from the estimated effect” (GRADE handbook for grading quality of evidence and strength of recommendations, 2013) (see Supplement 7 for additional information on GRADE). GRADE for each outcome is reported in detail in the Summary of Findings (SoF) Tables 2 and 3.

TABLE 2.

GRADE Summary of Findings table for child development outcomes

Summary of findings:
GRADE child development outcomes
Patient or population: improving child development outcomes for infants at increased likelihood or diagnosed with autism Intervention: very early support programs Comparison: community care as usual
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with community care as usual	Risk with very early support programs
Autism symptomatology assessed with: ADOS CSS Scale from: 1 to 10 follow‐up: range 23 months to 39 months		MD 0.08 lower (0.61 lower to 0.44 higher)	–	451 (6 RCTs)	⨁⨁⨁◯ Moderate a , b	Very early support programs probably result in little to no difference (MD −0.08, 95% CI −0.61 to 0.44) in autism symptomatology as measured by ADOS Calibrated Severity Score.
Cognitive outcomes assessed with: MSEL‐ELC, MSEL‐DQ follow‐up: range 22 months to 39 months	–	SMD 0.05 higher (0.19 lower to 0.29 higher)	–	538 (8 RCTs)	⨁⨁⨁◯ Moderate a	Very early support programs probably result in little to no difference (SMD 0.05, 95% CI −0.19 to 0.29) in cognitive outcomes as measured by MSEL‐ELC.
Receptive language assessed with: MSEL‐RL raw score, T score, age equivalent follow‐up: range 22 months to 39 months	–	SMD 0.04 higher (0.21 lower to 0.3 higher)	–	582 (9 RCTs)	⨁⨁◯◯ Low a , c	Very early support programs probably result in little to no difference (SMD 0.04, 95% CI −0.21 to 0.30) in receptive language as measured by MSEL‐RL.
Expressive language assessed with: MSEL‐EL: raw score, T score or age equivalent follow‐up: range 22 months to 39 months	–	SMD 0.06 higher (0.1 lower to 0.23 higher)	–	582 (9 RCTs)	⨁⨁⨁◯ Moderate a	Very early support programs probably result in little to no difference (SMD 0.06, 95% CI −0.10 to 0.23) in expressive language as measured by MSEL‐EL.
Adaptive skills; VABS‐2 Adaptive Behavior Composite (VABS‐2 ABC) follow‐up: range 22 months to 39 months	–	SMD 0.32 higher (0.05 higher to 0.59 higher)	–	212 (4 RCTs)	⨁◯◯◯ Very low a , d , e	Very early support programs may result in a small difference in adaptive skills: VABS2 Adaptive Behavior Composite (SMD 0.32 higher, 95% CI 0.05–0.59 higher) but the evidence is very uncertain.
Adaptive skills: VABS‐2 Communication assessed with: VABS‐2 Communication domain follow‐up: range 22 months to 39 months		MD 1.9 SS higher (1.22 lower to 5.02 higher)	–	328 (6 RCTs)	⨁◯◯◯ Very low d , e	Very early support programs may result in little to no difference (MD 1.90, 95% CI −1.22 to 5.02) in communication adaptive skills as measured by the VABS‐2 Communication domain but the evidence is very uncertain.
Adaptive skills: VABS‐2 Socialization assessed with: VABS‐2 Socialization domain follow‐up: range 22 months to 39 months		MD 2.06 SS higher (0.6 lower to 4.72 higher)	–	326 (6 RCTs)	⨁◯◯◯ Very low a , d , e	Very early support programs may result in little to no difference (MD 2.06, 95% CI −0.6 to 4.72) in socialization adaptive skills as measured by VABS‐2 Socialization domain but the evidence is very uncertain.

^* The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

Abbreviations: ADOS CSS, Autism Diagnostic Observation Schedule Calibrated Severity Score; CI, confidence interval; MD, mean difference; MSEL‐DQ, Mullen Scales of Early Learning‐Developmental Quotient; MSEL‐ELC, Mullen Scales of Early Learning‐Early Learning Composite; MSEL‐EL, Mullen Scales of Early Learning‐Expressive Language; MSEL‐RL, Mullen Scales of Early Learning‐Receptive Language; RCT, Randomized Control Trial; SMD, standardized mean difference; SS, standard score; VABS‐2, Vineland Adaptive Behavior Scales Edition 2.

Note: GRADE Working Group grades of 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.

^a Confidence intervals cross both null effect and small effect size.

^b Child development outcomes were not published from at least one study.

^c Inconsistency in both direction and size of results, with some confidence intervals almost not overlapping, as well as high statistical measures of heterogeneity (I squared).

^d All studies rated as high risk of bias due to detection bias.

^e Imprecision due to low numbers.

TABLE 3.

GRADE summary of findings neurocognitive markers

Summary of findings:
Neurocognitive markers of EEG and eye tracking
Patient or population: improving neurocognitive markers in infants at increased likelihood or diagnosed with autism Intervention: very early support programs Comparison: community care as usual
Outcomes	Impact	№ of participants (studies)	Certainty of the evidence (GRADE)
EEG: ERP responses to social versus non‐social stimuli (ERP) assessed with: P1, N170, Nc, P400 response amplitude and latency follow‐up: range 18 months to 77 months	Nc showed faster peak latency for ESDM group in comparison to CAU at 49–77 months but no group differences in Nc amplitude, or amplitude and latency for P1 or N170. P400 amplitude was greater at 12 months and peak latency faster at 18 months for PFR in comparison to CAU. There were no group differences in peak amplitude and latency at other time points.	50 (2 RCTs)	⨁◯◯◯ Very low a , b , c
EEG: Spectral analysis to social versus non‐social stimuli assessed with: Natural log theta ± alpha power bands follow‐up: range 18 months to 77 months	ESDM infants showed greater cortical activation for faces vs objects whereas CAU infants showed the greater activation for objects vs faces. PFR infants showed greater increase in cortical activation for faces over time (6–12 and 6–18 months) in comparison to CAU.	59 (2 RCTs)	⨁◯◯◯ Very low a , b , c
EEG: ERP responses to auditory oddball task assessed with: P100 and P300 response amplitude follow‐up: mean 15 months	P100 response amplitude EF = 0.54, MD = 0.96, 95% CI (−0.2644, 2.1844) p = 0.1202 P300 response amplitude EF = 0.38, MD = 1.24, CI (−1.023 to 3.51) p = 0.2744	35 (1 RCT)	⨁◯◯◯ Very low b , d , e
EEG: mu attenuation to grasping action familiar vs unfamiliar person follow‐up: mean 72 months	ESDM group showed significantly greater mu attenuation during familiar observation. p < 0.05 favors intervention	20 (1 RCT)	⨁◯◯◯ Very low a , b , c , d
Eye tracking: habituation times to social vs non‐social stimuli assessed with: Habituation times (msec) follow‐up: range 12 months to 18 months	Greater increase in habituation speed to faces vs objects in PFR group at both time intervals (6–12 and 6–18 months)	26 (1 RCT)	⨁◯◯◯ Very low a , b , c
Eye tracking: Gap‐overlap task assessed with: Change in gap effect (overlap‐baseline time in msec) over time (9–15 months) follow‐up: range 15 months to 27 months	Change in gap effect between 9 and 15 months showed moderate effect size that almost reached statistical significance [EF = 0.48 (95%CI −0.01, 1.02)] No between group differences at 27 months.	50 (1 RCT)	⨁⨁◯◯ Low b , e

Note: *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

Abbreviations: CAU, Care As usual; CI, confidence interval; EEG, electro‐encephalogram; EF, effect size; ERP, event related potential; ESDM, early start denver model; MD, mean difference; PFR, promoting first relationships; RCT, randomized control trial; SMD: standardized mean difference.

Note: GRADE Working Group grades of 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.

^a High risk of bias across most or all domains.

^b Very small numbers of participants.

^c Imprecision in sizes of between group differences on reported measures.

^d Risk of bias due to missing outcome data.

^e Imprecision in effect size due to wide confidence intervals, which include no effect and moderate effect.

We found that there was moderate certainty evidence that very early interventions probably result in little to no difference in autism symptomatology, cognitive or expressive language outcomes by 3 years of age (SoF Table 2). In addition, there was low certainty evidence that very early support programs do not improve receptive language (SoF Table 2). Down grading occurred due to inconsistency, with confidence intervals crossing both the null effect and small effect sizes. We found very low certainty evidence that very early interventions result in null or small effect sizes for adaptive skills (VABS‐II ABC, Communication and Socialization) due to small numbers of participants and detection bias (SoF Table 2). We found low to very low certainty evidence that very early interventions result in positive changes in neurocognitive markers (EEG and eye tracking), due to high risk of bias, inconsistency, and small numbers of participants (SoF Table 3).

DISCUSSION

This systematic review and meta‐analysis found that on blinded clinician assessed outcome measures, there was moderate certainty evidence that very early interventions do not lead to improvements in autism symptomatology, cognitive or expressive language outcomes by 3 years of age. In addition, there was low certainty evidence that very early support programs do not improve receptive language. This is consistent with a previous systematic review (Hampton & Rodriguez, 2021) and builds on the existing literature by using a standardized measure of certainty, GRADE. The evidence for neurocognitive markers was inconsistent, commensurate with the emerging nature of treatment responsive biomarkers in early autism biomarker research. A synthesis of the available outcome measures was rated as very low certainty due to small numbers of participants and methodological concerns. This was the first systematic review to examine neurocognitive markers. Biomarkers, specifically EEG and eye tracking assessments, may ultimately provide more sensitive and objective means of detecting the neurocognitive effects of early interventions, however, more rigorous research is needed, particularly pre‐registered analysis plans and accounting for missing data, such as comparisons of infants that did and did not contribute data to the analysis. There were no studies that were undertaken outside of the university setting or in socioeconomically vulnerable families, suggesting that the trials conducted to date represent efficacy and are likely to over‐estimate real‐world effectiveness (Nahmias et al., 2019).

To our knowledge, this is the first systematic review and meta‐analysis to pool outcomes from the same assessment measure (e.g. ADOS CSS, MSEL‐ELC, MSEL‐RL, MSEL‐EL, and VABS‐II) within each domain, as opposed to pooling measures by domain (e.g., receptive language). When pooling by domain, as previous reviews have done, the heterogeneity in outcomes measures and detection bias, due to combining blinded clinician and parent rated measures, makes findings difficult to interpret and a true effect from a specific assessment tool can be masked. This review thus gives greater certainty to the overall interpretation from this and another recent review (Hampton & Rodriguez, 2021): that short to medium term interventions (3 months to 2 years) commencing between 9 and 24 months of life do not lead to improvements in more generalized developmental skills in the short term (up to 24–39 months of age, or 2 years post‐intervention). The paucity of follow up beyond 39 months, or 2 years post‐intervention, leaves the question as to whether there may be as yet undetected longer‐term gains in child development unanswered. As the average age of diagnosis remains around 4 years, follow up to 39 months of age should not be considered a complete picture (van Hof et al., 2021). Paucity of follow up could be leading to either underestimating or overestimating the efficacy of very early interventions, as effects could accumulate, persist, or dissipate over time.

While very early interventions for infants and toddlers at increased likelihood of or diagnosed with autism have shown increasingly rigorous research methodology over the past 15 years, the inclusion of neurocognitive markers was not accompanied by a similar level of rigor, particularly with regards to protocol registration and pre‐specified analysis plans. We found that trial registrations for almost all studies lacked specific mention of EEG or eye tracking assessments and only one study included a pre‐specified analysis plan for neurocognitive markers, which was changed from a focus on social attention to non‐social attention without explanation (Green et al., 2015). There may be selection bias in the reporting of neurocognitive outcome measures in the current literature. Neurocognitive markers are difficult to standardize and often have high rates of missing data, so they are also likely underreported. There is room for improvement in the methodological rigor of neurocognitive markers in early autism research.

Strengths and limitations

The main strength of this systematic review is in its methodological rigor. We registered our review in PROSPERO and undertook extensive searches including clinical trial registries and sought unpublished data, which also allowed for comparison of pre‐specified analysis plans in comparison to reported findings. We included only RCTs and pooled effects by homogenous assessments rather than combining different outcomes measures, used the Cochrane RoB2 tool to assess risk of bias at the level of individual outcomes and provided an examination of the certainty of our findings using GRADE. We hope that by demonstrating the use of this methodology, that the use of these methods and tools will become more widely adopted in systematic reviews of autism interventions. Subsequent GRADE findings can then be utilized by policy makers in the Evidence‐to‐Decision making process to help inform Clinical Practice Guidelines for recommendations regarding autism interventions.

The present review was limited by the number of available studies and inability to access all unpublished data, which may have allowed more comprehensive subgroup analyses and greater precision in effect sizes. The ADOS CSS, MSEL, and VABS‐2 may lack sufficient sensitivity to change in terms of distal changes in child development. However it should be acknowledged that the ADOS CSS did show change in response to the two iBASIS intervention studies and other research (Carruthers et al., 2021). The standardized assessment process itself may be more challenging for autistic children to demonstrate their capabilities than for typically developing children (Akshoomoff, 2006), for example autistic children showed more off‐task behaviors in the MSEL assessment in comparison to typically developing children (Akshoomoff, 2006). This is likely true of most clinician administered developmental assessments and is also consistent with the difficulties that autistic children may experience in generalizing skills outside the intervention context (Carruthers et al., 2020). These factors may help to explain why very early interventions may lead to detectable change in parent–child interaction (which are context‐bound (Sandbank, Chow, et al., 2021)) but not translate into change in child developmental outcomes (Hampton & Rodriguez, 2021; Yoder et al., 2020). This highlights the importance of developing novel assessments that are rigorous, measure distal and/or generalized skills, and are sensitive to change.

Moderating factors such as intervention type and content, intensity of programs, earlier age of commencement, parent versus clinician mediated, baby sibling likelihood status (as lower pre‐test probability of later diagnosis of autism) versus showing signs or autism diagnosis, and female sex may still potentially have an influence on the efficacy of very early interventions, despite lack of evidence in older children for factors such as age of child or intensity of intervention. A more comprehensive analysis of moderating variables in neurocognitive markers was not possible and may be a future means of better understanding which interventions are most appropriate for different infants.

Future research

Although this review found limited effects of very early intervention on child developmental outcomes, the relative success of one intervention, iBASIS‐VIPP, for the alteration of autism symptoms in two rigorous RCTs suggests that parental sensitivity and responsiveness remain important early social intervention targets and that intervention for infants at increased likelihood of autism, either due to baby‐sibling status or showing early signs, may yet hold promise. The specific mechanisms of the iBASIS intervention that led to change in sensitive parental responding, parent–child interaction, and child development, in comparison to other very early interventions and intervention targets, are not yet known. In addition to parental responsiveness, new earlier targets, including non‐social targets and alterations in the infants' non‐social environment, may need to be explored to improve the ability of early interventions to impact on distal child developmental outcomes (Charman, 2019; Green & Garg, 2018; Kasari, 2019). The programs tested to date all commenced after 6 months, which may commence too late within the developmental trajectories of infants with a later diagnosis of autism, given the fact that a range of studies and modalities (Tanner & Dounavi, 2021) suggest that atypical development emerges within the first 6 months of life. Parental concerns (Sacrey et al., 2015), changes in parent‐infant interaction (Wan et al., 2012), motor skills (Choi et al., 2018; Estes, Gu, et al., 2015; LeBarton & Landa, 2019), visual reception (Estes, Gu, et al., 2015), visual attention (Jones & Klin, 2013; Shic et al., 2014), temperament and regulation (Clifford et al., 2013; del Rosario et al., 2014; Paterson et al., 2019), and MRI differences (Emerson et al., 2017; Scheinost et al., 2022; Shen et al., 2022) have all been documented before or emerging from 6 months. Shifting support to commence within the first 6 months of life may allow for a shift in intervention targets to the earliest autism signs, including attention, gross motor skills and visual reception (Harker et al., 2016; Wan et al., 2012) as well as early life context such as prenatal stress (Gliga et al., 2014; Whittingham et al., 2020), postnatal sensorimotor processing and the early postnatal environment (Gliga et al., 2014; Shen et al., 2022; Whittingham et al., 2020). There are currently two RCTs of interventions commencing antenatally underway, ENACT (Whittingham et al., 2020) and Baby AICES (ACTRN12620000019909, 2020) and a third yet to commence recruitment (NCT05104112), which provide the earliest possible window for harnessing neuroplasticity. Regardless of the timing of intervention, it is important to test the inclusion of non‐social and non‐language based targets such as attention shifting, gross or fine motor skills, regulation and sensory responsiveness. In addition, interventions may need to be more effectively individualized to the child (Green et al., 2022) and modified according to parental needs, for example around mental health.

As the genetic, neurodevelopmental and environment mechanisms of autism are better elucidated, a wider range of targets within the developmental cascade to autism should emerge, including those relevant to parents, clinicians, and the autistic community (Bal et al., 2018; Green & Garg, 2018). For now, the goals should be to explore earlier intervention possibilities as part of a sequential, individualized and flexible approach to supportive care, integrating new potential targets grounded in very early emerging developmental differences, and maintaining a focus on enhancing parental responsiveness. In addition, longer term follow‐up, to the age of 5 years, is urgently required, for greater diagnostic certainty, and to allow for the detection of effects that may accumulate, persist or dissipate with time.

CONCLUSION

Interest in very early interventions for infants and toddlers at high likelihood of or diagnosed with autism has led to a significant increase in methodologically rigorous RCTs and more standardized assessment protocols, including blinded clinician assessments and neurocognitive biomarkers (Simonoff, 2018). The evidence from rigorous RCTs to recommend very early interventions for alteration of early developmental trajectories and distal outcome measures of child development remains limited. Apart from two ESDM studies, the existing literature does not examine outcomes beyond 3 years of age. Overall, this review suggests that although important work has been done, the future gold‐standard very early intervention for infants and toddlers at increased likelihood of or diagnosed with autism is yet to be developed and will include novel and individualized intervention targets alongside the targeting of parental responsiveness.

FUNDING INFORMATION

AM was supported by a University of Queensland Graduate School Scholarship and Children's Hospital Foundation Top‐Up Grant. RB was supported by an Australian National Health and Medical Research Council Research Fellowship (1105038). These funding bodies had no role in study design; in the collection, analysis and interpretation of data; in the writing of the systematic review; and in the decision to submit the article for publication.

CONFLICT OF INTEREST STATEMENT

Dr Andrea McGlade, Dr Koa Whittingham, Dr Jacqui Barfoot and Prof Roslyn Boyd are key investigators in the ENACT trial, a very early support program for infants at high likelihood of autism. The ENACT support program was developed by Dr Koa Whittingham and Dr Andrea McGlade. ACTRN12618002046280.

Supporting information

Supplementary 1: PRISMA 2009 Flow Diagram.

Supplement 3.

Supplementary 4. Forest plots for clinician rated outcome measures.

Supplementary 7. Additional information on RoB2 and GRADE tools.

Supplementary 2. Excluded references.

Supplementary 5. RoB2 child development outcomes.

Supplementary 6. RoB2 neurocognitive markers.

McGlade, A. , Whittingham, K. , Barfoot, J. , Taylor, L. , & Boyd, R. N. (2023). Efficacy of very early interventions on neurodevelopmental outcomes for infants and toddlers at increased likelihood of or diagnosed with autism: A systematic review and meta‐analysis. Autism Research, 16(6), 1145–1160. 10.1002/aur.2924

DATA AVAILABILITY STATEMENT

The data that supports the findings of this study are available in the supplementary material of this article.