Early Start Denver Model effectiveness in young autistic children: a large multicentric randomised controlled trial in two European countries

Abstract

Background

Evidence regarding early interventions based on the Naturalistic Developmental Behavioral Interventions framework, such as the Early Start Denver Model (ESDM), suggests efficacy for autistic children. However, the effectiveness of ESDM across diverse cultural contexts remains under-researched.

Objective

To assess the effectiveness of ESDM compared with treatment as usual (TAU) on overall development in young children with autism spectrum disorder (ASD).

Method

This parallel, randomised controlled trial, using a modified Zelen design, was conducted in five Child and Adolescent Mental Health Services in France and Belgium. A total of 180 children aged 19–36 months, who met autism criteria and were referred by community professionals, were randomly assigned to either receive 12-hour weekly ESDM+TAU or TAU alone. The primary outcome was the change in developmental quotient (DQ) on the Mullen Scale of Early Learning, assessed blindly from baseline to 24 months post randomisation.

Findings

From September 2015 to March 2019, 180 children were randomly assigned to either ESDM+TAU (n=61, girls=21.7%) or TAU alone (n=119, girls=15.4%). Three children dropped out immediately after randomisation. Compared with TAU alone, children in the ESDM+TAU group did not significantly improve global DQ (endpoint mean difference 3.82 (95% CI −1.25 to 8.89), p=0.14).

Conclusions and clinical implications

Our findings suggest that ESDM+TAU cannot be universally recommended for young children with ASD. Further research is required to evaluate the long-term effectiveness of ESDM and identify subgroups that may benefit more, thereby guiding optimal implementation strategies.

Trial registration number

NCT02608333.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Naturalistic Developmental Behavioral Interventions (NDBI), such as the Early Start Denver Model (ESDM), demonstrate the strongest evidence of efficacy in autism, particularly in social communication.

WHAT THIS STUDY ADDS

• For the first time in France and Europe, this study evaluates the ESDM, an NDBI, in a new cultural context, across five independent Child and Adolescent Mental Health Services.

• The trial, conducted according to high-quality randomised controlled trial standards, includes a large sample of 180 community-referred autistic children.

• The ESDM did not show significantly greater effectiveness than standard care in improving overall child development outcomes.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Our findings suggest that ESDM combined with treatment as usual cannot be universally recommended for all young children with autism spectrum disorder (ASD).

• Further research is needed to assess the long-term effectiveness of ESDM and to identify subgroups of children with ASD who may derive specific benefits, thus informing more targeted and optimal implementation strategies.

Introduction

Autism diagnosis is associated with core challenges in social communication, restricted and repetitive interests and behaviours, and differences in sensory function. Individuals with autism spectrum disorder (ASD) exhibit a broad spectrum of associated difficulties such as language, intellectual development and coping with everyday life.¹ Pharmacological approaches may alleviate symptoms such as hyperactivity or irritability but not the core symptoms of autism.² Early non-pharmacological interventions may improve both, although evidence remains insufficient to support widespread implementation for all young children with ASD.

Among the wide range of early intervention approaches for ASD, including sensory-based, animal-assisted, cognitive–behavioural or developmental methods, Naturalistic Developmental Behavioral Interventions (NDBIs) approaches merge strategies from behavioural and developmental learning theories, offering a comprehensive framework to address developmental milestones.³ A recent meta-analysis assessed the effect of non-pharmacological treatments in randomised controlled trials (RCTs) previously conducted in children younger than 8 years of age. When focusing exclusively on outcomes not derived from caregiver reports—to avoid placebo-by-proxy bias—significant improvements in social communication were observed only for developmental interventions (Hedges’ g=0.31, 95% CI=0.13 to 0.49; p=0.003) and NDBI (95% CI=0.36, 0.23 to 0.49; p<0.001). Additionally, NDBI demonstrated a significant effect on core autism symptoms (95% CI=0.44, 0.20 to 0.68; p=0.002). When the meta-analysis was restricted to RCTs with a low risk of detection bias, only NDBIs showed efficacy for core autism symptoms.⁴

The same research team previously investigated whether certain characteristics of interventions or populations were associated with the effectiveness of NDBIs.⁵ Neither cumulative intervention intensity nor the type of interventionist (parents or professionals) was significantly related to the effect of NDBI. Additionally, no moderating effects were found for the mean chronological age (B=−0.06, p=0.256), mean language age of the sample at study entry (B=−0.02, p=0.758), or percentage of boys in each sample (B=−0.06, p=0.204). Due to the insufficient number of participants with severe autism symptomatology, the authors were unable to conduct a moderation analysis on autism severity. Recent studies have highlighted that children with severe or profound autism (representing 35.5% of ASDs according to Hughes et al) are often implicitly or explicitly excluded from the studies.⁶ Furthermore, most RCTs have been conducted in the USA. This raises concerns regarding the generalisability of these results to the broader ASD population and different cultural contexts. It is therefore advisable to continue evaluating these complex interventions in community-referred populations and different settings.

The Early Start Denver Model (ESDM) is one of the NDBI’s most frequently recommended models.7,9 However, it has never been assessed in RCT in French-speaking European countries.

To inform the generalisability of previous results in diverse environments and populations, we conducted a large-scale multicentre RCT to evaluate the effectiveness of ESDM in French-speaking European autistic children. We hypothesised that ESDM 12 hours per week in addition to treatment as usual (ESDM+TAU) would have a superior effect on the overall development of autistic children assessed by the Mullen Scale of Early Learning (MSEL) compared with TAU alone. A preliminary pilot study confirmed the feasibility of implementing this programme in France.¹⁰ The trial protocol was previously published.¹¹

Method

Study design

This study is a parallel RCT using a two-stage modified Zelen design.

The participants were recruited at five French-speaking sites in two European countries (Lyon-Vinatier, Lyon Saint Jean de Dieu, Strasbourg, and Versailles in France; Bruxelles in Belgium) after referral via clinicians from primary and secondary care (eg, speech pathologists, psychologists, paediatricians and child and adolescent psychiatrists) and declared in ClinicalTrials.gov ID NCT02608333. All centres were linked to the public health system, and families did not have to pay for the assessment or the intervention.

Participants

Inclusion criteria were children aged 15 months to 36 months and 30 days, meeting core autism criteria on the Autism Diagnostic Observation Schedule—second edition (ADOS-2) and the Toddler Autism Diagnosis Interview—Revised for toddlers,¹² and diagnosed by a multidisciplinary team. The developmental quotient (DQ) had to be 30 on the MSEL.¹³ Exclusion criteria were as follows: children suffering from a severe neurological or somatic disorder not allowing intensive intervention or carriers of Rett’s syndrome.

Clinicians referred families to participating clinical centres for diagnosis and intervention. If the diagnosis was confirmed and the children fulfilled the criteria for participation, an expert child psychiatrist provided information about the study. Parental written consent of eligible children was obtained according to the two-stage Zelen design.

Randomisation and masking

In this two-stage Zelen design, all parents of eligible children consented to participate in a 2-year observational study.^{14 15} Then, children of the observational study were randomised to intervention (ESDM+TAU) or control group (TAU), using an unequal ratio 1:2 intended to limit the cost and resources to implement the intervention. Only parents of children randomised in the intervention group were informed about the intervention and gave their consent for it. The adapted Zelen design aimed to avoid a strong feeling of disappointment and study withdrawal among parents of children not randomised to the intervention group,^{16 17} as well as to prevent potential contamination bias. This modified design was chosen over the conventional Zelen design to overcome ethical problems. In the conventional Zelen design, participants are not informed at all about their participation in a study, which is not acceptable.

The Hospices Civils de Lyon’s clinical trial unit (Lyon, France) used computer-generated randomisation, independent from assessment or intervention units. Randomisation was stratified, using minimisation, by site, age group (15–30, 31–36 months), ADOS-2 Calibrated Severity Score (CSS) at baseline (4–7, 8–10) and DQ MSEL at baseline (30–45, >45). The clinical trial unit emailed the allocation result to the expert child psychiatrist who enrolled and oversaw informing the family. Only participants who were randomised into the intervention group were called back. The expert child psychiatrist was not further involved in the study.

All study centres had a unit for the diagnosis and assessment of ASD and a unit for ESDM intervention. All researchers who performed the assessments were blinded to allocated groups. Dyadic Communication Measure for Autism and Brief Observation of Social Communication Change (BOSCC) assessments were cross coded from videotapes by blinded researchers who were not involved in ESDM intervention. Participant families and therapists cannot be masked to treatment allocation in the ESDM+TAU group. Parents were informed about the importance of not divulging treatment group allocation to preserve masking in assessment.

Interventions

ESDM intervention

The ESDM is extensively and comprehensively described in its manual.⁸

ESDM intervention took place 10 hours per week at the intervention centre in individual sessions (one therapist for one child) and 2 hours per week in the toddlers’ natural environment (home, nursery or preschool). For the latter, the therapy was alternately delivered by the ESDM therapist and the parent under the supervision of the ESDM therapist. The parents were thus trained in the ESDM model and had the opportunity to use it in everyday life outside the therapist’s presence.

The intensity of ESDM treatment was lower than in previous studies by Dawson et al and Rogers et al^{18 19}; however, the effect of intervention intensity (ranging from 1 to 25 hours per week for a period of 3 months to 2 years) was not found as an effect modifier in recent meta-regression in this type of intervention or more specifically in ESDM.^{5 9}

To describe the implementation of the ESDM intervention, the therapists recorded the number of hours of ESDM realised at the centre and home and the reasons for each child’s missed planned sessions.

Therapists and ESDM adherence

For ESDM intervention, qualified professionals, including speech therapists (19%), nurses (23%), educators (23%), psychologists (20%) and psychomotor therapists (15%), composed the intervention team. Before intervention in this RCT, the professionals received the official 4-day training followed by filmed supervision for fidelity rating. They ultimately validated therapist certification as mandated by the UC Davis MIND Institute. The certification was based on at least two 30 min uncut videos of interventions with two different children, in which the therapists had to demonstrate more than 80% fidelity to the ESDM model (see fidelity scales published in the ESDM manual).⁸ During the study, variability due to therapist effects was minimised via regular ESDM fidelity checking and clinical supervision by certified ESDM trainers.²⁰

Treatment as usual

Children with ASD were allowed to benefit from the usual interventions offered by available institutions and professionals working with children with autism at each site.

In France, treatment is usually mostly composed of unspecified and eclectic approaches without ESDM, which was poorly disseminated at the time of this RCT.

TAU and ESDM were provided over the entire study period (ie, 2 years).

Outcomes

The primary study outcome was the change in global DQ measured by the MSEL,¹³ from baseline to 24 months post randomisation. This metric was chosen because previous studies have used DQ and reported effects.^{9 19} DQ helps avoid the floor effect observed with the standardised score of the MSEL in populations with delay in development. The DQ may thus better capture improvements in this population. To facilitate comparison with other studies in the field of autism (not using DQ), MSEL standardised scores and developmental ages are presented in online supplemental material. Outcomes are detailed in online supplemental table S1 in paragraph ‘1 Outcomes’.

Sample size

Sample size and power calculation have been described previously.¹¹ Briefly, based on the results of the first RCT about ESDM,¹⁸ to detect a 15-point difference in DQ (SD=25) with a 10% non-consent (Zelen design) and a 10% dropout frequency, 180 children (120 in the control group and 60 in the intervention group) were estimated to provide 90% power with a bilateral alpha risk of 0.05.

Populations of interest

Population analyses were modified intention-to-treat 1 (m-ITT 1), m-ITT 2 and per protocol (PP). The m-ITT 1 included all randomised participants analysed in their assigned groups (including ESDM-enrolled children who refused the intervention after randomisation due to the Zelen design), excluding postrandomised participants without any data in all three visits (n=3, see figure 1 flowchart). The m-ITT 2 was defined as the m-ITT 1 with the additional exclusion of missing data on the principal outcome at 24 months. The PP was defined as the m-ITT 1 with major deviation excluded. The main analysis was performed with the m-ITT 1 population. The m-ITT 2 and PP are defined as sensitivity analyses.

Figure 1. Between-group differences for changes in outcomes from baseline to 24 months. (A) Principal outcomes from the Mullen Scales of Early Learning. The treatment effect measures on the forest plot are mean differences for each outcome. (B) Other outcomes. The treatment effect measures on the forest plot are mean differences for each outcome. ADOS-2, Autism Diagnostic Observation Schedule, second edition, BOSCC, Brief Observation of Social Communication Change, the higher the score, the more observable the autistic signs are; CSS, Comparison Severity Score, a higher score reflects a higher autism symptom severity; CSBS, Communication and Symbolic Behavior Scales, the higher the score, the higher the child’s skills are; DCMA, Dyadic Communication Measure for Autism, the higher the score, the more observable communication between the child and one parent is; DQ, developmental quotient; VABS-II, Vineland Adaptive Behavior Scales, second edition, the higher the score, the higher adaptive functioning in daily life is.

Statistical analysis

The primary and secondary outcomes planned for sequential hierarchical testing were analysed using a linear mixed model with a random intercept and adjustment for the randomisation–stratification variables and time. Missing data for the primary outcome were handled using the linear mixed model based on a missing-at-random hypothesis, except when the global DQ was only available at the baseline visit (T0), in which case a maximum bias was used to impute the change from baseline to 24 months post randomisation (see in online supplemental material in paragraph ‘2 Statistical Analysis Plan (SAP)’” section 8.1.3 p 28).

For the other outcomes not included in the main hierarchical analysis, no statistical inference tests were performed. This is a minor discrepancy with what was specified in the statistical analysis plan.

Results

Study population

From September 2015 to March 2019, 209 children were referred for this study. Of these, 180 were eligible, and their parents consented to participate. They were randomly assigned to ESDM+TAU (n=61) or TAU (n=119). Three participants withdrew immediately after randomisation, became unavailable (no answer by phone or email) and did not provide any data. These three participants were excluded from the main m-ITT 1 analysis. Two parents refused to participate in ESDM but continued in TAU, allowing, from their point of view, more schooling time in an ordinary preschool for their child. The flowchart is shown in online supplemental figure S1 in paragraph ‘3 Results’.

Table 1 presents children’s characteristics at baseline. The mean chronological age was 30 months, and 21.7% and 15.4% were girls in the ESDM+TAU and TAU groups, respectively.

Table 1. Baseline demographic and cognitive characteristics in modified intention-to-treat 1 population.

	ESDM+TAU (n=60)	TAU (n=117)
Centre, n (%)
Lyon	23 (38.3%)	44 (37.6%)
St Jean de Dieu	14 (23.3%)	30 (25.6%)
Versailles	10 (16.7%)	18 (15.4%)
Strasbourg	6 (10.0%)	13 (11.1%)
Bruxelles	7 (11.7%)	12 (10.3%)
Gender, n (%)
Male	47 (78.3%)	99 (84.6%)
Female	13 (21.7%)	18 (15.4%)
Age (in months), mean (SD)	30.1 (4.62)	30.1 (4.34)
Second parent in household (yes), n (%)	51 (87.9%)	96 (85.7%)
Number of children in household, median (IQR)	2.0 (2.0–3.0)	2.0 (1.0–3.0)
Different languages spoken at home (yes), n (%)	36 (62.1%)	68 (60.7%)
Family incomes inferior to 2500 per month* (yes), n (%)	21 (41.2%)	38 (40.0%)
Mother’s level of education—equivalence grade≥12† (yes), n (%)	42 (73.7%)	74 (67.9%)
Father’s level of education—Equivalence grade≥12† (yes), n (%)	38 (70.4%)	69 (66.3%)
Term birth (in weeks of amenorrhea), mean (SD)	38.5 (1.73)	39.3 (1.58)
MSEL DQ global, mean (SD)	50.2 (11.72)	50.0 (10.66)
Severity in autism measured with CSS of ADOS-2, mean (SD)	7.6 (1.47)	7.7 (1.61)

^*Median salary per family according to Insee (2021).

^†Grade based on the US system; grade 12 corresponds to the final year of secondary school in most parts of the world.

ADOS-2, Autism Diagnostic Observation Schedule, second edition; CSS, Comparison Severity Score; ESDM, Early Start Denver Model; MSEL DQ, Mullen Scales of Early Learning Developmental Quotient; TAU, treatment as usual.

Primary and secondary outcomes

The estimated difference in change of overall DQ score from baseline to 24 months (T2) between groups was 3.82 points (95% CI −1.25 to 8.89; p value=0.14). As the first hierarchy outcome was non-significant, further hierarchical outcomes were described without statistical inference tests in table 2 and figure 1A. In sensitivity analyses with m-ITT 2, the difference in change of overall DQ score from baseline to 24 months (T2) between groups was 5.43 points (95% CI −0.34 to 11.19). With the PP analysis set, this estimation was 5.17 points (95% CI −0.73 to 11.08).

Table 2. Outcomes in modified intention-to-treat 1 population.

	Baseline		12-month outcomes		24-month outcomes
	ESDM+TAU	TAU	ESDM+TAU	TAU	ESDM+TAU	TAU
	n=60	n=117	n=58	n=103	n=50	n=93
MSEL DQ global	50.2 (11.72)	50.0 (10.66)	55.0 (20.64)	53.1 (18.70)	60.2 (25.32)	55.3 (24.24)
MSEL DQ visual reception	58.6 (20.65)	57.3 (15.36)	63.7 (23.35)	60.6 (22.43)	67.5 (30.00)	63.1 (27.56)
MSEL DQ fine motricity	64.4 (15.66)	65.0 (13.91)	64.5 (20.98)	61.5 (18.02)	69.2 (25.89)	63.9 (25.21)
MSEL DQ receptive language	35.7 (12.66)	38.0 (15.02)	46.0 (24.43)	44.8 (21.40)	53.9 (27.94)	48.3 (27.11)
MSEL DQ expressive language	41.9 (12.14)	39.5 (13.01)	45.7 (21.15)	45.0 (20.50)	50.0 (24.62)	45.9 (23.38)
ADOS-2 module, n (%)
Module toddler	38 (63.3%)	67 (57.8%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)
Module niveau 1	22 (36.7%)	49 (42.2%)	52 (89.7%)	98 (94.2%)	34 (63.0%)	67 (72.0%)
Module niveau 2	0 (0.0%)	0 (0.0%)	6 (10.3%)	6 (5.8%)	19 (35.2%)	26 (28.0%)
Module niveau 3	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	1 (1.9%)	0 (0.0%)
ADOS-2—overall CSS	7.6 (1.47)	7.7 (1.61)	6.9 (2.00)	6.5 (1.83)	7.0 (1.77)	7.0 (1.88)
ADOS-2—social interaction CSS	7.7 (1.67)	7.6 (1.74)	6.7 (2.22)	6.3 (2.02)	6.9 (1.96)	6.8 (1.89)
ADOS—RRB CSS	7.3 (1.80)	7.7 (1.74)	7.8 (1.67)	7.4 (1.82)	7.1 (2.14)	7.3 (2.25)
BOSCC—overall score	32.4 (8.75)	29.9 (8.08)	23.2 (11.64)	25.3 (10.37)	21.9 (11.88)	22.6 (11.05)
BOSCC—social interaction	24.2 (6.79)	21.7 (6.28)	16.8 (8.70)	18.3 (8.13)	16.2 (8.77)	16.6 (8.49)
BOSCC—RRB	8.2 (3.87)	8.2 (2.90)	6.4 (3.85)	7.0 (3.62)	5.6 (3.98)	6.0 (3.33)
DCMA child initiation per min	0.2 (0.22)	0.2 (0.25)	0.5 (0.58)	0.4 (0.51)	0.7 (0.84)	0.6 (0.81)
DCMA child responses per min	1.2 (0.73)	1.3 (0.97)	2.4 (1.39)	2.0 (1.24)	2.8 (1.54)	2.4 (1.48)
DCMA child initiation/responses per min	1.4 (0.94)	1.6 (1.22)	3.0 (2.07)	2.5 (1.67)	3.6 (2.19)	3.2 (2.22)
VABS-II communication	66.3 (9.32)	68.0 (10.17)	69.8 (14.90)	70.2 (17.32)	71.1 (20.40)	69.2 (20.15)
VABS-II daily living skills	78.0 (11.28)	76.4 (11.05)	76.1 (12.70)	74.0 (12.81)	74.9 (17.11)	71.4 (16.39)
VABS-II socialisation	74.5 (7.60)	75.7 (8.40)	75.0 (10.90)	72.8 (11.47)	73.4 (14.35)	69.8 (15.90)
VABS-II motor skills	82.5 (10.47)	85.0 (9.42)	82.2 (13.10)	80.4 (11.25)	77.9 (16.86)	73.1 (18.95)
VABS-II Standard Composite Score	72.4 (7.54)	73.1 (7.90)	72.7 (11.43)	71.3 (11.65)	71.5 (16.06)	68.6 (16.19)
CSBS social communication	29.2 (6.85)	30.4 (8.47)	35.6 (6.79)	34.8 (7.93)	39.3 (6.09)	37.0 (7.94)
CSBS expressive speech	14.7 (9.18)	15.1 (9.31)	24.5 (12.05)	23.9 (12.09)	29.2 (13.08)	26.9 (12.14)
CSBS symbolic	23.4 (8.90)	24.1 (10.27)	35.4 (11.13)	33.1 (10.38)	39.3 (12.06)	36.9 (9.76)
DLPF total number of words produced	36.5 (90.51)	45.0 (104.12)	250.4 (306.06)	208.2 (261.34)	457.5 (422.93)	401.2 (430.80)

Outcomes are presented with mean and SD in brackets, except other precision.

ADOS-2, Autism Diagnostic Observation Schedule, second edition; BOSCC, Brief Observation of Social Communication Change; CSBS, Communication and Symbolic Behavior Scales; CSS, Calibrated Severity Score; DCMA, Dyadic Communication Measure for Autism; DLPF, Développement du Langage de Production Française (Development of French Production Language); ESDM, Early Start Denver Model; MSEL DQ, Mullen Scales of Early Learning Developmental Quotient; RRB, restricted and repetitive behaviour; TAU, treatment as usual; VABS-II, Vineland Adaptive Behavior Scales, second edition.

MSEL standardised scores and developmental ages are presented in online supplemental table S2 in paragraph ‘3 Results’.

Other prespecified outcomes

Prespecified analyses that were not included in the testing hierarchy are reported in table 2 and figure 1B with 95% CI estimates.

The between-group difference in change of autistic signs measured with the ADOS CSS from baseline to 24 months (T2) was 0.71 (95% CI −0.30 to 1.73). The difference between groups in the change of total score of autistic signs measured with the BOSCC was −3.66 (95% CI −6.33 to −1.00).

The change in CSBS social dimension score from baseline to T2 in the ESDM+TAU group differed on average from that in the TAU group by 1.68 points (95% CI 0.67 to 2.70). Similarly, the between-group difference in the change of standard score on the symbolism dimension of the CSBS from T0 to T2 was 1.40 points (95% CI 0.71 to 2.09) and the change in social functioning Vineland Adaptive Behavior Scales (VABS) subscore differed on average by 3.63 points (95% CI 0.15 to 7.12).

Safety and post hoc analysis

No serious adverse event was recorded during the study.

In a post hoc analysis, we estimated the developmental age gain during the study. It is estimated that a child who had made good progress and caught up some of the lost ground would have gained more than 2 years of developmental age over the 2 years studied. During the 2-year follow-up, 22 out of 93 (23.7%, 95% CI 15.5% to 33.6%) children in the TAU group vs 19 out of 50 (38.0%, 95% CI 24.7% to 52.8%) in the ESDM+TAU group gained more than 2 years in overall developmental age. The gain was majorly due to the gain in language.

Intervention and TAU

ESDM intervention was delivered without delay after randomisation (median 0.6 months (IQR 0.4–1.5)) and during a median time of 22.3 (IQR 21.4–23.3) months. Children received a median of 657.4 (IQR 588.5–733.5) hours in the clinic and 83.8 (IQR 66.9–109.0) hours at home over the 2 years. Thus, children in the ESDM+TAU group received a median of 746.4 (IQR 683.1–813.2) hours total over the 2 years, which is a median of 9.6 (IQR 9.0–10.3) hours per week of ESDM with a therapist.

At 24 months, parents estimated that they applied ESDM-like advice from the ESDM therapist at home (in the absence of a therapist) for a median time of 3 hours per week (IQR 1.0–7.0). On the other hand, 25/36 (69.4%) parents felt that they used ESDM principles more than 50% of the time when interacting with their child during the day.

During the 2-year follow-up, both groups received TAU. The amount of TAU received was higher in the TAU group compared with the ESDM+TAU group (median 164.5 (IQR 100.5–285.8) vs 52.4 (IQR 9.0–101.3) hours over the 2-year follow-up, respectively). Many children in both groups had therapy with speech-language therapist (107/112 (95.5%) in TAU vs 43/58 (74.1%)) in the ESDM+TAU group), psychomotor therapist (82/112 (73.2%) vs 22/58 (37.9 %)), psychologist (44/112 (39.3%) vs 10/58 (17.2%)) and educator (44/112 (39.3%) vs 9/58 (15.5%). Most of the interventions were conducted individually rather than in group settings with the children (for more information, please see online supplemental table S3 in paragraph ‘3 Results’). Few intervention hours were delivered using explicitly declared specific techniques, primarily involving applied behaviour analysis methods or alternative and augmentative communication systems (TAU group 0.0 (IQR 0.0–54.0) vs ESDM+TAU group 0.0 (IQR 0.0–18.0)). All other TAU interventions consisted of unspecified or eclectic therapies.

Most of the children went to (mainstream or specialised) preschool in the TAU group (104/112 (92.9%)) and the ESDM+TAU group (54/58 (93.1%)). The median preschool attendance time over the 2-year period was higher in the TAU group at 825.3 hours (IQR 434.5–1276.8) than in the ESDM+TAU group at 646.0 hours (IQR 342.0–879.0). Children in the TAU group got more time from the school support assistant (136.5 (0.0–811.5)) than the ESDM+TAU group (0.0 (IQR 0.0–456.0)) over the 2-year follow-up. Few children went to specialised preschools (12/112 (10.7%) vs 3/60 (5.2%) in the ESDM+TAU group).

Acceptability (scored 0–10) related to the constraints of the intervention was similar in both groups (ESDM+TAU with a median of 8.0 (5.0–9.0) vs TAU 7.0 (5.0–8.0)). Satisfaction about the intervention (scored 0–10) was superior in ESDM+TAU Group vs TAU group (9.0 (8.0–10.0) vs 8.0 (7.0–9.0).

Discussion

Principal findings

This multicentre RCT evaluated, for the first time in a large French-speaking European community population of autistic children, the effectiveness of an ESDM intervention applied for 12 hours per week for 2 years, including 10 hours per week by a therapist at the Child and Adolescent Mental Health Services facility and 2 hours at home.

There was no significant effect of ESDM combined with TAU on the primary outcome, that is, the child’s global development was measured with the MSEL scale by a psychologist blinded to the treatment group, compared with TAU alone. These outcomes were consistent across verbal and non-verbal DQ measurements, regardless of whether the assessment was 1 or 2 years after the start of the intervention.

For all other outcomes out of the hierarchy, only estimates with their 95% CIs were reported to inform on the precision of these estimates. According to the hierarchical analysis and given the numerous outcomes evaluated, the results on these secondary outcomes should be interpreted cautiously.

A slightly greater decrease in autistic signs as measured by BOSCC could be observed in the ESDM+TAU group compared with the TAU group (−3.66 (95% CI −6.33 to −1.00)), but this finding was not consistent with the ADOS-2 (CSS 0.71 (95% CI −0.30 to 1.73)). One may also question the clinical significance of such small differences and whether they fall within the measurement uncertainty.

Parents in the ESDM+TAU group reported slight improvements in social interaction and symbolic functioning as assessed by the CSBS, in social interaction as assessed by the VABS-2, and in word production as assessed by the DLFP. As in all early non-pharmacological interventions, parents cannot be blinded to the treatment given (tested intervention vs control). Then, the parents’ positive report on the effectiveness of this model should not be overlooked, but it could be due to a placebo effect linked to the enthusiasm for this model and the parents’ significantly greater satisfaction in obtaining it. This could also be due to the more positive view of this group because of the support they receive several times a week from a multidisciplinary team.²¹

Comparison with other studies

Contrary to recent recommendations or meta-analysis, the findings of this study did not show an effect of ESDM on the development of young autistic children.^{4 7 9} We were unable to replicate the results of the small randomised trial from 2010 (n=48), originally conducted by the authors who developed the model.¹⁸ In their initial trial, the difference in standardised scores between the two groups 2 years after inclusion was about 11 points compared with 1.3 points in our study, in favour of ESDM. There was also a clear effect on the standardised VABS score, with a difference of 7.2 points between the two groups (in favour of the ESDM group) vs −2.4 points (against the ESDM group) in our RCT.

In their more recent trial (n=118),¹⁹ a positive trend in language development was observed after 2 years of intensive intervention (T2–T4), with a difference between the two groups in equivalent age for expressive and receptive language of 3.1 months, compared with 2.6 months in favour of ESDM in the current RCT. No differences were found in this trial for the VABS or ADOS scores as in our study.

The absence of difference between the two groups could be explained by the characteristics of the children at inclusion. The children in this study were referred by professionals and practitioners in the community for diagnosis and early management. Notably, children referred early in such contexts tend to exhibit lower DQ scores.^{22 23} Compared with the global development score of the population in the previous randomised trial conducted by Rogers et al¹⁹ (MSEL mean score of 65), the population in our study demonstrated a lower score of 50 at the time of inclusion.¹⁹ A low level of initial global development could be a source of slower development and explain the lack of a significant effect of ESDM in our population.^{19 24}

Moreover, exploratory analysis showed that more children in the TAU+ESDM group (38%) than in the TAU group (23.7%) gained more than 2 years in overall development during the 2 years of follow-up. This led to the hypothesis that part of the population could benefit from the addition of ESDM in their therapy. These results appear encouraging for children with milder difficulties or those presenting factors associated with highly favourable developmental outcomes. Such children could potentially benefit even more from a higher number of therapy hours than those provided in this trial.²⁵

The absence of effect on the primary outcome could also be due to the high difference expected (15 points) in DQ between the two groups over the 2 years. This difference (equivalent to approximately one SD) was calculated based on the preliminary results of Dawson et al.¹⁸ However, this difference did not correspond to a minimal clinically important difference, which is unknown for most of the tools in autism.²⁶

Strengths and limitations

Our study has several strengths. It adheres to rigorous standards by incorporating randomisation and blinded assessments, aspects that are notably underrepresented in early intervention evaluation.²⁷ Our trial features a robust sample size (n=177) and is the first to provide a comprehensive evaluation of ESDM over a 2-year period independently of ESDM developers. The modified Zelen design was mainly used to limit study withdrawal for children not randomised into the intervention group, as 12 hours per week of ESDM was potentially highly desired by families of these children in France and Belgium at the time of the study.

The study also has some limitations. The children and their families could not be blinded to the treatment they received, but the evaluators were. One potential limitation of our study is the risk of differential attrition rates between groups (ie, more attrition in the intervention than in the control group), due to having knowledge of group assignment in the intervention group during the second consent stage of the Zelen design. However, our analysis showed minimal differential attrition between groups both immediately after consent and at later assessment points (78% vs 82% in m-ITT 2 analysis, 77% vs 79% in PP analysis), suggesting that this risk did not significantly impact our findings. Many families could not be contacted in time for the reassessment at T2 or could not come in for a reassessment because they were overloaded with everyday tasks. An imputation strategy for missing data was included in the analyses to limit this bias on the primary outcome. The DQ tool was chosen following the publication of Dawson et al¹⁸ and may be debatable. Still, other tools measuring social interactions were used to measure secondary and exploratory outcome measures.^{11 28}

Moreover, it would be valuable to assess the fatigability and daily learning capacity of young autistic children with low DQs. In the future, it would be beneficial to document all events, including the less severe ones.²⁹

Future research

Further research is needed to assess the long-term effectiveness of ESDM and to determine whether certain subgroups of children with ASD may derive greater benefit, thereby guiding optimal implementation strategies. An ongoing cost-effectiveness analysis planned alongside this clinical trial will also provide insights into the direct and indirect costs associated with ESDM and TAU for these children.

Clinical implications

Our study did not confirm the hypothesis that 12 hours per week of ESDM over 2 years is superior to TAU in promoting global development. Based on our findings, we cannot recommend adding ESDM to TAU for all children within the broad autism spectrum who are referred by community professionals to clinical centres. ESDM may not be generalisable to the entire ASD population, including children with severe autism. This raises concerns in the field of early intervention, as these children are often among the first to be identified and represent a group that we aim to support due to the severity of their condition.

Data availability statement

Data are available upon reasonable request.