Open access Original research

CHILD AND ADOLESCENT MENTAL HEALTH

Early Start Denver Model effectiveness in young

autistic children: a large multicentric randomised
controlled trial in two European countries
Marie-­Maude Geoffray ‍ ‍,1,2 Marie-­Joelle Oreve ‍ ‍,3 Lucie Jurek ‍ ‍,1,2
Sandrine Sonie ‍ ‍,4 Carmen Schroder ‍ ‍,5,6 Veronique Delvenne ‍ ‍,7
Sabine Manificat,1 Sandrine Touzet ‍ ‍,1,8 Jay Agathe ‍ ‍,1 Flavia Mengarelli ‍ ‍,1
Gallifet Natacha ‍ ‍,1 Nicolas Petit ‍ ‍,1,9 Mario Speranza ‍ ‍,10,11
Stéphane Bahrami ‍ ‍,12,13 Laetitia Bouveret ‍ ‍,14 Sara Linda Dochez ‍ ‍,1,15
Pauline Auphan ‍ ‍,1 Amelie Zelmar ‍ ‍,14 Bruno Falissard ‍ ‍,16 Sophie Carlier,17
Mikail Nourredine ‍ ‍,14 Angélique Denis ‍ ‍,18 Olivia Febvey-­Combes ‍ ‍14

► Additional supplemental ABSTRACT

material is published online Background Evidence regarding early interventions WHAT IS ALREADY KNOWN ON THIS TOPIC
only. To view, please visit the ⇒ Naturalistic Developmental Behavioral
journal online (https://​doi.​ based on the Naturalistic Developmental Behavioral
org/​10.​1136/​bmjment-​2024-​ Interventions framework, such as the Early Start Denver Interventions (NDBI), such as the Early Start
301424). Model (ESDM), suggests efficacy for autistic children. Denver Model (ESDM), demonstrate the
However, the effectiveness of ESDM across diverse strongest evidence of efficacy in autism,
For numbered affiliations see
end of article. cultural contexts remains under-­researched. particularly in social communication.
Objective To assess the effectiveness of ESDM
WHAT THIS STUDY ADDS
Correspondence to compared with treatment as usual (TAU) on overall
Professor Marie-­Maude development in young children with autism spectrum ⇒ For the first time in France and Europe, this
Geoffray, Le Vinatier Hospital disorder (ASD). study evaluates the ESDM, an NDBI, in a new
Center, Bron, Auvergne-­Rhône-­ cultural context, across five independent Child
Alpes, France; ​Marie-​maude.​ Method This parallel, randomised controlled trial, using
a modified Zelen design, was conducted in five Child and Adolescent Mental Health Services.
geoffray-​cassar@​ch-​le-​vinatier.​fr
and Adolescent Mental Health Services in France and ⇒ The trial, conducted according to high-­quality
Received 6 November 2024 Belgium. A total of 180 children aged 19–36 months, randomised controlled trial standards, includes
Accepted 1 April 2025
who met autism criteria and were referred by community a large sample of 180 community-­referred
professionals, were randomly assigned to either receive autistic children.
12-­hour weekly ESDM+TAU or TAU alone. The primary ⇒ The ESDM did not show significantly greater
outcome was the change in developmental quotient (DQ) effectiveness than standard care in improving
on the Mullen Scale of Early Learning, assessed blindly overall child development outcomes.
from baseline to 24 months post randomisation. HOW THIS STUDY MIGHT AFFECT RESEARCH,
Findings From September 2015 to March 2019, PRACTICE OR POLICY
180 children were randomly assigned to either
⇒ Our findings suggest that ESDM combined
ESDM+TAU (n=61, girls=21.7%) or TAU alone (n=119,
girls=15.4%). Three children dropped out immediately with treatment as usual cannot be universally
after randomisation. Compared with TAU alone, children recommended for all young children with
in the ESDM+TAU group did not significantly improve autism spectrum disorder (ASD).
⇒ Further research is needed to assess the long-­
global DQ (endpoint mean difference 3.82 (95% CI
−1.25 to 8.89), p=0.14). term effectiveness of ESDM and to identify
Conclusions and clinical implications Our subgroups of children with ASD who may derive
findings suggest that ESDM+TAU cannot be universally specific benefits, thus informing more targeted
recommended for young children with ASD. Further and optimal implementation strategies.
research is required to evaluate the long-­term effectiveness
of ESDM and identify subgroups that may benefit more,
thereby guiding optimal implementation strategies. function. Individuals with autism spectrum disorder
Trial registration number NCT02608333. (ASD) exhibit a broad spectrum of associated diffi-
© Author(s) (or their
employer(s)) 2025. Re-­use culties such as language, intellectual development
permitted under CC BY-­NC. and coping with everyday life.1 Pharmacological
Published by BMJ Group.
INTRODUCTION approaches may alleviate symptoms such as hyper-
To cite: Geoffray M-­M, Autism diagnosis is associated with core challenges activity or irritability but not the core symptoms of
Oreve M-­J, Jurek L, et al. BMJ in social communication, restricted and repetitive autism.2 Early non-­ pharmacological interventions
Ment Health 2025;28:1–8. interests and behaviours, and differences in sensory may improve both, although evidence remains
Geoffray M-­M, et al. BMJ Ment Health 2025;28:1–8. doi:10.1136/bmjment-2024-301424    1
 Open access
insufficient to support widespread implementation for all young pathologists, psychologists, paediatricians and child and adolescent
children with ASD. psychiatrists) and declared in ​ClinicalTrials.​gov ID NCT02608333.
Among the wide range of early intervention approaches All centres were linked to the public health system, and families did
for ASD, including sensory-­ based, animal-­ assisted, cognitive– not have to pay for the assessment or the intervention.
behavioural or developmental methods, Naturalistic Develop-
mental Behavioral Interventions (NDBIs) approaches merge Participants
strategies from behavioural and developmental learning theo- Inclusion criteria were children aged 15 months to 36 months
ries, offering a comprehensive framework to address develop- and 30 days, meeting core autism criteria on the Autism Diag-
mental milestones.3 A recent meta-­analysis assessed the effect nostic Observation Schedule—second edition (ADOS-­2) and the
of non-­ pharmacological treatments in randomised controlled Toddler Autism Diagnosis Interview—Revised for toddlers,12
trials (RCTs) previously conducted in children younger than 8 and diagnosed by a multidisciplinary team. The developmental
years of age. When focusing exclusively on outcomes not derived quotient (DQ) had to be 30 on the MSEL.13 Exclusion criteria
from caregiver reports—to avoid placebo-­by-­proxy bias—signif- were as follows: children suffering from a severe neurological or
icant improvements in social communication were observed somatic disorder not allowing intensive intervention or carriers
only for developmental interventions (Hedges’ g=0.31, 95% of Rett’s syndrome.
CI=0.13 to 0.49; p=0.003) and NDBI (95% CI=0.36, 0.23 to Clinicians referred families to participating clinical centres
0.49; p<0.001). Additionally, NDBI demonstrated a significant for diagnosis and intervention. If the diagnosis was confirmed
effect on core autism symptoms (95% CI=0.44, 0.20 to 0.68; and the children fulfilled the criteria for participation, an expert
p=0.002). When the meta-­analysis was restricted to RCTs with child psychiatrist provided information about the study. Parental
a low risk of detection bias, only NDBIs showed efficacy for core written consent of eligible children was obtained according to
autism symptoms.4 the two-­stage Zelen design.
The same research team previously investigated whether
certain characteristics of interventions or populations were Randomisation and masking
associated with the effectiveness of NDBIs.5 Neither cumulative In this two-­stage Zelen design, all parents of eligible children
intervention intensity nor the type of interventionist (parents or consented to participate in a 2-­ year observational study.14 15
professionals) was significantly related to the effect of NDBI. Then, children of the observational study were randomised to
Additionally, no moderating effects were found for the mean intervention (ESDM+TAU) or control group (TAU), using an
chronological age (B=−0.06, p=0.256), mean language age of unequal ratio 1:2 intended to limit the cost and resources to
the sample at study entry (B=−0.02, p=0.758), or percentage implement the intervention. Only parents of children randomised
of boys in each sample (B=−0.06, p=0.204). Due to the in the intervention group were informed about the intervention
insufficient number of participants with severe autism symp- and gave their consent for it. The adapted Zelen design aimed to
tomatology, the authors were unable to conduct a moderation avoid a strong feeling of disappointment and study withdrawal
analysis on autism severity. Recent studies have highlighted that among parents of children not randomised to the intervention
children with severe or profound autism (representing 35.5% of group,16 17 as well as to prevent potential contamination bias.
ASDs according to Hughes et al) are often implicitly or explicitly This modified design was chosen over the conventional Zelen
excluded from the studies.6 Furthermore, most RCTs have been design to overcome ethical problems. In the conventional Zelen
conducted in the USA. This raises concerns regarding the gener- design, participants are not informed at all about their participa-
alisability of these results to the broader ASD population and tion in a study, which is not acceptable.
different cultural contexts. It is therefore advisable to continue The Hospices Civils de Lyon’s clinical trial unit (Lyon, France)
evaluating these complex interventions in community-­referred used computer-­ generated randomisation, independent from
populations and different settings. assessment or intervention units. Randomisation was stratified,
The Early Start Denver Model (ESDM) is one of the NDBI’s using minimisation, by site, age group (15–30, 31–36 months),
most frequently recommended models.7–9 However, it has never ADOS-­2 Calibrated Severity Score (CSS) at baseline (4–7, 8–10)
been assessed in RCT in French-­speaking European countries. and DQ MSEL at baseline (30–45, >45). The clinical trial unit
To inform the generalisability of previous results in diverse emailed the allocation result to the expert child psychiatrist who
environments and populations, we conducted a large-­ scale enrolled and oversaw informing the family. Only participants who
multicentre RCT to evaluate the effectiveness of ESDM in were randomised into the intervention group were called back.
French-­ speaking European autistic children. We hypothesised The expert child psychiatrist was not further involved in the study.
that ESDM 12 hours per week in addition to treatment as usual All study centres had a unit for the diagnosis and assessment
(ESDM+TAU) would have a superior effect on the overall devel- of ASD and a unit for ESDM intervention. All researchers who
opment of autistic children assessed by the Mullen Scale of Early performed the assessments were blinded to allocated groups.
Learning (MSEL) compared with TAU alone. A preliminary pilot Dyadic Communication Measure for Autism and Brief Obser-
study confirmed the feasibility of implementing this programme vation of Social Communication Change (BOSCC) assessments
in France.10 The trial protocol was previously published.11 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
METHOD ESDM+TAU group. Parents were informed about the impor-
Study design tance of not divulging treatment group allocation to preserve
This study is a parallel RCT using a two-­stage modified Zelen masking in assessment.
design.
The participants were recruited at five French-­speaking sites in Interventions
two European countries (Lyon-­Vinatier, Lyon Saint Jean de Dieu, ESDM intervention
Strasbourg, and Versailles in France; Bruxelles in Belgium) after The ESDM is extensively and comprehensively described in its
referral via clinicians from primary and secondary care (eg, speech manual.8
2 Geoffray M-­M, et al. BMJ Ment Health 2025;28:1–8. doi:10.1136/bmjment-2024-301424
 Open access
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 environ-
ment (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 al18 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 Figure 1 Between-­group differences for changes in outcomes from
To describe the implementation of the ESDM intervention, baseline to 24 months. (A) Principal outcomes from the Mullen Scales
the therapists recorded the number of hours of ESDM realised of Early Learning. The treatment effect measures on the forest plot are
at the centre and home and the reasons for each child’s missed mean differences for each outcome. (B) Other outcomes. The treatment
planned sessions. effect measures on the forest plot are mean differences for each
outcome. ADOS-­2, Autism Diagnostic Observation Schedule, second
Therapists and ESDM adherence edition, BOSCC, Brief Observation of Social Communication Change,
For ESDM intervention, qualified professionals, including speech the higher the score, the more observable the autistic signs are; CSS,
therapists (19%), nurses (23%), educators (23%), psychologists Comparison Severity Score, a higher score reflects a higher autism
(20%) and psychomotor therapists (15%), composed the inter- symptom severity; CSBS, Communication and Symbolic Behavior Scales,
vention team. Before intervention in this RCT, the professionals the higher the score, the higher the child’s skills are; DCMA, Dyadic
received the official 4-­day training followed by filmed super- Communication Measure for Autism, the higher the score, the more
vision for fidelity rating. They ultimately validated therapist observable communication between the child and one parent is; DQ,
certification as mandated by the UC Davis MIND Institute. The developmental quotient; VABS-­II, Vineland Adaptive Behavior Scales,
certification was based on at least two 30 min uncut videos of second edition, the higher the score, the higher adaptive functioning in
interventions with two different children, in which the therapists daily life is.
had to demonstrate more than 80% fidelity to the ESDM model
(see fidelity scales published in the ESDM manual).8 During
180 children (120 in the control group and 60 in the interven-
the study, variability due to therapist effects was minimised via
tion group) were estimated to provide 90% power with a bilat-
regular ESDM fidelity checking and clinical supervision by certi-
eral alpha risk of 0.05.
fied ESDM trainers.20
Populations of interest
Treatment as usual Population analyses were modified intention-­to-­treat 1 (m-­ITT
Children with ASD were allowed to benefit from the usual 1), m-­ITT 2 and per protocol (PP). The m-­ITT 1 included all
interventions offered by available institutions and professionals randomised participants analysed in their assigned groups
working with children with autism at each site. (including ESDM-­enrolled children who refused the interven-
In France, treatment is usually mostly composed of unspeci- tion after randomisation due to the Zelen design), excluding
fied and eclectic approaches without ESDM, which was poorly postrandomised participants without any data in all three visits
disseminated at the time of this RCT. (n=3, see figure 1 flowchart). The m-­ITT 2 was defined as the
TAU and ESDM were provided over the entire study period m-­ITT 1 with the additional exclusion of missing data on the
(ie, 2 years). principal outcome at 24 months. The PP was defined as the
m-­ITT 1 with major deviation excluded. The main analysis was
Outcomes performed with the m-­ITT 1 population. The m-­ITT 2 and PP
The primary study outcome was the change in global DQ are defined as sensitivity analyses.
measured by the MSEL,13 from baseline to 24 months post
randomisation. This metric was chosen because previous studies Statistical analysis
have used DQ and reported effects.9 19 DQ helps avoid the floor The primary and secondary outcomes planned for sequential hier-
effect observed with the standardised score of the MSEL in archical testing were analysed using a linear mixed model with a
populations with delay in development. The DQ may thus better random intercept and adjustment for the randomisation–stratifi-
capture improvements in this population. To facilitate compar- cation variables and time. Missing data for the primary outcome
ison with other studies in the field of autism (not using DQ), were handled using the linear mixed model based on a missing-­
MSEL standardised scores and developmental ages are presented at-­random hypothesis, except when the global DQ was only
in online supplemental material. Outcomes are detailed in online available at the baseline visit (T0), in which case a maximum bias
supplemental table S1 in paragraph ‘1 Outcomes’. was used to impute the change from baseline to 24 months post
randomisation (see in online supplemental material in paragraph
Sample size ‘2 Statistical Analysis Plan (SAP)’” section 8.1.3 p 28).
Sample size and power calculation have been described previ- For the other outcomes not included in the main hierarchical
ously.11 Briefly, based on the results of the first RCT about analysis, no statistical inference tests were performed. This is a
ESDM,18 to detect a 15-­point difference in DQ (SD=25) with a minor discrepancy with what was specified in the statistical anal-
10% non-­consent (Zelen design) and a 10% dropout frequency, ysis plan.
Geoffray M-­M, et al. BMJ Ment Health 2025;28:1–8. doi:10.1136/bmjment-2024-301424 3
 Open access
MSEL standardised scores and developmental ages are
Table 1 Baseline demographic and cognitive characteristics in
presented in online supplemental table S2 in paragraph ‘3
modified intention-­to-­treat 1 population
Results’.
ESDM+TAU
(n=60) TAU (n=117)
Other prespecified outcomes
Centre, n (%)
Prespecified analyses that were not included in the testing
 Lyon 23 (38.3%) 44 (37.6%)
hierarchy are reported in table 2 and figure 1B with 95% CI
 St Jean de Dieu 14 (23.3%) 30 (25.6%)
estimates.
 Versailles 10 (16.7%) 18 (15.4%) The between-­ group difference in change of autistic signs
 Strasbourg 6 (10.0%) 13 (11.1%) measured with the ADOS CSS from baseline to 24 months (T2)
 Bruxelles 7 (11.7%) 12 (10.3%) was 0.71 (95% CI −0.30 to 1.73). The difference between
Gender, n (%) groups in the change of total score of autistic signs measured
 Male 47 (78.3%) 99 (84.6%) with the BOSCC was −3.66 (95% CI −6.33 to −1.00).
 Female 13 (21.7%) 18 (15.4%) The change in CSBS social dimension score from baseline to
Age (in months), mean (SD) 30.1 (4.62) 30.1 (4.34) T2 in the ESDM+TAU group differed on average from that in
Second parent in household (yes), n (%) 51 (87.9%) 96 (85.7%) the TAU group by 1.68 points (95% CI 0.67 to 2.70). Similarly,
Number of children in household, median (IQR) 2.0 (2.0–3.0) 2.0 (1.0–3.0) the between-­group difference in the change of standard score
Different languages spoken at home (yes), n (%) 36 (62.1%) 68 (60.7%) on the symbolism dimension of the CSBS from T0 to T2 was
Family incomes inferior to 2500 per month* (yes), 21 (41.2%) 38 (40.0%) 1.40 points (95% CI 0.71 to 2.09) and the change in social
n (%) functioning Vineland Adaptive Behavior Scales (VABS) subscore
Mother’s level of education—equivalence 42 (73.7%) 74 (67.9%) differed on average by 3.63 points (95% CI 0.15 to 7.12).
grade≥12† (yes), n (%)
Father’s level of education—Equivalence 38 (70.4%) 69 (66.3%) Safety and post hoc analysis
grade≥12† (yes), n (%)
No serious adverse event was recorded during the study.
Term birth (in weeks of amenorrhea), mean (SD) 38.5 (1.73) 39.3 (1.58) In a post hoc analysis, we estimated the developmental age
MSEL DQ global, mean (SD) 50.2 (11.72) 50.0 (10.66) gain during the study. It is estimated that a child who had made
Severity in autism measured with CSS of ADOS-­2, 7.6 (1.47) 7.7 (1.61) good progress and caught up some of the lost ground would
mean (SD) have gained more than 2 years of developmental age over the 2
*Median salary per family according to Insee (2021). years studied. During the 2-­year follow-­up, 22 out of 93 (23.7%,
†Grade based on the US system; grade 12 corresponds to the final year of 95% CI 15.5% to 33.6%) children in the TAU group vs 19 out of
secondary school in most parts of the world.
50 (38.0%, 95% CI 24.7% to 52.8%) in the ESDM+TAU group
ADOS-­2, Autism Diagnostic Observation Schedule, second edition; CSS, Comparison
Severity Score; ESDM, Early Start Denver Model; MSEL DQ, Mullen Scales of Early
gained more than 2 years in overall developmental age. The gain
Learning Developmental Quotient; TAU, treatment as usual. was majorly due to the gain in language.

Intervention and TAU

RESULTS ESDM intervention was delivered without delay after randomi-
Study population sation (median 0.6 months (IQR 0.4–1.5)) and during a median
From September 2015 to March 2019, 209 children were time of 22.3 (IQR 21.4–23.3) months. Children received a
referred for this study. Of these, 180 were eligible, and their median of 657.4 (IQR 588.5–733.5) hours in the clinic and 83.8
parents consented to participate. They were randomly assigned to (IQR 66.9–109.0) hours at home over the 2 years. Thus, chil-
ESDM+TAU (n=61) or TAU (n=119). Three participants with- dren in the ESDM+TAU group received a median of 746.4 (IQR
drew immediately after randomisation, became unavailable (no 683.1–813.2) hours total over the 2 years, which is a median of
answer by phone or email) and did not provide any data. These 9.6 (IQR 9.0–10.3) hours per week of ESDM with a therapist.
three participants were excluded from the main m-­ITT 1 anal- At 24 months, parents estimated that they applied ESDM-­like
ysis. Two parents refused to participate in ESDM but continued advice from the ESDM therapist at home (in the absence of a
in TAU, allowing, from their point of view, more schooling time therapist) for a median time of 3 hours per week (IQR 1.0–7.0).
in an ordinary preschool for their child. The flowchart is shown On the other hand, 25/36 (69.4%) parents felt that they used
in online supplemental figure S1 in paragraph ‘3 Results’. ESDM principles more than 50% of the time when interacting
Table 1 presents children’s characteristics at baseline. The with their child during the day.
mean chronological age was 30 months, and 21.7% and 15.4% During the 2-­year follow-­up, both groups received TAU. The
were girls in the ESDM+TAU and TAU groups, respectively. amount of TAU received was higher in the TAU group compared
with the ESDM+TAU group (median 164.5 (IQR 100.5–285.8)
Primary and secondary outcomes vs 52.4 (IQR 9.0–101.3) hours over the 2-­ year follow-­up,
The estimated difference in change of overall DQ score from respectively). Many children in both groups had therapy with
baseline to 24 months (T2) between groups was 3.82 points speech-­language therapist (107/112 (95.5%) in TAU vs 43/58
(95% CI −1.25 to 8.89; p value=0.14). As the first hierarchy (74.1%)) in the ESDM+TAU group), psychomotor therapist
outcome was non-­ significant, further hierarchical outcomes (82/112 (73.2%) vs 22/58 (37.9 %)), psychologist (44/112
were described without statistical inference tests in table 2 and (39.3%) vs 10/58 (17.2%)) and educator (44/112 (39.3%) vs
figure 1A. In sensitivity analyses with m-­ITT 2, the difference 9/58 (15.5%). Most of the interventions were conducted indi-
in change of overall DQ score from baseline to 24 months (T2) vidually rather than in group settings with the children (for
between groups was 5.43 points (95% CI −0.34 to 11.19). With more information, please see online supplemental table S3 in
the PP analysis set, this estimation was 5.17 points (95% CI paragraph ‘3 Results’). Few intervention hours were delivered
−0.73 to 11.08). using explicitly declared specific techniques, primarily involving
4 Geoffray M-­M, et al. BMJ Ment Health 2025;28:1–8. doi:10.1136/bmjment-2024-301424
 Open access

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.

applied behaviour analysis methods or alternative and augmen- DISCUSSION

tative communication systems (TAU group 0.0 (IQR 0.0–54.0) Principal findings
vs ESDM+TAU group 0.0 (IQR 0.0–18.0)). All other TAU inter- This multicentre RCT evaluated, for the first time in a large
ventions consisted of unspecified or eclectic therapies. French-­ speaking European community population of autistic
Most of the children went to (mainstream or special- children, the effectiveness of an ESDM intervention applied for
ised) preschool in the TAU group (104/112 (92.9%)) and the 12 hours per week for 2 years, including 10 hours per week by
ESDM+TAU group (54/58 (93.1%)). The median preschool a therapist at the Child and Adolescent Mental Health Services
attendance time over the 2-­ year period was higher in the facility and 2 hours at home.
TAU group at 825.3 hours (IQR 434.5–1276.8) than in the There was no significant effect of ESDM combined with TAU
ESDM+TAU group at 646.0 hours (IQR 342.0–879.0). Chil- on the primary outcome, that is, the child’s global development
dren in the TAU group got more time from the school support was measured with the MSEL scale by a psychologist blinded to
assistant (136.5 (0.0–811.5)) than the ESDM+TAU group (0.0 the treatment group, compared with TAU alone. These outcomes
(IQR 0.0–456.0)) over the 2-­year follow-­up. Few children went were consistent across verbal and non-­verbal DQ measurements,
to specialised preschools (12/112 (10.7%) vs 3/60 (5.2%) in the regardless of whether the assessment was 1 or 2 years after the
ESDM+TAU group). start of the intervention.
Acceptability (scored 0–10) related to the constraints of the For all other outcomes out of the hierarchy, only estimates
intervention was similar in both groups (ESDM+TAU with a with their 95% CIs were reported to inform on the precision
median of 8.0 (5.0–9.0) vs TAU 7.0 (5.0–8.0)). Satisfaction about of these estimates. According to the hierarchical analysis and
the intervention (scored 0–10) was superior in ESDM+TAU given the numerous outcomes evaluated, the results on these
Group vs TAU group (9.0 (8.0–10.0) vs 8.0 (7.0–9.0). 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%
Geoffray M-­M, et al. BMJ Ment Health 2025;28:1–8. doi:10.1136/bmjment-2024-301424 5
 Open access
CI −0.30 to 1.73)). One may also question the clinical signifi- correspond to a minimal clinically important difference, which
cance of such small differences and whether they fall within the is unknown for most of the tools in autism.26
measurement uncertainty.
Parents in the ESDM+TAU group reported slight improve- Strengths and limitations
ments in social interaction and symbolic functioning as assessed Our study has several strengths. It adheres to rigorous standards
by the CSBS, in social interaction as assessed by the VABS-­2, by incorporating randomisation and blinded assessments, aspects
and in word production as assessed by the DLFP. As in all early that are notably underrepresented in early intervention evalua-
non-­pharmacological interventions, parents cannot be blinded tion.27 Our trial features a robust sample size (n=177) and is
to the treatment given (tested intervention vs control). Then, the the first to provide a comprehensive evaluation of ESDM over a
parents’ positive report on the effectiveness of this model should 2-­year period independently of ESDM developers. The modified
not be overlooked, but it could be due to a placebo effect linked Zelen design was mainly used to limit study withdrawal for chil-
to the enthusiasm for this model and the parents’ significantly dren not randomised into the intervention group, as 12 hours
greater satisfaction in obtaining it. This could also be due to the per week of ESDM was potentially highly desired by families of
more positive view of this group because of the support they these children in France and Belgium at the time of the study.
receive several times a week from a multidisciplinary team.21 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
Comparison with other studies
risk of differential attrition rates between groups (ie, more attri-
Contrary to recent recommendations or meta-­ analysis, the
tion in the intervention than in the control group), due to having
findings of this study did not show an effect of ESDM on the
knowledge of group assignment in the intervention group during
development of young autistic children.4 7 9 We were unable to
the second consent stage of the Zelen design. However, our
replicate the results of the small randomised trial from 2010
analysis showed minimal differential attrition between groups
(n=48), originally conducted by the authors who developed
both immediately after consent and at later assessment points
the model.18 In their initial trial, the difference in standardised
(78% vs 82% in m-­ITT 2 analysis, 77% vs 79% in PP analysis),
scores between the two groups 2 years after inclusion was about
suggesting that this risk did not significantly impact our findings.
11 points compared with 1.3 points in our study, in favour of
Many families could not be contacted in time for the reassess-
ESDM. There was also a clear effect on the standardised VABS
ment at T2 or could not come in for a reassessment because they
score, with a difference of 7.2 points between the two groups (in
were overloaded with everyday tasks. An imputation strategy for
favour of the ESDM group) vs −2.4 points (against the ESDM missing data was included in the analyses to limit this bias on the
group) in our RCT. primary outcome. The DQ tool was chosen following the publi-
In their more recent trial (n=118),19 a positive trend in cation of Dawson et al18 and may be debatable. Still, other tools
language development was observed after 2 years of intensive measuring social interactions were used to measure secondary
intervention (T2–T4), with a difference between the two groups and exploratory outcome measures.11 28
in equivalent age for expressive and receptive language of 3.1 Moreover, it would be valuable to assess the fatigability and
months, compared with 2.6 months in favour of ESDM in the daily learning capacity of young autistic children with low DQs.
current RCT. No differences were found in this trial for the In the future, it would be beneficial to document all events,
VABS or ADOS scores as in our study. including the less severe ones.29
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 Future research
practitioners in the community for diagnosis and early manage- Further research is needed to assess the long-­term effectiveness of
ment. Notably, children referred early in such contexts tend to ESDM and to determine whether certain subgroups of children
exhibit lower DQ scores.22 23 Compared with the global devel- with ASD may derive greater benefit, thereby guiding optimal
opment score of the population in the previous randomised trial implementation strategies. An ongoing cost-­effectiveness anal-
conducted by Rogers et al19 (MSEL mean score of 65), the popu- ysis planned alongside this clinical trial will also provide insights
lation in our study demonstrated a lower score of 50 at the time into the direct and indirect costs associated with ESDM and TAU
of inclusion.19 A low level of initial global development could be for these children.
a source of slower development and explain the lack of a signif-
icant effect of ESDM in our population.19 24 CLINICAL IMPLICATIONS
Moreover, exploratory analysis showed that more children in Our study did not confirm the hypothesis that 12 hours per week
the TAU+ESDM group (38%) than in the TAU group (23.7%) of ESDM over 2 years is superior to TAU in promoting global
gained more than 2 years in overall development during the 2 development. Based on our findings, we cannot recommend
years of follow-­up. This led to the hypothesis that part of the adding ESDM to TAU for all children within the broad autism
population could benefit from the addition of ESDM in their spectrum who are referred by community professionals to clin-
therapy. These results appear encouraging for children with ical centres. ESDM may not be generalisable to the entire ASD
milder difficulties or those presenting factors associated with population, including children with severe autism. This raises
highly favourable developmental outcomes. Such children could concerns in the field of early intervention, as these children are
potentially benefit even more from a higher number of therapy often among the first to be identified and represent a group that
hours than those provided in this trial.25 we aim to support due to the severity of their condition.
The absence of effect on the primary outcome could also be
due to the high difference expected (15 points) in DQ between Author affiliations
1
CH Le Vinatier, Bron, Auvergne-­Rhône-­Alpes, France
the two groups over the 2 years. This difference (equivalent to 2
INSERM 1290 RESearch on HealthcAre PErformance Lyon 1, Lyon, France
approximately one SD) was calculated based on the prelimi- 3
Service universitaire de psychiatrie de l’enfant et de l’adolescent, CH Versailles, Le
nary results of Dawson et al.18 However, this difference did not Chesnay, Île-­de-­France, France

6 Geoffray M-­M, et al. BMJ Ment Health 2025;28:1–8. doi:10.1136/bmjment-2024-301424

 Open access
4
Centre de ressources autisme Rhône-­Alpes, CH Le Vinatier, Bron, Auvergne-­Rhône-­ Lucie Jurek http://orcid.org/0000-0001-7568-568X
Alpes, France Sandrine Sonie http://orcid.org/0000-0003-2694-6307
5
Department for Child and Adolescent Psychiatry, CHU de Strasbourg, Strasbourg, Carmen Schroder http://orcid.org/0000-0001-7264-2086
Grand Est, France Veronique Delvenne http://orcid.org/0000-0001-6644-3208
6
Institute for Cellular and Integrative Neurosciences, CNRS UPR 3122, Strasbourg, Sandrine Touzet http://orcid.org/0000-0002-0354-7920
France Jay Agathe http://orcid.org/0000-0002-8343-5426
7
Child and adolescent psychiatry, HUDERF, Brussels, Belgium Flavia Mengarelli http://orcid.org/0009-0009-2265-7218
8
Université de Lyon, Lyon, France Gallifet Natacha http://orcid.org/0009-0002-0444-295X
9
Université Claude Bernard Lyon 1, Villeurbanne, France Nicolas Petit http://orcid.org/0000-0002-9910-2706
10
Public Health, Paris-­Saclay University, Versailles, France Mario Speranza http://orcid.org/0000-0002-1785-5764
11 Stéphane Bahrami http://orcid.org/0009-0003-2546-950X
Child and Adolescent Psychiatry, Versailles Hospital Center, Versailles, France
12 Laetitia Bouveret http://orcid.org/0009-0002-7339-3119
Versailles Saint-­Quentin-­en-­Yvelines University Faculty of Medicine, Montigny-­le-­
Bretonneux, France Sara Linda Dochez http://orcid.org/0009-0004-5365-4080
13
Paris-­Saclay University, Saclay, Île-­de-­France, France Pauline Auphan http://orcid.org/0009-0002-7548-3484
14
CHU Lyon, Lyon, Auvergne-­Rhône-­Alpes, France Amelie Zelmar http://orcid.org/0009-0000-2937-2711
15
Neurodéveloppemental research, Institut des sciences cognitives Marc Jeannerod, Bruno Falissard http://orcid.org/0000-0002-2418-4954
Bron, Rhône-­Alpes, France Mikail Nourredine http://orcid.org/0000-0002-0952-0091
16
Université Paris-­Saclay, Gif-­sur-­Yvette, Île-­de-­France, France Angélique Denis http://orcid.org/0000-0003-4117-1714
17
Hopital Universitaire de Bruxelles, Laeken, Bruxelles, Belgium Olivia Febvey-­Combes http://orcid.org/0000-0002-4888-0815
18
Service Biostatistiques, CHU Lyon, Lyon, Auvergne-­Rhône-­Alpes, France
X Lucie Jurek @JurekLucie

Acknowledgements We thank all the intervention teams (Lyon CH le Vinatier,

REFERENCES
1 Zeidan J, Fombonne E, Scorah J, et al. Global prevalence of autism: A systematic
Lyon Saint-­Jean de Dieu, Strasbourg, Versailles, Bruxelles) and families who took part
review update. Autism Res 2022;15:778–90.
in this study.
2 Siafis S, Çıray O, Wu H, et al. Pharmacological and dietary-­supplement treatments
Contributors AD and MN: statistical analysis. ST, M-­MG, SB, BF and AZ designed for autism spectrum disorder: a systematic review and network meta-­analysis. Mol
the project, and ST, OF-­C and AZ led the scientific coordination of the trial. M-­MG Autism 2022;13:10.
and GN were responsible for ESDM implementation, staff training and supervision. 3 Schreibman L, Dawson G, Stahmer AC, et al. Naturalistic Developmental Behavioral
AD, AZ and LB were responsible for the randomisation procedure, statistical plan Interventions: Empirically Validated Treatments for Autism Spectrum Disorder. J Autism
and responsible for the data management. AD and MN were responsible for the Dev Disord 2015;45:2411–28.
final statistical analysis plan and statistics. M-­MG, LJ, MN, OF-­C and SC drafted 4 Sandbank M, Bottema-­Beutel K, Crowley LaPoint S, et al. Autism intervention meta-­
the protocol. M-­JO, SS, SC, MS, M-­MG and VD were responsible for recruitment analysis of early childhood studies (Project AIM): updated systematic review and
and inclusion as well as study center supervision; FM, PA, SLD, JA and NP were secondary analysis. BMJ 2023;383:e076733.
responsible for assessment and data curation. All authors critically review and give 5 Crank JE, Sandbank M, Dunham K, et al. Understanding the Effects of Naturalistic
feedback to the article. M-­MG is responsible for the overall content as guarantor. Developmental Behavioral Interventions: A Project AIM Meta-­analysis. Autism Res
2021;14:817–34.
Funding This study is supported by a grant from the Direction Générale de l’offre
6 Hughes MM, Shaw KA, DiRienzo M, et al. The Prevalence and Characteristics of
de soin (PREPS 14-­0533) and a grant from the Fondation de France 2015-­013B.
Children With Profound Autism, 15 Sites, United States, 2000-­2016. Public Health Rep
Competing interests None declared. 2023;138:971–80.
Patient consent for publication Consent obtained from parent(s)/guardian(s). 7 Hirota T, King BH. Autism Spectrum Disorder: A Review. JAMA 2023;329:157–68.
8 Rogers S, Dawson G. Early start denver model for young children with autism.
Ethics approval This study involves human participants and was approved by Promoting language, learning, and engagement. Guilford, 2012.
the French Institutional Review Board (reference no. 2015-­013 B—’Comité de 9 Wang Z, Loh SC, Tian J, et al. A meta-­analysis of the effect of the Early Start
Protection de Personnes’), the French Data Protection Agency (reference DR-­ Denver Model in children with autism spectrum disorder. Int J Dev Disabil
2015-­319—“Commission Nationale Informatique et Liberté”) and the Belgium 2022;68:587–97.
Institutional Review Board (reference P2016/554, reference CCB B406201630678— 10 Geoffray M-­M, Denis A, Mengarelli F, et al. Using ESDM 12 hours per week in children
’Comité d’Ethique Erasme-­ULB’). Participants gave informed consent to participate in with autism spectrum disorder: feasibility and results of an observational study.
the study before taking part. Psychiatr Danub 2019;31:333–9.
Provenance and peer review Not commissioned; externally peer reviewed. 11 Touzet S, Occelli P, Schröder C, et al. Impact of the Early Start Denver Model on
the cognitive level of children with autism spectrum disorder: study protocol
Data availability statement Data are available upon reasonable request. Trial-­ for a randomised controlled trial using a two-­stage Zelen design. BMJ Open
related anonymised participant data will be made available on reasonable request 2017;7:e014730.
to the principal investigator (MMG, ​mmgeoffray@​hotmail.​com, ​marie-­​maude.​ 12 Kim SH, Thurm A, Shumway S, et al. Multisite study of new autism diagnostic
geoffraycassar@​ch-­​le-­​vinatier.​fr); data access will be made available following interview-­revised (ADI-­R) algorithms for toddlers and young preschoolers. J Autism
publication by email after a signed data sharing agreement has been made with Dev Disord 2013;43:1527–38.
the study team and with an ethics waiver. The trial protocol is published and our 13 Mullen EM. Mullen scales of early learning. AGS Circle Pines, MN, 1995.
statistical analysis plan is available in the online supplemental material 1. 14 Campbell R, Peters T, Grant C, et al. Adapting the randomized consent (Zelen) design
Supplemental material This content has been supplied by the author(s). It for trials of behavioural interventions for chronic disease: feasibility study. J Health
has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have Serv Res Policy 2005;10:220–5.
been peer-­reviewed. Any opinions or recommendations discussed are solely those 15 Simon GE, Shortreed SM, DeBar LL. Zelen design clinical trials: why, when, and how.
of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and Trials 2021;22:541.
responsibility arising from any reliance placed on the content. Where the content 16 Adamson J, Cockayne S, Puffer S, et al. Review of randomised trials using the post-­
includes any translated material, BMJ does not warrant the accuracy and reliability randomised consent (Zelen’s) design. Contemp Clin Trials 2006;27:305–19.
of the translations (including but not limited to local regulations, clinical guidelines, 17 Land J, McCourt O, Heinrich M, et al. The adapted Zelen was a feasible design to trial
terminology, drug names and drug dosages), and is not responsible for any error exercise in myeloma survivors. J Clin Epidemiol 2020;125:76–83.
and/or omissions arising from translation and adaptation or otherwise. 18 Dawson G, Rogers S, Munson J, et al. Randomized, controlled trial of an
intervention for toddlers with autism: the Early Start Denver Model. Pediatrics
Open access This is an open access article distributed in accordance with the 2010;125:e17:e17–23:.
Creative Commons Attribution Non Commercial (CC BY-­NC 4.0) license, which 19 Rogers SJ, Estes A, Lord C, et al. A Multisite Randomized Controlled Two-­Phase Trial
permits others to distribute, remix, adapt, build upon this work non-­commercially, of the Early Start Denver Model Compared to Treatment as Usual. J Am Acad Child
and license their derivative works on different terms, provided the original work is Adolesc Psychiatry 2019;58:853–65.
properly cited, appropriate credit is given, any changes made indicated, and the use 20 Zitter A, Rinn H, Szapuova Z, et al. Does Treatment Fidelity of the Early Start Denver
is non-­commercial. See: http://creativecommons.org/licenses/by-nc/4.0/. Model Impact Skill Acquisition in Young Children with Autism? J Autism Dev Disord
2023;53:1618–28.
ORCID iDs 21 Bent CA, Pellicano E, Iacono T, et al. Perspectives from parents of autistic
Marie-­Maude Geoffray http://orcid.org/0000-0002-5528-3613 children on participating in early intervention and associated research. Autism
Marie-­Joelle Oreve http://orcid.org/0000-0003-3354-1025 2023;27:1295–306.

Geoffray M-­M, et al. BMJ Ment Health 2025;28:1–8. doi:10.1136/bmjment-2024-301424 7

 Open access
22 Denisova K, Lin Z. The importance of low IQ to early diagnosis of autism. Autism Res 26 Krause KR, Hetrick SE, Courtney DB, et al. How much is enough? Considering
2023;16:122–42. minimally important change in youth mental health outcomes. Lancet Psychiatry
23 Shattuck PT, Durkin M, Maenner M, et al. Timing of identification among children with 2022;9:992–8.
an autism spectrum disorder: findings from a population-­based surveillance study. J 27 French L, Kennedy EMM. Annual Research Review: Early intervention for infants and
Am Acad Child Adolesc Psychiatry 2009;48:474–83. young children with, or at-­risk of, autism spectrum disorder: a systematic review. J
24 Bishop SL, Guthrie W, Coffing M, et al. Convergent validity of the Mullen Scales of Child Psychol Psychiatry 2018;59:444–56.
Early Learning and the differential ability scales in children with autism spectrum 28 McConachie H, Parr JR, Glod M, et al. Systematic review of tools to measure
disorders. Am J Intellect Dev Disabil 2011;116:331–43. outcomes for young children with autism spectrum disorder. Health Technol Assess
25 Yoder P, Rogers S, Estes A, et al. Interaction of Treatment Intensity and Autism Severity 2015;19:1–506.
on Frequency and Maturity of Spontaneous Communication in Toddlers with Autism 29 Bottema-­Beutel K, Crowley S, Sandbank M, et al. Adverse event reporting in
Spectrum Disorder. Autism Res 2020;13:1902–12. intervention research for young autistic children. Autism 2021;25:322–35.

8 Geoffray M-­M, et al. BMJ Ment Health 2025;28:1–8. doi:10.1136/bmjment-2024-301424