Pilot RCT of Early-Intervention Delivered in Inclusive Vs. Autism-Specific Settings: Blinded Proximal Outcomes from LENA Recordings and Classroom Footage

Background: Little experimental research addresses the relative benefits of different intervention approaches for children with autism or differential effects of specific delivery methods. Further, understanding relative benefits is complicated by a lack of outcome measures that are both meaningful and sensitive to change over time among young children with autism who may also present with substantial developmental delays.

Objectives: We examined growth in proximal outcome measures in toddlers and pre-schoolers with autism during receipt of a manualized early intervention. Children were randomly assigned to receive the Group-Early Start Denver Model (G-ESDM; Vivanti et al., 2017) within either mainstream inclusive classrooms or specialised autism-specific classrooms within a community childcare service. We hypothesised that children receiving their intervention within mainstream settings might make greater or more rapid gains in proximal outcome measures than those in the autism-specific setting, given the opportunity afforded for regular interaction with typically-developing peers, alongside the specialist input from classroom staff.

Methods: We randomly assigned 29 preschoolers with ASD (aged 18-36 months) to receive intervention across the school calendar year either in classrooms that included 1) only children with ASD (n=15; i.e., Autism-Specific Setting) or 2) mostly typically-developing children (n=14; i.e., Inclusive Setting). Two proximal measures were sampled at the start and end of the school year, and at three intermediate times. Language ENvironment Analysis (LENA) was used to record rates of spontaneous child vocalisation from 40-minute semi-structured interaction samples, with automated data extraction. Following Clifford et al. (2010) we collected footage of free-play and snack time sessions to sample spontaneous child intentional communication bids within classrooms, and coding by blinded researchers demonstrated excellent inter-rater agreement (ICC=.93).

Results: Children showed increased rates of spontaneous vocalisation and intentional communication bids – including both initiations and responses toward others – across the intervention year, particularly within the first half of the year. There was no differential effect of randomisation group. However, children randomised to the inclusive setting showed greater rates of intentional communication at the start of the year and maintained this advantage over time.

Conclusions: Spontaneous child vocalisations, recorded and extracted automatically via LENA, and intentional communication bids, coded from naturalistic 10-minute classroom behaviour samples, appear to be sensitive measures of growth in children’s skills over time. Advantages of these methods include the ability for repeated sampling across an intervention period and automated extraction (LENA) and blinded coding (Classroom Footage) with high inter-rater agreement. While we did not observe differential effects of randomisation, we may have been underpowered to do so with this relatively small pilot sample. Replication with a larger sample would also permit the use of more sophisticated method of growth analysis and allow examination of potential moderators and mediators of outcome.