Measuring Change in Social Interaction during a Music-Based Intervention for Children with Autism: Behavioral Coding of Therapist-Child Joint Engagement

Background: Standardized methods for assessing social abilities in children with autism are often insufficient to monitor dynamic social interactions (Peper et al. 2016). Current tools typically assess deficits and often rely on parent report or advanced verbal skills. Tools capturing the dynamics of social interaction through direct observation of behaviour may improve sensitivity in evaluation of treatment outcomes (Cunningham, 2012). In a randomized controlled trial (RCT), we recently showed that 8-12 weeks of music intervention can improve auditory-motor connectivity and parent-reported social communication in school-age children with autism (Sharda et al, 2018). These findings suggest that collaborative music-making can enhance communication. However, there is limited clarity on the specific aspects of music-based interventions that are beneficial. It has been proposed that joint engagement between therapist and child in treatment settings may drive social benefits (Spiro and Himberg, 2016) but its specific role is poorly understood due to lack of suitable tools (Mössler, 2017).

Objectives: Our aims were to 1) identify a behavioral coding scheme appropriate for capturing changes in levels of triadic engagement between a therapist and school-aged child with ASD around intervention activities, 2) apply this scheme to session video from one-on-one music or a control play therapy, 3) evaluate potential changes in joint engagement over the course of music versus play therapy, 4) evaluate whether initial level of joint engagement predicts response to treatment.

Methods: An engagement coding scheme, adapted from Adamson et al., (2004), was used to assess levels of engagement in video-taped sessions of 6-12 year-old children with autism undergoing music or play-based intervention (Sharda et al., 2018). Seven mutually exclusive engagement state codes were employed (Fig.2a). For each intervention activity, duration of time spent in each state was coded using BORIS software (Friard & Gamba, 2016; Fig.1). Three raters were trained using a training manual in two phases, for a total of 70 hours of training, coding and discussion on intended coding targets, until they achieved acceptable inter-rater reliability (IRR). All raters were blind to session number and were not involved in the RCT. Twelve additional participant sessions (each with 4 activities making a total of 48 clips) were coded by each rater in order to measure IRR using intraclass correlation coefficient (ICC; one-way-single unit, agreement) for subsequent independent coding of videos.

Results: The ICCs for 4 (Coordinated joint, Supported Joint, Object engagement, Non- task-relevant object engagement) out 7 of engagement codes was >.79 (p<.001; Fig.2b). The remaining three state codes (Person only, Other, Unengaged) occurred quite rarely. Following the procedure in Adamson et al (2004) , we pooled these together resulting in an IRR=0.71 (p<.001) but excluded them from further analysis.

Conclusions: Measuring joint engagement through direct observation may provide a more sensitive tool for measuring response to behavioural interventions. Our adapted coding scheme had high reliability for key codes reflecting joint engagement with therapist and intervention activities. In ongoing work, we are applying this coding scheme to longitudinal intervention data from music and play-based interventions in autism to identify mechanisms of treatment-related change.