Identifying Predictors of Successful Peer Engagement for Toddlers with Autism

Background: Deciding how and when to combine interventions are everyday questions faced by community programs. Yet to date, intervention research provides few answers regarding how best to tailor individual children’s programming. This study aims to respond to such questions posed by a community center-based program for toddlers with autism that offers two social communication interventions: a 1:1 social communication intervention (JASPER: Kasari et al., 2004) and adapted JASPER that includes a peer (jasPEER). It’s possible that toddlers may need 1:1 intervention to best succeed in group jasPEER. This study aims to explore whether children’s baseline skills may inform such questions.

Objectives: To explore predictors of successful peer engagement for children who received the jasPEER intervention in order to inform future services.

Methods:

Participants. Fifty toddlers (mean age 32.6 months, 80% male, and 84% minority) who received jasPEER were included.

Teacher-Child Interaction (TCX): Ten-minute teacher-child interactions were coded for children’s time in engagement states: unengaged, person (e.g. engaged with another person only), object (e.g. engaged only with objects), joint engagement (e.g., notices teacher and shared activity). Verbal/nonverbal joint attention initiations (JA), and percentage of time in play levels (simple, combination, pre-symbolic, and symbolic) are also coded from the TCX.

Mullen Scales of Early Learning (MSEL): Expressive and receptive language subscales (age equivalency) were collected by independent assessors.

Peer Observation. Five-minute unsupported child-peer interactions were coded for engagement in 1 minute intervals. Children were considered “successfully engaged” with peers if they demonstrated awareness (parallel aware) or coordination of the peer (joint engagement) for at least 20% of time.

Results:

Random forest (Breiman, 2001), a machine learning technique, was used to identify potential predictors of successful peer engagement. While random forest has been shown to provide high level of prediction accuracy, the forest cannot provide objective decision rules or criteria needed to inform clinicians in determining which type of children would have successful peer engagement. Hence, decision rules were identified using the Classification and Regression Tree (CART: Breiman et al., 1984), a statistical learning technique designed to develop decision rules based on recursive partitioning of predictors, and generate cutoff values within predictors.

MSEL expressive language was the strongest predictor of successful peer engagement (accuracy 78%; p<0.001) where children who had >18.5 months of expressive language were more likely to show peer engagement while children who demonstrated <10% combination play were less likely to show peer engagement. Further, children who showed >10% combination play, and who spend less than 14% time object engaged are more likely to demonstrate peer engagement (accuracy 84%, p<0.001).

Conclusions: Expressive language remains the best predictor of peer engagement for toddlers with autism. However, the presence of functional play skills is also important. Children who struggled with basic combination play or who become overly focused on toys to the exclusion of others were less likely to share a play interaction with peers. Children with this profile may benefit from 1:1 intervention to ready them for peer interactions.