Combining Multiple Eye Tracking Measures at 6 Months in Infant Siblings: Associations with Outcomes

Background: Atypical scanning patterns towards social information are evident by 6-7 months in infant siblings of children with ASD who later develop ASD themselves (Chawarska, Macari, & Shic, 2013; Elison et al., 2013; Jones & Klin, 2013; Shic, Macari, & Chawarska, 2014). However, it is not yet clear how different variables uncovered by eye tracking can be related to one another and to developmental outcomes.

Objectives:   To use data reduction techniques to develop more compact representations of the latent structure associated with visual scanning strategies evidenced by high risk infant siblings of children with ASD at 6 months of age. To associate this structure with later outcomes, both directly and via clustering.

Methods: Eye-tracking data were collected from 6-month-old high-risk infant siblings (N=85). Primary eye tracking tasks included the Dyadic Bid (DB) Probe (Chawarska, Macari, & Shic, 2012) and the Speaking Face (SF) Task (Shic, Macari, & Chawarska, 2014). Region of interest variables reflecting time looking towards the Scene, Faces, and Eye-Mouth ratio were dependent measures. Missing dependent measures caused by movement or inattention were imputed with bootstrap methods using augmented data from up to 6 other experimental conditions. This reconstructed dataset was subject to principal component analysis (PCA) using parallel analysis to identify number of components followed by hierarchical clustering with the number of clusters determined via majority vote via NbClust (Charrad & Ghazzali, 2014). Outcomes included diagnoses of ASD (N=12), atypical development (ATYP, N=34, e.g. language delay), or typical development (TYP, N=39) at 24 or 36 months of age. Pearson’s correlation analysis was used to explore phenotypic relationships between 6-month eye tracking variables and clinical outcomes at 24 months.

Results:   Two principal components (PCs) were identified, with the first PC reflecting decreased looking at the scene and at faces in both the DB and SF tasks and decreased looking at the eyes in SF task and increased looking at eyes in the DB task. This PC correlated with ADOS 1 composite and total scores (r=.32 to .36, p<.01). Hierarchical clustering on the two PCs revealed 3 clusters. One cluster showed poor overall attention (non-lookers, NON, N=35), another showed greater looking at eyes (EYE, N=20), and the last more looking at mouths (MOU, N=30). The MOU group was predominantly female (62% female) and showed the fewest autism symptoms (ADOS 1 TOT = 5.1), whereas the NON and EYE groups were more male (80% and 75% male) and showed more symptoms (TOT=7.2, 9.2). Children with outcomes of ASD were primarily in the NON cluster (58.3%); children with TYP outcomes were more often in the MOU cluster (46.2%).

Conclusions:

6-month old eye tracking phenotypic data combined using data reduction techniques and augmented with clustering can provide insight into the developmental relationships between visual scanning strategies early in development and the heterogeneous outcomes associated with ASD. Results suggest that gender as well specific looking patterns to the mouth or eyes in differing contexts may provide clues regarding developmental trajectories.