Subtyping Toddlers with ASD Based on Their Scanning Patterns in Response to Dyadic Bids for Attention

Background: Heterogeneity of syndrome expression is a well-known but poorly understood phenomenon in ASD.  Our previous study (Chawarska et al., in revisions), revealed impaired visual responses to scenes containing child-directed speech and eye-contact (dyadic cues). The analysis also revealed marked variability in performance amongst toddlers with ASD.  The mechanisms supporting responsiveness to dyadic cues are present shortly after birth and are highly consequential for development of social cognition and communication.  Hence, examining variability in performance in toddlers with ASD in this domain may have important implications for identifying meaningful subtypes within the spectrum, as well as for design of non-invasive early screening methods and identification of novel targets for treatment. 

Objectives: To examine for presence of subgroups amongst toddlers with ASD based on their scanning patterns.

Methods: Fixation patterns were recorded in 57 18-24-month-old toddlers with ASD in response to a video of an actress emulating bids for dyadic attention.  Their diagnoses were ascertained at 3 years.  Hierarchical clustering using Euclidean distance and Ward’s method was applied to four eye-tracking variables: the percentage of looking time at the scene, and percentages of looking time at the Person, Toys, and Background.   Bootstrapping methods, in which subjects are randomly resampled to estimate the accuracy of statistical measures, were performed to assess cluster stability.

Results: Clustering analysis identified three subgroups amongst toddlers with ASD. Subsequent analysis evaluated symptom severity and levels of verbal and nonverbal functioning in each of the three clusters. The clusters were: (1) toddlers who displayed limited attention to the entire scene in general and to the speaker, who had low cognitive skills and severe symptoms of autism (32% of toddlers); (2) toddlers who had no difficulties in attending to the scene, but who spent very little time monitoring the speaker, and who were higher functioning and had less severe autism symptoms than Cluster 1 (28%); (3) toddlers with scanning patterns comparable to those of DD and TD controls, who had similar clinical profiles to Cluster 2, but  less abnormal language profiles (40%).  77% of subjects were assigned to the same cluster in at least 80% of bootstrapped samples, demonstrating that the identified clusters are remarkably consistent.

Conclusions: The cluster analysis captured complex dependencies between visual scanning and severity of cognitive deficits, language impairment, and social disability in a sample of toddlers with ASD.  Almost 2/3 of the sample exhibited limited attention to the speaker, approximately half of whom had major difficulties in attending to complex scenes in general.  However, 40% of the sample performed similarly to their TD and DD peers, at least when global indices of performance were considered.  It is not clear if the scanning pattern observed in this group is equivalent to the typical pattern observed in control groups with regard to underlying mechanisms.  As work in our lab and by others have demonstrated, what appears to be a typical pattern of responses might sometimes be driven by atypical or compensatory mechanisms, thus further investigation into this phenomenon is necessary.