Validation of Temporally-Sensitive Eye-Tracking Indices of Social Disability As Treatment Endpoints in School-Age Children with ASD

Background: An important goal for biomedical research in autism is to develop effective treatments for ameliorating social disability in individuals on the autism spectrum. To develop effective treatments for social disability requires effective measures of social disability to be used as treatment endpoints. While percentage of time spent looking at socially-relevant, spatially-defined content regions, such as the eyes of other people, can serve as promising quantifiers of social disability (e.g., Rice et al., 2009), a critical feature of typical social interaction is the ability to look not only at socially-relevant content, but to look at the right content at the right moments in time. The goal of the current research is to test the extent to which eye-tracking-based, time-varying measures of visual scanning can serve as successful outcome measures to assess the efficacy of new treatments. These measures quantify both when and at what individuals look when viewing social interaction.

Objectives: This study will assess the reliability and validity of time-varying measures of visual scanning as treatment endpoints in school-age children with ASD.

Methods: We assessed the use of content-based and time-varying eye-tracking measures, collected during free viewing of naturalistic videos of social interaction, in terms of their content, construct, and convergent validity, as well as their general appropriateness for measuring social disability; their reliability, precision, and internal consistency; as well as their interpretability and patient acceptability. Content-based measures quantified looking to eyes, mouth, body, and object regions; time-varying measures quantified the probability of looking at the same location at the same time as typical viewers. To assess the range and distribution of scores found in the general population, we collected normative eye-tracking data from typically developing children (TD, N=42, mean age=9.61 years, FSIQ: 73-140) and also collected comparison values in a large and heterogeneous sample of children with ASD (N=128, mean age=10.12 years, FSIQ: 32-149).

Results: Content-based measures of visual attention demonstrate both high reliability and strong agreement for eyes, mouth, body, and object regions: ICC > 0.5, all p < 0.001 for ASD. Time-varying measures of visual scanning are more robust than content-based measures: ICC= 0.719, p < 0.001 for ASD. Preliminary analyses show promising convergent validity of ASD individuals’ time-varying scores with calibrated severity scores on the Autism Diagnostic Observation Schedule (ADOS) across the entire sample, despite marked heterogeneity in age and FSIQ, r= -0.357, p < 0.001.  

Conclusions: Temporally-sensitive measures of visual scanning during free-viewing of naturalistic scenes of social interaction can serve as effective and valid quantifiers of social disability. In the current analyses, we focused on measures—normed relative to typically-developing peers—of looking to the ‘right’ content at the ‘right’ moments in time. Future analyses will explore whether these measures can also effectively stratify subtypes within our heterogeneous ASD sample, and whether analysis of those sub-groups will improve convergent validity with expert-clinician assessments like the ADOS. Rigorous assessment of promising eye-tracking measures will provide us with multiple dimensions with which to assess the effectiveness of treatments aimed at treating social deficits in ASD.