Performance of Eye-Tracking-Based Assays for Early Identification and Prediction of Developmental Functioning in ASD

Background: Autism Spectrum Disorder (ASD) is a common neurodevelopmental disorder with widespread and often debilitating impact. Although early detection is considered one of the strongest positive predictors of improved long-term outcome, average age of diagnosis in the US remains stubbornly delayed—until after 4 years in 3 consecutive CDC surveillance cohorts—primarily because of limited access to expert clinicians.

Objectives: The goal of this study was to test the extent to which performance-based objective measures of a child’s social visual engagement, collected via eye-tracking technology between the ages of 16 and 30 months, could (a) match the diagnostic determination of expert clinicians using gold standard diagnostic instruments (ADOS), and (b) could quantify varying levels of autism symptom severity, verbal ability, and nonverbal ability relative to standardized assessments (ADOS and Mullen Scales of Early Learning).

Methods:

In N=326 toddlers between the ages of 16-30 months, including those referred for clinical concerns for ASD and those without suspected concerns for ASD, we conducted a prospective, double-blinded, within-subject comparison study of the performance of eye-tracking-based assays as diagnostic classifier (ASD vs Non-ASD) and as predictors of developmental function. The sample was split into training sample (N=218) and independent testing sample (N=108). Eye-tracking measures quantified moment-by-moment visual scanning to scenes of peer social interaction.

Results: In the independent testing sample, ASD status was classified with 80.4% sensitivity and 80.6% specificity; positive predictive value was 75.5%, negative predictive value was 84.7%, accuracy was 80.6%, and AUC was 0.87. Eye-tracking assays were significantly predictive of levels of social disability (R=-0.735, p<0.001, relative to ADOS total score) and of both verbal and nonverbal cognitive ability (R=0.573, p<0.001 and R=0.390, p=0.030, respectively, relative to scores on the Mullen).

Conclusions: For optimizing long-term outcomes in ASD and attenuating long-term burdens, two of the most important factors are early diagnosis and early intervention. Because the majority of parents are concerned about the developmental status of their children by two years, and because reliable and stable diagnosis can be made at that time, the American Academy of Pediatrics recommends universal screening at months 18 and 24; however, population-based studies find that the median age of identification in the U.S. is far later. One key factor in that delay is limited access to experienced clinicians, because current screening and assessment tools depend on qualitative judgments regarding the presence or absence of social-communicative symptoms, symptoms which can vary in both onset and severity and which can be masked by spurts in cognitive and language function. The present results show that performance-based objective measures of a child’s social visual engagement, collected via eye-tracking technology, can effectively match the diagnostic determination of expert clinicians and can quantify variation in autism symptom severity, verbal ability, and nonverbal ability. These results represent an important first step in advancing cost-effective procedures to augment access to diagnostic services where these are not readily available, thus promoting the goal of community-wide earlier identification of ASD, a key factor in promoting optimal long-term outcome.