Developing a Framework to Advance and Accommodate Behavioral Variability in Toddler Autism Screening: A Mixed-Methods Approach to Improve Autism Screening Accuracy

Autism Spectrum Disorder (ASD) is a prevalent, lifelong condition that manifests early in life and is associated with significant social communication challenges1–3 and economic needs1 for children with ASD and their families. Access to early, ASD-specific services can profoundly improve long-term outcomes for children with ASD4–9 and depends on timely detection of the condition. However, the predictive accuracy of ASD screening is well below recommended standards for screening tools10,15,21. As a result, a formal ASD diagnosis is often not conferred until preschool or school-age22,23. Critically, these delays impede access to ASD-specialized care within the time-period of increased brain plasticity25 during which services are most likely to attenuate developmental challenges, thereby reducing the need for more costly and restrictive services later in life8.
Screening tools, which typically rely on parental report, may lack predictive accuracy because they do not sufficiently account for heterogeneity in social communication expression across individuals. Variation in the expression of ASD-specific social communication behaviors have been widely noted29,46–51, and some social communication behaviors used in ASD screening perform significantly differently across a range of individuals31,32,36–38. Additionally, existing phenomenological heterogeneity has been observed via clinicians in the development of social communication emergence across prodromal individuals over time44, 53–58. However, current screeners employ methods that do not sufficiently capture this existing heterogeneity, and this heterogeneity has not been precisely characterized as prospectively observed and reported by caregivers.
The primary objective of this proposal is to generate a framework for developing highly predictive parental- report screening measures to improve the accuracy of autism screening. This will be achieved by i) developing ASD screening items that reflect the heterogeneity in social communication expression across individuals; ii) using these items in data collection via ecological momentary assessment (EMA); and iii) employing Machine Learning (ML) to develop prototype algorithms that maximize ASD prediction across individuals. EMA provides real-time in-context monitoring at time-periods proximal to target dynamics60–62. EMA is therefore an ideal choice to capture the heterogeneity in developmental trajectories of social-communication behaviors over time as observed and reported by caregivers. ML develops prediction algorithms that can differentially weigh and combine ASD indicators based on individual differences and temporal specificity and is an optimal approach for algorithm development in the presence of developmental heterogeneity76 but is underutilized in ASD screening efforts77. Through the proposed research and training plan, the applicant will develop the skills and expertise needed to make a substantive contribution to ASD research as an independent clinical scientist.

Grant Number: 5F31MH133296-03
NIH Institute/Center: NIH
Principal Investigator: Shana Attar