Infant Communication Measures Predict Preschool ASD Symptom Severity and ASD Diagnosis

Poster Presentation
Friday, May 3, 2019: 11:30 AM-1:30 PM
Room: 710 (Palais des congres de Montreal)
L. R. Watson1, S. Nowell2, L. Turner-Brown3, D. Garrido4, M. DuBay5, R. Grzadzinski6, G. Baranek7 and E. Crais1, (1)Department of Allied Health Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC, (2)University of North Carolina at Chapel Hill, Carrboro, NC, (3)UNC TEACCH Autism Program, University of North Carolina at Chapel Hill, Chapel Hill, NC, (4)University of Granada, Granada, Spain, (5)University of North Carolina, Chapel Hill, NC, (6)Carolina Institute for Developmental Disabilities, University of North Carolina, Chapel Hill, NC, (7)Chan Division of Occupational Science and Occupational Therapy, University of Southern California, Los Angeles, CA

Recent screening tools have expanded our potential to identify infants at greater likelihood of later autism spectrum disorder diagnosis (GL-ASD), and have utility for both clinicians and researchers. However, the time lag between a positive infant screening and a conclusive ASD diagnosis presents challenges, including caregiver psychological burdens due to uncertainty about their child’s diagnosis, delayed access to ASD-specific services requiring a diagnosis, and slower, more costly progress in research when study aims require knowledge of diagnostic outcomes. Thus, researchers continue to seek infant markers that more accurately predict ASD diagnostic outcomes.


  1. Evaluate the extent to which infant communication measures are associated with severity of preschool ASD symptoms in the Social Affect (SA) and Restricted and Repetitive Behaviors (RRB) domains
  2. Test the accuracy of infant communication variables in classifying preschool ASD-related diagnostic outcomes


Preschool assessments provided diagnostic outcomes for 52 children (aged 36 – 70 months) identified through community screening at 12 months as at GL-ASD. Children were assessed initially at 13.7 months, pursuant to an intervention trial. Infant communication measures included the Communication and Symbolic Behavior Scales (CSBS) Social, Speech, and Symbolic composites; Mullen Scales of Early Learning (MSEL) Receptive and Expressive Language and Visual Reception scales; and frequencies of Canonical Vocalizations and Directed Vocalizations coded from 30 minutes of adult-infant interaction. Preschool outcomes were the Autism Diagnostic Observation Schedule (ADOS) Calibrated Severity Scores (CSS) for the SA and RRB domains (available for 45 children), and final clinical classification (ASD or non-ASD).


Associations between infant communication measures and preschool ADOS SA and RRB CSS scores were examined, controlling for MSEL Visual Reception scores and intervention group. Infant CSBS Speech and MSEL Expressive Language had significant partial correlations with preschool SA CSS (both rs = -.39). Three infant communication variables had significant partial correlations with RRB CSS: MSEL Receptive Language (r = -.31), Canonical Vocalizations (r = -.41), and Directed Vocalizations (r = -.49). Multiple other partial correlations between infant communication measures and preschool SA or RRB scores were marginally significant. We evaluated an iterative series of binomial logistic regression models, using variables significantly or marginally significantly correlated with CSS scores, to derive a parsimonious model predicting preschool diagnostic classification from infant CSBS Social, MSEL Receptive and Expressive Language, and Directed Vocalizations. The model yielded a χ2of 9.81 (p= .044) in the omnibus test. The model correctly classified 79% of the sample – 31/34 (91%) of NonASD children but only 10/18 (56%) of children with ASD.


Multiple measures of infant communication correlated with preschool SA and/or RRB symptoms. Such measures may collectively support earlier “best estimate” diagnoses of infants at GL-ASD, although none of the measures we tested had significant unique discriminative power within the model. The current model requires validation with an independent sample. Further, our model misclassified 44% of the children eventually diagnosed with ASD as “NonASD,” suggesting that ongoing monitoring and reassessments are warranted for screen-positive infants even when initial communication assessments do not support an ASD diagnosis.