Divergent Autonomic Response Profiles in Early Childhood Autism

Background: The heterogeneous presentation of Autism Spectrum Disorder (ASD) presents challenges for clinical care and research, including complicating the development of methods to assess clinically relevant indicators of prognosis (long-term functional outcomes) and treatment response. There is a need for measures that can mark clinically meaningful individual differences in underlying features or characteristics related to short-term responses to treatment and long-term functional outcomes. Physiologic measures of autonomic regulation may be promising biomarkers for prognosis and outcome.

Objectives: To utilize autonomic responses to index individual differences in social and regulatory responses of children with autism during standardized behavioral assays of social engagement and emotion regulation.

Methods: Participants were 104 children, 66 with ASD and 38 typically developing controls (age range 2–7; M= 4.3 years, 79% male). IQ was assessed using the Stanford Binet-5 and scores for the ASD group varied widely (range: 40-122; M=70.6; SD=23.4). ECG was acquired at 1kHz during standardized behavioral assays designed to reflect levels of social engagement (responses to opportunities for social interaction) and emotion regulation (responses to frustrating tasks such as a sabotaged toy). Heart rate (HR) and RSA were derived from the ECG. RSA was defined as HR variability in the frequency range of respiration (spectral method; frequency band: .24 – 1.04 Hz). Latent cluster analyses were performed using the HR/RSA responses during social engagement tasks and frustration tasks for the ASD group using a Gaussian Mixture Model method. Models were evaluated using Bayesian Information Criteria (BIC). Missing data was addressed using multiple imputation (predictive mean matching). Resulting ASD clusters were compared to typical controls.

Results: A 2-cluster solution was found for HR/RSA responses to Frustration Episodes. A single “cluster” was found for the HR/RSA responses to the Social Engagement tasks. BIC indicated that the 2-cluster solution for the Frustration responses was a very good fit for the data and visual inspection of cluster plots indicated very good separation between the two resulting ASD classes. As expected based on latent classifications, the ASD Class 1 children had lower RSA and higher HR than ASD Class 2 children (p’s < 0.001). While the overall ASD group did not differ from the typical group, the ASD Class 1 and 2 children differed from typicals in opposite directions. ASD Class 1 children had higher HR than typicals (p < 0.001), whereas ASD Class 2 children had lower HR during frustration events. ASD Class 2 children had higher RSA than typicals (p < 0.001).

Conclusions: Our results identified two subsets of children with ASD who have qualitatively different autonomic response profiles during mildly frustrating events. Physiologic response profiles during standardized behavioral assays can be used to dissect ASD heterogeneity into clinically meaningful subgroups. If validated, these biomarkers may help to predict behavioral outcomes and/or treatment response. Ongoing analyses will test the relationship between frustration and social engagement responses, and will investigate whether the ASD classes differ on measures of behavioral functioning.