Streamlining Autism Diagnostics: A Machine Learning Approach to Simplifying the ADOS-2 for Accessible ASD Assessment

Abstract

The Autism Diagnostic Observation Schedule-2 (ADOS-2) is the current gold standard  assessment tool for supporting the diagnosis of Autism Spectrum Disorder (ASD). However, the  administration of the ADOS-2 is resource-intensive, requiring specialized training and  significant financial investment, leading to delays in diagnosis and limited availability. As a  result, many children face significant obstacles in obtaining a timely ASD diagnosis, crucial for  accessing early intervention services. This study analyzed a dataset of ADOS-2 assessments taken  across 108 children utilizing a simple machine learning approach in order to simplify the  assessment down to the most predictive features. Utilizing recursive feature elimination (RFE)  with a multivariate linear regression, six key features from the ADOS-2 assessment were  identified as playing a critical role in predicting the overall ASD severity score: echolalia, quality of response, stereotyped words, conversation, eye contact, and quality of rapport. The optimized  model also demonstrated a robust predictive capability, achieving a Pearson correlation  coefficient greater than 0.9 after bootstrapping the algorithm to run 1,000 times. These findings  suggest that a streamlined approach to ASD diagnosis may greatly enhance diagnostic efficiency  and improve accessibility for early intervention. Future work should focus on validating these  findings in a clinical setting, with a larger dataset, and integrating them with other measures of  ASD in children.