A systematic statistical analysis of Q-CHAT using DSM-5 criteria and predictive modeling on data from polish toddlers

Autism-Spectrum-Disorder (ASD) is characterized by notable difficulties in social interactions, communication and conducts, with screening or diagnosis primarily based on behavioral observations. The Quantitative-Checklist-for-Autism-in-Toddlers (Q-CHAT) and its variants are widely used in this context and have been the focus of numerous studies. However, important gaps remain regarding the alignment of the Q-CHAT with DSM-5 criteria and its empirical-assessment. Addressing these gaps contributes to the evaluation of the Q-CHAT’s effectiveness, suitability and reliability as a screening tool for ASD. This study presents a systematic mapping of Q-CHAT items on DSM-5 criteria and evaluates their association with target-feature (ASD Yes/No) using a labeled secondary dataset of 252 Polish toddlers. It further identifies the most suitable data-driven machine-learning-model for ASD prediction based on both original and refined feature sets. A two-step systematic mapping process revealed underrepresentation of certain DSM-5 subdomains in Q-CHAT, particularly “relational-communication”. An integrated feature-selection-strategy using Chi-square analysis, and Recursive-Feature-Elimination with Random-Forest (RF) identified eleven items of Q-CHAT that were not consistently supported by both selection criteria. Finally, four machine-learning models, namely Decision-Tree, Random-Forest, Support-Vector-Machine, and Logistic-Regression were compared using both original and refined set of features. Results showed that, the refined feature set achieved comparable performance to the original feature set across all models except Decision-Tree, suggesting that the excluded items had minimal overall impact on predictive performance. RF achieved 92% of accuracy, 91% of specificity and 93% of precision, sensitivity and F1-score, demonstrating strong predictive performance. The proposed process can be adopted for analyzing other practicing screening tools of ASD. Such investigations will help in standardization of items/tools across DSM-5 criteria.