Prediction of Autism at 3 Years from Behavioural and Developmental Measures in High-Risk Infants: A Longitudinal...

Area of Science:

• Neurodevelopmental disorders
• Machine learning in healthcare

Background:

• Autism Spectrum Disorder (ASD) diagnosis relies on behavioral observation.
• Early identification is crucial for timely intervention and improved outcomes.
• Predicting individual ASD trajectories remains challenging.

Purpose of the Study:

• To enhance early prediction of Autism Spectrum Disorder (ASD) outcome.
• To integrate multiple behavioral and developmental measures using machine learning.
• To assess prediction accuracy at different early time points.

Main Methods:

• Longitudinal study of high-risk (HR) siblings and low-risk (LR) controls from 8 to 36 months.
• Utilized Mullen Scales of Early Learning, Vineland Adaptive Behavior Scales, and early ASD symptoms.
• Applied machine learning algorithms for classification of ASD and atypical development.

Main Results:

• At 8 months, machine learning classified HR-ASD at chance level; broader atypical development had 69.2% AUC.
• At 14 months, ASD and broader atypical development were classified with approximately 71% AUC.
• Longitudinal analysis showed LR and HR-Typical groups had better development and fewer ASD symptoms than HR-Atypical and HR-ASD groups.

Conclusions:

• Machine learning offers moderate accuracy for predicting ASD and atypical development by 14 months.
• Integrating multiple developmental measures improves early ASD outcome prediction.
• Further research is needed to enhance prediction accuracy at earlier stages.