A predictive model for paediatric autism screening.

Area of Science:

• Neurodevelopmental disorders
• Global mental health

Background:

• Autism spectrum disorder (ASD) presents challenges in social interaction, communication, and behavior.
• Current screening tools like the Modified Checklist for Autism in Toddlers show cultural variability in sensitivity, particularly in regions like Sri Lanka.
• Low- and middle-income countries face significant barriers to early ASD diagnosis due to a scarcity of mental health specialists.

Purpose of the Study:

• To propose a novel, culturally sensitive mobile application for autism spectrum disorder screening.
• To address the diagnostic gap in low- and middle-income countries by leveraging technology.
• To facilitate early identification and intervention for autism spectrum disorder.

Main Methods:

• Development of a mobile application integrating an intelligent machine learning model.
• Utilization of a clinically validated symptom checklist, the Pictorial Autism Assessment Schedule (PAAS).
• Training and evaluation of machine learning models on PAAS data, including feature selection for optimization.

Main Results:

• The random forest classifier demonstrated optimal predictive performance with an area under the receiver operating characteristic curve of 0.98.
• Feature selection identified redundant questions within the PAAS, enabling a more streamlined screening process.
• The study successfully developed and validated a machine learning model for a mobile screening tool.

Conclusions:

• The proposed mobile application offers a promising solution for culturally sensitive autism spectrum disorder screening in resource-limited settings.
• Early detection of autism spectrum disorder is crucial for timely intervention and improved developmental outcomes.
• Machine learning integration can enhance the efficiency and accuracy of autism spectrum disorder screening tools globally.