Predicting autism spectrum disorder using maternal risk factors: A multi-center machine learning study

Area of Science:

• Neurodevelopmental Disorders
• Environmental Health
• Computational Biology

Background:

• Autism spectrum disorder (ASD) has a complex etiology influenced by maternal environmental factors.
• Previous research has limited the scope of maternal risk factors considered and multi-center studies.
• Machine learning offers a promising approach to integrate diverse maternal factors for ASD prediction.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting ASD risk based on maternal factors across preconception, perinatal, and postnatal periods.
• To identify key maternal risk factors contributing to ASD development.
• To assess the clinical applicability of a predictive model for identifying high-risk populations.

Main Methods:

• Developed five predictive models using 57 maternal risk factors from a multi-center cohort (1232 ASD, 1090 typically developing).
• Utilized extreme gradient boosting, achieving 66.2% accuracy on an external validation cohort (266 ASD, 353 typically developing).
• Employed Shapley values to determine the importance of maternal risk factors, calculating ASD risk scores and risk groups.

Main Results:

• The extreme gradient boosting model demonstrated the best performance in predicting ASD.
• Unstable maternal emotions and lack of multivitamin supplementation were identified as the most significant risk factors.
• A high-risk group, identified by the model, showed a significantly increased risk of ASD compared to the low-risk group.

Conclusions:

• Machine learning models effectively predict ASD risk using maternal factors, offering insights into emotional and nutritional influences.
• The developed model holds potential for clinical application in early identification of high-risk individuals for ASD.
• This study underscores the importance of considering a comprehensive set of maternal factors throughout pregnancy for ASD risk assessment.