Machine Learning Prediction of Autism Spectrum Disorder Diagnosis Through Linking Mothers’ and Children’s Electronic Health Records

Autism spectrum disorder (ASD) is a neuro-developmental disorder typically diagnosed in children. Early detection of ASD, particularly in girls who are often diagnosed later than boys, can significantly aid long-term development. This study aimed to develop machine learning models for predicting ASD diagnosis in children, including both boys and girls, using child-mother linked electronic health records (EHRs) data from a large clinical research network. Model features included children and mothers’ risk factors in EHRs, particularly maternal health factors. We tested XGBoost and logistic regression with Random Oversampling (ROS) and Random Undersampling (RUS) to address imbalanced data. Logistic regression with ROS, considering a three-year observation window for children’s risk factors, achieved the best performance for predicting ASD among the overall study population (AUROC = 0.824), boys (AUROC = 0.782), and girls (AUROC = 0.791). SHAP values were calculated to quantify the impact of important clinical and sociodemographic risk factors.