Machine Learning Model Can Predict Autism Spectrum Disorder

XGBoost model demonstrated strong performance with AUROC of 0.895 and sensitivity and specificity of 0.805 and 0.829

By Elana Gotkine HealthDay Reporter

MONDAY, Aug. 19, 2024 (HealthDay News) — In a diagnostic study, machine learning (ML) can predict autism spectrum disorder (ASD), according to a study published online Aug. 19 in JAMA Network Open.

Shyam Sundar Rajagopalan, Ph.D., from the Karolinska Institutet in Stockholm, and colleagues developed and validated an ML model for predicting ASD using a minimal set of features from background and medical information. A retrospective analysis of the Simons Foundation Powering Autism Research for Knowledge database was conducted, including data from 30,660 participants (15,330 with and 15,330 without ASD). Generalizable ML prediction models were developed using four algorithms: logistic regression, decision tree, random forest, and eXtreme Gradient Boosting (XGBoost).

The researchers found that the XGBoost model demonstrated strong performance, with an area under the receiver operating characteristic curve (AUROC) score of 0.895; sensitivity and specificity of 0.805 and 0.829, respectively; and a positive predictive value of 0.897. The most important predictors were developmental milestones and eating behavior. An AUROC of 0.790 was seen on validation on independent cohorts, indicating good generalizability.

“Early medical information in child care clinics can be used to screen for those with a higher probability of being diagnosed with ASD,” the authors write.