An Interpretable Machine Learning Model for Predicting Intellectual Disability in Children With Cerebral Palsy

Area of Science:

• Neuroscience
• Machine Learning
• Pediatrics

Background:

• Intellectual disability (ID) affects 45% of children with cerebral palsy (CP).
• Early ID identification in CP is challenging due to motor and communication impairments.
• This study focuses on predicting ID risk in children with CP.

Purpose of the Study:

• Develop and validate an interpretable machine learning (ML) framework.
• Predict intellectual disability (ID) risk in children with cerebral palsy (CP).
• Provide a decision-support tool for timely interventions.

Main Methods:

• Retrospective registry-based study of 807 children with CP.
• Used clinical and neuroimaging data available by age 2.
• Trained and compared eight ML algorithms, using SHAP for interpretability.

Main Results:

• Optimized ML models achieved an AUC of 0.813.
• Key predictors included inability to sit independently by age 2, epilepsy, spastic quadriplegia, and severe GMFCS levels.
• SHAP analysis provided global and individual risk insights.

Conclusions:

• The transparent ML framework is a reliable decision-support tool.
• Translates complex ML output into clinically understandable insights.
• Facilitates personalized, timely neurodevelopmental interventions for high-risk children.