Complication Risk Classification in Children and Adolescents With Type 1 Diabetes: Interpretable Machine Learning...

Area of Science:

• Pediatric Endocrinology
• Clinical Decision Support Systems
• Health Informatics

Background:

• Type 1 diabetes (T1D) complications pose significant health risks for children and adolescents.
• Existing clinical data lacks interpretable tools for effective risk stratification in this demographic.
• There's a need for risk assessment tools aligned with local clinical guidelines.

Purpose of the Study:

• To develop a clinically interpretable model for classifying T1D complication risks (acute, chronic, low).
• To utilize real-world pediatric T1D data and expert rules from the Saudi Diabetes Clinical Practice Guidelines.
• To create a tool supporting early detection and management of T1D complications.

Main Methods:

• A dataset of 306 pediatric T1D patients was preprocessed and engineered.
• Risk labels were derived from the Saudi Diabetes Clinical Practice Guidelines.
• A hybrid feature selection approach (SHAP + exhaustive) identified key predictors.
• A decision tree model was trained and optimized using cross-validation (F1-score metric).

Main Results:

• The final model achieved a high mean F1-score of 0.9876 with low variance (0.0189).
• Only five clinical features were needed: BMI, hypoglycemia, disease duration, HbA1c, and impaired glucose metabolism.
• The decision tree provided a transparent logic path, enhancing interpretability.

Conclusions:

• A simple, interpretable model guided by national guidelines can effectively predict T1D complication risks in pediatric patients.
• The model's strong performance and clarity make it suitable for clinical decision support systems.
• This approach supports a move towards predictive and personalized diabetes management.