Machine learning approach to analyzing complex care coordination patterns for medically complex children.

Area of Science:

• Pediatric Healthcare
• Artificial Intelligence in Medicine
• Health Informatics

Background:

• The population of children with medical complexity (CMC) in Japan exceeds 20,000, with significant international implications.
• Care coordination for CMC is challenging due to the need to address growth, school attendance, and limited coordinator experience.
• Investigated the potential of machine learning (ML) to support care coordination for CMC.

Purpose of the Study:

• To assess the feasibility of using ML models to predict care needs for CMC.
• To identify critical factors influencing service utilization and care patterns in CMC.
• To evaluate the effectiveness of AI in enhancing care coordination for CMC.

Main Methods:

• Developed three gradient boosting prediction models using 1,042 care sheets from a pediatric respite facility.
• Models predicted ventilator use, home doctor utilization, and age to assess care requirements.
• Utilized SHAP values for model interpretability and performed ten-fold cross-validation to mitigate sampling bias.

Main Results:

• Achieved high F1 scores (0.83, 0.826, 0.829) for the prediction models.
• Identified significant predictors such as equipment maintenance for ventilator use and complex service utilization for home doctor needs.
• Demonstrated varying factors influencing care needs across different school-age transitions.

Conclusions:

• AI shows significant potential as a tool for identifying crucial factors in CMC services and improving care coordination.
• Emphasized the importance of digital healthcare information preservation for CMC.
• ML models offer a valuable approach to support care coordination for children with medical complexity.