Explainable machine-learning predictions for complications after pediatric congenital heart surgery

Area of Science:

• Pediatric Cardiology
• Machine Learning in Medicine
• Surgical Outcomes Research

Background:

• Pediatric congenital heart surgery outcomes require improvement.
• Accurate prediction of postoperative complications is crucial for timely intervention.
• Existing risk models lack sufficient predictive power and interpretability.

Purpose of the Study:

• To develop an interpretable machine learning model for predicting postoperative complications in pediatric congenital heart surgery.
• To integrate diverse data sources including demographics, surgical details, and intraoperative blood pressure.
• To enhance clinical decision-making and improve patient prognosis.

Main Methods:

• Developed an interpretable machine learning model using patient demographics, surgery-specific features, and intraoperative blood pressure data.
• Extracted time-series features from blood pressure data using k-means and dynamic time warping.
• Employed the SHAP framework for model interpretability and explanation of predictions.

Main Results:

• The developed model demonstrated superior performance in both binary and multi-label classification compared to existing risk models.
• The model provides explanations for its predictions, enhancing clinical understanding of complication risks.
• Achieved high accuracy in predicting postoperative complications, facilitating prompt therapeutic adjustments.

Conclusions:

• The interpretable machine learning model offers a reliable tool for predicting complications in pediatric congenital heart surgery.
• Model interpretability fosters clinical trust and provides actionable insights for managing patient risk.
• This approach has the potential to significantly improve treatment quality and patient prognosis.