Model-driven survival prediction after congenital heart surgery

Area of Science:

• Cardiology
• Medical Informatics
• Pediatric Surgery

Background:

• Congenital heart surgery carries significant risks.
• Accurate postoperative risk assessment is crucial for patient management and safety.
• Existing risk prediction models often rely on preoperative data, limiting their utility for immediate postoperative adjustments.

Purpose of the Study:

• To develop and validate a machine-learning model for predicting 30-day postoperative survival in congenital heart surgery patients.
• To utilize easily accessible peri- and postoperative parameters for enhanced risk stratification.
• To improve the accuracy of risk assessment compared to traditional preoperative methods.

Main Methods:

• A bicentric retrospective analysis of 1765 congenital heart surgery procedures (2014-2019).
• Training and testing a machine-learning model using parameters including the STAT mortality score, age, aortic cross-clamp time, and postoperative lactate levels.
• Evaluating model performance using Area Under the Curve (AUC), sensitivity, and specificity.

Main Results:

• The developed model achieved a high Area Under the Curve (AUC) of 94.86%, with 89.48% specificity and 85.00% sensitivity.
• STAT mortality score and aortic cross-clamp time were highly significant predictors of mortality.
• Postoperative lactate levels provided critical information regarding mortality risk.
• The complete model, incorporating pre-, intra-, and 24-hour postoperative data, demonstrated a 53.5% reduction in prediction error compared to the STAT score alone.

Conclusions:

• The machine-learning model accurately predicts postoperative survival following congenital heart surgery.
• Postoperative risk assessment using this model halves the prediction error compared to preoperative assessments.
• Enhanced identification of high-risk patients can lead to improved preventive strategies and increased patient safety.