Machine Learning-Based Risk Stratification for Sudden Cardiac Death Using Clinical and Device-Derived Data.

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Sudden cardiac death (SCD) poses a significant clinical challenge.
• Implantable cardioverter-defibrillators (ICDs) are the primary preventive measure for SCD.
• Current patient selection for ICDs relies on imperfect risk markers.

Purpose of the Study:

• To evaluate the potential of machine learning (ML) models in improving SCD risk prediction.
• To utilize tabular clinical data, including ECG and ICD-derived features, for enhanced risk assessment.
• To refine patient selection for ICD implantation using advanced predictive analytics.

Main Methods:

• Trained various ML models (Random Forest, Naive Bayes, Logistic Regression, Voting Classifiers) on diverse patient data.
• Included demographic, clinical, laboratory, and device-derived variables.
• Optimized models for F2-score to prioritize high-risk patient detection and used SHAP values for interpretability.

Main Results:

• The Random Forest model achieved the highest F2-score (0.74) and high recall (97.30%).
• Voting Classifiers demonstrated the best overall discrimination (AUC-ROC 0.76).
• ML models successfully identified known and potential SCD predictors, confirming their predictive capability.

Conclusions:

• Machine learning models show significant potential for refining patient selection for ICDs.
• The high recall achieved by ML models indicates improved detection of high-risk individuals for SCD.
• ML-driven risk prediction offers a promising avenue to enhance cardiovascular care and reduce SCD incidence.