Machine Learning-Based Prediction of Three-Year Heart Failure and Mortality After Premature Ventricular Contraction...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Long-term outcomes, including heart failure and mortality, following catheter ablation for premature ventricular contractions (PVCs) are not well understood.
• Effective risk stratification models are needed to identify patients at higher risk for adverse events post-ablation.

Purpose of the Study:

• To develop and compare machine learning models for predicting three-year heart failure and mortality after PVC ablation.
• To assess the performance of different algorithms in handling class imbalance in predicting rare events.

Main Methods:

• Retrospective analysis of 4195 adult patients undergoing PVC ablation using a nationwide claims database.
• Application of synthetic minority over-sampling technique (SMOTE) and random over-sampling examples (ROSE) to address class imbalance.
• Comparison of logistic regression, decision tree, random forest, XGBoost, and LightGBM, evaluated using ROC AUC and PR curves.

Main Results:

• LightGBM with ROSE achieved the highest ROC AUC (0.822) for predicting three-year heart failure.
• Logistic regression and LightGBM with ROSE demonstrated comparable performance (ROC AUCs 0.886 and 0.882) for three-year mortality prediction.
• Key predictors identified include age, prior heart failure, malignancy, and end-stage renal disease.

Conclusions:

• Machine learning models, especially LightGBM with ROSE, provide robust and clinically interpretable risk stratification after PVC ablation.
• These models can be integrated into electronic health records for improved patient management.
• Further external validation and local threshold optimization are recommended.