Predicting Left Ventricular Ejection Fraction Recovery After Percutaneous Coronary Intervention in Patients With...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Accurate prediction of left ventricular ejection fraction (LVEF) recovery post-percutaneous coronary intervention (PCI) is vital for managing patients with chronic coronary syndrome (CCS).
• Current prediction methods may lack the precision needed for optimal clinical decision-making.

Purpose of the Study:

• To develop and compare various machine learning (ML) models for predicting LVEF recovery after PCI.
• To identify the most significant clinical and laboratory features influencing LVEF recovery.

Main Methods:

• Retrospective analysis of 520 CCS patients using a clinical database.
• Development of 48 ML models by combining 3 feature selection strategies (LASSO, RFE, all features) with 4 algorithms (XGBoost, LightGBM, CatBoost, Random Forest).
• Model performance evaluation using 10-fold cross-validation, AUC, decision curve analysis, and calibration plots.

Main Results:

• Models combining feature selection with XGBoost achieved high predictive performance (AUC up to 0.93).
• Key predictors identified include uric acid, platelets, hematocrit, brain natriuretic peptide, and baseline LVEF.
• Recursive feature elimination (RFE) with XGBoost showed the highest AUC for predicting good recovery in patients with preserved LVEF.

Conclusions:

• Machine learning models with feature selection demonstrate robust predictive capabilities for LVEF recovery post-PCI.
• Interpretable ML models can enhance clinical decision-making and improve the management of CCS patients.
• The study highlights the potential of ML in personalizing cardiovascular care.