Explainable Artificial Intelligence-driven Risk Assessment for Malignant Ventricular Arrhythmia and Mortality in...

Area of Science:

• Cardiology
• Artificial Intelligence
• Predictive Analytics

Background:

• Malignant ventricular arrhythmia (MVA) is a critical complication post-acute myocardial infarction (AMI), frequently leading to sudden cardiac death.
• Early identification and intervention are crucial for managing MVA risk in AMI patients.

Purpose of the Study:

• To develop and validate interpretable artificial intelligence (AI) models for predicting MVA and in-hospital death following AMI.
• To assess the performance of various AI models, including XGBoost, LightGBM, and Random Forest, in a large patient cohort.

Main Methods:

• Utilized data from 4471 patients across two medical centers for model development and validation.
• Developed seven AI models using nested 5-fold cross-validation.
• Evaluated predictive performance via AUROC curves, calibration curves, and decision analysis curves.

Main Results:

• The XGBoost model achieved an AUROC of 0.792 for the composite endpoint (MVA and in-hospital death) in the validation group.
• LightGBM excelled in MVA prediction (AUROC = 0.827), while Random Forest was superior for mortality prediction (AUROC = 0.784).
• External validation demonstrated robust performance, with the XGBoost model achieving an AUROC of 0.726 for the primary endpoint.

Conclusions:

• An interpretable AI framework integrating multimodel analysis was developed for AMI risk management.
• The validated AI system provides clinicians with a tool for personalized risk assessment, potentially improving patient outcomes.
• The AI framework facilitates early intervention strategies through real-time risk assessment capabilities.