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

Malignant ventricular arrhythmia (MVA) is a severe complication that can occur after acute myocardial infarction, often leading to sudden cardiac death.

A total of 4471 patients from 2 medical centers were included in this study. The primary endpoint was a composite of MVA and in-hospital death. Seven state-of-the-art artificial intelligence (AI) models were developed and optimized by nested 5-fold cross-validation. Predictive performance was evaluated using the area under the receiver-operating characteristic (AUROC) curve, the calibration curve, and the decision analysis curve.

Among the enrolled patients, 3456 were assessed for model development and validation and 1015 patients from another medical center were asssessed for external validation. In the validation group, the eXtreme Gradient Boosting (XGBoost) model achieved the highest AUROC of 0.792 (95% confidence interval [CI] 0.740-0.845) for the composite endpoint. The Light Gradient Boosting Machine (LightGBM) model demonstrated superior performance for MVA prediction (AUROC = 0.827, 95% CI 0.768-0.885), whereas the Random Forest (RF) model outperformed the others for mortality prediction (AUROC = 0.784, 95% CI 0.720-0.848). In the external validation group, the AUROC of the XGBoost model with 15 variables for predicting the primary endpoint event was 0.726. The AUROCs were 0.704 for the LightGBM model with 15 variables for predicting MVA and 0.823 for the RF model with 20 variables for predicting in-hospital death. The Web-based prediction system showed real-time risk assessment capabilities.

Our study presents an interpretable AI framework integrating multimodel analysis for acute myocardial infarction risk management. The system offers clinicians a validated tool for personalized risk assessment that can potentially improve patient outcomes through early intervention strategies.