Abstract
BACKGROUND: Acute aortic dissection is a life-threatening condition, and malperfusion significantly exacerbates the prognosis of patients diagnosed with type A Acute aortic dissection (ATAAD). Current risk assessment tools often fail to consider the impact of nutritional and inflammatory factors, limiting their predictive accuracy. The aim of this study was to develop a machine learning model that integrates nutritional and inflammatory indices to predict 1-year mortality in ATAAD patients with malperfusion. METHODS: This retrospective study included 433 ATAAD patients with malperfusion from Henan Provincial Chest Hospital (August 2020 to June 2023). Four machine learning models-logistic regression, XGBoost, random forest, and deep neural network-were developed to predict 1-year mortality using inflammatory and nutritional laboratory values, indices, and other clinical variables. Model training employed stratified 5-fold cross-validation and SMOTE for imbalanced data. The area under the receiver operating characteristic (ROC AUC) and other performance metrics were used to evaluate model efficacy, while SHAP values were computed to interpret feature importance. RESULTS: Among 433 ATAAD patients with malperfusion, the random forest model used inflammatory and nutritional laboratory values to achieve the highest discrimination (AUC = 0.8242, 95% CI 0.7095-0.9219), while the XGBoost model performed best with inflammatory and nutritional indices (AUC = 0.7334, 95% CI 0.6115-0.8488). Calibration curves and Brier scores indicated good agreement between predicted and observed outcomes. Decision curve analysis demonstrated consistent net benefit for random forest and XGBoost models across clinically relevant threshold probabilities. Feature importance and SHAP analyses identified albumin, platelet count, total cholesterol, and C-reactive protein as consistently influential predictors. CONCLUSION: Nutritional and inflammatory factors significantly contribute to the 1-year mortality risk of ATAAD patients with malperfusion. Machine learning models that incorporate these factors, particularly random forest and XGBoost, can effectively stratify patient risk and support clinical decision-making. These findings underscore the importance of a comprehensive approach to risk assessment that includes metabolic and inflammatory markers to enhance patient outcomes and guide personalized interventions.