Abstract
Acute Type A Aortic Dissection (TAAD) is a life-threatening cardiovascular emergency, and early identification of high-risk patients for rupture is critical for optimizing surgical resource allocation. Existing studies predominantly focus on postoperative mortality prediction, lacking tools for in-hospital rupture risk assessment upon admission, with limitations including small sample sizes, feature redundancy, and single-algorithm bias. This study proposes an innovative framework integrating Boruta feature selection, Conditional Tabular Generative Adversarial Network (CTGAN) for data augmentation, and Blending ensemble strategy to enhance predictive performance. Utilizing 200 original TAAD cases, CTGAN synthesized 900 high-fidelity samples. Key features (e.g., CKMB, lactate) were selected via Boruta, and a Blending ensemble model combining eight base models (e.g., Random Forest, XGBoost) was developed. The model’s performance was evaluated using AUC, sensitivity, and F1-score. The Blending ensemble model achieved an AUC of 0.978, sensitivity of 0.920, and F1-score of 0.919, outperforming individual models. The study addresses small-sample constraints through CTGAN and leverages Blending to harness complementary strengths of diverse algorithms, providing a high-accuracy and interpretable tool for emergency triage. This framework fills the gap in TAAD rupture risk prediction and offers insights for clinical decision-making in resource-limited settings. The integration of Boruta feature selection, CTGAN data augmentation, and Blending ensemble strategy enhances model robustness and interpretability, making it a valuable tool for clinical applications.