Interpretable prognostic modeling for long-term survival of Type A aortic dissection patients using support vector...

Area of Science:

• Cardiovascular Surgery
• Machine Learning in Medicine
• Aortic Dissection Research

Background:

• Type A aortic dissection (TAAD) poses significant long-term survival challenges.
• Accurate prediction of TAAD patient outcomes is crucial for effective treatment planning.

Purpose of the Study:

• To develop a reliable and interpretable machine learning (ML) model for predicting long-term survival in Type A aortic dissection (TAAD) patients.
• To identify key prognostic factors influencing survival in TAAD.

Main Methods:

• Retrospective review of TAAD patient data undergoing open surgical repair.
• Utilized LASSO Cox regression for prognostic factor identification and Support Vector Machine (SVM) for predictive modeling.
• Employed SHapley Additive exPlanation (SHAP) values for model interpretability.

Main Results:

• A robust SVM model was developed, demonstrating excellent performance across training and testing datasets (AUCs ranging from 0.85 to 0.91).
• Key predictors identified include operation time, cardiopulmonary bypass (CPB) duration, aortic cross-clamp (ACC) time, age, plasma transfusion volume, creatinine, and white blood cell (WBC) count.
• The model showed strong clinical applicability with no significant overfitting.

Conclusions:

• An interpretable SVM-based predictive model for TAAD long-term survival was successfully developed.
• The model provides accurate, precise, and robust identification of high-risk patients, offering valuable clinical evidence for improved patient management.