Long-term mortality prediction after operations for type A ascending aortic dissection.

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Limited long-term mortality prediction studies exist for acute aortic dissection (AAD) Type A.
• Advanced computational models like neural networks (NN) and support vector machines (SVM) offer potential for improved discriminatory power over traditional methods.

Purpose of the Study:

• To evaluate the efficacy of NN and SVM in predicting long-term mortality following AAD Type A.
• To identify key risk factors contributing to long-term mortality in AAD Type A patients using novel predictive models.

Main Methods:

• Utilized 32 risk factors identified from literature and prior short-term outcome studies.
• Trained and validated NN and SVM models on a cohort of 235 AAD Type A patients.
• Assessed model discrimination using receiver operating characteristic area under the curve (AUC) and classification performance metrics.

Main Results:

• Identified 5 key predictors of long-term mortality: post-operative chronic renal failure, circulatory arrest time, ascending aorta/hemi-arch surgery type, extracorporeal circulation time, and Marfan habitus.
• NN models demonstrated excellent training and validation accuracy (AUC ~0.870) but higher classification errors for deceased patients.
• SVM models showed promising results, with training achieving perfect discrimination (AUC 1.0) and validation yielding an error rate of 22% (AUC 0.821).

Conclusions:

• NN and SVM effectively identified operative and immediate post-operative factors, along with Marfan habitus, as significant predictors of long-term mortality in AAD Type A.
• The identified combination of factors can be practically applied to accurately assess post-operative death risk.