Predicting the Aortic Aneurysm Postoperative Risks Based on Russian Integrated Data.

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Thoracic aortic aneurysm (TAA) operations carry significant postoperative risks.
• Unstructured medical records contain valuable data for predicting patient outcomes.
• Integrating data from multiple centers enhances model generalizability.

Purpose of the Study:

• To extract features from Russian unstructured medical records for TAA patients.
• To predict postoperative complications using machine learning algorithms.
• To evaluate the performance of different algorithms in risk prediction.

Main Methods:

• Feature extraction from 72 variables in unstructured Russian medical records.
• Integration of datasets from two distinct medical centers.
• Application of machine learning algorithms, including XGBoost, for prediction of 8 target complications: Mortality, Temporary Neurological Deficit (TND), Permanent Neurological Deficit (PND), Prolonged Lung Ventilation (LV), Renal Replacement Therapy (RRT), Bleeding, Myocardial Infarction (MI), and Multiple Organ Failure (MOF).

Main Results:

• XGBoost algorithm achieved the highest performance across most target variables.
• F-measure scores ranged from 0.74 to 0.95 for XGBoost predictions.
• The predictive performance is comparable to existing studies in cardiovascular postoperative risk prediction.

Conclusions:

• Machine learning, particularly XGBoost, is effective for predicting postoperative complications in TAA surgery.
• Feature extraction from unstructured data is a viable approach for enhancing risk prediction models.
• This method offers a valuable tool for improving patient care and outcomes in complex cardiovascular operations.