Prediction of spinal cord injury after thoracoabdominal aortic repair by machine learning method

Background: Spinal cord injury (SCI) is a common and serious complication after thoracoabdominal aortic repair, which significantly increases the mortality and disability. Most of the existing studies were based on traditional modeling to predict the risk factors of SCI with surgical data. An improved machine learning method used comprehensive perioperative data-driven learning to predict the occurrence of SCI. Methods: 270 patients were included in the study, and 66 potentially meaningful variables were selected from 84 original variables as input data to drive machine learning. The improved machine learning algorithm (PSO-FLXGBoost) was used to stratify and rank the risk factors, and finally SHAP was used to assign attributional values for interpretation. Findings: This PSO-FLXGBoost model has a high AUC (0.895) and a good prediction effect. The SHAP analysis of the four most important variables showed that lower intraoperative hemoglobin (＜70g/L), higher preoperative D-Dimer (＞6ug/ml) and platelet (＞250*10^9/L), and prolonged operation time (＞500min) can significantly increase the probability of SCI. Interpretation: An improved machine learning method was developed to predict SCI after TAAR. The research demonstrated that intraoperative hemoglobin, preoperative coagulation status and operation time are crucial for SCI prediction. Funding: This study was funded by Beijing Municipal Natural Science Foundation (Grant No. 7224341), Beijing Science and Technology Program (NO. Z191100007619042) and CAMS Innovation Fund for Medical Sciences (2022-I2M-C&T-B-039). Declaration of Interest: None. Ethics Approval Statement: The protocol was approved by the Ethics Committee of Fuwai Hospital, National Center for Cardiovascular Diseases. (Ethical approval number: 2022-1745) Informed consent was waived because the study was retrospective.