Using a Novel Machine-Learning Algorithm as an Auxiliary Approach to Predict the Transfusion Volume in Mitral Valve Surgery

Background: Blood transfusion is an indispensable supportive therapy. It plays a pivotal role in the perioperative management of cardiac surgery. The aim of this study was to develop a model for predicting the transfusion volume in isolated mitral valve surgery. Methods: We gathered data from 677 patients undergoing isolated mitral valve surgery with and without simultaneous tricuspid valve operation. The dataset was partitioned into a training dataset (70%) and a testing dataset (30%). We evaluated 18 machine-learning algorithms, incorporating inputs from 36 demographic and perioperative features. Additionally, the performance of multiple linear regressions was compared with machine-learning algorithms. CatBoost was selected for further analysis, and Shapley additive explanation (SHAP) values were employed to evaluate feature importance. Finally, we explored the impact of various features on the accuracy of CatBoost by analyzing the reasons for misjudgment. Results: CatBoost outperformed all 18 machine learning algorithms with an R-squared value of 0.420, mean absolute error of 0.702, mean squared error of 1.208, and root mean squared error of 1.090, surpassing multiple linear regression. The analysis of the testing group achieved 72.5% accuracy. SHAP identified 20 pertinent features influencing transfusion volume. No significant differences were observed between correctly and incorrectly predicted groups in tricuspid valve repair, American Society of Anesthesiologists classification, or platelet count. Conclusion: CatBoost effectively predicts the intraoperative transfusion volume in mitral valve surgery, aiding clinicians in transfusion decision-making and enhancing patient care.