LI Xuhua 1 , CHEN Hao 1 , BAI Qiyuan 1 , LIU Shidong 2 , ZHANG Yalan 2 , LIU Hongxu 1 , SONG Bing 1,2
  • 1. The First Clinical Medical College of Lanzhou University, Lanzhou, 730000, P. R. China;
  • 2. Department of Cardiovascular Surgery, The First Hospital of Lanzhou University, Lanzhou, 730000, P. R. China;
SONG Bing, Email: songbinldyy@163.com
Heart valve diseases; cardiopulmonary bypass; mechanical heart valves; acute respiratory distress syndrome; risk factors; prediction model; machine learning; artificial intelligence
10.7507/1007-4848.202510048
Export PDF Favorites Scan Get Citation

Abstract Full text Figures/Tables References Cited by

Objective To develop a predictive model for acute respiratory distress syndrome (ARDS) following cardiac mechanical valve replacement under cardiopulmonary bypass (CPB) using artificial intelligence algorithms, providing a novel method for early identification of high-risk ARDS patients. Methods Patients undergoing CPB-assisted cardiac mechanical valve replacement surgery in the Department of Cardiovascular Surgery at the First Hospital of Lanzhou University from January 2023 to March 2025 were retrospectively and consecutively enrolled. Data processing and model construction were performed using Python software. Variables with missing data proportions ≥30% were excluded, while multiple imputation combined with sensitivity analysis and standardization was applied to the remaining dataset. The dataset was randomly partitioned into training (70%) and testing (30%) sets. Feature selection was conducted using the Boruta algorithm and least absolute shrinkage and selection operator regression. The synthetic minority over-sampling technique edited nearest neighbors (SMOTEEN) algorithm was applied to balance samples in the training set. Six machine learning models, including random forest, light gradient boosting machine, extreme gradient boosting, categorical boosting (CatBoost), gradient boosting decision tree, and logistic regression, were developed through 5-fold nested cross-validation for parameter optimization. Model performance was evaluated via area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, average precision, recall rate, and F1 score. The optimal model was determined based on AUC values and validated through Hosmer-Lemeshow (HL) goodness-of-fit test. Decision curve analysis was performed for all models, while SHAP algorithm was employed for feature interpretation and visualization. External validation was conducted using clinical data from patients who underwent CPB-assisted mechanical valve replacement between April 1 and October 1, 2025. Results A total of 352 patients were included [training set: n=246, 135 males, 111 females, aged (51.71±11.03) years; testing set: n=106, 62 males, 44 females, aged (53.27±9.67) years], with 34 (9.7%) patients developing early ARDS in ICU. Key predictors included cardioplegia duration, right atrial transverse diameter, right ventricular transverse diameter, indirect bilirubin, and rewarming time. The CatBoost model demonstrated superior performance (AUC=0.828) with HL test P=0.64. In the single-center temporal validation cohort [n=41, 25 males, 16 females, aged (52.18±10.56) years], the CatBoost model achieved AUC=0.771. Conclusion Cardiac arrest duration, right atrial transverse diameter, right ventricular transverse diameter, indirect bilirubin, and rewarming time are identified as critical factors influencing postoperative ARDS development after CPB-assisted mechanical valve replacement. The CatBoost model exhibits excellent accuracy, consistency, and clinical applicability.

Copyright ? the editorial department of Chinese Journal of Clinical Thoracic and Cardiovascular Surgery of West China Medical Publisher. All rights reserved