Table 7_Machine learning prediction of ARDS after heart valve surgery: development and validation in Northwest China.docx

ObjectiveTo develop an AI-based predictive model for acute respiratory distress syndrome (ARDS) following cardiopulmonary bypass (CPB)-assisted heart valve replacement (HVR) to enable early identification of high-risk patients.
MethodsWe retrospectively analyzed 400 patients who underwent CPB-assisted HVR between January 2023 and February 2025. After data preprocessing and feature selection, the dataset was split into training (n = 280) and test (n = 120) sets. Multiple machine learning models were developed and optimized, with XGBoost emerging as the optimal model based on training performance.
ResultsAmong 400 patients, 56 (14%) developed ARDS postoperatively. Key predictors included Age, absolute monocyte count,right atrial transverse diameter, intraoperative blood loss, platelet count, main pulmonary artery diameter. The XGBoost model achieved excellent performance with an AUC of 0.853 and demonstrated good calibration (HL test p > 0.05).
ConclusionThe XGBoost model accurately predicts ARDS risk following CPB-assisted HVR using six clinically relevant predictors, providing a valuable tool for early risk stratification and potential intervention in high-risk patients.

**研究目的**：构建基于人工智能的体外循环（CPB）辅助心脏瓣膜置换术（HVR）后急性呼吸窘迫综合征（ARDS）的预测模型，以实现高危患者的早期识别。
**研究方法**：本研究回顾性分析2023年1月至2025年2月期间接受CPB辅助HVR的400例患者临床资料。经数据预处理与特征筛选后，将数据集划分为训练集（n=280）与测试集（n=120）。构建并优化了多种机器学习模型，基于训练集性能评估结果，最终选定XGBoost模型为最优模型。
**研究结果**：400例患者中，共有56例（14%）在术后发生ARDS。关键预测因子包括年龄、单核细胞绝对计数、右心房横径、术中失血量、血小板计数及主肺动脉直径。XGBoost模型展现出优异的预测性能，受试者工作特征曲线下面积（AUC）为0.853，且校准性能良好（霍斯默-莱梅肖检验，P>0.05）。
**研究结论**：基于6项临床相关预测因子构建的XGBoost模型可精准预测CPB辅助HVR术后ARDS发生风险，可为高危患者的早期风险分层与潜在干预措施提供极具价值的辅助工具。