Zhang Zheng, Duan Yi, Li Zuozhi, Gao Zhifeng, Zhang Huan
Department of Anesthesiology, Beijing Tsinghua Changgung Hospital, School of Clinical Medicine, Tsinghua Medicine, Tsinghua University, Beijing, 102218, China.
Department of Cardiology, Fuwai Hospital, National Center for Cardiovascular Disease, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
BMC Cardiovasc Disord. 2025 Aug 6;25(1):581. doi: 10.1186/s12872-025-05026-7.
The objective of this study is to construct an explainable machine learning predictive model for high intraoperative blood pressure variability(IBPV) based on preoperative characteristics, to enhance intraoperative circulatory management and surgical outcomes. This study utilized a retrospective observational design, employing the eXtreme Gradient Boosting (XGBoost) algorithm to create a predictive model for high IBPV.
The data for the study were obtained from the central operating room of a major hospital in Beijing, China, covering the period from March 2016 to April 2022. A total of 37,756 noncardiac surgeries were included in the analysis. The dataset comprised demographic, preoperative laboratory, and diagnostic information. Selection criteria included all noncardiac surgeries with complete preoperative data. High IBPV was defined as a coefficient of variation exceeding 20% during the surgical procedure. The main outcome measure was the prediction of high IBPV, assessed using the area under the receiver operating characteristic curve (AUC), accuracy, and specificity.
The XGBoost-based model achieved an accuracy of 0.81 and a specificity of 0.99, with a moderate discriminative ability (AUC = 0.60). SHAP analysis identified age and American Society of Anesthesiologists (ASA) classification as the top positive predictors for high IBPV, with maximum SHAP values of 0.4 and 0.2, respectively. Preoperative plasma albumin level was the key negative predictor, with a maximum SHAP value of -0.6. Interactions between preoperative blood calcium and age, and weight and age, were also influential. The model quantified individual high IBPV risk probabilities and variable contributions.
The XGBoost model effectively identifies significant predictors of high IBPV, including age, ASA classification, and plasma albumin levels, and is capable of estimating individual risk probabilities. However, external validation of the model in different clinical settings and populations is needed to further confirm its predictive performance and generalizability.
NCT05698433.
The online version contains supplementary material available at 10.1186/s12872-025-05026-7.
本研究的目的是基于术前特征构建一个可解释的机器学习预测模型，用于预测术中高血压变异性（IBPV），以加强术中循环管理并改善手术结果。本研究采用回顾性观察设计，运用极端梯度提升（XGBoost）算法创建高IBPV预测模型。
本研究的数据来自中国北京一家大型医院的中心手术室，涵盖2016年3月至2022年4月期间。分析共纳入37756例非心脏手术。数据集包括人口统计学、术前实验室检查和诊断信息。入选标准为所有术前数据完整的非心脏手术。高IBPV定义为手术过程中变异系数超过20%。主要结局指标是高IBPV的预测，采用受试者操作特征曲线下面积（AUC）、准确率和特异性进行评估。
基于XGBoost的模型准确率为0.81，特异性为0.99，具有中等判别能力（AUC = 0.60）。SHAP分析确定年龄和美国麻醉医师协会（ASA）分级是高IBPV的前两大正向预测因素，最大SHAP值分别为0.4和0.2。术前血浆白蛋白水平是关键的负向预测因素，最大SHAP值为 -0.6。术前血钙与年龄、体重与年龄之间的相互作用也有影响。该模型量化了个体高IBPV风险概率和变量贡献。
XGBoost模型有效地识别了高IBPV的重要预测因素，包括年龄、ASA分级和血浆白蛋白水平，并能够估计个体风险概率。然而，需要在不同临床环境和人群中对该模型进行外部验证，以进一步确认其预测性能和通用性。
NCT05698433。
在线版本包含可在10.1186/s12872-025-05026-7获取的补充材料。
