Chen Ming, Zhang Dingyu
Department of Anesthesiology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, 430022, China.
Department of Anesthesiology, Institute of Anesthesia and Critical Care Medicine, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, No. 1277, Jiefang Avenue, Wuhan, 430022, China.
BMC Med Inform Decis Mak. 2025 Feb 22;25(1):96. doi: 10.1186/s12911-025-02930-y.
Post-induction hypotension (PIH) increases surgical complications including myocardial injury, acute kidney injury, delirium, stroke, prolonged hospitalization, and endangerment of the patient's life. Machine learning is an effective tool to analyze large amounts of data and identify perioperative complication factors. This study aims to identify risk factors for PIH and develop predictive models to support anesthesia management.
A dataset of 5406 patients was analyzed using machine learning methods. Logistic regression, random forest, XGBoost, and neural network models were compared. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC), calibration curves, and decision curve analysis (DCA).
The logistic regression model achieved an AUROC of 0.74 (95% CI: 0.71-0.77), outperforming the random forest (AUROC: 0.71), XGBoost (AUROC: 0.72), and neural network (AUROC: 0.72) models. In terms of calibration, logistic regression demonstrated superior performance, as reflected by Brier Scores and calibration curves, followed by XGBoost, random forest, and neural network. Decision curve analysis indicated that the logistic regression model provided the greatest clinical utility among all models. Baseline blood pressure, age, sex, type of surgery, platelet count, and certain anesthesia-inducing drugs were identified as important features.
This study provides a valuable tool for personalized preoperative risk assessment and customized anesthesia management, allowing for early intervention and improved patient outcomes. Integration of machine learning models into electronic medical record systems can facilitate real-time risk assessment and prediction.
诱导后低血压（PIH）会增加手术并发症的发生，包括心肌损伤、急性肾损伤、谵妄、中风、住院时间延长以及危及患者生命。机器学习是分析大量数据并识别围手术期并发症因素的有效工具。本研究旨在识别PIH的危险因素并开发预测模型以支持麻醉管理。
使用机器学习方法分析了5406例患者的数据集。比较了逻辑回归、随机森林、XGBoost和神经网络模型。使用受试者操作特征曲线下面积（AUROC）、校准曲线和决策曲线分析（DCA）评估模型性能。
逻辑回归模型的AUROC为0.74（95%CI：0.71 - 0.77），优于随机森林（AUROC：0.71）、XGBoost（AUROC：0.72）和神经网络（AUROC：0.72）模型。在校准方面，逻辑回归表现出卓越性能，这在Brier评分和校准曲线中得到体现，其次是XGBoost、随机森林和神经网络。决策曲线分析表明，逻辑回归模型在所有模型中提供了最大的临床实用性。基线血压、年龄、性别、手术类型、血小板计数和某些诱导麻醉药物被确定为重要特征。
本研究为个性化术前风险评估和定制麻醉管理提供了有价值的工具，允许进行早期干预并改善患者预后。将机器学习模型集成到电子病历系统中可以促进实时风险评估和预测。
