Katsin Maksim, Glebov Maxim, Berkenstadt Haim, Orkin Dina, Portnoy Yotam, Shuchami Adi, Yaniv-Rosenfeld Amit, Lazebnik Teddy
Department of Anesthesiology, Sheba Medical Center, 21 Emek Dotan 11, Ramat Gan, Israel.
Faculty of Medicine, Tel-Aviv University, Tel-Aviv, Israel.
J Clin Monit Comput. 2025 May 5. doi: 10.1007/s10877-025-01295-x.
Arterial hypotension is a common and often unintended event during surgery under general anesthesia, associated with increased postoperative complications, such as kidney injury, myocardial injury, and stroke. Postinduction hypotension (PIH) is influenced by patient-specific factors, chronic medication use, and anesthetic induction regimens. Traditional predictive models struggle with this complexity, making machine learning (ML) a promising alternative due to its ability to handle complex datasets and identify hidden patterns. This study aimed to develop and validate an ML-based model for predicting PIH and identifying key clinical predictors. A retrospective cohort study of 20,309 adult patients undergoing non-obstetric surgery under general anesthesia with intravenous induction was conducted. The primary outcome was the occurrence of PIH, defined as mean arterial pressure (MAP) < 55 mmHg within 10 min post-induction. Data were split into training and validation sets using k-fold cross-validation. The model's predictive performance was evaluated using the area under the receiver operating characteristic curve (AUC), and feature importance was assessed using SHapley Additive exPlanations (SHAP) values. PIH occurred in 4,948 patients (24.4%). Key predictors included preinduction systolic and mean arterial pressures, propofol dose, and beta-blocker use. The ML model achieved an AUC of 0.732 in predicting PIH. The ML-based model demonstrated significant predictive capability for PIH, identifying key clinical predictors. This model holds the potential for improving preoperative planning and patient risk stratification. However, further validation through prospective studies is necessary to confirm these findings.
动脉低血压是全身麻醉手术期间常见且往往意外发生的事件，与术后并发症增加有关，如肾损伤、心肌损伤和中风。诱导后低血压（PIH）受患者特定因素、慢性药物使用和麻醉诱导方案的影响。传统的预测模型难以应对这种复杂性，由于机器学习（ML）有能力处理复杂数据集并识别隐藏模式，因此成为一种有前景的替代方法。本研究旨在开发并验证一种基于ML的模型，用于预测PIH并识别关键临床预测因素。对20309例接受全身麻醉静脉诱导下非产科手术的成年患者进行了一项回顾性队列研究。主要结局是PIH的发生，定义为诱导后10分钟内平均动脉压（MAP）<55 mmHg。使用k折交叉验证将数据分为训练集和验证集。使用受试者工作特征曲线下面积（AUC）评估模型的预测性能，并使用SHapley加性解释（SHAP）值评估特征重要性。4948例患者（24.4%）发生了PIH。关键预测因素包括诱导前收缩压和平均动脉压、丙泊酚剂量和β受体阻滞剂的使用。该ML模型在预测PIH方面的AUC为0.732。基于ML的模型对PIH具有显著的预测能力，识别出了关键临床预测因素。该模型具有改善术前规划和患者风险分层的潜力。然而，需要通过前瞻性研究进行进一步验证以证实这些发现。
