Zhang Kaiwen, Jiao Bo, Sun Jiaoli, Zhang Xianwei, Zhang Guanglei, Li Ningbo, Liu Baowen, Zhou Zhiqiang
Department of Anesthesiology and Pain Medicine, Hubei Key Laboratory of Geriatric Anesthesia and Perioperative Brain Health, and Wuhan Clinical Research Center for Geriatric Anesthesia, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, People's Republic of China.
Int J Surg. 2025 Jun 1;111(6):3859-3875. doi: 10.1097/JS9.0000000000002354. Epub 2025 Apr 3.
Early identification of high-risk factors for inadequate analgesia and adverse reactions in obstetric patients is critical for improving outcomes. This study developed a machine learning model to predict these factors and optimize anesthesia management in obstetric surgery.
This prospective study included 763 obstetric patients who underwent elective cesarean delivery between December 2023 and May 2024. A machine-learning model for postoperative analgesia and adverse reactions (MLPIAAR) was constructed using 42 variables categorized into preoperative (18), intraoperative (12), and postoperative (12) factors. Ten algorithms were applied for model development. Model performance was optimized through 10-fold cross-validation and GridSearchCV. Evaluation metrics included receiver operating characteristic, area under the curve, accuracy, and precision. Model interpretability was enhanced using SHapley Additive exPlanations (SHAP). Regression analysis explored PCA compression frequency, consumption, and satisfaction of PCA factors. Validation was performed using predicted values and scatter plots.
Our study found 24.25% of patients experienced inadequate postoperative analgesia within 24 h. Postoperative nausea also occurred in 23.20% of patients. Simultaneously using dexamethasone and flurbiprofen axetil reduced both risks. However, pregnancy-induced hypertension and intraoperative shivering increased nausea risk. Postoperative vomiting occurred in 11.80% of patients, primarily associated with intravenous PCA type. Hydromorphone PCA showed a higher vomiting incidence than nalbuphine PCA. Intraoperative nausea and vomiting increased postoperative vomiting risk. Patients receiving diclofenac sodium and those with longer surgeries exhibited higher PCA compression frequencies and consumption. Higher preoperative Edinburgh Postnatal Depression Scale scores correlated with increased PCA consumption.
The MLPIAAR model predicted high-risk factors for inadequate analgesia, nausea, and vomiting in obstetric patients. Dexamethasone and flurbiprofen axetil reduced these risks, while hydromorphone PCA increased vomiting risk compared to nalbuphine PCA. Machine learning and SHAP are valuable for optimizing anesthesia and analgesia management in obstetrics.
早期识别产科患者镇痛不足和不良反应的高危因素对于改善预后至关重要。本研究开发了一种机器学习模型来预测这些因素并优化产科手术中的麻醉管理。
这项前瞻性研究纳入了2023年12月至2024年5月期间接受择期剖宫产的763例产科患者。使用42个变量构建了术后镇痛和不良反应的机器学习模型（MLPIAAR），这些变量分为术前（18个）、术中（12个）和术后（12个）因素。应用十种算法进行模型开发。通过十折交叉验证和GridSearchCV优化模型性能。评估指标包括受试者工作特征曲线、曲线下面积、准确性和精确性。使用SHapley加法解释（SHAP）增强模型可解释性。回归分析探讨了PCA因素的压缩频率、消耗量和满意度。使用预测值和散点图进行验证。
我们的研究发现24.25%的患者在24小时内术后镇痛不足。23.20%的患者还出现了术后恶心。同时使用地塞米松和氟比洛芬酯可降低这两种风险。然而，妊娠期高血压和术中寒战会增加恶心风险。11.80%的患者出现术后呕吐，主要与静脉PCA类型有关。氢吗啡酮PCA的呕吐发生率高于纳布啡PCA。术中恶心和呕吐会增加术后呕吐风险。接受双氯芬酸钠的患者和手术时间较长的患者PCA压缩频率和消耗量较高。术前爱丁堡产后抑郁量表得分较高与PCA消耗量增加相关。
MLPIAAR模型预测了产科患者镇痛不足、恶心和呕吐的高危因素。地塞米松和氟比洛芬酯降低了这些风险，而与纳布啡PCA相比，氢吗啡酮PCA增加了呕吐风险。机器学习和SHAP对于优化产科麻醉和镇痛管理具有重要价值。
