Department of Breast Surgery, Hubei Cancer Hospital, Tongji Medical College, Huazhong University of Science and Technology, Hubei Provincial Clinical Research Center for Breast Cancer, Wuhan Clinical Research Center for Breast Cancer, Wuhan, Hubei, China.
Department Gynecology, Cancer Biology Research Center, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei, PR China.
Medicine (Baltimore). 2024 Oct 4;103(40):e39038. doi: 10.1097/MD.0000000000039038.
Intraoperative hypothermia is one of the most common adverse events related to surgery, and clinical practice has been severely underestimated. In view of this, this study aims to build a practical intraoperative hypothermia prediction model for clinical decision-making assistance. We retrospectively collected clinical data of patients who underwent gynecological laparoscopic surgery from June 2018 to May 2023, and constructed a multimodal algorithm prediction model based on this data. For the construction of the prediction model, all data are randomly divided into a training queue (70%) and a testing queue (30%), and then 3 types of machine learning algorithms are used, namely: random forest, artificial neural network, and generalized linear regression. The effectiveness evaluation of all predictive models relies on the comprehensive evaluation of the net benefit method using the area under the receiver operating characteristic curve, calibration curve, and decision curve analysis. Finally, 1517 screened patients were filtered and 1429 participants were included for the construction of the predictive model. Among these, anesthesia time, pneumoperitoneum time, pneumoperitoneum flow rate, surgical time, intraoperative infusion, and room temperature were independent risk factors for intraoperative hypothermia and were listed as predictive variables. The random forest model algorithm combines 7 candidate variables to achieve optimal predictive performance in 2 queues, with an area under the curve of 0.893 and 0.887 and a 95% confidence interval of 0.835 to 0.951 and 0.829 to 0.945, respectively. The prediction efficiency of other prediction models is 0.783 and 0.821, with a 95% confidence interval of 0.725 to 0.841 and 0.763 to 0.879, respectively. The intraoperative hypothermia prediction model based on machine learning has satisfactory predictive performance, especially in random forests. This interpretable prediction model helps doctors evaluate the risk of intraoperative hypothermia, optimize clinical decision-making, and improve patient prognosis.
术中低体温是与手术相关的最常见不良事件之一，但临床实践对此严重低估。有鉴于此，本研究旨在构建一种实用的术中低体温预测模型，以辅助临床决策。我们回顾性收集了 2018 年 6 月至 2023 年 5 月期间接受妇科腹腔镜手术的患者的临床数据，并基于这些数据构建了一个多模态算法预测模型。为了构建预测模型，所有数据随机分为训练队列（70%）和测试队列（30%），然后使用 3 种机器学习算法，即随机森林、人工神经网络和广义线性回归。所有预测模型的有效性评估都依赖于使用接受者操作特征曲线、校准曲线和决策曲线分析的净收益方法进行综合评估。最后，筛选出 1517 名患者，纳入 1429 名参与者构建预测模型。其中，麻醉时间、气腹时间、气腹流量、手术时间、术中输液和室温是术中低体温的独立危险因素，被列为预测变量。随机森林模型算法结合 7 个候选变量，在两个队列中实现了最佳预测性能，曲线下面积分别为 0.893 和 0.887，95%置信区间分别为 0.835 至 0.951 和 0.829 至 0.945。其他预测模型的预测效率分别为 0.783 和 0.821，95%置信区间分别为 0.725 至 0.841 和 0.763 至 0.879。基于机器学习的术中低体温预测模型具有令人满意的预测性能，尤其是在随机森林中。这种可解释的预测模型有助于医生评估术中低体温的风险，优化临床决策，并改善患者预后。
