基于决策树法与 Logistic回归的肩关节镜手术患者术中低体温风险预测模型的构建 [中文引用][英文引用]

目的 应用决策树法与 Logistic 回归构建肩关节镜手术患者术中低体温风险预测模型，并比较两种模型的预测效能，为准确识别高风险患者提供参考。方法 采取回顾性研究，选取 2023 年 4 月—2025 年 4 月在本院接受肩关节镜手术的 340 例患者的临床资料，根据术中是否发生低体温分为低温组（n=146）和对照组（n=194）。采用单因素分析和二元Logistics 回归分析肩关节镜术患者术中低体温的影响因素，并绘制列线图和决策树图，并采用受试者工作特征（receiver operatorcharacteristic，ROC）曲线评价模型的预测效能。结果 340 例患者完成研究。146 例肩关节镜术患者术中发生低体温，发生率为42.94%。Logistic回归分析结果显示，身体质量指数<18.5kg/m（2 OR=8.076，95%CI：4.113~15.858）、美国麻醉医师协会（AmericanSociety of Anesthesiologists，ASA）分 级 为 Ⅱ~Ⅲ 级（OR=5.071，95%CI：2.630~9.778）、入 室 体 温 <36.5 ℃（OR=3.282，95%CI：1.865~5.774）、灌洗量≥ 12L（OR=4.123，95%CI：2.333~7.287）、手术时间≥ 120min（OR=11.650，95%CI：6.073~22.348）是肩关节镜术患者发生术中低体温的危险因素（均 P<0.05）；根据危险因素构建的决策树模型深度为 5，共 25 个节点，13 个终端节点，模型以手术时间、身体质量指数、灌洗量、ASA 分级、入室体温为解释变量，其中手术时间为最重要的预测因子；Logistic 回归模型的ROC 曲线下面积（area under the curve，AUC）为 0.846（95%CI：0.803~0.889），决策树模型的 AUC 为 0.838（95%CI：0.794~0.883），两者比较，差异无统计学意义（Z=0.552，P=0.581）。校准曲线和决策曲线显示，Logistic 回归模型和决策树模型具有较高的校准度与临床实用性。结论 身体质量指数 <18.5kg/m2、ASA 分级为Ⅱ~Ⅲ级、入室体温 <36.5℃、灌洗量≥ 12L、手术时间≥ 120min是肩关节镜术患者术中低体温的危险因素，据此构建的 Logistic 回归模型和决策树模型均有较高的预测效能。

Objective To establish the risk prediction models based on decision tree and Logistic regression approaches for intraoperative hypothermia among the patients undergoing arthroscopic shoulder surgery，to identify high-risk patients and compare the models. Methods Clinical data of 340 patients who underwent arthroscopic shoulder surgery in our hospital from April 2023 to April 2025 were selected and assigned into a hypothermia group（n=146）and a control group（n=194）according to occurrence of intraoperative hypothermia. Univariate and binary Logistic regression analyses were conducted to identify risk factors，of which nomogram and decision tree diagrams were constructed. Model performance was evaluated by receiver operating characteristic（ROC） curves. Results All 340 patients completed the study，and 146 of them developed intraoperative hypothermia，accounting for 42.94%. Logistics regression showed that body mass index of 18.5kg/m2 （OR=8.076，95%CI：4.113~15.858），ASA grade II-III（OR=5.071，95%CI： 2.630~9.778），body temperature at admission<36.5℃（OR=3.282，95%CI：1.865~5.774），lavage volume ≥12L（OR=4.123，95%CI：2.333~ 7.287），operation time ≥120 minutes（OR=11.650，95%CI：6.073~22.348）were the independent risk factors（all P<0.05）. The decision tree model（depth=5，25 nodes，13 terminal nodes）gave operative time，body mass index，lavage volume，ASA classification and body temperature at admission as the explanatory variables，with operation time being the most important predictor. Areas under the ROC curve（AUC）was 0.846（95%CI：0.803~0.889）of the Logistic regression model and 0.838（95%CI 0.794-0.883）of decision-tree model，without a significant difference between the two models（Z=0.552，P=0.581）. Calibration curve and decision curve analyses showed that the two models both had high calibration and clinical practicability. Conclusion Body mass index <18.5kg/m2 ，ASA grade II-III，body temperature at admission <36.5℃，lavage volume ≥12L，and operative time ≥120 minutes are the independent factors of intraoperative hypothermia in patients undergoing arthroscopic shoulder surgery. Both Logistics regression and decision tree models constructed with these variables demonstrate high predictive performance.