基于CART决策树模型构建胸腔镜下食管癌根治术后发生谵妄的风险预测模型  

Construction of a Risk Prediction Model for the Occurrence of Delirium After Thoracoscopic Radical Esophageal Cancer Surgery Based on the CART Decision Tree Model

作　　者：沈冠红 徐汉丽 SHEN Guanhong;XU Hanli(Department of Cardiothoracic and Vascular Surgery,No.904 Hospital of People's Liberation Army Logistics Support Unit,Wuxi,Jiangsu 213000,China)

机构地区：[1]中国人民解放军联勤保障部队第九〇四医院心胸血管外科,江苏无锡213000

出　　处：《转化医学杂志》2024年第9期1374-1380,共7页Translational Medicine Journal

基　　金：无锡市科学技术协会项目(KX-23-CO95)。

摘　　要：目的探讨胸腔镜下食管癌(EC)根治术后患者发生谵妄(POD)的影响因素,采用分类和回归树(CART)算法,构建胸腔镜下EC根治术后发生POD的决策树预测模型。方法选取2023年1月至2024年1月中国人民解放军联勤保障部队第九〇四医院心胸外科行胸腔镜下EC根治术的患者为研究对象,依据患者术后是否发生POD将其分为POD组与非POD组。收集所有受试者的临床资料、实验室指标以及手术相关指标。通过单因素分析和多因素分析筛选出与术后POD发生存在显著相关的风险因素,应用SPSS Modeler软件构建胸腔镜下EC根治术后发生POD的CART决策树模型,并分析决策树模型的预测效能。结果本研究共纳入194例行胸腔镜下EC根治术的患者,其中未发生POD患者104例,发生POD患者为90例,POD发生率为46.39%;POD组与非POD组患者在年龄、过度饮酒史、深度镇静等方面差异有统计学意义(P均<0.05);POD组血红蛋白水平低于非POD组(P<0.05);POD组术后疼痛评分、术中低血压患者比例高于非POD组(P均<0.05);多因素Logistic回归分析结果显示,高龄、过度饮酒史、深度镇静、低血红蛋白水平、术后疼痛评分、术中低血压为胸腔镜下EC根治术后发生POD的危险因素(P均<0.05);将单因素分析中差异具有统计学意义的6个项目纳入决策树模型,共筛选出解释变量5个,分别为血红蛋白、术中低血压、年龄、术后疼痛评分、过度饮酒史。决策树模型共5层,7个终末节点,其中血红蛋白水平是最重要的预测因子;胸腔镜下EC根治术后发生POD决策树模型的曲线下面积(AUC)为0.837(95%CI:0.777~0.886),Logistic回归模型的AUC为0.777(95%CI:0.712~0.834),决策树模型在预测胸腔镜下EC根治术后POD的发生方面具有更高的准确性(P=0.009)。结论高龄、过度饮酒史、术后深度镇静、低血红蛋白水平、术后疼痛评分、术中低血压为胸腔镜下EC根治术后发生POD的危险因素。基于Objective To investigate the influencing factors of postoperative delirium(POD)after thoracoscopic radical surgery for esophageal cancer(EC),and to construct a decision tree prediction model for POD occurrence using the classification and regression tree(CART)algorithm.Methods A total of 194 patients who underwent thoracoscopic radical EC surgery at the Department of Cardiothoracic Surgery,No.904 Hospital of the PLA Logistics Support Unit from January 2023 to January 2024 were enrolled and divided into the POD and non-POD groups based on postoperative outcomes.Clinical data,laboratory indices,and surgery-related parameters were collected.Risk factors significantly associated with POD were identified through univariate and multivariate analyses,and the CART decision tree model for POD prediction was constructed using SPSS Modeler software.The predictive efficacy of the model was evaluated.Results Among the 194 patients,90(46.39%)developed POD,while 104 did not.Significant differences were observed between the POD and non-POD groups in terms of age,excessive alcohol consumption history,and deep sedation(P<0.05).Hemoglobin levels were lower in the POD group compared to the non-POD group(P<0.05),and postoperative pain scores and the proportion of patients with intraoperative hypotension were higher in the POD group(P<0.05).Multivariate logistic regression identified advanced age,excessive alcohol consumption history,deep sedation,low hemoglobin levels,postoperative pain scores,and intraoperative hypotension as risk factors for POD(P<0.05).Six variables showing significant differences in univariate analysis were included in the CART model,which screened five explanatory variables:hemoglobin,intraoperative hypotension,age,postoperative pain score,and excessive alcohol consumption history.The decision tree model comprised 5 levels and 7 terminal nodes,with hemoglobin being the most critical predictor.The AUC of the CART decision tree model was 0.837(95%CI:0.777–0.886),higher than the AUC of the logistic regression mod

关 键 词：CART决策树模型 胸腔镜手术 食管癌根治术 术后谵妄 风险预测模型 

分 类 号：R735.1[医药卫生—肿瘤]