列线图与CART决策树模型对膝关节置换术后急性疼痛风险预测中的效能比较  

Comparison of efficacy of nomogram and classification and regression trees(CART)decision tree model in predicting risk of acute postoperative pain(APP)after total knee arthroplasty(TKA)

作　　者：马超 韩影 程旻桦 MA Chao;HAN Ying;CHENG Minhua(Department of Pain Medicine,Nanjing Gulou Hospital,Nanjing 210008,China;Department of Critical Care Medicine,Nanjing Gulou Hospital,Nanjing 210008,China)

机构地区：[1]南京鼓楼医院疼痛医学科,南京210008 [2]南京鼓楼医院重症医学科,南京210008

出　　处：《新疆医科大学学报》2025年第2期195-202,共8页Journal of Xinjiang Medical University

基　　金：江苏省科技基础研究青年基金项目(BK20210026)。

摘　　要：目的分别构建预测膝关节置换术(TKA)后急性疼痛(APP)风险的列线图与分类与回归树(CART)决策树模型,并比较两种模型在对TKA后APP风险预测中的预测效能。方法以274例膝关节骨性关节炎(KOA)患者为研究对象,均于2018年3月至2024年4月在本院进行TKA治疗,根据术后是否发生APP将患者分为APP组(n=98)和非APP组(n=176),对两组患者进行单因素分析。根据单因素分析结果进行Logistic回归分析TKA后APP的危险因素,根据危险因素绘制列线图模型;根据单因素分析结果进行CART决策树模型建立。绘制两种模型的受试者工作特征(ROC)曲线并对两种模型的预测效能进行DeLong检验。结果单因素分析结果显示,两组患者在年龄、体质指数(BMI)、糖尿病、西安大略和麦克马斯特大学骨关节炎指数(WOMAC)、术前疼痛灾难化量表(PCS)评分、术前视觉模拟评分(VAS)、止血带使用时间、神经阻滞、术后使用镇痛泵方面比较差异具有统计学意义(P<0.05)。多因素Logistic回归分析结果显示,BMI≥25 kg/m^(2)、糖尿病、PCS评分≥27分、VAS评分≥5分、术后未使用镇痛泵为TKA后APP的独立危险因素(P<0.05)。基于多因素Logistic回归结果采用R软件绘制列线图模型。将单因素分析中差异具有统计学意义的相关因素纳入CART决策树模型,最终模型筛选出5个特征,包括BMI≥25 kg/m^(2)、糖尿病、WOMAC≥48分、术前使用神经阻滞、未使用术后镇痛泵。绘制两种模型的ROC曲线,结果显示列线图模型和CART决策树模型的AUC分别为0.858和0.911,灵敏度分别为81.88%和86.34%,特异度分别为82.91%和87.62%,阳性预测值分别为75.43%和80.69%,阴性预测值分别为82.94%和89.27%,预测准确率分别为83.31%和89.75%。两种模型AUC值相比差异具有统计学意义(Z=9.864,P<0.001)。结论两种模型均对TKA后APP风险具有较好的预测效能,CART决策树预测效能优于列线图模型。Objective To construct nomogram and classification and regression trees(CART)decision tree models for predicting acute postoperative pain(APP)risk after total knee arthroplasty(TKA),and to compare their predictive abilities.Methods A retrospective analysis was conducted on 274 patients with knee osteoarthritis who underwent TKA in the hospital from March 2018 to April 2024.The patients were divided into an APP group(n=98)and a non-APP group(n=176)based on whether they experienced APP postoperatively.Univariate analysis was performed on both groups.Logistic regression analysis identified risk factors for APP following TKA based on univariate analysis results;these factors were used to create a nomogram model.Subsequently,a CART decision tree model was established using significant variables from the univariate analysis.Receiver operating characteristic(ROC)curves were plotted for both models,and DeLong's test compared their predictive performances.Results Univariate analysis revealed statistically significant differences between the 2 groups regarding age,body mass index(BMI),diabetes,Western Ontario and McMaster Universities Osteoarthritis Index(WOMAC),preoperative Pain Catastrophizing Scale(PCS)score,preoperative Visual Analog Scale(VAS)score,tourniquet usage time,nerve block utilization and postoperative patient-controlled analgesia(PCA)(P<0.05).Multivariate Logistic regression analysis indicated that BMI≥25 kg/m^(2),diabetes,PCS score≥27,VAS score≥5,and lack of postoperative PCA use were independent risk factors for APP after TKA(P<0.05).A nomogram model was developed using R software based on multivariate Logistic regression outcomes.The final CART decision tree model included 5 features selected through univariate analysis:BMI≥25 kg/m^(2),diabetes,WOMAC≥48 points,preoperative nerve block use and absence of postoperative PCA.ROC curve analysis showed that the area under the curve(AUC)values for the nomogram model and CART decision tree were 0.858 and 0.911 respectively,with sensitivity of 81.88%and

关 键 词：膝关节置换术 术后急性疼痛 预测效能 列线图模型 CART决策树模型 

分 类 号：R684[医药卫生—骨科学]