基于生物电阻抗分析的膝关节退变程度预测模型构建及评价  

Construction and evaluation of the prediction model of knee degeneration based on bioelectrial impedance analysis

作　　者：王梦琪 陈泓伯 鲁寒 王翠 邹子秋 梁叶田 陈可欣 金仕达 刘沛源 王宇光[3] 尚少梅[1] Wang Mengqi;Chen Hongbo;Lu Han;Wang Cui;Zou Ziqiu;Liang Yetian;Chen Kexin;Jin Shida;Liu Peiyuan;Wang Yuguang;Shang Shaomei(School of Nursing,Peking University,Beijing 100191,China;Office of Nursing Education,the Fourth Hospital of Hebei Medical University,Shijiazhuang 050011,China;Peking University Hospital of Stomatology&National Engineering Laboratory for Digital and Material Technology of Stomatology,Beijing 100081,China)

机构地区：[1]北京大学护理学院,北京100191 [2]河北医科大学第四医院护理教学办公室,石家庄050011 [3]北京大学口腔医院口腔数字化医疗技术和材料国家工程实验室,北京100081

出　　处：《中华现代护理杂志》2023年第1期7-13,共7页Chinese Journal of Modern Nursing

基　　金：国家自然科学基金(81972158,82102661);国家重点研发计划(2020YFC2008801)。

摘　　要：目的基于生物电阻抗指标构建膝关节骨关节炎患者的膝关节退变程度预测模型,并对模型的预测能力和应用效能进行评价。方法本研究为横断面调查研究。采用便利抽样法选取2021年5—7月石家庄市裕强社区卫生服务中心参与体检的124例社区居家膝关节骨关节炎患者的248个膝关节为对象建立模型,根据Kellgren-Lawrence(K-L)分级系统将其分为K-L1级(19个)、K-L2级(103个)、K-L3级(96个)、K-L4级(30个)4组。通过方差分析或Kruskal-Wallis检验筛选纳入模型的指标,利用支持向量机建立膝关节退变程度预测模型,采用网格参数寻优法对模型进行优化。采用受试者工作特征(ROC)曲线下面积及敏感度、特异度、准确率、阳性预测值、阴性预测值评价模型的预测效能。结果纳入模型的指标包括年龄、合并症、合并腰/背/髋部疼痛、从事高危职业、西安大略和麦克马斯特大学骨关节炎指数(WOMAC)-疼痛、WOMAC-功能、电抗值和相位角。训练集中模型的ROC曲线下面积为0.999,预测准确率为0.920,95%置信区间为0.868~0.957;测试集中模型的ROC曲线下面积为0.833,预测准确率为0.682,95%置信区间为0.572~0.780。结论本研究构建的膝关节退变程度预测模型具有较好的预测能力且易于使用,可作为膝关节骨关节炎患者的膝关节退变程度的筛查工具。Objective To construct the prediction model of knee degeneration in patients with knee osteoarthritis based on bioelectrical impedance index,and evaluate the prediction performance and application efficiency of the model.Methods This was a cross-sectional study.From May to July 2021,248 knee joints of 124 patients with knee osteoarthritis at home from Shijiazhuang Yuqiang Community Health Service Center who participated in physical examination were selected by convenience sampling to establish the model.According to Kellgren-Lawrence(K-L)grading system,the knee joints were divided into four groups,namely K-L1(n=19),K-L2(n=103),K-L3(n=96),and K-L4(n=30).The indicators included in the model were selected through analysis of variance or Kruskal-Wallis test,and a prediction model of knee degeneration was established using support vector machine,and the model was optimized using grid parameter optimization method.The prediction performance of the model was evaluated by the area under the receiver operating characteristic(ROC)curve,sensitivity,specificity,accuracy,positive predictive value and negative predictive value.Results The indicators in the model included age,complications,lumbar/back/hip pain,high-risk occupation,Western Ontario and McMaster Universities Osteoarthritis Index(WOMAC)-pain,WOMAC-function,capacitive reactance and phase angle.The area under the ROC curve of the training set model was 0.999,the prediction accuracy was 0.920,and the 95%confidence interval was 0.868 to 0.957.The area under the ROC curve of the test set model was 0.833,the prediction accuracy was 0.682,and the 95%confidence interval was 0.572 to 0.780.Conclusions The prediction model of knee degeneration has good prediction performance and is easy to use,which can be used as a screening tool for knee degeneration in patients with knee osteoarthritis.

关 键 词：骨关节炎 膝 膝关节退变程度 生物电阻抗 支持向量机 预测模型 

分 类 号：R684.3[医药卫生—骨科学]