肝胆手术手术部位感染风险预测模型的构建

发布时间：2018-03-26 02:41

  本文选题：肝胆手术　切入点：手术部位感染　出处：《石河子大学》2017年硕士论文

【摘要】：目的:通过构建肝胆术后手术部位感染风险预测模型,为筛选手术部位感染高危人群提供技术支持。方法:回顾性收集2013年1月—2015年12月国内6家医院1145例肝胆手术病例资料,通过逐份查阅电子病例收集包括患者人口学特征,合并基础疾病情况,手术相关信息,抗菌药物使用情况及术后感染情况五个方面信息。将所有病例按照7:3的比例随机分为建模组和验证组,建模组用来建立模型,验证组用来验证模型。在建模组,所有变量先进行单因素分析,单因素分析中有统计学意义的变量纳入多元Logistic回归分析,采用Back—LR法建立风险预测模型。利用验证组数据对模型进行验证,以Hosmer-Lemeshow反映模型的符合程度,以受试者工作特征曲线(ROC)及其曲线下的面积(AUC)反应模型的判别能力。ROC曲线中取Youden指数最大值为最佳阈值,对模型进行回代验证。根据Logistic模型中各风险因素的权重β系数建立评分表,按各评分感染率将患者进行风险分层,≥5分为高风险组,3-4分为中风险组,0-2分为低风险组。结果:本研究共纳入肝胆手术病例1145例,术后发生手术部位感染143例,总感染率为12.49%,其中表浅切口感染89例(7.77%)、深部切口感染17例(1.48%)、器官腔隙感染37例(3.23%)。通过单因素与多因素分析,最终纳入风险预测模型的变量包括低蛋白血症、高血压、术前合并胆道感染、术前腹部手术史、手术切口分类、手术持续时间、ASA分级和吻合口瘘8个指标。验证结果显示:所建立模型的ROC曲线下面积(AUC)为0.851,优于NNIS风险指数预测效力(AUC为0.731);ROC曲线中最佳阈值为0.139,验证组回代结果:模型灵敏度68.29%,特异度86.01%,阳性预测值41.18%,阴性预测值94.98%,总准确率为83.79%,说明模型的判别能力很好。将模型转化为风险评分,手术切口分类赋值2分,吻合口瘘赋值3分,其余六个变量赋值1分。将患者进行危险分层,高中低风险组手术患者所占的比例分别为3.42%,13.45%,83.13%,感染率分别为85.71%,30.91%,6.47%。结论:低蛋白血症、高血压、术前合并胆道感染、术前腹部手术史、手术切口分类、手术持续时间、ASA分级和吻合口瘘是肝胆手术手术部位感染的独立风险因素。利用肝胆手术患者病例资料建立的手术部位感染风险预测模型效力较高,对于肝胆手术发生感染的术前评估及群体监测均有一定的应用价值。
[Abstract]:Objective: to provide technical support for screening high risk population of postoperative hepatobiliary infection by constructing a prediction model of postoperative infection in hepatobiliary surgery. Methods: 1145 cases of hepatobiliary surgery were collected retrospectively from January 2013 to December 2015 in 6 hospitals in China. The electronic case collection includes demographic features of patients, combined with underlying diseases, and operation-related information. All cases were randomly divided into modeling group and verification group according to the proportion of 7:3. The modeling group was used to build the model, and the validation group was used to verify the model. All variables were analyzed by single factor analysis, the variables with statistical significance in single factor analysis were included in multivariate Logistic regression analysis, and the risk prediction model was established by Back-LR method. The model was validated by validation group data. The maximum value of Youden exponent was chosen as the best threshold in the discriminant ability of the response model of Hosmer-Lemeshow and its area under the curve, according to the degree of coincidence of the model reflected by Hosmer-Lemeshow, and the maximum value of Youden exponent was taken as the optimum threshold in the response model of the test subjects' operating characteristic curve and its area. According to the weight 尾 coefficient of each risk factor in Logistic model, the scoring table was established, and the risk was stratified according to the infection rate of each score. 鈮，

本文编号：1665971