基于机器学习的老年患者腹腔镜胆囊切除术后感染预测模型的构建  

Construction of a machine learning-based predictive model for postoperative infections in elderly patients undergoing laparoscopic cholecystectomy

作　　者：胡晔 黄磊 吴慧 张琳琳[1,2] 王红霞 尹存林 HU Ye;HUANG Lei;WU Hui;ZHANG Linlin;WANG Hongxia;YIN Cunlin(Department of Pharmacy,the First People's Hospital of YanchengJiangsu Yancheng 224000,China;Deparment of PharmacyYancheng First Hospital,Afiliated Hospital of Nanjing University Medical School,Jiangsu Yancheng 224000,China)

机构地区：[1]盐城市第一人民医院药学部,江苏盐城224000 [2]南京大学医学院附属盐城第一医院药学部,江苏盐城224000

出　　处：《临床药物治疗杂志》2024年第6期66-70,共5页Clinical Medication Journal

基　　金：江苏省研究型医院学会精益化用药科研基金(JY202211)。

摘　　要：目的构建基于机器学习的老年患者腹腔镜胆囊切除术后感染预测模型,筛选影响术后感染的因素,为药学服务提供参考。方法回顾性纳入2021年12月至2023年1月于盐城市第一人民医院普外科行腹腔镜胆囊切除术的老年患者,根据术后是否感染分为感染组和未感染组。采用SPSS Modeler 18.0软件,将患者按照7∶3比例随机分为训练集和验证集。基于训练集,构建随机森林、贝叶斯网络、支持向量机和分类回归树共4种机器学习模型;使用验证集评估各模型性能。结果最终纳入385例患者,其中感染组62例,未感染组323例;感染组与未感染组的糖尿病史、围手术期有无胆囊急性炎症,术前白细胞计数、白蛋白水平、总蛋白水平,手术时长共6个方面比较,差异均有统计学意义(P<0.05);随机森林模型受试者工作特征曲线下面积为0.811,特异度为0.874,准确率为84.07%,为最终模型;变量重要性排序依次为术前白蛋白水平、白细胞计数,糖尿病史,围手术期有无胆囊急性炎症,手术时长,术前总蛋白水平。结论基于随机森林模型可以较好地预测老年患者腹腔镜胆囊切除术后感染情况,临床药师应结合重要变量特征变化开展药学服务。Objective To construct a machine learning-based predictive model for infections after laparoscopic cholecys-tectomy in elderly patients and identify the important variables affecting postoperative infection,providing insights for pharma-ceutical care.Methods This retrospective study included patients who underwent laparoscopic cholecystectomy in the Depart-ment of General Surgery of the First People's Hospital of Yancheng from December 2021 to January 2023.Patients were divid-ed into infected and non-infected groups according to their postoperative outcomes.SPss Modeler 18.O software was used to randomly divide the patients into a training set and a validation set in a 7:3 ratio.Four machine learning models-random for-est,bayesian network,support vector machine,and classification and regression treewere developed using the training set,and their performances were evaluated with the validation set.Results A total of 385 patients were finally included,including 62 patients in the infected group and 323 patients in the non-infected group.There were significant differences in the history of diabetes,acute gallbladder inflammation in perioperative period,preoperative white blood cell count,albumin level,total protein level,and operation duration between the infected group and the non-infected group(P<0.05).The random forest model,with an area under the receiver operating characteristic curve of 0.811,a specificity of 0.874,and an accuracy of 0.841,was selected as the final model.Variable importance was ranked as follows:preoperative albumin level,white blood cell count,history of diabetes,acute gallbladder inflammation in perioperative period,operation duration and preoperative total protein levels.Conclusion The random forest model effectively predicts postoperative infections in elderly undergoing laparoscopic cholecystectomy.Clinical pharmacists should tailor pharmaceutical services based on changes in important variable characteristics.

关 键 词：胆囊切除术 药学服务 机器学习 随机森林模型 合理用药 

分 类 号：R969.3[医药卫生—药理学]