采用机器学习方法建立Ⅰ类切口手术患者使用抗菌药物合理性的评价模型
|
摘要
目的:建立Ⅰ类切口手术患者使用抗菌药物合理性的评价模型,为临床药师进行处方点评提供依据。方法:以某院2017年1月1日-2017年12月31日住院的432例Ⅰ类切口手术患者作为研究对象,提取患者年龄、院内感染、使用抗菌药物的种类数等多项诊疗信息;结合临床药师对患者预防及治疗使用抗菌药物合理性的点评结果,以抗菌药物类别(预防或治疗使用)为因变量、患者诊疗信息为自变量,分别利用机器学习方法中的非条件Logistic回归法和支持向量机法,将药师点评结果转化为机器学习模型可识别的客观指标,建立Ⅰ类切口手术患者使用抗菌药物合理性的分类判别模型,并以灵敏度、特异度、约登指数等为指标,另采用61例Ⅰ类切口手术患者样本对建立的模型进行验证;收集使用模型前(人工点评,2017年1-12月)、后(2018年1-10月)Ⅰ类切口手术患者使用抗菌药物处方合理性点评情况,对模型效果进行应用评价。结果:以非条件Logistic回归法建立的模型的灵敏度为65.63%、特异度为75.00%、约登指数为40.63%;以支持向量机法建立的模型主要参数gamma为0.01、cost为10,其灵敏度为92.19%、特异度为87.50%、约登指数为79.69%,支持向量机法建立的模型优于非条件Logistic回归法;对采用支持向量机法建立的模型进行验证,结果灵敏度为100%、特异度为88.57%、约登指数为88.57%;与使用模型前比较,使用模型后处方点评比例由69.44%升高到100%,抗菌药物预防使用率由23.84%下降到16.43%,品种选用合理率由37.86%升高到54.39%,使用疗程由5.01 d缩短到3.26 d。结论:应用机器学习方法中的支持向量机法建立Ⅰ类切口手术患者使用抗菌药物的评价模型,可实现处方点评全覆盖,提高Ⅰ类切口手术患者使用抗菌药物的合理水平,同时为药师的处方点评工作提供了新思路。
OBJECTIVE:To establish antibiotics use rationality evaluation model in type Ⅰ incision surgery patients,and to provide reference for prescription review of clinical pharmacists. METHODS:Totally 432 inpatients underwent type Ⅰ surgical incision in a hospital from Jan. 1 st-Dec. 31 st,2017 were selected as the research objects. The information of diagnosis and treatment including age,nosocomial infection,the number of kinds of antibiotics used were extracted. Based on the results of clinical pharmacists' comments on the antibiotics use rationality in patients' prevention and treatment,non-conditional Logistic regression and support vector machine(SVM)in machine learning method were used to convert clinical pharmacists' comments into objective index that can be recognized by the machine learning model,using categories of antibiotics(preventive or therapeutic use) as dependent variables and the patient's diagnosis and treatment information as independent variables. Classification and identification model was established for antibiotics use rationality in type Ⅰ incision surgery patients. Using sensitivity,specificity and Youden index as indexes,established mode was validated on the other 61 samples of type Ⅰ incision surgery patients. The rationality of antibiotics prescriptions in type Ⅰ incision surgery patients before(by manual review,Jan.-Dec. 2017)and after(Jan.-Oct. 2018) using the model were collected, and the effects of the model were evaluated. RESULTS:The sensitivity,specificity and Youden index of non-conditional Logistic regression model were 65.63%,75.00% and 40.63%,respectively. Main parameters of the model established by SVM included gamma 0.01,cost 10,sensitivity 92.19%,specificity87.50%,Youden index 79.69%. The model established by SVM was better than non-conditional Logistic regression. SVM was used to validate established mode,and sensitivity,specificity and Youden index were 100%,88.57% and 88.57%,respectively.Compared with before using the model,the evaluation ratio increased from 69.44% to 100%,the rate of prophylactic use of antibiotics decreased from 23.84% to 16.43%,the rate of rational drug type selection increased from 37.86% to 54.39%,and treatment course shortened from 5.01 days to 3.26 days after using the model. CONCLUSIONS:Established antibiotics use rationality evaluation model in type Ⅰ incision surgery patients by SVM in machine learning method fully covers all the patients,promotes rational use of antibiotics in typeⅠincision surgery patients,and provides a new idea for pharmacist prescription comment.
OBJECTIVE:To establish antibiotics use rationality evaluation model in type Ⅰ incision surgery patients,and to provide reference for prescription review of clinical pharmacists. METHODS:Totally 432 inpatients underwent type Ⅰ surgical incision in a hospital from Jan. 1 st-Dec. 31 st,2017 were selected as the research objects. The information of diagnosis and treatment including age,nosocomial infection,the number of kinds of antibiotics used were extracted. Based on the results of clinical pharmacists' comments on the antibiotics use rationality in patients' prevention and treatment,non-conditional Logistic regression and support vector machine(SVM)in machine learning method were used to convert clinical pharmacists' comments into objective index that can be recognized by the machine learning model,using categories of antibiotics(preventive or therapeutic use) as dependent variables and the patient's diagnosis and treatment information as independent variables. Classification and identification model was established for antibiotics use rationality in type Ⅰ incision surgery patients. Using sensitivity,specificity and Youden index as indexes,established mode was validated on the other 61 samples of type Ⅰ incision surgery patients. The rationality of antibiotics prescriptions in type Ⅰ incision surgery patients before(by manual review,Jan.-Dec. 2017)and after(Jan.-Oct. 2018) using the model were collected, and the effects of the model were evaluated. RESULTS:The sensitivity,specificity and Youden index of non-conditional Logistic regression model were 65.63%,75.00% and 40.63%,respectively. Main parameters of the model established by SVM included gamma 0.01,cost 10,sensitivity 92.19%,specificity87.50%,Youden index 79.69%. The model established by SVM was better than non-conditional Logistic regression. SVM was used to validate established mode,and sensitivity,specificity and Youden index were 100%,88.57% and 88.57%,respectively.Compared with before using the model,the evaluation ratio increased from 69.44% to 100%,the rate of prophylactic use of antibiotics decreased from 23.84% to 16.43%,the rate of rational drug type selection increased from 37.86% to 54.39%,and treatment course shortened from 5.01 days to 3.26 days after using the model. CONCLUSIONS:Established antibiotics use rationality evaluation model in type Ⅰ incision surgery patients by SVM in machine learning method fully covers all the patients,promotes rational use of antibiotics in typeⅠincision surgery patients,and provides a new idea for pharmacist prescription comment.
引文
[1]宋攀,黄垂国,李云龙,等.抗菌药物在外科的不合理应用及对策[J].中华医院感染学杂志,2017,27(3):717-720.
[2]程模,杨培洪,何霖.专项处方点评模式对我院清洁手术围术期抗菌药物预防使用的干预效果评价[J].中国药房,2018,29(3):293-297.
[3]陈猛,舒志兵,周月红.某“三甲”综合性医院腹腔镜手术围手术期抗菌药物预防使用情况调查及合理性评价[J]中国药房,2018,29(8):1065-1068.
[4]廖丽娜,李鑫,黄菲,等.处方前置审核系统在保障患者用药安全中的作用[J].中国医院管理,2018,38(10):23-25.
[5]高燕灵,扶玲,钟秀娟,等.合理用药监测系统存在的问题及对策探讨[J].中国药房,2015,26(22):3159-3161.
[6]徐芳琴,郭超,陆杨,等.合理用药决策系统对我院妇产科抗菌药物使用的干预效果分析[J].中国药房,2016,27(8):1133-1135.
[7]朱芳芳,沈怡,廖丽娜,等.我院《临床合理用药智能化管理解决方案》的应用与实践[J].中国药房,2016,27(25):3528-3531.
[8]曾小辉,彭涛,高月琴.基于前列腺影响报告和数据系统第2版的机器学习模型诊断高级别前列腺癌[J].中国医学影像技术,2018,34(12):1852-1856.
[9]傅裕,鲍迎秋,谢沂伯,等.人工智能技术在皮肤病辅助诊断的应用研究[J].中国数字医学,2018,13(10):29-31,38.
[10]杜明伦,黄君君,马香,等.大肠杆菌基因组中重叠基因注释的机器学习优化方法[J].中国生物化学与分子生物学报,2018,34(8):861-867.
[11]唐佑萍,卓绮雯,叶霭莹,等.数据挖掘技术在我院合理用药管理中的应用[J].中国数字医学,2013,8(9):110-112.
[12]韩蓉,吴俊.基于数据挖掘技术的门诊合理用药分析[J].中国数字医学,2014,9(2):18-20.
[13]国家卫生和计划生育委员会.抗菌药物临床应用指导原则(2015年版)[S].2015-07-24.
[14]宋花玲,贺佳,黄品贤,等.ROC曲线下面积估计的参数法与非参数法的应用研究[J].第二军医大学学报,2006,27(7):726-728.
[15]田锋奇,于卫江,胡伟.我院2014-2015年住院患者用药医嘱点评与分析[J].中国药房,2016,27(17):2315-2317.
[16]邱晓春,韦灵玉,张琼.我院2012-2014年抗菌药物医嘱点评分析[J].中国药房,2015,26(35):4919-4922.
[17]刘方园,王水花,张煜东.支持向量机模型与应用综述[J].计算机系统应用,2018,27(4):1-9.
[18]傅翔,杨樟卫,陈盛新.数据挖掘在合理用药信息分析中的应用[J].药学实践杂志,2009,27(6):411-413,433.
[19]黎颖然,郭少青,杜生,等.Pnodiver数据挖掘系统下临床药学工作的实践应用[J].中国数字医学,2015,10(4):75-76,103.
[2]程模,杨培洪,何霖.专项处方点评模式对我院清洁手术围术期抗菌药物预防使用的干预效果评价[J].中国药房,2018,29(3):293-297.
[3]陈猛,舒志兵,周月红.某“三甲”综合性医院腹腔镜手术围手术期抗菌药物预防使用情况调查及合理性评价[J]中国药房,2018,29(8):1065-1068.
[4]廖丽娜,李鑫,黄菲,等.处方前置审核系统在保障患者用药安全中的作用[J].中国医院管理,2018,38(10):23-25.
[5]高燕灵,扶玲,钟秀娟,等.合理用药监测系统存在的问题及对策探讨[J].中国药房,2015,26(22):3159-3161.
[6]徐芳琴,郭超,陆杨,等.合理用药决策系统对我院妇产科抗菌药物使用的干预效果分析[J].中国药房,2016,27(8):1133-1135.
[7]朱芳芳,沈怡,廖丽娜,等.我院《临床合理用药智能化管理解决方案》的应用与实践[J].中国药房,2016,27(25):3528-3531.
[8]曾小辉,彭涛,高月琴.基于前列腺影响报告和数据系统第2版的机器学习模型诊断高级别前列腺癌[J].中国医学影像技术,2018,34(12):1852-1856.
[9]傅裕,鲍迎秋,谢沂伯,等.人工智能技术在皮肤病辅助诊断的应用研究[J].中国数字医学,2018,13(10):29-31,38.
[10]杜明伦,黄君君,马香,等.大肠杆菌基因组中重叠基因注释的机器学习优化方法[J].中国生物化学与分子生物学报,2018,34(8):861-867.
[11]唐佑萍,卓绮雯,叶霭莹,等.数据挖掘技术在我院合理用药管理中的应用[J].中国数字医学,2013,8(9):110-112.
[12]韩蓉,吴俊.基于数据挖掘技术的门诊合理用药分析[J].中国数字医学,2014,9(2):18-20.
[13]国家卫生和计划生育委员会.抗菌药物临床应用指导原则(2015年版)[S].2015-07-24.
[14]宋花玲,贺佳,黄品贤,等.ROC曲线下面积估计的参数法与非参数法的应用研究[J].第二军医大学学报,2006,27(7):726-728.
[15]田锋奇,于卫江,胡伟.我院2014-2015年住院患者用药医嘱点评与分析[J].中国药房,2016,27(17):2315-2317.
[16]邱晓春,韦灵玉,张琼.我院2012-2014年抗菌药物医嘱点评分析[J].中国药房,2015,26(35):4919-4922.
[17]刘方园,王水花,张煜东.支持向量机模型与应用综述[J].计算机系统应用,2018,27(4):1-9.
[18]傅翔,杨樟卫,陈盛新.数据挖掘在合理用药信息分析中的应用[J].药学实践杂志,2009,27(6):411-413,433.
[19]黎颖然,郭少青,杜生,等.Pnodiver数据挖掘系统下临床药学工作的实践应用[J].中国数字医学,2015,10(4):75-76,103.