改进近邻传播聚类算法的抗生素处方行为

Antibiotic Prescribing Behavior of Physicians Based on Improved Affinity Propagation Clustering

通过有效分析某医院抗生素处方行为,提出了一种基于数据仓库改进的近邻传播聚类方法,利用数据仓库透视出所需数据,而利用改进的近邻传播聚类方法在数据仓库的基础上聚类出具有代表性医生的抗生素处方数据,找到抗生素处方行为的影响因素以及规范医生处方行为的评价指标.采用2012年浙江省某三甲医院信息系统中的抗生素处方数据,对数据进行横断面研究,首先对提取的数据建立数据仓库,之后利用改进的近邻传播聚类算法对数据降维和分类,得到医生的抗生素处方行为的训练集和测试集,最后利用SAS9.1软件的多因素方差分析和配对非参数检验,分析抗生素处方行为的影响因素以及评价指标.结果表明,不同科室、不同月份、不同抗生素种类对抗生素处方数据有显著影响,而在该医院青霉素、头孢菌素随季节变化有显著差异,可作为医生的处方评价指标.

A method based on improved affinity propagation clustering of data warehouse is proposed by analyzing physicians＇ antibiotic prescribing practices in this article. Loading pivottable by the data warehouse, we can select a representative physicians＇ antibiotic prescription data and find the internal factors of antibiotic prescribing behavior of physicians through improved affinity propagation. The data is from the antibiotic prescription of information system of a first-class hospital in Zhejiang province in 2012, which is studied through cross-sectional. Firstly, we build the data warehouse, then cluster and reduce the dimensionality of data set by improved affinity propagation clustering to get the training set and test set. Finally, the results show that different departments, months and types of antibiotics have significant impacts on the data of antibiotic prescriptions. There are significant differences of cephalosporin and penicillin with seasonal changes in this hospital which can be used as evaluation index of the doctor＇s prescription.