基于文本分类技术的住院患者药源性变态反应自动监测模块研究

Study on automatic monitoring module of inpatient drug-induced allergy based on text categorization technology

目的:利用医疗电子病历中的文本信息开展住院患者用药安全性评价,为住院患者ADR监测提供新方法。方法:在已有的主动监测系统基础上,设计、开发基于文本分类技术的住院患者药源性变态反应自动监测模块,利用优质文本进行分类算法的机器学习。结果:完成包括事件配置器、特征词集、自然语言处理器、文本分类器、结果展示器5部分的主动监测模块的开发;试用于头孢哌酮舒巴坦用药患者的主动监测,结果显示629例患者中出现变态反应的阳性预测值达到44.44%(4/9),其真实世界发生率0.64%,与说明书中所列的发生率0.68%相近。结论:本研究建立了文本信息主动监测方法,阳性预测值可通过特征词集、分类规则的深入研究加以改善。

Objective： To evaluate drug safety of inpatients by text information in electronic healthcare records, andprovide the new method for inpatients ADR monitoring. Methods： On the basis of active monitoring system, automatic monitoringmodule of inpatient drug-induced allergy based on text categorization technology was designed and developed, and the machinelearning of categorization algorithm was carried out through qualified text. Results： The active monitoring module consists of theevent configurator, the representative feature set, the natural language processors, the text categorizer and the result display unit. Themodule had been tested, and the positive predictive value was 44.44% （4/9） in the 629 users of cefoperazone and sulbactam. Thereal world incidence rate was 0.64%, which was similar with the incidence rate （0.68%） in the drug directions. Conclusion： Textinformation active monitoring method had been built and the positive predictive value could be improved by further study of therepresentative feature set and categorization rules.