护理学报2018，Vol.25Issue(3)：1-4,4.DOI:10.16460/j.issn1008-9969.2018.03.001
护理不良事件非结构上报内容的自然语言处理及效果比较
Natural Language Processing of Non-structured Text of Nursing Adverse Events and Its Effect
摘要
Abstract
Objective To verify the feasibility of natural language processing (NLP) in non-structured text of the nursing adverse events. Methods From Jan 2013 to Dec 2016, 1,599 cases of pressure ulcers adverse events were included. Sensitivity, specificity, agreement and positive predictive value were calculated to evaluate the identification of nursing adverse event factors through NLP and manual annotation (MA). Results The marked rate of NLP group (88.56%) was significantly higher than that of MA group (63.77%)(P<0.001). The sensitivity of NLP group reached 87.19%, but the specificity (9.03%) and the accuracy (62.79%) were lower. Iterative analysis of NLP group showed that the sensitivity, positive predictive value and agreement would boost with the increasing amount of training data (χ2=2607.603, P<0.001). There were 16 factors related to the reported text of the nursing adverse events through the analysis of the location of the annotations, 7 of which had the sensitivity higher than 70%. Conclusion NLP can effectively analyze and identify unstructured escalation texts of the nursing adverse events and the iterative training set can improve the recognition accuracy of NLP. It is of great significance for the big data analytics and artificial intelligence of nursing adverse events.关键词
风险管理/不良事件报告/自然语言处理/人工智能Key words
risk management/incident reporting/natural language processing/artificial intelligence分类
医药卫生引用本文复制引用
宋杰,章洁,高远,皮红英..护理不良事件非结构上报内容的自然语言处理及效果比较[J].护理学报,2018,25(3):1-4,4.基金项目
中国人民解放军总医院医疗大数据研发项目(2016MBD-005) （2016MBD-005）
