自然语言处理技术在临床检查合理性质控中的应用研究
尹丽丽 ,范正磊 ,徐畅 ,张群 ,王浩 ,韩影
作者信息
- 徐州仁慈医院 江苏 徐州 221000
摘要
目的:针对临床检查中多开、漏开、错开及禁忌证检查等不合理现象,构建基于自然语言处理的临床检查合理性校验系统,实现检查单开具的实时智能审核。方法:以徐州仁慈医院2023年185 137人次检查数据为研究对象,联合多部门制订临床检查合理性规则库。采用临床BERT模型,即临床双向编码器表示模型(clinical bidirectional encoder representations from transformers,Clinical BERT),从病历文本中自动提取患者关键信息,基于Profile架构开发校验平台并嵌入电子病历系统,实现患者信息与规则库的实时匹配校验。结果:Clinical BERT模型实体识别F1值达94.7%,系统响应时间1.8 s,规则匹配精准率92.3%。系统应用后,试点科室检查不合理率降幅达63.2%。其中多开检查减少41.3%,漏开检查减少35.8%,禁忌证检查实现100%拦截。结论:基于自然语言处理(natural language processing,NLP)的临床检查合理性校验系统可有效识别病历关键信息,实现检查单实时智能校验,显著降低临床检查不合理率,具有良好推广应用价值。
Research on the Application of Natural Language Processing Technology in the Rationality Quality Control of Clinical Examinations
Lili Yin , Zhenglei Fan , Chang Xu , Hao Wang , Ying Han
Author Information
- Xuzhou Renci Hospital, Xuzhou 221000, Jiangsu Province, China
Abstract
Objective: To address unreasonable clinical examinations such as over-ordering, under-ordering, wrong-ordering,and examinations with contraindications, a clinical examination appropriateness verification system based on natural languageprocessing was constructed to achieve real-time intelligent review of examination orders. Methods: A total of 185 137examination records from Xuzhou Renci Hospital in 2023 were selected as the research subjects, and a clinical examination appropriateness rule base was developed jointly by multiple departments. The clinical bidirectional encoder representations from transformers (Clinical BERT) model was used to automatically extract key patient information from medical record texts. A verification platform was developed based on the Profile architecture and embedded into the electronic medical record system to achieve real-time matching and verification between patient information and the rule base. Results: The entity recognition F1 score of the Clinical BERT model reached 94.7%, the system response time was 1.8 s, and the rule matching precision was 92.3%. After system application, the unreasonable examination rate in pilot departments decreased by 63.2%.Among them, over-ordering of examinations decreased by 41.3%, under-ordering decreased by 35.8%, and examinations with contraindications achieved 100% interception. Conclusion: The clinical examination appropriateness verification system based on natural language processing (NLP) can effectively identify key information in medical records, achieve realtime intelligent verification of examination orders, and significantly reduce the rate of unreasonable clinical examinations,demonstrating good value for promotion and application.
