A query interface for clinical research with Chinese electronic health record using Natural Language Processing

Abstract Background: The recruitment of clinical trials is a challenging task. Traditionally, the time-consuming task is accomplished manually or assisted by form-based tools. Electronic health record (EHR) contains comprehensive information which can speed up the process. The development of natural language processing (NLP) makes it possible to reduce manual effort. To our knowledge, there are almost no query interfaces based on NLP techniques for Chinese EHR.Methods: A query interface based on NLP was developed. Firstly, we collected and annotated eligibility criteria(EC) from the Chinese Clinical Trial Registry. Then, an information extraction(IE) system was developed to parse them including named entity recognition(NER), relation classification(RC), and concept matching(CM). Next, the extracted information was post-processed(PP) to generate formal queries. Finally, we built a natural language query interface based on the system.Results: 4691 stroke-related EC were collected for annotation. 91.23% F1 score was achieved for the NER task. For the RC task, 92.75% F1 score was achieved. For the CM task, 90.78% accuracy was achieved and an evaluation result showed 42.42% of entities from NER results can be matched with EHR. And a natural language query interface has been implemented and applied in clinical research about stroke. Conclusions: We build a query interface for Chinese EHR based on NLP techniques. The proposed information extraction pipeline can support medical professionals to reduce the information gap with minimal human effort when interacting with Chinese EHR.