Incremental Semantic Enrichment of Narrative Content in Electronic Health Records

Semantic interoperability is a major desideratum in health care for computer-based documentation and communication through electronic health records. They require structured data ideally represented via standardized information models, e.g., HL7 RIM or openEHR, connected to standardized terminologies or ontologies, e.g., SNOMED CT or LOINC. But since natural language is seen by health professionals as their most natural and effective form of expression, semantically interoperable architectures must adequately deal with unstructured data and be seamlessly integrated into the workflows of health professionals. Therefore we propose a self-learning natural language processing system, which automatically segments input narratives into sections, detects contexts such as negations, and assigns terminology codes. To be usable in clinical contexts, the system must properly handle the idiosyncratic medical language and grammar and spelling errors of narratives produced in everyday clinical practice. We present user interaction items that are important to make the interaction with the system as easy as possible to reach high acceptance among health professionals.