Disease ontology, defined as a causal chain of abnormal states, is believed to be a valuable knowledge base in medical information systems. Automatic mapping between electronic health records (EHR) and disease ontology is indispensable for applying disease ontology in real clinical settings. Based on an analysis of ontologies of 148 chronic diseases, approximately 41% of abnormal states require information extraction from clinical narratives. This paper presents a semi-automatic framework to identify abnormal states in clinical narratives. This framework aims to effectively build mapping modules between EHR and disease ontology. We show that the proposed method is effective in data mapping for 18%-33% of the abnormal states in the ontologies of chronic diseases. Moreover, we analyze the abnormal states for which our method is invalid in extracting information from clinical narratives.
Keywords: Machine learning; Natural language processing; Ontology.