Mapping Clinical Narrative Texts of Patient Discharge Summaries to UMLS Concepts

Abstract

Patient discharge summaries are typical unstructured text which is not amenable for processing by an automated system. For an efficient electronic healthcare system, free-form medical texts generated at various subsystems need to be mapped into standard codes like ICD-10, SNOMED-CT, etc. The Unified Medical Language System (UMLS) unifies these standard codes into a set of concepts identified by a Concept Unique Identifier (CUI). In this paper, we have used NLP techniques to map clinical narrative texts found in discharge summaries to CUIs. We have developed a Matcher algorithm to match the clinical text strings to that of UMLS and thereby extract the concepts. We achieve 70% similarity between the set of concepts generated using our Matcher algorithm to that of a gold standard tool.