Information extraction from narrative data

Abstract

Interesting and exciting work is being done in the medical field to advance clinical decision-support (CDS) technology. One major focus of research is finding ways to access relevant information from a patient’s electronic health record (EHR) and linking that data to CDS tools. The two types of information found in the EHR are coded data and narrative (non-coded) data. Coded data contain structured data elements that are easily queried, such as International Classification of Diseases codes, laboratory values, and medication names. For example, searching an EHR to determine if a patient has taken simvastatin might be as simple as entering the drug name as a search term. However, much of the desired knowledge in the EHR may be, in effect, locked away in narrative text, from which it is more difficult to access data relevant to a particular information need. Narrative text is written in free form (e.g., phrases or sentences in the body of a clinical note). When narrative data such as chart notes are entered into the record, data-retrieval problems arise in relation to the use of synonyms and jargon, misspellings, multiple abbreviations of the same term, and differences between providers in documentation style. Natural language processing (NLP), a method of automated data extraction, holds the potential to unlock the data of interest contained within narrative text.