An open-source, interactive Java-based system for rapid encoding of significant events in the ICU using the unified medical language system

TL;DR: An interactive Java-based application that encodes significant clinical events by retrieving the clinical concepts that most closely match a free-text input phrase by integrating a spell-checker and a personalized dictionary.

Abstract: We present an interactive Java-based application that encodes significant clinical events by retrieving the clinical concepts that most closely match a free-text input phrase. Events are coded using a subset of the National Library of Medicine's Unified Medical Language System (UMLS), a standard, freely available collection of medical vocabularies for identifying diseases, symptoms, and other clinical concepts. Some common difficulties that occur in the process of coding are spelling errors, abbreviation ambiguities, and combinations of events for which there is no UMLS code. We address these issues by integrating a spell-checker and a personalized dictionary allowing each user of the system to keep a unique list of frequently used abbreviations. Additionally, utilizing the UMLS concept hierarchy and using "fuzzy" string matching both help to find the UMLS concepts that most closely match the free text description. Using almost 1500 medical phrases manually extracted from a random selection of over 300 nursing notes, three clinicians scored the accuracy of the coding algorithm while the program monitored the distributions of search times. For each medication, disease or sign/symptom group, the algorithm successfully coded 97%, 92% and 77% of the phrases, respectively. Although our application is written to support the MIMIC II project (previously described at Computers in Cardiology 2002), similar applications will be important components of future real-time medical decision support systems. The use of a flexible API and freely available dictionaries facilitates open source distribution and integration.