The U-M School of Dentistry has employed natural language processing, machine learning, and graph database technology to parse our plain-text patient histories and connect them to the patient's treatment history and a standard tree of health care concepts, the Unified Medical Language System (UMLS). By associating free text with computable knowledge, we have unlocked a trove of previously unusable information about our patient population.
Applications of this tool include identifying communities of patients with similar health and dental care patterns for more targeted precision care, searching for populations of patients for research, identifying gaps in the medical record, improvements to our history-collection processes and forms, and reconciling patient histories at the School of Dentistry with those at Michigan Medicine.
This talk will describe the process and tools used to build the knowledge graph and demonstrate example applications of the tool. We will also discuss our plans for continuing improvement and extension. Technologies used include: Neo4j, MetaMap, MySQL, CLAMP, and Python with scikit-learn.
