Automated knowledge extraction from the UMLS.

This paper presents our work in extracting disease-chemical relationship knowledge from the UMLS Co-occurrence table (MRCOC) using an automated method. We evaluated the quality of the knowledge from UMLS MRCOC by comparing it with knowledge from other sources: For disease-lab chemical relationships, knowledge was obtained from a decision support system (DXplain) and our own knowledge base of medical terminology (MED) through automated processes. For disease-drug chemical relationships, knowledge was manually acquired from the medical literature. Evaluations showed that the UMLS MRCOC knowledge has good sensitivity, especially regarding disease-drug relationships. We are using this knowledge to produce disease-specific views of patients' electronic patient record.