Analysis of medical student content searches that resulted in unidentified UMLS concepts.

Many authors have reported on the use of the Unified Medical Language System (UMLS) to match concepts in free text. Unmatched search strings may be due to misspellings, concepts not in the UMLS, or searches for words not expected to be in the UMLS (e.g., names of people or places). We mapped search strings from a full-text, concept-based curriculum database to UMLSconcepts and performed a failure analysis. The majority of unmatched text strings were medically related (71.7%). Unrecognized abbreviations (11.6%)and misspellings (38.5%) were the most common causes of unmatched medically related searches. Content searching must take these searches into account for completeness.