Advancing Practice Instruction and Innovation Through Informatics

Presented at the 1999 APIII Conference Return to 1999 Abstract Index

DERMATOPATHOLOGY FALSE NEGATIVE TERMS IN UMLS

Baltimore Veteran's Administration Medical Center
Pathology and Laboratory Medicine Service
Baltimore, Maryland
G. William Moore, MD, PhD

Grace F. Kao, M.D1,2; G. William Moore, MD, PhD1,3,4
1Pathology and Laboratory Medicine Service, Veterans Affairs Maryland Health Care System, Baltimore, Maryland
2Division of Dermatology and Department of Medicine, George Washington University School of Medicine, Washington, D.C.
3Department of Pathology, University of Maryland School of Medicine, Baltimore, Maryland
4Department of Pathology, The Johns Hopkins Medical Institutions, Baltimore, Maryland

Background: The Unified Medical Language System Metathesaurus (UMLS-M) of the U. S. National Library of Medicine is the most comprehensive, publicly-available list of standardized medical terminology in the world. Published false-negative rates for the UMLS-M applied to general medical text are reported as one to two percent. This study examines the false-negative rate of UMLS-M concepts in publicly-available electronic dermatopathology text.

Design: The entire collection of uncopyrighted dermatopathology image-legends from the the Armed Forces Institute of Pathology Electronic Fascicles (AFIP-EF) on melanocytic and non-melanocytic tumors of the skin was encoded into the UMLS, via a computer translation program that parses and maps plain-text into UMLS terms by the barrier word method, in which natural-language medical text is processed as a sequence of MEDICALLY-SIGNIFICANT TERMS, linked together with grammatical objects, or BARRIER WORDS. A medically-significant term, present in the image-legend but not captured by the program, was defined as a FALSE NEGATIVE. Conversely, a term captured by the program but not present in the text, was defined as a FALSE POSITIVE. By design, the program captured no false positives. Ambiguous terms and compound terms containing subconcepts were indexed redundantly.

Results: There were 446 image-legends from the AFIP-EFs on melanocytic and non-melanocytic tumors of the skin, 2,140 distinct words, and 8,334 UMLS concept-terms. Each image-legend yielded an average of 18.7 UMLS concept-terms per legend, ranging in frequency from three concept-terms in the least-indexed legend to 57 concept-terms in the most-indexed legend. There were 260 false-negative terms, for a false-negative rate of 3.1%. False-negative UMLS concepts tended to be descriptive terms in dermatopathology that characterize microscopic findings.

Conclusion: The UMLS has a false-negative rate for dermatopathology concepts of 3.1%, similar to the reported findings for general medical text. This study supports the view that the UMLS is a nearly-comprehensive metathesaurus for dermatopathology text. Missing concepts could be suggested for incorporation into future updates of the UMLS.