Related Experiment Videos
Extracting medical knowledge for a coded problem list vocabulary from the UMLS Knowledge Sources
J W Hales1, K M Schoeffler, D P Kessler
1Division of Medical Informatics, Duke University Medical Center, Durham, North Carolina, USA.
Introduction:
The Unified Medical Language System (UMLS) Knowledge Sources embody a rich source of medical knowledge. We sought to extract a portion of this knowledge by incorporating information about relationships between UMLS concepts into an existing problem list vocabulary.
Methods:
We matched terms from the coded problem list of The Medical Record (TMR), a computer-based patient record system, with those found in the UMLS Metathesaurus. Those UMLS concepts that participate in 'parent' relationships with the matched TMR concepts were translated back into TMR codes and the relationship information was retained for integration into the coded problem list of TMR.
Results:
Of the coded problems currently in use in TMR, 67% (1627/2436) could be matched by normalized string matches to the UMLS Knowledge Sources. Of these matched TMR concepts, 91% (1488/1627) participated in at least one UMLS-identified parent relationship but only 28% of the matched concepts (454/1627) participated in parent relationships that already matched to a TMR code. As a result, although 67% of TMR codes were matched to UMLS concepts, only 19% of our original problem list (454/2436) could be augmented by relationship information contained in UMLS without improving the rate of matches or adding additional UMLS concepts as coded problems in TMR.
Conclusion:
This study illustrates the rapid decline in overall rates of matching that result from a multiplicative effect of successive matches of terms to concepts, concepts to relationships and concepts back to entry terms. This effect will hamper any effort to extract relationship knowledge from the UMLS for incorporation into an entry vocabulary that is not already one of the source vocabularies of the UMLS Metathesaurus.