Combining matching strategies for medical vocabulary

Heterogeneity among knowledge representation systems poses a significant obstacle for the efficient information sharing across health care institutions. This happens because, despite the consensus that the use of standard terminology should be enforced, many institutions still use legacy independently developed systems or make adaptations to standard code systems to be used locally. Mappings across local and standard medical vocabulary are therefore needed in a range of medical domain scenarios. Many of those are related to the translation of guidelines, protocols, and patient reports to enable better continuity of patient care. Query rewriting and query expansion is also often necessary to enable searches across medical institutions.

Providing such mappings manually although is an extremely cost and time intensive task. Matcher software can offer support to a human specialist in the task of establishing those mappings by suggesting pairs of concepts which, according to given matching criteria or method, are similar.

Within the context of this project, the performance of several matching criteria and methods of combination of these criteria are evaluated. This evaluation is performed using portions of UMLS source vocabularies. The relationship between vocabulary structure and the precision of matching methods is also subject of analysis.

Another important outcome of this research is the development of a flexible medical vocabulary matcher. It provides the human domain specialist a library of lexical, structural and semantic similarity metrics and allows her to choose how to combine this metrics according to the characteristics of the vocabularies to be mapped. A list of mappings between the vocabularies is created within a series of interactions with the user, which chooses to accept or reject the suggestions made by the Matcher.

Research Team

  • M.Sc. Mariana Kessler Mortoluzzi

 

Funding:

This research is funded by a DAAD grant.