Filtering the UMLS ® Metathesaurus ® for MetaMap 2011 Edition Francois

The MetaMap program’s purpose is to discover the Metathesaurus concepts referred to in arbitrary text. A given Metathesaurus concept can have many alternative names (Metathesaurus strings) which originate in the many source vocabularies included in the Metathesaurus. As the number of strings has grown over the years, MetaMap’s performance has suffered. In the 2011AA version of the Metathesaurus, for example, the Metathesaurus includes 5,875,828 English strings, 5,812,906 (98.92%) of them distinct, comprising 2,394,524 concepts. There are 8.9% more English strings and 9.1% more concepts than in the 2010AA edition. Many of the strings in the Metathesaurus are of little value to MetaMap for one of four reasons: 1. Some strings are virtually indistinguishable from each other; for efficiency, only one representative of a set of indistinguishable strings is needed. 2. Some strings either represent general, nonmedical concepts, are unnecessarily ambiguous, or have been found to be problematic for some other reason. 3. Some strings have an assigned type in their vocabulary because they have a form (e.g., an idiosyncratic abbreviation) that is highly unlikely to appear in regular text. 4. Some strings, including lengthy descriptions of things such as procedures, health activities or medical devices, are so complicated that it is again unlikely to find them in normal text.