Classifying diseases with respect to anatomy: a study in SNOMED CT.

Anatomy is a major organizing principle for dis-eases. In the formal definitions provided by SNOMED CT, for example, the role 'finding site' relates disorders to anatomical entities. This study investigates SNOMED CT and compares the anatomy-based classification of diseases supported by the role finding site to the anatomy-based classification of diseases provided by subsumption (is-a) relations between diseases. For each of the 3,540 anatomical entities associated with disorders,, we compared two sets of disorders: first, the set of disorders associated with any descendant of the anatomical entity under investigation (ANAT); second, the set of dis-orders corresponding to the union of the descendants of the disorders associated with the anatomical entity under investigation (TAXO). The ANAT and TAXO sets were different for 1,231 anatomical entities (35%). In 607 cases, the overlap between ANAT and TAXO was less than 50%. When a difference was found, the TAXO set was always a subset of the ANAT set. Among the 1,025,904 subsumption relations among disorders generated by the ANAT approach, 40% were not present in TAXO. This approach helps identify missing classes and taxonomic relations in existing ontologies. It can be generalized to other kinds of partitions of biomedical ontologies.