Recently, we have seen a new generation of medical terminologies, satisfying the popular desiderata of Cimino. Examples are SNOMED CT (theSystematized Nomenclature of Medicine - Clinical Terms), the National Cancer Institute Thesaurus (NCIT), Kaiser's Convergent Medical Terminology (CMT), the Veteran Administration's internal enterprise terminology, the Foundational Model of Anatomy (FMA) and the Medical Entities Dictionary (MED). These terminologies are of substantial size and complexity, which makes errors unavoidable. Therefore, auditing them is essential, because terminologies underlie decision-support systems, clinical patient records, healthcare administrative systems, etc. We propose to design auditing techniques that utilize the features of terminologies satisfying the desiderata, such as systematic relationship inheritance. Our techniques will be based on dividing a large collection of concepts, in two steps into smaller units of structurally and semantically similar concepts that can be more easily comprehended and audited. The units are Areas, which contain sets of concepts with the same relationships (structure) and Partial Areas, which contain concepts from an Area that are all descendants of one single root concept (uniform semantics). We will develop two levels of abstraction networks (Area Taxonomy and Partial-Area Taxonomy) that support compact views, auditing, maintenance and comprehension. An intermediate partition using obtainment-pattern regions refines the Partial-Area taxonomy. Advanced partitioning methodologies will be developed to handle some more complex cases. These auditing methodologies will be general and applicable to terminologies conforming to Cimino's desiderata that allow a relationship to be introduced at multiple points in the IS-Ahierarchy, a feature lacking in the MED and its previous analyses. A sample of errors in SNOMED and NCIT is presented. We will use SNOMED and the NCIT's genomics hierarchies as test-beds for our methodologies. We will perform a rigorous evaluation of our results of auditing SNOMED and NCIT and our hypothesis about the high likelihood of errors in small Partial-Areas and strict inheritance regions, and our auditing methodology. This will be done involving human subjects, our Ph.D.students, Drs.Cimino and Hripcsak of Columbia, Dr. Elhanan of 3M Medical Systems, Dr. Spackman of SNOMED and two editors from the American Academy of Ophthalmologists (AAO).
