We present preliminary results for the application of a procedure that detects and corrects errors in concept definitions of a local interface vocabulary with SNOMED CT as its reference vocabulary. Using the relations inferred by SNOROCKET we detected redundant fully defined concepts, but also we detected suspected patterns where concepts had redundant inferred relations.
Our procedure detected errors in 1.63% of the whole vocabulary, the primary type of error was produced by duplications since these concepts did not exist when the knowledge modeler asserted them. Using these results, we implemented a GUI to track patterns and correct errors. Our procedure contributes to the quality assurance of our local interface vocabulary since errors in the hierarchies can compromise interoperability and meaningful use of the vocabulary. Our approach could be used by thesaurus implementers to detect suspected patterns, grouping them, and offer a centralized interface to correct them.