Auditing complex concepts of SNOMED using a refined hierarchical abstraction network

Area of Science:

• Medical Informatics
• Ontology Engineering
• Knowledge Representation

Background:

• Auditing large terminologies such as SNOMED CT presents significant challenges.
• Complex concepts, difficult to model, require specialized identification techniques.
• Overlapping concepts, defined as those in intersections of multiple partial-areas, are a focus of this work.

Purpose of the Study:

• To develop and evaluate a novel auditing methodology for complex, overlapping concepts within SNOMED CT.
• To leverage a disjoint partial-area taxonomy for a structured, top-down review of these concepts.
• To investigate error ratios associated with overlapping concepts in SNOMED CT.

Main Methods:

• Development of a disjoint partial-area taxonomy from partitioned overlapping concepts.
• Implementation of a top-down auditing approach within semantically uniform groups.
• Application of the methodology to SNOMED CT's Specimen hierarchy across two releases.
• Statistical analysis using the double bootstrap and Fisher's exact test.

Main Results:

• The novel auditing methodology was applied to SNOMED CT's Specimen hierarchy.
• Auditing overlapping concepts, particularly roots of partial-areas, yielded a statistically significant higher proportion of errors.
• Hypotheses regarding error ratios for different types of overlapping concepts were formulated and tested.

Conclusions:

• The developed disjoint partial-area taxonomy provides a basis for an effective auditing methodology.
• The top-down, semantically grouped review effectively identifies complex concepts requiring attention.
• The findings indicate a higher error rate in overlapping concepts within SNOMED CT, warranting focused auditing efforts.