Analysis of readability and structural accuracy in SNOMED CT

Area of Science:

• Biomedical Informatics
• Ontology Engineering
• Computational Linguistics

Background:

• Growing adoption of ontologies in biomedical research necessitates robust quality assurance.
• Existing quality assurance processes for ontologies like Gene Ontology and SNOMED CT are insufficient for detecting all modeling flaws.
• There is a need for efficient and effective quality assurance methods for biomedical ontologies.

Purpose of the Study:

• To propose quantitative metrics for assessing the readability and structural accuracy of biomedical ontologies.
• To leverage lexical regularities within ontology content for quality analysis.
• To provide a framework for identifying modeling flaws and improving ontology engineering.

Main Methods:

• Developed quantitative metrics based on lexical regularities in ontology content.
• Readability metrics assess the ratio of labels, descriptions, and synonyms.
• Structural accuracy metrics evaluate adherence to best practices: lexically suggest locally define (LSLD) and systematic naming.

Main Results:

• Applied metrics to various SNOMED CT versions, showing stability over time but sensitivity to modeling changes.
• LSLD metric increased from 0.27 to 0.31; systematic naming metric remained around 0.17.
• Analysis of SNOMED CT July 2019 release revealed varying adherence to structural accuracy criteria across hierarchies (LSLD: 0-0.92, systematic naming: 0.08-1), identifying non-compliant cases.

Conclusions:

• Generated valuable information for ontology engineering through novel readability and structural accuracy metrics.
• Demonstrated the utility of lexical regularities for defining structural accuracy metrics.
• Provided quality assurance insights for SNOMED CT, highlighting areas for improvement.