Bridging the Granularity Gap in Family History Information Extracted from Clinical Narratives

Area of Science:

• Medical Informatics
• Genomics and Computational Biology
• Clinical Epidemiology

Background:

• Family history (FH) is vital for assessing disease risk and guiding prevention strategies.
• Integrating FH data from electronic health records (EHRs) into analytics is hindered by a lack of standardization.
• Automating the alignment of FH concepts is necessary for consistent data utilization.

Purpose of the Study:

• To automatically align FH concepts from clinical text to established disease classification resources.
• To evaluate the mapping coverage and granularity of FH concepts across different terminological systems.
• To address the challenge of standardizing NLP-derived FH information for clinical applications.

Main Methods:

• Utilized the Unified Medical Language System (UMLS) to map FH concepts.
• Employed parent and broader/alike relationships within UMLS for concept alignment.
• Compared mapping performance against five resources: Clinical Classification System (CCS), Phecode, Comparative Toxicogenomics Database (CTD), Human phenotype ontology, and Human disease ontology (HDO).

Main Results:

• Achieved high mapping coverage of FH concepts using UMLS.
• Comparative Toxicogenomics Database (CTD) demonstrated the highest coverage (93%) of FH concepts.
• Human disease ontology (HDO) offered the coarsest granularity, while CCS provided the finest granularity of FH disease categories.

Conclusions:

• The study successfully demonstrated a method for standardizing NLP-derived FH data.
• Leveraging UMLS and established ontologies can mitigate granularity challenges in FH analytics.
• This approach enhances the utility of EHR-derived FH information for disease risk assessment and prevention.