Mapping Phenotypic Information in Heterogeneous Textual Sources to a Domain-Specific Terminological Resource

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Clinical Data Mining

Background:

• Biomedical literature and Electronic Health Records (EHRs) contain valuable, yet distinct, disease-phenotype information.
• Variability in textual expression of phenotypes across sources hinders automatic information integration.
• Bridging these diverse text types is crucial for discovering novel disease-phenotype associations.

Purpose of the Study:

• To develop and evaluate PhenoNorm, a novel method for automatically linking phenotype mentions to standardized biomedical concepts.
• To address the challenge of diverse terminologies and writing styles in EHRs and literature.
• To facilitate the integration of phenotype information from disparate sources for enhanced biomedical research.

Main Methods:

• Developed PhenoNorm, integrating multiple similarity measures for concept linking.
• Utilized the PhenoCHF corpus, containing EHR narratives and literature annotated for congestive heart failure (CHF) phenotypes.
• Evaluated PhenoNorm against alternative methods using an expert-verified enriched corpus.

Main Results:

• PhenoNorm accurately links phenotype mentions to UMLS Metathesaurus concepts.
• The method demonstrated superior performance compared to alternative approaches on the PhenoCHF dataset.
• PhenoNorm showed robust performance and favorable accuracy in linking diverse medical concepts across broader subject areas.

Conclusions:

• PhenoNorm effectively bridges the gap between different text types (EHRs and literature) for phenotype information.
• The method facilitates the automatic integration of valuable clinical and research data.
• PhenoNorm offers a versatile tool for enhancing biomedical data analysis and knowledge discovery.