Using SemRep to label semantic relations extracted from clinical text.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Biomedical Data Science

Background:

• Understanding semantic relatedness between medical concepts is crucial for analyzing clinical reports and biomedical literature.
• Existing methods often rely on co-occurrence in text, but the nature of the association requires deeper investigation.

Purpose of the Study:

• To investigate the relationship between semantic relatedness of medical concepts in clinical reports and their associations in biomedical literature.
• To determine if labeled relations from Medline abstracts can clarify associations observed in clinical data.

Main Methods:

• Utilized a corpus of 800,000 inpatient clinical notes to assess concept association strength.
• Employed the SemRep database, which extracts labeled relations from Medline abstracts, as a source of predicate types.
• Analyzed the frequency of different predicate types for concept pairs within the SemRep database.

Main Results:

• Identified that the same medical concept pairs can have multiple predicate types in SemRep, often with varying frequencies.
• Demonstrated that predicate type frequency information from SemRep is beneficial for labeling semantic relations derived from semantic relatedness and similarity measures.
• Found a correlation between distribution-based measures in clinical reports and labeled relations in biomedical literature.

Conclusions:

• Predicate type frequencies from SemRep provide valuable insights into the nature of semantic relations between medical concepts.
• Combining clinical report data with labeled biomedical literature relations enhances the understanding of medical concept associations.
• This approach offers a robust method for semantic relation labeling in clinical and biomedical text analysis.