Extracting Family History Information From Electronic Health Records: Natural Language Processing Analysis.

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Biomedical Data Science

Background:

• Family history (FH) is crucial for diagnosing and treating genetic disorders but is often unstructured in clinical notes.
• Accessing FH data from free-text clinical notes remains a challenge.

Purpose of the Study:

• To develop automated natural language processing (NLP) methods for extracting family history (FH) data from clinical text.
• To improve the accessibility and utility of FH information for patient care.

Main Methods:

• Utilized transformer models for disease mention extraction.
• Employed rule-based methods and coreference resolution for family member (FM) information extraction.
• Evaluated transfer learning strategies and performed error analysis for system optimization.

Main Results:

• Achieved an 81.63% F1 score for disease and FM extraction on a public dataset, significantly improving over baseline.
• Outperformed the median F1 score of 76.59% from the 2019 N2C2/Open Health NLP Shared Task.
• Demonstrated statistically significant improvements (P<.001) in information extraction accuracy.

Conclusions:

• The developed NLP approach, combining named entity recognition and hybrid FM detection, achieved high effectiveness in FH extraction.
• The system's performance was comparable to top-performing systems in the 2019 N2C2 FH extraction challenge.
• The method shows promise for enhancing clinical decision-making through accessible family history data.