A comprehensive study of mobility functioning information in clinical notes: Entity hierarchy, corpus annotation, and...

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Rehabilitation Medicine

Background:

• Secondary use of Electronic Health Records (EHRs) traditionally focuses on diseases and drugs, overlooking patient function.
• Patient function is a critical health indicator, yet it remains underutilized in health status assessments.
• The World Health Organization's International Classification of Functioning, Disability and Health (ICF) provides a standard for describing function, but its application in EHRs is limited.

Purpose of the Study:

• To pioneer the first comprehensive analysis and identification of functioning concepts within the Mobility domain of the ICF.
• To develop and evaluate a high-performance machine learning model for recognizing mobility-related entities in clinical notes.

Main Methods:

• A hierarchical order of mobility-related entities (types, relations, attributes, values) was induced from physical therapy notes.
• A gold standard corpus of 14,281 nested entity mentions was manually curated by domain experts from 400 clinical notes.
• An Ensemble machine learning model for named entity recognition (NER) was trained and evaluated on the curated corpus.

Main Results:

• High inter-annotator agreement (92.3% F1-score, 96.6% Cohen's kappa) was achieved in corpus curation.
• The developed Ensemble NER model achieved an average F1-score of 84.90% for exact entity matching.
• This performance surpassed popular NER methods like CRF, RNN, and BERT.

Conclusions:

• Mobility functioning information can be reliably extracted from clinical notes using advanced sequence labeling methods.
• The study demonstrates the feasibility of identifying ICF functioning concepts in EHR data.
• This approach can be extended to identify functioning concepts in other ICF domains.