Clinical Note Structural Knowledge Improves Word Sense Disambiguation.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Documentation

Background:

• Clinical notes contain numerous ambiguous medical abbreviations requiring disambiguation.
• Existing learning-based approaches leverage contextual knowledge but underutilize clinical note structures.
• Previous methods for structural analysis of notes lacked semantic meaningfulness.

Purpose of the Study:

• To develop and evaluate a novel approach for medical abbreviation disambiguation using semantically rich representations of clinical note structures.
• To investigate the utility of semantic types from the Unified Medical Language System (UMLS) for enhancing abbreviation sense disambiguation.

Main Methods:

• Representing clinical note structure using semantic types from the Unified Medical Language System (UMLS).
• Integrating these semantic representations as features alongside N-grams for abbreviation disambiguation.
• Evaluating the approach using three machine learning models on two distinct datasets.

Main Results:

• The proposed feature augmentation using UMLS semantic types consistently improved model performance for abbreviation disambiguation.
• The optimal F1 score achieved was 0.93, demonstrating significant enhancement over baseline methods.
• The study confirms the value of semantically meaningful structural information in clinical notes.

Conclusions:

• Representing clinical note structure with UMLS semantic types is an effective strategy for improving medical abbreviation disambiguation.
• This method offers a more semantically grounded approach compared to previous structural analyses.
• The findings support the integration of structured clinical information into natural language processing models for healthcare applications.