MedJEx: A Medical Jargon Extraction Model with Wiki's Hyperlink Span and Contextualized Masked Language Model Score

Area of Science:

• Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• Electronic Health Records (EHRs) contain complex medical jargon that can hinder patient comprehension.
• Identifying and simplifying this jargon is crucial for effective patient-provider communication and health literacy.

Purpose of the Study:

• To develop and evaluate a novel Natural Language Processing (NLP) application, MedJEx, for identifying potentially difficult medical jargon within EHR notes.
• To introduce MedJ, a new, publicly available dataset of expert-annotated medical jargon from EHR notes.

Main Methods:

• Creation of the MedJ dataset comprising expert-annotated medical jargon from over 18,000 EHR note sentences.
• Development of the MedJEx model, incorporating training on an auxiliary Wikipedia hyperlink span dataset and fine-tuning on MedJ data.
• Utilizing a contextualized masked language model score to enhance jargon detection.

Main Results:

• The MedJEx model demonstrated superior performance compared to existing state-of-the-art NLP models in medical jargon extraction.
• Training on auxiliary Wikipedia hyperlink span datasets improved the performance of MedJEx and positively impacted six out of eight biomedical named entity recognition benchmark datasets.
• A contextualized masked language model score proved beneficial for identifying domain-specific unfamiliar jargon terms.

Conclusions:

• The developed MedJEx model and MedJ dataset offer a significant advancement in automatically identifying patient-incomprehensible medical jargon from EHRs.
• Public availability of both MedJ and MedJEx facilitates further research and development in clinical NLP and patient communication tools.
• The findings highlight the effectiveness of leveraging external knowledge sources like Wikipedia for enhancing specialized NLP tasks.