Improving Layman Readability of Clinical Narratives with Unsupervised Synonym Replacement

Area of Science:

• Natural Language Processing
• Medical Informatics
• Machine Learning

Background:

• Clinical narratives contain complex medical terminology that can be difficult for patients to understand.
• Patient comprehension of medical information is crucial for adherence and health outcomes.
• Existing tools for simplifying medical text are limited in scope and effectiveness.

Purpose of the Study:

• To develop and evaluate a prototype tool that assists laypersons in comprehending clinical narratives.
• To leverage unsupervised machine learning for identifying and replacing difficult medical terms with simpler alternatives.

Main Methods:

• Utilized unsupervised machine learning on two large unlabeled text corpora: a clinical corpus and a general domain corpus.
• Created a joint semantic word-space model to extract easier-to-understand word alternatives.
• Evaluated the tool's performance using two domain experts and calculated inter-rater agreement.

Main Results:

• The prototype tool successfully suggested acceptable layperson alternatives for 55.51% of difficult words when providing ten suggestions per word.
• Domain expert evaluation indicated the tool's potential for improving patient understanding of clinical documents.

Conclusions:

• The developed tool shows promise in simplifying clinical narratives for patients.
• Future work will incorporate supervised machine learning and further manual evaluation to enhance the tool's performance and accuracy.