README: Bridging Medical Jargon and Lay Understanding for Patient Education through Data-Centric NLP.

Area of Science:

• Natural Language Processing
• Health Informatics
• Patient Education

Background:

• Healthcare is increasingly patient-centric, emphasizing self-care and education.
• Electronic Health Records (EHR) contain complex medical jargon hindering patient understanding.
• Bridging the medical knowledge gap is crucial for effective patient-centered care.

Purpose of the Study:

• To develop an automated system for generating lay definitions of medical terms.
• To simplify complex medical information within EHRs for improved patient comprehension.
• To advance patient-centric healthcare through accessible medical language.

Main Methods:

• Created the README dataset: 50,000+ medical term/lay definition pairs with expert annotations.
• Engineered a Human-AI pipeline for data filtering, augmentation, and selection to enhance data quality.
• Utilized Retrieval-Augmented Generation with fine-tuned open-source models trained on the README dataset.

Main Results:

• Open-source models fine-tuned on high-quality data achieve performance comparable to or better than state-of-the-art closed-source models.
• Demonstrated the effectiveness of the README dataset and Human-AI pipeline in improving model output quality.
• Reduced hallucinations and enhanced the accuracy of generated lay definitions.

Conclusions:

• Automated generation of lay definitions is feasible and effective for simplifying medical jargon.
• High-quality, curated datasets enable open-source models to compete with advanced proprietary LLMs.
• This work significantly contributes to closing the patient knowledge gap and supporting patient-centered healthcare.