Bridging the Gap in Health Literacy: Harnessing the Power of Large Language Models to Generate Plain Language Summaries from Biomedical Texts

Abstract

AbstractHealth literacy is essential for individuals to navigate the healthcare system and make informed decisions about their health. Low health literacy levels have been associated with negative health outcomes, particularly among older populations and those financially restricted or with lower educational attainment. Plain language summaries (PLS) are an effective tool to bridge the gap in health literacy by simplifying content found in biomedical and clinical documents, in turn, allowing the general audience to truly understand health-related documentation. However, translating biomedical texts to PLS is time-consuming and challenging, for which they are rarely accessible by those who need them. We assessed the performance of Natural Language Processing (NLP) for systematizing plain language identification and Large Language Models (LLMs), Generative Pre-trained Transformer (GPT) 3.5 and GPT 4, for automating PLS generation from biomedical texts. The classification model achieved high precision (97·2%) in identifying if a text is written in plain language. GPT 4, a state-of-the-art LLM, successfully generated PLS that were semantically equivalent to those generated by domain experts and which were rated high in accuracy, readability, completeness, and usefulness. Our findings demonstrate the value of using LLMs and NLP to translate biomedical texts into plain language summaries, and their potential to be used as a supporting tool for healthcare stakeholders to empower patients and the general audience to understand healthcare information and make informed healthcare decisions.