Enhancing Patient Education on Cardiovascular Rehabilitation with Large Language Models

Abstract

Introduction: There are barriers that exist for individuals to adhere to cardiovascular rehabilitation programs. A key driver to patient adherence is appropriately educating patients. A growing education tool is using large language models to answer patient questions.

Methods: The primary objective of this study was to evaluate the readability quality of educational responses provided by large language models for questions regarding cardiac rehabilitation using Gunning Fog, Flesh Kincaid, and Flesch Reading Ease scores.

Results: The findings of this study demonstrate that the mean Gunning Fog, Flesch Kincaid, and Flesch Reading Ease scores do not meet US grade reading level recommendations across three models: ChatGPT 3.5, Copilot, and Gemini. The Gemini and Copilot models demonstrated greater ease of readability compared to ChatGPT 3.5.

Conclusions: Large language models could serve as educational tools on cardiovascular rehabilitation, but there remains a need to improve the text readability for these to effectively educate patients.