GPT-4 Enhances Readability of Online Patient Materials for Cardiac Amyloidosis

BackgroundCardiac amyloidosis (CA) is a rare, progressive condition that requires complex management, highlighting the importance of accessible and understandable patient education materials (PEMs). The American Medical Association (AMA) recommends that PEMs be written at or below a 6th-grade reading level; however, PEMs for CA often exceed this recommendation. GPT-4, a large language model (LLM), is increasingly being studied for its ability to enhance PEM readability. Materials and MethodsInstitutional PEMs were sourced from the websites of the top 10 institutions ranked by the 2022-2023 US News & World Report for "Best Hospitals for Cardiology, Heart & Vascular Surgery." GPT-4 (version updated 20 July 2023) was prompted with "please explain the following in simpler terms" along with each institutional PEM to produce revised responses. The readability of both the institutional and GPT-4 revised PEMs was evaluated using validated readability formulas through readable.com. Finally, a board-certified cardiologist reviewed institutional and revised PEMs to assess for changes in accuracy and comprehensiveness. ResultsA total of 86 PEMs were analyzed. None of the institutional PEMs met the recommended 6th-grade reading level and had a median Flesch-Kincaid Grade Level (FKGL) of 10.9 (high school freshman; IQR: 9.2, 12.6, p<0.001). In comparison, GPT-4 revised PEMs reduced the median FKGL to 7.8 (seventh grade; IQR: 7.0, 8.8, p<0.001), with 14/86 (16.3%) achieving at least a 6th-grade reading level. All GPT-4 revised PEMs 86/86 (100%) were accurate after review, with 9/86 (10.5%) deemed more comprehensive and 3/86 (3.5%) deemed less comprehensive than the institutional materials. ConclusionsGPT-4 significantly improved the readability of institutional PEMs for CA while maintaining accuracy and, in some cases, enhancing comprehensiveness. These findings underscore the potential of LLMs to bridge health literacy gaps by simplifying complex medical information without compromising content integrity. However, further research is needed to validate our findings and assess patient comprehension, real-world efficacy, and the impact of AI-driven education on clinical outcomes.