Related Experiment Video
Updated: Jun 8, 2026
A Protocol for Comprehensive Assessment of Bulbar Dysfunction in Amyotrophic Lateral Sclerosis ALS
Published on: February 21, 2011
Improving the Readability of Spasmodic Dysphonia Patient Education Materials Using ChatGPT-4o Mini: A Cross-Sectional
Kelly Clevenger1, Bonnie K Slavych2, Richard I Zraick1
1School of Communication Sciences and Disorders, College of Health Professions and Sciences, University of Central Florida, Orlando, Florida.
Purpose:
Online patient education materials (PEMs) for spasmodic dysphonia (SD) often exceed recommended readability levels, creating barriers to patient comprehension. This cross-sectional study examined (a) the current readability of SD-related PEMs, (b) whether the artificial intelligence (AI) tool ChatGPT-4o mini improves readability when instructed to write at a sixth-grade level, and (c) the extent to which essential clinical information is preserved after AI-based revision.
Method:
Fourteen SD-related websites were identified through a systematic Google search, yielding 38 PEMs addressing three core domains: What is SD?, What causes SD?, and How is SD treated? For each PEM, corresponding sixth-grade level responses were generated using ChatGPT-4o mini. Original and AI-revised texts were analyzed using four established readability indices (Flesch-Kincaid Grade Level, Flesch Reading Ease, Gunning Fog Index, and SMOG). Paired-samples t-tests assessed differences in readability. A structured content fidelity analysis was then performed at the statement level to evaluate specificity reduction (SR), dropped content (DC), added content (AC), and overall fidelity using F/P/X classification thresholds.
Results:
Original PEMs required reading levels ranging from approximately 10th grade to early college. ChatGPT-4o mini significantly improved readability across all indices (P <0.001; large effect sizes). Despite readability gains, fidelity analysis of 407 statement pairs revealed notable information loss: 57% of changes were DC, 29% were SR, and 5% were AC. Overall, 21% of PEMs were fully preserved (F), 50% were partially preserved (P), and 29% were not preserved (X), primarily due to omission of clinical detail or reduced specificity.
Conclusion:
ChatGPT-4o mini substantially improves the readability of SD-related PEMs but often at the cost of reduced clinical specificity or loss of essential information. AI tools may assist in producing accessible drafts, but expert oversight remains necessary to ensure accuracy and completeness. Future research should examine how AI-revised materials affect patient comprehension and decision-making.
