Improving The Readability Of Multilingual Online Pediatric Cataract Education Materials Using Large Language Models: A Comparative And Prompt-Driven Evaluation

To systematically evaluate the effectiveness of large language models (LLMs) in improving the readability of online patient education materials (PEMs) on pediatric cataracts through multilingual adaptation, web-based content retrieval, and prompt engineering.

Web-based analysis of publicly available online patient education materials.

A total of 103 pediatric cataract–related PEMs in multiple languages were collected from diverse online sources. Three advanced LLMs—ChatGPT-4o, Gemini 2.0, and DeepSeek-R1—were applied to revise and simplify the materials. Readability levels of original and LLM-processed PEMs were assessed using multiple established readability formulas. In addition, different prompt engineering strategies, including chain-of-thought–based approaches, were evaluated for their impact on readability optimization.

PEMs generated directly by LLMs exceeded a 10th-grade reading level. Compared with Google search results, PEMs retrieved via LLM web-browsing showed better content quality but remained difficult to read. After LLM-based conversion, Google-derived PEMs showed significant readability improvement across all models. DeepSeek-R1 achieved the greatest reduction, lowering the mean reading level from 10.59 ± 2.20 to 7.01 ± 0.91 (P < 0.001). Prompt engineering significantly affected outcomes, with Zero-shot Chain-of-Thought (APE) achieving target readability below the sixth-grade level. LLM-generated and converted Chinese PEMs consistently met recommended readability standards.

LLMs can substantially enhance the readability of multilingual online pediatric cataract PEMs. The integration of web-based retrieval and optimized prompt engineering further improves performance, highlighting a scalable strategy for patient-centered health communication.