Using large language models to generate child-friendly education materials on myopia.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Education

Background:

• Myopia management requires clear, accessible patient education for children and parents.
• Large language models (LLMs) offer a potential tool for generating such educational content.

Purpose of the Study:

• To evaluate the efficacy of LLMs in creating patient education materials for childhood myopia.
• To compare the performance of different LLMs and prompts in generating accurate and age-appropriate myopia information.

Main Methods:

• Thirty-five common myopia questions were used with two prompts (adult- and child-directed).
• Five ophthalmologists rated LLM responses on correctness, completeness, conciseness, and potential harm.
• Readability was assessed using Flesch-Kincaid scores.

Main Results:

• ChatGPT 4o received the highest ratings for correctness and conciseness across both prompts.
• ChatGPT 4o demonstrated superior completeness for child-directed responses.
• Child-directed prompts resulted in significantly more readable content for all evaluated LLMs.

Conclusions:

• LLMs, especially ChatGPT 4o, show promise for creating myopia patient education materials for diverse age groups.
• While effective, LLMs require further development for complex medical information and clinical application.