- My Library
- 2026 Issues
- Issue 100
- Sticks and stones: bias and readability…
Sticks and stones: bias and readability assessment in large language model–generated patient education for anterior cruciate injury
Key Points
- Large language models generated ACL patient education materials demonstrated minimal sex, gender, ethnicity, or socioeconomic status related bias but exceeded recommended readability levels.
BACKGROUND & OBJECTIVE
Anterior cruciate ligament (ACL) injuries remain common in adolescent and sporting populations and are associated with significant physical and psychological consequences. Alongside this, increased wait times and financial cost of healthcare could drive patients and their families to increasingly seek health-related information online before consulting a clinician, with large language models (LLMs) emerging as a novel source of patient education.
While these tools may improve access to information, concerns exist regarding bias, readability, tone, and completeness of AI-generated patient education materials (PEMs). Previous work has shown that online PEMs frequently exceed recommended reading levels and may negatively influence health literacy and clinical engagement (1,2).
The primary aim of this study was to examine sex, gender, ethnicity, and socioeconomic status (SES) related bias in LLM generated ACL patient education. A secondary aim was to assess readability, understandability, actionability, accuracy, and emotional tone of these materials.
This reinforces the importance of early, clinician led communication that provides clear education on injury expectations, prepares patients for challenges and setbacks, and support throughout the rehabilitation process.
METHODS
- This cross-sectional study evaluated ACL PEMs generated by four widely used, free-access LLMs: ChatGPT-4o, Google Gemini 1.5, Claude 3.5 Sonnet, and Llama 3.2.