Large language models for primary care ophthalmic education: a systematic review

Area of Science:

• Ophthalmology
• Medical Education
• Artificial Intelligence

Background:

• Primary care physicians (PCPs) require enhanced ophthalmology training and support for eye-health screening.
• Generative AI, specifically large language models (LLMs), presents a potential solution for improving PCPs' ophthalmic knowledge and skills.
• Current LLM applications in primary care ophthalmology lack comprehensive validation regarding educational effectiveness, reproducibility, and safety.

Purpose of the Study:

• To systematically review studies evaluating LLMs in ophthalmic education, training, assessment, and primary care clinical support.
• To identify the current evidence base, applications, limitations, and recommended safeguards for LLM use in this domain.

Main Methods:

• Systematic review of literature from PubMed, Web of Science, and Scopus (Jan 2020-Dec 2025).
• Search terms combined LLMs/generative AI, ophthalmology, and education/assessment.
• Citation chaining and independent data extraction by two reviewers.

Main Results:

• Evidence is largely based on vignette studies and controlled settings; real-world validation is scarce.
• LLMs can act as 'cognitive apprenticeship' tools for clinical reasoning, differential diagnosis, and referral decisions.
• Applications include virtual patient interviews and documentation support; multimodal models struggle with image interpretation. Limitations include heterogeneity, bias, and hallucination risks.

Conclusions:

• LLMs can enhance primary care ophthalmic education and support if task scopes are clear and safeguards are robust, augmenting but not replacing expert judgment.
• Future research should focus on pragmatic trials, implementation studies, and standardized evaluations to establish safe and effective LLM integration pathways.