Abstract / Summary
Resource constraints and low health literacy hinder the effectiveness of primary eye care screening. Here, we developed EyeSeek, a specialized large language model (LLM) to provide residents with personalized screening interpretations and guidance. We introduced a novel abstention-driven iterative learning framework to enhance reliability by detecting uncertainty, alongside a role-playing strategy for tailored communication. Findings demonstrated that our method can abstain and seek external answers when handling queries beyond its knowledge boundary, which reduces hallucination. In expert evaluation, EyeSeek outperformed several LLMs and primary care physicians in multiple dimensions. Readability analysis confirmed that EyeSeek adapted responses to Grade 3 ~ 8 levels, enabling more accessible communication for residents with varying educational backgrounds. Furthermore, we performed a single-center real-world prospective study comparing referral adherence between the EyeSeek-assisted group ( n = 84) and the unassisted group ( n = 86). The EyeSeek-assisted group demonstrated higher referral adherence ( P = 0.037) and improved health literacy compared with the unassisted group, with high user satisfaction. Given its multifaceted performance, EyeSeek holds promise as an adaptable digital solution to support primary eye care in community settings.