Can Artificial Intelligence Replace The Ophthalmologist In Patient Education? A Comparative Analysis In Blepharitis
Published 2026 - 44th Congress of the ESCRS
Reference: POCOR429 | Type: Free paper | DOI: 10.82333/312q-r578
Authors: Dr Daniel Bahir* 1 , A/Prof. Audrey Rostov 2 , Dr Yumna Busool Abu Eta 1 , Dr Shirin Hamed Azzam 1 , A/Prof. David Lockington 3 , A/Prof. Joshua C Teichman 4 , A/Prof. Artemis Matsou 5 , A/Prof. Clara C Chan 4 , Dr Elad Shvartz 6 , A/Prof. Michael Mimouni 7
1Ophthalmology,Tzafon Medical Center,Poriya,Israel, 2Bellevue Precision Vision,Bellevue, WA,United States, 3Ophthalmology,Gartnavel General Hospital,Glasgow,United Kingdom, 4Department of Ophthalmology and Vision Sciences, University of Toronto,Toronto,Canada, 5Ophthalmology,Queen Victoria Hospital,East Grinstead,United Kingdom, 6zrieli Faculty of Medicine, Bar Ilan University,Safed,Israel, 7Ophthalmology,Rambam Health Care Campus,Haifa,Israel
Purpose
To evaluate whether widely used artificial intelligence chatbots can match expert ophthalmologists in delivering accurate and clinically appropriate patient education in blepharitis, and to assess their potential role in real-world anterior segment practice.
Setting
International, multicenter expert-based comparative study.
Methods
Thirteen clinically relevant blepharitis patient questions were developed through a structured consensus process by anterior segment specialists. Responses were generated by GPT-based models and Gemini-based models and compared with responses from expert ophthalmologists. All responses were anonymized and independently graded by a blinded expert panel using a 7-point Likert scale assessing correctness and clinical educational value. Non-parametric statistical analyses were performed to compare performance between groups.
Results
Experts achieved the highest mean scores for correctness (6.3) and clinical education value (6.4), significantly outperforming all AI models (P < 0.01).
Expert ophthalmologists achieved the highest scores for both correctness and clinical educational value, significantly outperforming all AI models (P < 0.01). AI systems demonstrated strong performance in definition-based and straightforward informational questions but showed reduced accuracy and contextual adaptation in management-oriented and scenario-based queries. Variability in clinical nuance and individualized recommendations was observed across AI-generated responses.
Conclusions
Large language model–based AI systems demonstrate promising capabilities in basic ophthalmic patient education but remain inconsistent in complex, context-dependent clinical scenarios. While AI may serve as an adjunctive educational tool, it cannot currently replace physician-guided counseling in anterior segment practice. Given the increasing reliance of patients on AI-generated medical information prior to consultation, careful integration with human oversight is essential.