PREDICTING POSTOPERATIVE OUTCOMES IN FULL-THICKNESS MACULAR HOLE REPAIR SURGERY: ChatGPT Versus Clinical Decision

Area of Science:

• Ophthalmology
• Artificial Intelligence in Medicine
• Surgical Outcomes

Background:

• Full-thickness macular hole (FTMH) surgery aims to restore vision by closing the hole.
• Accurate prediction of surgical outcomes is crucial for patient management and expectations.
• The role of artificial intelligence (AI) in predicting ophthalmic surgical results is an emerging area.

Purpose of the Study:

• To assess ChatGPT-5's ability to predict long-term anatomical and functional outcomes after FTMH surgery.
• To compare ChatGPT-5's predictive performance against senior retina specialists.
• To evaluate the real-world accuracy of AI predictions versus actual patient results.

Main Methods:

• Retrospective analysis of 50 eyes undergoing pars plana vitrectomy for FTMH.
• Inputting de-identified clinical data and OCT images into ChatGPT-5 for outcome prediction.
• Comparing AI predictions with actual 12-month outcomes and specialist assessments.

Main Results:

• Anatomical prediction accuracy: 90% for ChatGPT-5 vs. 72-86% for specialists.
• Functional outcome prediction accuracy: 66% for ChatGPT-5 vs. 42-44% for specialists.
• ChatGPT-5 showed optimism bias, overestimating closure and visual recovery, particularly in non-closure cases.

Conclusions:

• ChatGPT-5 demonstrates high apparent accuracy in predicting FTMH surgery outcomes.
• The AI model's predictions are influenced by optimism bias, overestimating positive results.
• Clinical judgment is lacking; larger prospective studies are required prior to potential clinical use.