This retrospective study evaluated how well ChatGPT-5 could predict long-term anatomical closure and visual outcomes after full-thickness macular hole (FTMH) surgery, comparing its predictions with those of senior retinal specialists and actual 12-month results in 50 eyes. ChatGPT showed higher apparent accuracy than specialists in predicting anatomical closure (90% vs. 72–86%), but it tended to overestimate closure and visual improvement, reflecting an optimism bias. For functional outcomes (visual acuity), ChatGPT’s predictions were more accurate than specialists’ but still imprecise, especially for stable or worsened vision. The authors conclude that while generative AI shows promise in outcome prediction, it currently lacks clinical judgment and needs further validation before routine clinical use.

Full Article