Artificial Intelligence In Predicting Postoperative Outcomes After Anterior Segment And Ocular Adnexal Surgery: A Systematic Review
Published 2026 - 44th Congress of the ESCRS
Reference: POOTH003 | Type: Free paper | DOI: 10.82333/b3v5-5666
Authors: Dr Abdullah Al-Ani* 1 , Mayar Alkhawaja 2 , Lucy Yang 3 , David Mikhail 2 , Liam Connors 4 , Daniel Milad 5 , Vivian Hill 1 , Karim Punja 1 , Fiona Costello 1 , Ahmed Al-Ghoul 1 , Patrick Gooi 1
1University of Calgary,Calgary,Canada, 2University of Toronto,Toronto,Canada, 3University of Alberta,Edmonton,Canada, 4McMaster University,Collingwood,Canada, 5University of Montreal,Montreal,Canada
Purpose
Artificial intelligence (AI) is increasingly applied to support decision-making in ophthalmology. This systematic review evaluates the ability of AI models to predict postoperative outcomes following anterior segment and ocular adnexal surgical procedures and compares their performance with conventional statistical approaches when available.
Setting
Systematic review of published studies evaluating AI-based prediction models for anterior segment and ocular adnexal surgical outcomes.
Methods
A comprehensive search of MEDLINE, Embase, Cochrane Library, Compendex, Web of Science, Scopus, ProQuest Dissertations and Theses, IEEE Xplore, and grey literature was conducted to June 3, 2025. Studies evaluating AI models predicting postoperative outcomes in anterior segment or adnexal surgery were included; studies limited to intraocular lens power calculation or refractive outcomes were excluded. Risk of bias was assessed using PROBAST.
Results
Twenty-eight studies were included (n=2,871,898 eyes). Most procedures involved corneal and ocular surface surgery (50%), followed by glaucoma (25%) and cataract/oculoplastic surgery (25%). Tree-based algorithms (54%) and deep learning models (43%) predominated. Among 14 studies comparing AI with traditional methods, 13 (93%) reported superior AI performance. Reported AUCs ranged from 0.33 to 0.99. Seventy-one percent of studies had high risk of bias, and only 11% incorporated external validation.
Conclusions
AI models show promise in predicting postoperative outcomes following anterior segment and adnexal surgery and frequently outperform conventional statistical approaches. However, most studies demonstrate substantial methodological limitations, including a high risk of bias and limited external validation. Standardized reporting and robust external validation are required before widespread clinical integration.