Artificial Intelligence for Predicting Postoperative Outcomes in Anterior Segment and Ocular Adnexa Surgery: A Systematic Review.

TOPIC: Artificial intelligence (AI) is increasingly applied to support decision-making in ophthalmology. This review evaluates the ability of AI to predict postoperative outcomes after anterior segment and ocular adnexal procedures.

CLINICAL RELEVANCE: The prognostic application of AI algorithms, capable of handling multimodal, complex data, and modeling nonlinear relationships, may enhance perioperative patient counseling, surgical planning, and outcomes.

METHODS

A comprehensive search was conducted across MEDLINE, Embase, the Cochrane Library, Compendex, Web of Science, Scopus, ProQuest Dissertations and Theses, and gray literature. Studies that investigated only intraocular lens power calculation and refractive outcomes were excluded. Methodological quality was assessed using the Prediction Model Risk of Bias Assessment Tool.

RESULTS

Twenty-eight studies were included (n = 2 871 898 eyes). Most procedures were corneal and ocular surface (50%), followed by glaucoma (25%) and cataract/oculoplastic procedures (25%). Tree-based algorithms (57%) dominated the representation layer, followed by deep learning algorithms (43%). Fourteen studies examined how AI performance compared with traditional prediction paradigms; in 13 (93%), AI outperformed traditional methods. Areas under the curve ranged from 0.33 to 0.99, reflecting substantial variation in outcome definitions and model design. The Prediction Model Risk of Bias Assessment Tool assessment indicated that 75% of studies had high risk of bias, primarily due to insufficient reporting of model transparency and calibration metrics. Only 11% of studies incorporated external validation.

CONCLUSIONS

Artificial intelligence models are promising in predicting anterior segment and adnexal ophthalmic surgical outcomes. While these models generally outperform conventional methods, methodological limitations persist. Future studies should incorporate external validation, standardized reporting, and assessment of workflow integration.

FINANCIAL DISCLOSURES: The author has no/the authors have no proprietary or commercial interest in any materials discussed in this article.