A new systematic review published in npj Digital Medicine suggests that artificial intelligence could play a significant role in predicting glaucoma progression, potentially helping physicians identify patients at highest risk of vision loss earlier and more accurately. However, researchers caution that barriers remain before AI tools can be reliably implemented in routine clinical care.

The study analyzed 43 unique studies evaluating AI models designed to forecast glaucoma conversion, disease deterioration, and the need for surgery. According to the review, many AI systems demonstrated moderate-to-good predictive performance across several clinical endpoints, raising optimism that machine learning could help ophthalmologists personalize treatment strategies and identify high-risk patients sooner.¹

The authors reviewed studies published since 2014 across MEDLINE, Embase, Web of Science, Cochrane CENTRAL, and arXiv databases. Two independent reviewers screened eligible studies and assessed risk of bias using the QUADAS-2 evaluation framework.

Despite encouraging findings, the researchers identified substantial weaknesses in the current body of evidence. Among the most significant concerns were inconsistent reporting standards, heterogeneity in study design, limited transparency in AI development, and poor generalizability across patient populations and healthcare settings.

The study noted that many models were trained and validated using narrow datasets, potentially limiting their ability to perform reliably in real-world clinical environments. The authors warned that without standardized methodologies and broader validation, AI systems risk underperforming when applied across diverse patient populations.

To address these shortcomings, the researchers proposed what they describe as the first glaucoma-specific set of recommendations aimed at improving study quality and accelerating clinical translation. The proposed framework emphasizes robust study design, transparent reporting, external validation, and improved interpretability of AI algorithms.

The authors argue that stronger reporting and validation standards will be critical if AI-based glaucoma prediction systems are to gain physician trust and regulatory acceptance.

Reference

1. Liang YG, Fan L, Teixeira-Pinto A, Liew G, White AJR. A systematic review of AI for predicting glaucoma progression: challenges and recommendations towards clinical implementation. npj Digit Med. 2026;9:Article 477. doi:10.1038/s41746-025-02321-7