Clinical Report: AI Applications in Forecasting Glaucoma Progression

Overview

This systematic review analyzed 43 unique study cohorts involving over 202,000 subjects to evaluate artificial intelligence (AI) models predicting glaucoma progression. The findings highlight AI's promising predictive accuracy, challenges in model transparency, integration into clinical workflows, and the need for standardized reporting to facilitate clinical adoption.

Background

Glaucoma remains the leading cause of irreversible blindness worldwide, with an increasing number of diagnosed cases despite a decline in age-standardized prevalence. Early and accurate prediction of glaucoma progression is critical to preserving vision and guiding treatment intensity. Traditional risk factors lack sufficient predictive granularity, prompting interest in AI approaches, including machine learning and deep learning, which have shown advantages in complex clinical tasks such as image processing and time series prediction. Recent research has focused on leveraging AI to forecast glaucoma progression, though challenges remain in heterogeneity of definitions, model interpretability, and clinical implementation.

Data Highlights

A total of 43 unique study cohorts from six countries were included, encompassing over 202,207 subjects. The majority of studies originated from the United States (28 studies), followed by South Korea (8), Japan (3), China (2), Egypt (1), and Lithuania (1). The largest cohort included 39,090 patients, with a median cohort size of 1,231 subjects. A total of 46 reports were qualitatively synthesized to assess AI model performance and study characteristics.

Key Findings

• AI models have demonstrated promising accuracy in predicting glaucoma progression, approaching performance levels seen in diagnostic AI tasks.
• There is significant heterogeneity in study designs, definitions of progression, and AI methodologies, complicating direct comparisons.
• Challenges include limited transparency of AI model outputs, difficulties integrating AI tools into existing clinical workflows, and inconsistent reporting standards.
• Most studies originated from the United States, with fewer from other countries, indicating potential geographic research concentration.
• Large-scale cohorts and multi-center data improve model robustness but are not yet standard across studies.
• Regulatory approval and rigorous validation remain necessary steps before clinical implementation of AI progression prediction tools.

Clinical Implications

AI models hold potential to enhance personalized glaucoma management by enabling early identification of patients at high risk for rapid progression, thereby guiding intensified monitoring and treatment. However, clinicians should be cautious due to current limitations in model transparency and integration. Adoption will require standardized reporting, validation in diverse populations, and seamless incorporation into clinical workflows to ensure safety and efficacy.

Conclusion

AI-based forecasting of glaucoma progression shows considerable promise but faces challenges related to heterogeneity, interpretability, and clinical integration. Addressing these issues through standardized methodologies and validation will be essential for translating AI advances into improved patient outcomes.

References

1. Tham et al. 2014 -- Global Prevalence of Glaucoma and Projections
2. He et al. 2020 -- AI in Ophthalmology: Current Status and Future Perspectives
3. Medeiros et al. 2019 -- Predicting Glaucoma Progression Using AI
4. FDA 2018 -- Approval of AI Devices for Diabetic Retinopathy Detection