Machine Learning-Derived Baseline Visual Field Patterns Predict Future Glaucoma Onset in the Ocular Hypertension...

Area of Science:

• Ophthalmology
• Machine Learning
• Data Science

Background:

• The Ocular Hypertension Treatment Study (OHTS) identified risk factors for primary open-angle glaucoma (POAG), such as pattern standard deviation (PSD).
• Archetypal analysis, an unsupervised machine learning technique, presents a novel method for risk stratification by uncovering patterns in baseline visual fields (VFs).

Purpose of the Study:

• To explore the utility of archetypal analysis in identifying patterns in baseline visual fields (VFs) for risk stratification of primary open-angle glaucoma (POAG) in patients with ocular hypertension.
• To compare the predictive performance of an archetypal analysis-based model against traditional risk factors like PSD.

Main Methods:

• Archetypal analysis was applied to 8494 baseline 24-2 VFs from the OHTS dataset.
• A penalized Cox proportional hazards model identified discriminative archetypes (ATs) associated with POAG onset and VF progression.
• The AT-based model's predictive accuracy (C-index) was evaluated and compared to the established OHTS model.

Main Results:

• An optimal model identified 19 archetypes (ATs), with AT9, AT11, and AT7 being the most prevalent.
• The AT-based model achieved a C-index of 0.75 for POAG onset prediction.
• Specific ATs (AT5, AT8, AT15, AT17) were significantly associated with increased POAG risk, while AT5, AT10, and AT14 correlated with rapid VF progression. PSD significance diminished in the presence of high-risk ATs.

Conclusions:

• Baseline VFs contain subtle patterns predictive of POAG onset and progression in ocular hypertension patients.
• Archetypal analysis provides interpretable, glaucoma-specific insights, potentially enhancing clinical risk assessment.
• The predictive value of PSD for POAG is influenced by the presence of specific high-risk archetypal patterns at baseline.