AI-GUIDED ENDPOINT SELECTION FOR NEUROPROTECTION TRIALS IN GLAUCOMA.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Standard Automated Perimetry (SAP) is crucial for monitoring glaucoma progression.
• FDA acceptance of SAP as a trial endpoint requires stringent criteria for progression.
• Identifying high-risk visual field locations for clinical trials remains a challenge.

Purpose of the Study:

• To develop an attention-based graph neural network (GNN) to predict visual field points most likely to deteriorate (High-5).
• To validate the GNN model's ability to identify progression and improve clinical trial efficiency.

Main Methods:

• Developed an attention-based GNN model using baseline SAP data.
• Trained and validated the model across three large datasets: BPOR, DGR, and UWHVF.
• Compared the predictive performance of High-5 locations against Mean Deviation (MD) and Low-5 points.

Main Results:

• High-5 points showed significantly faster deterioration rates in progressors compared to Low-5 and MD across all cohorts.
• The GNN model demonstrated superior discrimination of progressors from non-progressors (AUC 0.883-0.937).
• High-5 identified progression in nearly all cases, leading to an estimated 42% reduction in required trial size.

Conclusions:

• The GNN-based framework enables data-driven identification of high-risk SAP locations.
• This approach aligns with regulatory definitions of progression and enhances trial sensitivity.
• The High-5 method substantially improves the efficiency and power of glaucoma clinical trials.