Proactive Decision Support for Glaucoma Treatment: Predicting Surgical Interventions with Clinically Available Data.

Area of Science:

• Ophthalmology
• Medical Informatics
• Artificial Intelligence

Background:

• Glaucoma management requires timely surgical intervention decisions.
• Predicting the need for glaucoma surgery is crucial for patient outcomes.
• Current prediction methods may not fully leverage multi-modal data.

Purpose of the Study:

• To develop and evaluate multi-modal machine learning models for predicting glaucoma surgical interventions.
• To assess the accuracy of these models in forecasting surgical needs at different time points.
• To explore the utility of integrating diverse patient data for enhanced predictive capabilities.

Main Methods:

• Utilized a longitudinal ophthalmic dataset including 369 surgical and 592 non-surgical glaucoma patients.
• Incorporated multi-modal data: demographics, medical history, clinical measurements, optical coherence tomography (OCT), and visual field (VF) testing.
• Trained and validated machine learning models on independent datasets from a separate study site.

Main Results:

• Models achieved high predictive accuracy, with Area Under the Curve (AUC) of 0.93, 0.92, and 0.93 for 1, 2, and 3 years prior to surgery, respectively.
• Demonstrated strong performance with high sensitivity (up to 0.89) and specificity (up to 0.91) at 0.80 precision.
• Successfully predicted surgical interventions up to three years before the procedure.

Conclusions:

• Multi-modal machine learning models can accurately predict the need for glaucoma surgery well in advance.
• Integrating diverse data sources significantly enhances predictive power for glaucoma intervention.
• These models represent a promising tool for proactive management and timely surgical planning in glaucoma care.