Opportunities for Improving Glaucoma Clinical Trials via Deep Learning-Based Identification of Patients with Low...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma clinical trials require substantial sample sizes due to patient variability.
• Identifying patients with low future visual field (VF) variability can optimize trial design.
• Deep learning models (DLMs) offer potential for predicting disease progression patterns.

Purpose of the Study:

• To develop and evaluate a DLM for forecasting eyes with low future VF variability.
• To assess the impact of using such a DLM on sample size requirements for neuroprotective trials.

Main Methods:

• A retrospective cohort study and simulation approach was used.
• DLMs were trained on baseline VF, OCT, and clinical data to predict low VF variability.
• Sample size estimations were performed for all eyes, low variability eyes, and DLM-predicted low variability eyes.

Main Results:

• DLMs achieved an AUC of 0.73 (DLM1) and 0.82 (DLM2) in forecasting low VF variability.
• Using DLM-predicted low variability eyes reduced sample size by 30-38% for detecting treatment effects.
• The models utilized baseline VF/OCT/clinical data to predict future variability.

Conclusions:

• DLMs can accurately forecast eyes with low VF variability from a single baseline visit.
• This predictive capability can significantly reduce sample size requirements for glaucoma trials.
• The findings suggest a potential to decrease the burden of future glaucoma clinical trials.