A Bayesian Hierarchical Longitudinal Model for Estimation of Central Visual Field Rates of Change in Glaucoma

Area of Science:

• Ophthalmology
• Medical Statistics
• Computational Biology

Background:

• Glaucoma monitoring relies on visual field (VF) testing.
• Low visual field sensitivities (≤19 dB) reduce test reliability.
• Accurate detection of glaucoma progression is crucial for timely intervention.

Purpose of the Study:

• To evaluate longitudinal variability in central 10-degree VF measurements.
• To assess the impact of baseline sensitivity on VF measurement reliability.
• To compare a Bayesian hierarchical model with simple linear regression for analyzing VF changes.

Main Methods:

• Included 124 glaucoma patients with central or advanced VF damage and >2 years follow-up.
• Utilized a Bayesian linear model to estimate pointwise change rates.
• Compared model performance using simulations with varying residual standard deviations (SD).

Main Results:

• Residual standard deviations were significantly higher for baseline sensitivities of 5-20 dB compared to ≥25 dB.
• The Bayesian model identified more significant negative slopes than simple linear regression, especially with low residual SD.
• Mean global slopes were -0.21 dB/year (Bayesian) and -0.36 dB/year (SLR).

Conclusions:

• High residual variability in VFs with baseline sensitivity of 5-20 dB significantly impedes glaucoma progression detection.
• Excluding points with sensitivity ≤20 dB from analytical models may enhance the efficiency of detecting perimetric progression.