Rates of Glaucoma Progression Derived from Linear Mixed Models Using Varied Random Effect Distributions

Area of Science:

• Ophthalmology
• Statistical modeling

Background:

• Glaucoma is a leading cause of irreversible vision loss.
• Accurate estimation of visual field loss progression is crucial for patient management.
• Conventional statistical models may not optimally capture the complex progression patterns in glaucoma.

Purpose of the Study:

• To compare the performance of linear mixed models (LMMs) with varying random effect distributions (Gaussian, Student's t, log-gamma) for estimating visual field loss rates in glaucoma.
• To evaluate the accuracy of these models in identifying glaucoma progression.

Main Methods:

• Retrospective analysis of standard automated perimetry (SAP) data from the Duke Glaucoma Registry.
• Modeling of visual field mean deviation (MD) values using ordinary least squares (OLS) and LMMs with different random effect distributions.
• Model fit assessed by Watanabe-Akaike information criterion (WAIC); simulation studies to evaluate predictive accuracy.

Main Results:

• The log-gamma (LG) distribution demonstrated superior model fit (lowest WAIC).
• LG models identified progression earlier and showed greater accuracy in predicting slopes compared to OLS.
• Gaussian models underestimated progression rates, particularly in fast and catastrophic progressors.

Conclusions:

• LMMs incorporating the LG distribution are more effective for estimating SAP MD loss rates in glaucoma patients.
• Utilizing the LG distribution enhances the accuracy of identifying high-risk glaucoma progressors, facilitating timely interventions to mitigate vision loss.