A SPATIALLY VARYING HIERARCHICAL RANDOM EFFECTS MODEL FOR LONGITUDINAL MACULAR STRUCTURAL DATA IN GLAUCOMA PATIENTS

Area of Science:

• Ophthalmology
• Biostatistics
• Medical Imaging

Background:

• Glaucoma monitoring relies on longitudinal macular thickness measurements.
• Current methods use simple linear regression, limiting precision.
• Disease progression can lead to irreversible vision loss.

Purpose of the Study:

• To develop a novel Bayesian hierarchical model for precise glaucoma progression monitoring.
• To improve the estimation of macular thickness changes over time.
• To reduce errors in predicting future vision loss.

Main Methods:

• A Bayesian hierarchical model with spatially varying coefficients was developed.
• Multivariate Gaussian process priors with Matérn cross-covariance functions were employed.
• Visit effects were incorporated to account for imaging-specific errors.

Main Results:

• The novel model demonstrated improved precision in estimating slopes of macular thickness.
• Including visit effects significantly reduced prediction errors for future measurements.
• The model showed a greatly improved fit compared to standard methods.

Conclusions:

• The proposed Bayesian model offers a more accurate approach to monitoring glaucoma progression.
• Accurate monitoring of macular thickness is crucial for preventing vision loss.
• The model's ability to incorporate visit effects enhances predictive accuracy.