Bayesian hierarchical modeling of longitudinal glaucomatous visual fields using a two-stage approach

Area of Science:

• Ophthalmology
• Biostatistics
• Computational Statistics

Background:

• Glaucoma is a leading cause of global blindness, necessitating early detection of visual field deterioration.
• Accurate modeling of visual field progression is crucial for optimizing treatment strategies and preventing vision loss.
• Traditional one-stage Bayesian methods using Markov chain Monte Carlo (MCMC) sampling face computational limitations with large, complex datasets.

Purpose of the Study:

• To address computational challenges in analyzing longitudinal glaucoma data using a two-stage Bayesian approach.
• To develop and extend a model for predicting visual field (VF) progression in glaucoma patients.
• To enable early detection of deterioration and inform optimized treatment strategies.

Main Methods:

• Application and extension of a recently proposed two-stage Bayesian approach for hierarchical longitudinal data.
• Modeling of glaucoma patient data, accounting for random effects, censoring, and high measurement variability.
• Incorporation of tools for model evaluation within the extended two-stage framework.

Main Results:

• Successfully applied the two-stage Bayesian approach to overcome computational difficulties encountered with a one-stage method.
• Developed a model capable of predicting visual field progression in longitudinal glaucoma data.
• Enhanced the two-stage approach with integrated model evaluation tools.

Conclusions:

• The two-stage Bayesian approach provides a computationally feasible solution for complex hierarchical longitudinal models in ophthalmology.
• This method facilitates accurate prediction of visual field progression, aiding in early glaucoma detection and management.
• The extended approach offers robust tools for model evaluation, improving the reliability of glaucoma data analysis.