How useful is population data for informing visual field progression rate estimation?

Area of Science:

• Ophthalmology
• Biostatistics
• Medical Imaging

Background:

• Bayesian estimators utilize population-level data (prior distributions) to refine individual rate estimates.
• Glaucoma progression is influenced by various risk factors, including intraocular pressure (IOP) management.
• Understanding the impact of risk factors on prior distributions is crucial for optimizing Bayesian methods.

Purpose of the Study:

• To evaluate the benefit of incorporating IOP treatment status into Bayesian prior distributions for visual field progression.
• To assess the potential gains from developing priors that account for other significant glaucoma risk factors.

Main Methods:

• Prior distributions were created using published data from treated or untreated glaucoma patients.
• Simulated visual field data (mean deviation values) were generated based on true progression rates.
• Rates were estimated using linear regression and Bayesian methods, comparing matched-prior and unmatched-prior conditions.

Main Results:

• For short visual field series, the matched-prior Bayesian approach showed a median error of 0.02 dB/y, compared to 0.20 dB/y for the unmatched-prior and 0.00 dB/y for linear regression.
• Positive predictive values for identifying rapid progression (<-1 dB/y) were 0.46 (matched-prior), 0.42 (unmatched-prior), and 0.38 (regression).
• Negative predictive values were high across all methods (0.93-0.95). Differences diminished with longer data series.

Conclusions:

• The performance enhancement of Bayesian estimators for visual field progression is modest, even with priors that include major risk factors like IOP treatment.
• The study suggests limited benefits from developing priors for other risk factors, indicating a plateau in performance gains.