Improving Visual Field Forecasting by Correcting for the Effects of Poor Visual Field Reliability

Area of Science:

• Ophthalmology
• Computational Science

Background:

• Visual field (VF) testing is crucial for monitoring glaucoma and other optic neuropathies.
• Accurate prediction of future VF mean deviation (MD) is essential for timely intervention.
• Current methods may not fully account for test reliability, potentially impacting predictive accuracy.

Purpose of the Study:

• To develop and evaluate methods for accurately forecasting future reliable VF MD values.
• To assess the impact of correcting for test unreliability on predictive model performance.

Main Methods:

• Four linear regression techniques (standard, unfiltered, corrected, weighted) were applied to VF data from 5939 eyes.
• Models were fitted using all VFs except the final one to predict the final VF MD.
• Model error was calculated by comparing predicted vs. actual final VF MD, with performance analyzed across reliability subgroups.

Main Results:

• The standard Humphrey Field Analyzer (HFA) method performed worst, with higher average residuals compared to other methods.
• The weighted regression method demonstrated the best performance, outperforming the standard method significantly in certain reliability subgroups.
• Despite its accuracy, the weighted method had a relatively large average 95% prediction interval (7.67 dB).

Conclusions:

• Incorporating all VFs, including unreliable ones, enhances the predictive power for future reliable VFs.
• Correcting for VF reliability demonstrably improves the accuracy of predictive models.
• These VF correction methods may enable earlier detection of VF worsening in clinical practice.