Enhancement of Visual Field Predictions with Pointwise Exponential Regression (PER) and Pointwise Linear Regression...

Area of Science:

• Ophthalmology
• Medical Technology
• Data Science

Background:

• Glaucoma management relies on predicting visual field (VF) worsening.
• Accurate forecasting of VF changes is crucial for timely treatment adjustments.
• Current prediction models may benefit from improved data processing techniques.

Purpose of the Study:

• To evaluate the efficacy of threshold smoothing algorithms in enhancing the prediction accuracy of glaucoma visual field (VF) progression.
• To compare different smoothing methods and regression models for forecasting VF changes.

Main Methods:

• The study included 798 primary open-angle glaucoma patients with at least 6 years of follow-up and 8 VF examinations.
• Threshold data from the initial follow-up period were smoothed using nearest neighbor (NN), Garway-Heath, Glaucoma Hemifield Test (GHT), and correlation-based methods.
• Smoothed and unsmoothed data were analyzed using pointwise exponential regression (PER) and pointwise linear regression (PLR) models.
• Model performance was assessed by comparing predicted versus observed thresholds using root mean square error (RMSE).

Main Results:

• Smoothed visual field (VF) data demonstrated superior predictive performance compared to unsmoothed data across both regression models.
• The nearest neighbor (NN) smoothing algorithm yielded the lowest RMSE values, indicating the most accurate predictions.
• Pointwise exponential regression (PER) consistently outperformed pointwise linear regression (PLR) in prediction accuracy.

Conclusions:

• Threshold smoothing algorithms significantly enhance the prediction of visual field (VF) worsening in glaucoma patients.
• Nearest neighbor (NN) smoothing combined with pointwise exponential regression (PER) provides the most accurate forecasting of VF changes.
• Implementing smoothing algorithms in VF analysis aids in more precise treatment decisions for glaucoma management.