Predictive modeling of glaucomatous optic neuropathy progression rate using machine learning methods

PURPOSE. To develop a method for individualized prediction of the rate of progression in moderate and advanced primary open-angle glaucoma (POAG). METHODS. The study included patients with confirmed moderate and advanced POAG who had been under observation for at least 36 months. Prediction of the rate of visual function loss was performed using modern machine learning methods, specifically Ranked PLS-DA, which is highly resistant to multicollinearity and allows for the ordered nature of classes to be taken into account. Two sets of input data were considered: a complete set of 34 variables and an optimized set of 20 variables, including demographic, functional, structural, and vascular indicators. For model optimization and validation, a test dataset was artificially generated using Procrustes Cross-Validation (PCV) method. Model performance was evaluated using specific metrics: sensitivity, specificity, total efficiency (TEFF), and area under the ROC curve (AUC). RESULTS. The optimized set of variables improved model sensitivity (0.93 vs 0.78) while maintaining high specificity (0.78). Total efficiency on the test dataset was 0.77 for the reduced set, with an AUC of 0.9. The model not only distinguished patients with fast, moderate, and slow progression rates but also identified «borderline» cases requiring closer monitoring. Analysis of individual variable contributions revealed key predictors influencing prognostic accuracy: age, retinal nerve fiber layer thickness and ganglion cell complex thickness, peripapillary vessel density, and parafoveal retinal thickness. These findings underscore the importance of a comprehensive approach to assessing the risk of irreversible changes in visual functions. CONCLUSION. The developed Ranked PLS-DA model demonstrated high effectiveness in stratifying patients with moderate and advanced glaucoma according to progression rate. The model may serve as a reliable basis for individualized monitoring and therapy in routine clinical practice.