Machine learning model to diagnose PCG.

Purpose
To explore the relationship between oxidative stress biomarkers and primary congenital glaucoma (PCG).
Methods
This case-control study included 40 PCG patients and 38 matched controls. Serum total antioxidant status (TAS), as well as superoxide dismutase (SOD), malondialdehyde (MDA), reactive oxygen species (ROS) and hydrogen peroxide (H2O2) levels were measured, along with eye and body exams. Logistic regression analysis was performed for PCG risk factors and machine learning model biomarker diagnosis.
Results
In the PCG group, H2O2 and MDA levels were notably higher than in controls (p < 0.001, p = 0.020), while TAS levels were significantly lower (p = 0.043). Adjusting for age and gender, the serum TAS (OR = 0.07, 95% CI 0.01–0.85, p = 0.037), H2O2 (OR = 1.21, 95% CI 1.09–1.35, p = 0.001) and MDA (OR = 1.17, 95% CI 1.00–1.34, p = 0.034) were determined to be independent risk/protective factors for PCG. Pearson analysis revealed significant negative correlations: SOD with anterior chamber depth (r = −0.445, p = 0.012) and H2O2 with mean deviation values for the visual field (r = −0.412, p = 0.041). Positive correlations were also significant: MDA with axial length (AL (r = 0.576, p = 0.002). The XGBoost or KNN model using TAS alone achieved the highest AUC (0.74) in five-fold cross-validation.
Conclusion
The decrease in TAS levels and the increase in H2O2 and MDA levels are found to be correlated with PCG, and the results indicate that oxidative stress plays a part in congenital glaucoma onset.