A multimodal model of aqueous humor small RNA and clinical features shows limited discrimination between primary open-angle and pseudoexfoliative glaucoma

Abstract Glaucoma is a leading cause of irreversible blindness worldwide and encompasses diseases with different clinical presentations, including primary open-angle glaucoma (POAG) and pseudoexfoliative glaucoma (PXG). We investigated whether smallRNA profiles in aqueous humor, considered together with clinical data and fundus images, could provide candidate biomarkers for glaucoma in a machine-learning framework. This prospective, single-center, cross-sectional study included 70 participants: 31 with POAG, 19 with PXG, and 20 cataract controls. The leakage-controlled POAG-versus-PXG analysis comprised 50 independent participants. Under repeated cross-validation with all data-dependent steps confined to the training folds, the AUC was 0.555 (95% CI, 0.392–0.711) for the smallRNA-only model and 0.608 (95% CI, 0.452–0.764) for the model combining smallRNA with clinical and demographic variables. Thus, once information leakage was controlled, the models showed limited ability to distinguish POAG from PXG and did not clearly perform better than chance. SmallRNA profiles remain of biological interest, but this cohort does not establish their predictive usefulness.