Multicenter Validation of Metabolomic Fingerprints for Accurate Diagnosis, Subtyping, and Severity Stratification of Glaucoma

Abstract

Timely and accurate diagnosis of primary glaucoma, along with reliable subtype and severity stratification, remains a major clinical challenge. Here, we develop a serum-based metabolomic fingerprint strategy that leverages flower-like hierarchical metal oxide heterojunctions as the matrix for laser desorption/ionization mass spectrometry, combined with a neural network algorithm. A total of 591 serum samples from two independent hospital cohorts were analyzed. In the internal test set, the model achieved exceptionally high diagnostic performance, with accuracy, F1 score, precision, and recall all reaching 1.000. External validation further confirmed its robustness, with an area under the curve (AUC) value of 1.000 and classification accuracy, F1 score, and recall each at 0.990. Subtype classification for primary angle-closure glaucoma (PACG) achieved an accuracy of 97.6%. Severity assessment of severe glaucoma showed strong performance, with an AUC of 0.990 and accuracy of 0.831. These results support the applicability of the proposed approach for precise glaucoma diagnosis and longitudinal monitoring across multicenter clinical cohorts.