Multimodal Prediction of Primary Open-Angle Glaucoma Using Polygenic Risk Scores and Clinical Features in a High-Risk African Ancestry Cohort

SUMMARY We developed four polygenic risk scores (PGS) for primary open-angle glaucoma (POAG), which is the leading cause of irreversible blindness worldwide and remains undiagnosed in over half of patients. We constructed two genome-wide PGS using genome wide association study from African ancestry subjects: 1) the Primary Open-Angle African Ancestry Glaucoma Genetics (POAAGG) study (N = 7,031; POAAGG PGS) and an African ancestry GWAS (N = 11,275; MEGA PGS). We also derived two selected loci PGS from six multi-ancestry glaucoma GWAS and weighted these scores using African ancestry effect sizes (PGS616 and PGS526). In an independent training cohort (N = 271), the curated loci-based score PGS526 demonstrated the strongest standalone performance (mean AUC = 0.668), outperforming the genome-wide PGS constructed using PRS-CS. Integration with baseline demographic features (age and gender) further improved prediction, with the base + PGS616 model achieving a peak AUC of 0.806 with support vector machine model. Clinical enrichment in an independent suspect cohort (N = 1,013) showed that higher predicted genetic risk was significantly associated with elevated intraocular pressure, larger cup-to-disc ratio, and thinner retinal nerve fiber layer, which are all POAG diagnostic features. Leveraging inter-eye asymmetry, PGS further enhanced early disease discrimination, improving AUC from 0.823 to 0.862 for ΔIOP, from 0.769 to 0.817 for ΔCDR, and from 0.790 to 0.831 for ΔRNFL. These results demonstrate that PGS enhances prediction in phenotype-rich settings and enables accurate risk stratification in deep phenotype-limited cohorts, supporting earlier glaucoma detection. GRAPHICAL ABSTRACT HIGHLIGHTS Ancestry-matched polygenic scores improve POAG risk prediction in African ancestry populations Curated loci-based PGS outperform genome-wide scores in machine learning models Genetic risk integrates with minimal demographics to achieve AUC up to 0.806 Predicted risk aligns with optic nerve damage and inter-eye asymmetry in suspects