Multi-trait polygenic risk scores improve genomic prediction of atrial fibrillation across diverse ancestries

Polygenic scores (PGSs) can improve atrial fibrillation (AF) risk prediction, both alone and alongside clinical scores, offering potential for guiding targeted screening. However, their limited accuracy and cross-ancestry transferability remain major barriers to clinical translation. Here, we explored several multi-PGS approaches to generate ancestry-optimized PGSs for AF, and tested these in independent samples from the All of Us Research Program. Our ancestry-specific multi-trait approach, which leverages correlated traits, in particular outperformed the current gold-standard PGS among Asian (OR/SD = 1.76 [1.56–1.99]; AUROC = 0.637; AUPRC = 0.055), Admixed American (1.45 [1.38–1.53]; 0.595; 0.054) and African ancestry groups (1.39 [1.32–1.45]; 0.573; 0.064). Although predictive accuracy remained highest among Europeans (1.89 [1.85–1.93]; 0.646; 0.157) - in whom our PGS explained ~ 50% of SNP-heritability - our multi-trait approach yielded relatively larger gains in non-European populations. Improved risk stratification was also observed at PGS extremes, identifying a substantial proportion of European individuals with risk comparable to rare TTN variants (e.g., 5.8% with > 4-fold odds). Overall, our ancestry-tailored multi-trait PGSs advance equitable AF risk prediction and provide a foundation for implementation.