Proteomic Signatures for Risk Prediction of Atrial Fibrillation
Circulation, 2025
Park H., Norby F., Kim D., Jang E., Yu H., Kim T., Uhm J., Sung J., Pak H., Lee M., Yang P., Joung B.
| Disease area | Application area | Sample type | Products |
|---|---|---|---|
CVD | Patient Stratification | Plasma |  Olink Explore 3072/384 |
Abstract
BACKGROUND:
Proteomic signatures might improve disease prediction and enable targeted disease prevention and management. We explored whether a protein risk score derived from large-scale proteomics data improves risk prediction of atrial fibrillation (AF).
METHODS:
A total of 51 680 individuals with 1459 unique plasma protein measurements and without a history of AF were included from the UKB-PPP (UK Biobank Pharma Proteomics Project). A protein risk score was developed with lasso-penalized Cox regression from a random subset of 70% (36 176 individuals, 54.4% women, 2155 events) and was tested on the remaining 30% (15 504 individuals, 54.4% women, 910 events). The protein risk score was externally replicated with the ARIC study (Atherosclerosis Risk in Communities; 11 012 individuals, 54.8% women, 1260 events).
RESULTS:
The protein risk score formula developed from the UKB-PPP derivation set was composed of 165 unique plasma proteins, and 15 of them were associated with atrial remodeling. In the UKB-PPP test set, a 1-SD increase in protein risk score was associated with a hazard ratio of 2.20 (95% CI, 2.05–2.41) for incident AF. The C index for a model including CHARGE-AF (Cohorts for Heart and Aging Research in Genomic Epidemiology Atrial Fibrillation), NT-proBNP (N-terminal B-type natriuretic peptide), polygenic risk score, and protein risk score was 0.816 (95% CI, 0.802–0.829) compared with 0.771 (95% CI, 0.755–0.787) for a model including CHARGE-AF, NT-proBNP, and polygenic risk score (C-index change, 0.044 [95% CI, 0.039–0.055]). Protein risk score added to CHARGE-AF, NT-proBNP, and polygenic risk score resulted in a risk reclassification of 5.4% (95% CI, 2.9%–7.9%) with a 5-year risk threshold of 5%. In the decision curve, the predicted net benefit before and after the addition of protein risk score to a model including CHARGE-AF, NT-proBNP, and polygenic risk score was 3.8 and 5.4 per 1000 people, respectively, at a 5-year risk threshold of 5%. External replication of a protein risk score in the ARIC study showed consistent improvement in risk stratification of AF.
CONCLUSIONS:
Protein risk score derived from a single plasma sample improved risk prediction of AF. Further research using proteomic signatures in AF screening and prevention is needed.