Large-scale plasma proteomics profiles for predicting atrial fibrillation associated stroke risk
S Huang, W FengAbstract
Background
The accuracy of stroke risk stratification in atrial fibrillation (AF) patients remains limited due to the lack of reliable molecular biomarkers. This study aimed to identify circulating proteins associated with stroke risk in AF patients and develop an integrated proteomics-based prediction model.
Methods
We included 709 AF participants from the UK Biobank Pharma Proteomics Project (UKB-PPP) (median age 64 years, 30.5% female), with a median follow-up of 15 years. Stroke-related proteins were screened using multivariate Cox regression, least absolute shrinkage and selection operator (LASSO) Cox regression models, and machine learning modeling. Mendelian randomization (MR) analysis further validated the potential causal association between key proteins and stroke. Finally, a protein model was constructed, and its discrimination against the clinical model was assessed by C-index and time-dependent area under the receiver operating characteristic curve (AUC), with advantages verified through model performance comparison.
Results
Five core proteins for post-AF stroke were identified: EGFR, ATP6V1D, NTF4, AMN, and DCBLD2. MR validation showed that genetically higher levels of AMN and EGFR were associated with reduced stroke risk (both P< 0.05), while DCBLD2 was linked to increased risk (P< 0.05). The protein-based model achieved a C-index of 0.734 (95% CI: 0.675–0.789), outperforming the reference model (age and sex: 0.610 [95% CI: 0.550–0.671]) and clinical model (0.663 [95% CI: 0.600–0.725]). Time-dependent AUC values at 3, 5, and 8 years were 0.765, 0.722, and 0.753, respectively—all higher than the clinical model. The model demonstrated significant discrimination improvement (8-year integrated discrimination improvement [IDI] = 0.073, continuous net reclassification improvement [NRI] = 0.356). Subgroup analysis revealed strong performance in male and older age subgroups.
Conclusion
This study identified five core proteins for stroke in AF. The protein-based model outperformed clinical models, and key proteins showed causal associations with stroke risk—providing new targets and tools for optimizing precision stroke prevention in AF patients.stroke