AHA Scientific Sessions 2026 Abstract · Nov 2, 2026
Proteomic Profiling for Identification of Residual Coronary Artery Disease Risk in Clinically and Genetically Low-Risk Individuals
By Sui Y, Supriami K, Kim MS et al.
Introduction: Coronary artery disease (CAD) remains a leading cause of mortality worldwide. Polygenic risk scores (PRS) and contemporary clinical prediction models identify high-risk individuals, yet substantial residual CAD risk persists among individuals classified as low risk. Whether plasma proteomics can identify residual CAD risk beyond established genetic and clinical tools remains unclear. Hypothesis: A proteomic risk score derived from individuals classified as low genetic and clinical risk independently improves CAD risk prediction beyond PRS and a contemporary clinical risk model. Methods: Among UK Biobank participants with plasma proteomic profiling (~2,900 proteins measured using Olink), a low-risk group was defined as individuals below the median of both a CAD polygenic score (GPSMult) and a Predicting Risk of Cardiovascular Disease Events (PREVENT Base ) score (N=9,437; 82 incident CAD events). Proteome-wide analyses followed by data-driven feature selection were used to reduce multicollinearity and overfitting. A proteomic risk score was constructed using Cox proportional hazards modeling. Discrimination was evaluated using Harrell's C-index with 300-fold bootstrapping. Independent associations were assessed after adjustment for PRS, PREVENT score, and 10 principal components. Results: Addition of the proteomic score to PREVENT+PRS significantly improved discrimination among low-risk individuals (ΔC-index=0.071, 95% CI 0.035-0.113, P<0.001), increasing the C-index from 0.708 to 0.779 ( Figure A ). In multivariable Cox regression adjusting for PRS, PREVENT score, and ancestry principal components, the proteomic score remained independently associated with incident CAD in the low-risk group (HR=2.25, 95% CI 1.90-2.66, P=4.42х10 ⁻ ² ¹ ). Individuals classified as low risk by both PRS and clinical criteria demonstrated markedly different cumulative CAD incidence across proteomic risk score strata ( Figure B ). Key proteomic signals implicated inflammatory, immune, and metabolic pathways relevant to atherosclerosis biology. In comparison, incremental proteomic improvement was smaller in a higher-risk reference group (N=27,353; 1,587 events; ΔC-index=0.012, 95% CI 0.007-0.017, P<0.001), although the proteomic score remained independently associated with incident CAD (HR=1.29, 95% CI 1.23-1.35, P=3.89х10 ⁻ ² ⁵ ). Conclusions: Among individuals deemed low risk by both PRS and a contemporary clinical model, a plasma proteomic score identified substantial residual CAD risk not captured by either genetic or clinical information. These findings support the potential role of proteomic profiling in refining cardiovascular risk stratification beyond established genetic and clinical risk assessment tools. Ongoing analyses are validating the performance of the proteomic score in an external dataset.