Large-Scale Plasma Proteomics Profiles for Predicting Atrial Fibrillation Associated Stroke Risk : Type of manuscript: Original Research.
Stroke is a major complication of atrial fibrillation (AF), and risk prediction using the congestive heart failure, hypertension, age, diabetes, stroke, vascular disease, and sex category score (CHA₂DS₂-VASc) remains limited by residual heterogeneity. We aimed to identify plasma proteins associated with post-AF stroke and evaluate whether a protein score provides incremental predictive information beyond CHA₂DS₂-VASc.
We analyzed 709 AF participants from the UK Biobank Pharma Proteomics Project, with 76 incident strokes. Stroke-related proteins were identified using multivariable Cox regression, least absolute shrinkage and selection operator (LASSO) Cox regression, and machine-learning approaches. A five-protein score was constructed, and its incremental value beyond CHA₂DS₂-VASc was assessed by discrimination, calibration, reclassification, clinical net benefit, and 1000-bootstrap internal validation. Mendelian randomization served as supportive genetic evidence.
Five core proteins were identified: epidermal growth factor receptor (EGFR), V-type proton ATPase subunit D (ATP6V1D), neurotrophin 4 (NTF4), amnionless (AMN), and discoidin, CUB and LCCL domain-containing protein 2 (DCBLD2). Adding the five-protein score to CHA₂DS₂-VASc improved discrimination, increasing the concordance index from 0.681 to 0.768. The combined model had 3-, 5-, and 8-year receiver operating characteristic areas of 0.799, 0.775, and 0.806, respectively, and showed favorable five-year prediction error and calibration, with a Brier score of 0.0347, calibration intercept of -0.068, and calibration slope of 0.968. The score improved continuous net reclassification improvement (0.459; P = 0.028). Mendelian randomization provided supportive genetic evidence for AMN, EGFR, and DCBLD2.
The five-protein score provided incremental predictive information beyond CHA₂DS₂-VASc for post-AF stroke risk assessment.
We analyzed 709 AF participants from the UK Biobank Pharma Proteomics Project, with 76 incident strokes. Stroke-related proteins were identified using multivariable Cox regression, least absolute shrinkage and selection operator (LASSO) Cox regression, and machine-learning approaches. A five-protein score was constructed, and its incremental value beyond CHA₂DS₂-VASc was assessed by discrimination, calibration, reclassification, clinical net benefit, and 1000-bootstrap internal validation. Mendelian randomization served as supportive genetic evidence.
Five core proteins were identified: epidermal growth factor receptor (EGFR), V-type proton ATPase subunit D (ATP6V1D), neurotrophin 4 (NTF4), amnionless (AMN), and discoidin, CUB and LCCL domain-containing protein 2 (DCBLD2). Adding the five-protein score to CHA₂DS₂-VASc improved discrimination, increasing the concordance index from 0.681 to 0.768. The combined model had 3-, 5-, and 8-year receiver operating characteristic areas of 0.799, 0.775, and 0.806, respectively, and showed favorable five-year prediction error and calibration, with a Brier score of 0.0347, calibration intercept of -0.068, and calibration slope of 0.968. The score improved continuous net reclassification improvement (0.459; P = 0.028). Mendelian randomization provided supportive genetic evidence for AMN, EGFR, and DCBLD2.
The five-protein score provided incremental predictive information beyond CHA₂DS₂-VASc for post-AF stroke risk assessment.
Authors
Huang Huang, Feng Feng, Wu Wu, Liang Liang, Gao Gao, Wang Wang, Tang Tang, Wu Wu, Fang Fang, Yang Yang, Tiemuerniyazi Tiemuerniyazi, Lin Lin, Duan Duan, Xu Xu, Mao Mao, Zhao Zhao, Hu Hu, Duan Duan, Feng Feng
View on Pubmed