Data-driven discovery and validation of circulating blood-based biomarkers associated with prevalent atrial fibrillation
Data-driven discovery and validation of circulating blood-based biomarkers associated with prevalent atrial fibrillation
复制标题
DOI：
10.1093/eurheartj/ehy815
复制
发表时间：
2019-04-21
影响因子：
39.3
通讯作者：
Fabritz, Larissa
中科院分区：
文献类型：
--
作者：
Chua, Winnie;Purmah, Yanish;Fabritz, Larissa
目标 未被发现的心房颤动 (AF) 是一个主要的健康问题。与 AF 相关的血液生物标志物可以简化筛选患者的选择，并进一步为正在进行的 AF 分层预防和治疗研究提供信息。 方法和结果 对转诊至医院的 638 名连续患者的 40 种常见心血管生物标志物进行了量化 [平均+/-标准差年龄 70 +/- 12 岁，398 名 (62%) 男性，294 名 (46%) 患有 AF] 已知 AF 或 >= 2 CHA(2)DS(2)-VASc 危险因素。通过 7 天的心电图监测排除了阵发性或无声性 AF。使用具有前向选择和机器学习算法的逻辑回归来确定与 AF 相关的临床危险因素、成像参数和生物标志物。心房颤动与年龄显着相关[自举比值比 (OR) 每年 = 1.060，95% 置信区间 (1.04-1.10)； P = 0.001]，男性 [OR = 2.022 (1.28-3.56)； P = 0.008]，体重指数[BMI，每单位OR = 1.060（1.02-1.12）； P = 0.003]，脑钠肽升高 [BNP，OR 每倍变化 = 1.293 (1.11-1.63)； P = 0.002]，成纤维细胞生长因子-23 [FGF-23，OR = 1.667 (1.36-2.34) 升高； P = 0.001]，并减少 TNF 相关凋亡诱导的配体受体 2 [TRAIL-R2，OR = 0.242 (0.14-0.32)； P = 0.001]，但其他生物标志物则不然。与单独的临床危险因素相比，生物标志物改善了 AF 的预测（净重分类改善 = 0.178；P < 0.001）。逻辑回归和机器学习在验证过程中都能很好地预测 AF [受试者-操作曲线下面积分别 = 0.684 (0.62-0.75) 和 0.697 (0.63-0.76)]。结论 三个简单的临床危险因素（年龄、性别和 BMI）和两个生物标志物（BNP 升高和 FGF-23 升高）可识别 AF 患者。需要进一步的研究来阐明 AF 的 FGF-23 依赖性机制。
Aims Undetected atrial fibrillation (AF) is a major health concern. Blood biomarkers associated with AF could simplify patient selection for screening and further inform ongoing research towards stratified prevention and treatment of AF.Methods and results Forty common cardiovascular biomarkers were quantified in 638 consecutive patients referred to hospital [mean +/- standard deviation age 70 +/- 12 years, 398 (62%) male, 294 (46%) with AF] with known AF or >= 2 CHA(2)DS(2)-VASc risk factors. Paroxysmal or silent AF was ruled out by 7-day ECG monitoring. Logistic regression with forward selection and machine learning algorithms were used to determine clinical risk factors, imaging parameters, and biomarkers associated with AF. Atrial fibrillation was significantly associated with age [bootstrapped odds ratio (OR) per year = 1.060, 95% confidence interval (1.04-1.10); P = 0.001], male sex [OR = 2.022 (1.28-3.56); P = 0.008], body mass index [BMI, OR per unit = 1.060 (1.02-1.12); P = 0.003], elevated brain natriuretic peptide [BNP, OR per fold change = 1.293 (1.11-1.63); P = 0.002], elevated fibroblast growth factor-23 [FGF-23, OR = 1.667 (1.36-2.34); P = 0.001], and reduced TNF-related apoptosis-induced ligand-receptor 2 [TRAIL-R2, OR = 0.242 (0.14-0.32); P = 0.001], but not other biomarkers. Biomarkers improved the prediction of AF compared with clinical risk factors alone (net reclassification improvement = 0.178; P < 0.001). Both logistic regression and machine learning predicted AF well during validation [area under the receiver-operator curve = 0.684 (0.62-0.75) and 0.697 (0.63-0.76), respectively].Conclusion Three simple clinical risk factors (age, sex, and BMI) and two biomarkers (elevated BNP and elevated FGF-23) identify patients with AF. Further research is warranted to elucidate FGF-23 dependent mechanisms of AF.