64 A multiple blood biomarker model for identifying patients with prevalent AF. (17th July 2020)

• Record Type: Journal Article
• Authors: Chua, Winnie; Cardoso, Victor; Purmah, Yanish; Crijns, Harry; Schotten, Ulrich; Guasch, Eduard; Sinner, Moritz; Hatem, Stephane; Casadei, Barbara; Mont, Lluis; Kastner, Peter; Ziegler, Andre; Gkoutos, Georgios; Kirchhof, Paulus; Fabritz, Larissa
• Is Part Of: Heart. Volume 106(2020)Supplement 2
• Page Start: A50
• Page End: A51
• Publication Date: 2020-07-17
• Subjects: Atrial fibrillation -- Biomarkers -- Screening; Heart -- Diseases -- Treatment -- Periodicals; Cardiology -- Periodicals; 616.12
• DOI: 10.1136/heartjnl-2020-BCS.64
• Languages: English
• ISSNs: 1355-6037

Abstract

Abstract : Background: Biomarkers reflecting different biological pathways have been associated with atrial fibrillation (AF). These discoveries motivate the consideration of a multiple biomarker model in the context of AF screening to improve detection. Objective: We assessed a selection of clinical characteristics and biomarkers known to be associated with AF as established in literature, to identify a mechanism-based combination of markers for simplifying patient selection for screening. Methods and Results: 1485 patients presenting acutely to hospital (median age [Q1, Q3] 69 [60, 78] years; 60% male; 45% with AF) with either diagnosed AF or ≥2 CHA2DS2-VASc risk factors (silent AF ruled out by 7-day ECG monitoring) were analysed. From EDTA plasma, 12 known cardiovascular biomarkers selected from published literature were quantified at a single centre on a high-throughput platform (Roche Diagnostics GMBH, DE). After adjustment for known confounders (age, sex, BMI, eGFR, heart failure, stroke/TIA, hypertension status), 6 biomarkers remained univariately associated with increased odds of AF (BMP10, ANG2, NTproBNP, IGFBP7, FGF23, CA125; see Figure ). A model which simultaneously considered clinical characteristics and biomarkers was developed in a discovery cohort (n = 933) randomly sampled from all included patients (60:40 discovery:validation) and subsequently validated on the remaining patients (n = 552) using both logistic regression and machine learning methodologies for comparison. Using regression with backward statistical selection, an optimism-adjusted model of Age, Sex, BMI, BMP10, ANG2, and FGF23 was found to discriminate between patients with and without AF with an area under the ROC curve, AUC, of 0.743 [95% confidence interval, 0.712, 0.775], corroborated by machine learning (AUC 0.760 [95%CI 0.746, 0.764]). Performance was similar in the validation cohort for regression (AUC 0.719 [0.677, 0.762]) and machine learning (AUC 0.733 [95%CI 0.691, 0.775]). In a sensitivity analysis using biomarker quartiles instead of absolute values, an additional biomarker was selected: NTproBNP. Conclusion: In our analysis of known markers of AF, a combination of 3 simple clinical characteristics (Age, Sex, BMI) and 3 biomarkers (BMP10, ANG2, and FGF23) robustly discriminated between patients with diagnosed AF and sinus rhythm patients with cardiovascular risks in both discovery and validation cohorts. Biomarkers implicate pathways related to inflammation (BMP10), fibrosis (FGF23) and hypoxia (ANG2), known to be associated with AF. Prospective studies can examine if AF screening with multiple biomarkers has the potential of identifying patients who could benefit from further ECG monitoring. Conflict of Interest: None.