The following biomarkers were included in the analyses: total N-terminal pro-B type natriuretic peptide (total NT proBNP), N-terminal pro-B type natriuretic peptide (NT-proBNP), angiopoietin 2 (Ang2), bone morphogenic protein-10 (BMP10), Dickkopf-related protein-3 (DKK3), endothelial cell specific molecule-1 (ESM1), fatty acid-binding protein 3 (FABP3), fibroblast growth factor 23 (FGF23), growth differentiation factor-15 (GDF15), insulin-like growth factor-binding protein-7 (IGFBP7), and myosin binding protein C3 (MYPBC3). The characteristics of the biomarkers (e.g. an exploratory analysis of nine circulating biomarkers, total NT-proBNP and NT-proBNP), are described in Additional file 1: Appendix 2 Assays of circulating biomarkers and detection of recurrent AF during follow-up, and detectability in Table 1.

Per protocol, for handling and processing the samples were left in the local lab for up to 1 h before freezing at − 70 °C. Samples stored locally were transferred to the core lab every year.

Biomarkers were assayed under blind conditions in a laboratory at Roche Diagnostics, Penzberg, Germany. Total NT-proBNP (Roche Diagnostics GmbH, Mannheim) was measured on a Cobas Elecsys Immunoanalyzer with a prototype sandwich immunoassay for use in exploratory research. This detects any NT-proBNP that is not O-glycosylated (position S44). The lower limit of detection is 6.5 pg/mL; within-run and between-run precisions are ≤ 1.2% and ≤ 2.5%.

Primary objective of this study is to evaluate whether deglycosylated total NT-proBNP, NT-proBNP and other nine circulating biomarkers are associated to AF, both ongoing or recurrent; the secondary endpoint is the relationship with first hospitalization for cardiovascular reasons. Continuous variables are expressed as mean ± SD if normally distributed or median and interquartile range [IQR] if not normally distributed; categorical variables were reported as absolute numbers and percentages. Differences between groups of patients with and without ongoing or recurrent AF were assessed with one-way ANOVA, Kruskal–Wallis test or χ² test, as appropriate. The correlations between circulating biomarkers were analyzed with Spearman's rank correlation coefficient.

The association of each biomarker with ongoing AF at the follow-up visit was assessed on the basis of the area under the curve (AUC) of ROC analysis, followed by logistic regression analysis adjusted for variables significantly associated with AF in univariate analysis: heart failure, LVEF < 40% or both, history of hypertension, AF episode with LA dilatation, and oral anticoagulant use. Biomarkers which had an AUC < 70% and were not independently associated with ongoing atrial fibrillation were then excluded from further analyses except BMP10. This biomarker was carried included in the analyses after an authoritative study on its mechanistic involvement in the pathophysiology of AF and its specific production by atrial tissue [16]

Kaplan–Meier curves and log-rank tests were used to assess differences in the AF recurrence-free survival according to biomarker baseline values above or below the median. Biomarkers were modelled as continuous variables (expressed as 1 SD increment) as linearity was tested by restricted cubic splines. Cox proportional univariable and adjusted hazard models were constructed to assess the prediction of AF recurrence. Cardiovascular hospitalization was predicted similarly. Cox analyses were adjusted for covariates selected on the basis of univariate analysis; for AF recurrence, sex and two or more episodes of AF in the six months before inclusion in GISSI-AF. For cardiovascular hospitalization, systolic blood pressure, history of hypertension, peripheral artery disease and smoking. The c-index derived from the multivariable models was used to assess the improvement in the prognostic model including either Total NT-proBNP or NT-proBNP in the adjusted model. Comparisons between the areas under the ROC curves were performed with the use of U-statistics [17]. All probability values are two-tailed and p-values were corrected for multiple testing by means of the False Discovery Rate (FDR-correction). A p < 0.05 was considered significant. Data were analyzed using SPSS Version 25 (IBM SPSS, Armonk, NY) and SAS Version 9.4.

The study comprised 382 patients, and 1038 plasma samples were analyzed (baseline, 382; six-month follow-up 325; 12-month follow-up 331). The baseline characteristics of this population were similar to those of the 1442 patients enrolled in the main GISSI-AF study [15]. In brief, age was 67.6 ± 9.1 years, 142 (37.2%) were women; 154 (40.8%) had had two or more episodes of AF in the previous 6 months and 336 (88.0%) had undergone cardioversion for AF in the previous two weeks. There was a history of hypertension in 324 (84.8%), history of stroke 15 (3.9%), and heart failure (HF) or LVEF < 40% in 42 (11.0%). The study treatment, valsartan, was given to 186 patients (48.7%), ACE-inhibitors to 206 (53.9%), beta-blockers to 114 (29.8%) and aldosterone blockers to 20 (5.2%) (Additional file 1: Table 1 Baseline clinical, eletrocardiographic and echocardiograpic characteristics of all patients and according to ongoing AF and AF recurrence).

During follow-up (median 365 days, range 5–373 days), 203 (53.1%) patients had at least one newly diagnosed episode of AF and 113 (29.6%) had more than one. The median number of episodes of recurrent AF per patient was 3 (range 2–27). During the one-year follow-up one patient died, and the incidence of hospitalization was 18.7% for any reason and 16.3% for CV reasons. As there were no significant differences between valsartan and placebo for any of the circulating biomarkers and echocardiographic variables, the whole cohort of 382 patients was analyzed, irrespective of the treatment.

Concentrations of total NT-proBNP and the other biomarkers at baseline and at each visit are shown in Table 1; medians were within the reference range except for total NT-proBNP and NT-pro BNP, where the levels were above the normal range. Median concentrations of all biomarkers were either stable or decreased slightly over the 12-month follow-up. Linear correlations between different biomarkers showed r values ranging from 0.90 for total NT-proBNP—NT-proBNP, to 0.11 for Ang2—FABP2 (Additional file 1: Table 2 Correlation coefficients for biomarker concentrations at baseline).

At 6- and 12-months follow-up, respectively 34/325 (10.5%) and 45/331 (13.6%) patients had ongoing AF in the 12-lead ECG. Concentrations of total NT-proBNP, NT-proBNP, Ang2, BMP10, DKK3, FGF23 and MyBPC3 were significantly higher in those with AF at both visits. Natriuretic peptides and Ang2 showed by far the largest significant increases in the patients with AF at 6 and 12 months (P < 0.001) (Table 2).

ROC analyses were done to assess the strength of the relation between biomarkers and ongoing AF (i.e. biomarkers assayed while an AF rhythm was recorded in a 12-lead ECG during a clinical visit). The AUCs were highest at 6 and 12 months, (Table 2) for total NT-proBNP (AUC_6M 0.78 and AUC_12M 0.77), NT-proBNP (AUC_6M 0.87 and AUC_12M 0.86), Ang2 (AUC_6M 0.73 and AUC_12M 0.82); for BMP10 AUC_6M was 0.71 and AUC_12M, 0.63. Multivariable logistic regression models estimated the association between these four biomarkers (total NT-proBNP, NT-proBNP, Ang2 and BMP10) with ongoing AF. The association was significant for all four after adjustment for clinical variables that were significant in univariable analysis (Table 3).

Total NT-proBNP, NT-proBNP, Ang2 and BMP10 were then included in analyses for the prediction of AF recurrence. In one year of follow-up, 203 patients (53.1%) experienced at least one episode of AF recurrence, mostly identified in telemetric device recordings. Clinical characteristics and treatments of patients with at least one AF recurrence and with any recurrence were previously published [5] (see also Additional file 1: Table 1 Baseline clinical, eletrocardiographic and echocardiograpic characteristics of all patients and according to ongoing AF and AF recurrence). Males and patients with more than two episodes of AF in the six months prior to inclusion in the study were more likely to suffer at least one AF recurrence.

Baseline concentrations of biomarkers in patients with or without AF recurrence were similar for all four biomarkers (Additional file 1: Table 3. Concentrations of circulating biomarkers at baseline in patients with or without AF recurrence). Kaplan Meier curves showing the incidence of AF recurrence in relation to the baseline biomarker concentration below or above the median indicate that curves started to diverge within the first week for total NT-proBNP, NT-proBNP and at one month for Ang2 and BMP10 (Fig. 1). Restricted cubic spline analysis showed linear associations for all four biomarkers with the recurrence of atrial fibrillation (Fig. 2), p > 0.05 for non-linearity. The predictive power for AF recurrence of baseline concentrations of total NT-proBNP, NT-proBNP, Ang2, and BMP10 was assessed using Cox proportional hazard models. After adjustment for clinical variables only total NT-proBNP (HR 1.19, 95%CI 1.04–1.36, p_fdr = 0.026) and NT-proBNP (HR 1.19, 95%CI 1.06–1.36, p_fdr = 0.016) significantly predicted AF recurrence (Table 4). The areas under the ROC curves based on the multivariate Cox models had a C-index of 0.822 for Total NT-proBNP and 0.829 for NT-proBNP, the difference was not statistically significant (P = 0.263) between the two models (Additional file 1: Fig. 1A ROC curves for first episode of recurrent atrial fibrillation. Curves are based on multivariable Cox survival model including standardized biomarkers ̶ total-NT-proBNP and NT-proBNP ̶ concentration). The risk of AF recurrence was analyzed in patients by quartiles of concentrations of the biomarkers at baseline. None of the quartiles were significantly different from the reference quartile after correction for multiple testing (Additional file 1: Table 4 Risk of AF recurrence for patients in the second, third and fourth quartiles of total-NT-proBNP, NT-proBNP, Ang2, BMP10 concentrations, compared to the lowest quartile.).

Total NT-proBNP was associated with first hospitalization for CV reasons in multivariable Cox proportional hazards regression analysis (HR 1.57; 95%CI 1.29–1.90; p_fdr = 1.2 × 10^–5); the association in this analysis was also significant for NT-proBNP after adjusting for confounders, (HR 1.57; 95%CI 1.34–1.84; p_fdr = 1.2 × 10^–6) and Ang2 (HR 1.34; 95%CI 1.04–1.73; p = 0.029), but not for BMP10 (Table 5). The areas under the ROC curves based on the multivariate Cox models had a C-index of 0.725 for Total NT-proBNP and 0.742 for NT-proBNP, the difference was not statistically significant (P = 0.056) between the two models (Additional file 1: Fig. 1B ROC curves for first hospitalization for CV reasons. Curves are based on multivariable Cox survival model including standardized biomarkers ̶ total-NT-proBNP and NT-proBNP ̶ concentration). The risk of first hospitalization for CV reasons was significantly higher only for patients in the fourth quartile of NT-proBNP compared to those in the lowest quartile (Additional file 1: Table 5 Risk for first hospitalization for CV reasons for patients in the second, third and fourth quartiles of total-NT-proBNP, NT-proBNP, Ang2, BMP10 concentrations, compared to the lowest quartile.).

The main strengths of this study are: (a) it was a multicenter randomized clinical trial with concomitant serial echocardiography and circulating biomarkers analyzed centrally; and (b) trans-telephonic electrocardiographic monitoring enabled us to record and identify AF recurrence efficiently during the 12-month follow-up. Another strength is the simultaneous analysis of total NT-proBNP with other biomarkers involved in different pathophysiological mechanisms, some of them assessed for the first time for AF diagnosis or as predictors of AF recurrence (Ang2, BMP10). The potential added value of total NT-proBNP to the benchmark biomarker NT-proBNP was assessed from different dimensions of performance, as recently proposed for the evaluation of new biomarkers [20].

Our results cannot apply to all patients with AF since at baseline the GISSI-AF patients were in sinus rhythm and had a lower rate of co-morbidities than patients with AF in real life or in other cohorts with a higher frequency of persistent or permanent AF [1, 2, 18]. This was due to compliance with the strict eligibility criteria of the trial. The mean CHADS2 modified score was indeed very low, averaging 1.41 ± 0.84 in the whole population, compatible with the low morbidity rate of the patients selected [22]. The low frequency of deaths after 12 months (only one) and thromboembolic events (three) in the GISSI-AF sub-study could not be considered for the outcome analysis. The small number of patients with ongoing AF at a follow-up visit is a limitation for the association of the biomarkers with the diagnosis of AF.