Relations of Biomarkers of Distinct Pathophysiological Pathways and Atrial Fibrillation Incidence in the Community

Abstract

Background

Biomarkers of multiple pathophysiological pathways have been related to incident atrial fibrillation (AF), but their predictive ability remains controversial.

Methods and Results

In 3120 Framingham cohort participants (mean age 58.4±9.7, 54% women), we related 10 biomarkers representing inflammation (C-reactive protein [CRP], fibrinogen), neurohormonal activation (B-type natriuretic peptide [BNP], N-terminal pro-atrial natriuretic peptide), oxidative stress (homocysteine), renin-angiotensin-aldosterone system (renin, aldosterone), thrombosis and endothelial function (D-dimer, plasminogen-activator inhibitor 1 [PAI-1]), and microvascular damage (urine albumin excretion, n=2673) with incident AF (n=209, 40% women) over a median follow-up of 9.7 years (range 0.05–12.8 years).

In multivariable-adjusted analyses, the biomarker panel was associated with incident AF (P<0.0001). In stepwise selection models (P<0.01 for entry and retention), log-transformed BNP, hazard ratio [HR] per standard deviation 1.62 (95% confidence interval [CI] 1.41–1.85, P<0.0001), and CRP, HR 1.25 (95% CI 1.07–1.45, P=0.004), were chosen.

The addition of BNP to variables recently combined in a risk score for AF increased the C-statistic from 0.78 (95%CI 0.75–0.81 to 0.80 (95% CI 0.78–0.83), and showed an integrated discrimination improvement of 0.03 (95% CI 0.02–0.04, P<0.0001) with 34.9% relative improvement in reclassification analysis. The combined analysis of BNP and CRP did not appreciably improve risk prediction over the model incorporating BNP in addition to the risk factors.

Conclusions

BNP is a predictor of incident AF and improves risk stratification based on well-established clinical risk factors. Whether knowledge of BNP concentrations may be used to target individuals at risk of AF for more intensive monitoring or primary prevention needs further investigation.

It is anticipated that over the next four decades the prevalence of atrial fibrillation (AF) will increase dramatically due to an aging population, improved therapies and longer survival with heart disease.^1,2 AF is associated with higher rates of stroke and hospitalization,^3,4 diminished quality of life,⁵ and significant mortality.⁶ Identifying risk factors for developing AF is an important epidemiologic task with potential implications for public health,^7,8 and research in this respect has been prioritized by the National Heart Lung and Blood Institute (http://www.nhlbi.nih.gov/meetings/workshops/prevent-af.htm).

Well-established clinical risk factors for AF besides age and sex are body mass index, hypertension, and cardiovascular disease, including valvular disease and heart failure.^9–11 However, these risk factors do not explain all cases of AF, suggesting a need for improvement in risk prediction and understanding of the pathophysiology of AF.¹² Blood and urinary biomarkers are potential tools to enhance AF risk prediction and to provide insights into the disease’s pathophysiology. Based on biological plausibility, and prior reports, biomarkers were chosen to represent distinct pathophysiological pathways including inflammation (C-reactive protein [CRP], fibrinogen),^13,14 neurohormonal activation (B-type natriuretic peptide [BNP], N-terminal pro-atrial natriuretic peptide [N-ANP]),^15,16 oxidative stress and endothelial dysfunction (homocysteine),¹⁷ renin-angiotensin-aldosterone system (renin, aldosterone),¹⁶ thrombosis and endothelial function (D-dimer, plasminogen-activator inhibitor [PAI] 1),^18,19 and microvascular damage (urine albumin excretion).²⁰ We hypothesized that the combined analysis of these biomarkers identifies a small panel of distinct biomarkers that are associated with new-onset AF and will improve risk stratification beyond clinical risk factors.

Materials and Methods

Study Sample

The Framingham Offspring study enrolled 5124 individuals in the early 1970s with regular follow-up every four to eight years.²¹ Participants (n=3532) who attended the sixth examination cycle (1995–1998) were eligible for analysis. For the present study attendees were excluded based on any missing biomarker measurements (n=270), incomplete or missing follow-up (n=1), prevalent AF (n=106), serum creatinine greater than two (n=18) or missing covariate data (n=17). As a result, data on 3120 participants were available for analysis (urinary albumin, n=2673). Boston University Medical Center Institutional Review Board approved the study protocols and participants provided informed consent at each examination. The authors had full access to the data and take responsibility for its integrity. All authors have read and agreed to the manuscript as written.

Clinical Evaluations

Regular cardiovascular health assessments at the Framingham Heart Study clinic include cardiac risk factor documentation during a physician-administered interview and physical examination. Valvular heart disease was considered present if ≥ three out of six systolic, or any diastolic murmur was detected on auscultation. Heart failure was diagnosed by the endpoint adjudication committee on the basis of previously published criteria.²² Hypertension medication was determined by self-report. The average of two seated systolic blood pressure measurements obtained by a Heart study physician constituted the examination blood pressure.

Atrial Fibrillation Verification

The participants’ physician office visits and hospitalization records were collected. The diagnosis of AF was based on atrial fibrillation or atrial flutter present on electrocardiogram tracings and information from hospital or outpatient records or Framingham Study clinic exams. For Offspring participants, biennial health history updates included a routine question on AF. Incident AF cases underwent review and two Framingham cardiologists had to agree on the diagnosis.¹²

Biomarker determination

Blood samples were obtained routinely from fasting participants and processed immediately. The measurement characteristics of the biomarkers have been described previously.²³ Plasma biomarkers comprised high-sensitivity CRP, D-dimer, fibrinogen, BNP, N-ANP, renin, PAI-1 and homocysteine. Aldosterone was measured from serum. Urinary albumin and creatinine were determined on spot morning specimen. Assay details are provided in the Supplement. Mean inter-assay coefficients of variation were less than 13% for natriuretic peptides and less than 10% for other biomarkers.

Echocardiography

Attendees at examination cycle 6 routinely received transthoracic echocardiography. M-mode left atrial diameter, wall thickness (sum of diastolic interventricular septum and left ventricular posterior wall) and a measure of systolic function (left ventricular fractional shortening);²⁴ echocardiographic measurements were available at the baseline examination on 2289 attendees.

Statistical Analyses

Biomarkers were natural logarithmically transformed and standardized (mean of 0 and standard deviation of 1) for analyses. For multivariable-adjusted models, we selected AF risk factors that have been reported in association with incident AF and recently have been incorporated in a weighted risk score for individualized risk prediction of AF. The current sample constitutes a subsample of the risk score derivation sample. The variables of the risk algorithm comprised: age (at baseline examination six), sex, body mass index, systolic blood pressure, electrocardiographic PR interval, hypertension treatment, heart valve disease (heart murmur), and heart failure.¹² Multivariable-adjusted proportional hazards regression models were estimated relating the biomarkers to incident AF.²⁵ The proportional hazards assumption was examined using a Kolmogorov-type supremum test based on cumulative sums of martingale-based residuals over follow-up times and covariate values.²⁶ For primary analyses, we used a stepwise procedure to select biomarkers associated with AF at a conservative two-sided significance threshold of P<0.01 for entry and retention in the model,²⁷ with age, sex, and clinical covariates forced in. The regression coefficients presented are per standard deviation (SD) increase in log-transformed biomarkers. For the final model, covariate-adjusted cumulative AF incidence estimates for tertiles of biomarker score (calculated as coefficient-weighted sums of standardized biomarkers associated with AF incidence) were estimated and plotted graphically. We assessed C-statistics to describe discrimination of the baseline model and the model including selected biomarker(s).²⁸ Calibration was calculated for deciles of risk using a modified Hosmer-Lemeshow statistic for survival analysis.²⁹ We assessed net reclassification improvement for predefined 10-year AF risk categories (<5%, 5–10%, >10%),¹² integrated discrimination improvement as well as the relative integrated discrimination improvement and reclassification calibration.³⁰ The statistical metrics to assess reclassification are an area of intense development. We also tested the newly introduced reclassification-calibration, a method that also accounts for censored data.³¹

To establish the value of the retained biomarker(s) as a potential clinical tool in risk prediction, we reran the proportional hazards models using as a baseline “covariate” a recently developed risk score function for 10-year incidence of AF (http://www.framinghamheartstudy.org/risk/atrial.html). Any event outside the 10-year time frame was censored (n=6 cases). Censored data were treated as non-events.

Secondary Analyses

For the final model incorporating the biomarkers significantly associated with incident AF, we assessed potential effect modification by age and sex by a global likelihood ratio test. We further explored whether the association of the selected biomarkers with incident AF was mediated by heart murmur or interim heart failure. In addition, bivariate correlation coefficients for the (log) biomarkers were calculated. We also present the data for each biomarker analyzed separately in multivariable-adjusted models. For the final model, we adjusted for echocardiographic variables (left atrial size, left ventricular wall thickness, and left ventricular fractional shortening) to explore whether the relations between biomarkers and incident AF were mediated by cardiac structure and function measures. Analyses were conducted using SAS version 8.1 (Cary, North Carolina, http://www.sas.com/presscenter/guidelines.html).

Results

Participant Characteristics

Our study sample had an overall mean age of 58.4±9.7 years, and 54% were women. In Table 1 the baseline characteristics for individuals who developed AF and those free of AF during follow-up are provided. During a median 9.7 years follow-up (maximum 12.8 years) until November 2007, 209 incident AF cases (40.0% among women) occurred (n=166 in urinary albumin-to-creatinine ratio subset).

Table 1.

Baseline Characteristics by Incident AF Status

	AF Status in Follow-up
Clinical Characteristics*	No AF (n=2911)	Incident AF (n=209)
Age, years	57.8±9.5	66.3±8.6
Women, No (%)	1608 (55)	84 (40)
Body mass index, kg/m2	27.9±5.2	28.7±5.9
Systolic pressure, mm Hg	128±18	137±22
Hypertension treatment, No (%)	751 (26)	105 (50)
ECG PR-interval, msec	163±23	170±27
Significant murmur, No (%)	68 (2)	14 (7)
Prevalent heart failure, No (%)	12 (<1)	5 (2)
Biomarkers, median (25th, 75th percentile)
C-reactive protein, mg/l	2.0 (0.9, 4.6)	3.0 (1.3, 7.1)
Fibrinogen, mg/dl	329 (288, 378)	351 (306, 401)
B-type natriuretic peptide, pg/ml	7.7 (4.0, 16.7)	21.3 (8.2, 46.0)
N-terminal pro-atrial natriuretic peptide, pmol/l	311 (218, 444)	459 (323, 753)
Renin, mU/l	12.0 (7.0, 21.0)	10.0 (5.0, 21.0)
Aldosterone, ng/dl	10.0 (7.0, 14.0)	10.0 (7.0, 14.0)
Homocysteine, mmol/l	9.0 (7.4, 11.0)	9.8 (8.2, 12.3)
D-dimer, ng/ml	311 (200, 462)	435 (300, 636)
Plasminogen-activator inhibitor 1, mg/ml	22.6 (14.2, 33.8)	26.3 (19.3, 39. 6)
Urinary albumin-to-creatinine ratio**	6.1 (2.7, 14.2)	8.4 (3.0, 20.0)
Echocardiographic Variables***
Fractional shortening	0.37±0.06	0.36±0.07
Left atrial size, cm	3.92±0.50	4.27±0.63
Left ventricular wall thickness, mm	1.89±0.24	2.04±0.33

Age was at entry of the follow-up period.

^*Clinical characteristics expressed as mean±SD or number (%).

^**Available for a subset of 2507 subjects without and 166 subjects with incident AF.

^***Available for a subset of 2154 subjects without and 135 subjects with incident AF.

Biomarkers and AF incidence

We confirmed the validity of the proportionality of hazards assumption for the variables in the selected models. In models adjusted for established risk factors, the biomarkers as a panel were associated with incident AF (P<0.0001). In multivariable models for single biomarkers in relation to incident AF, N-ANP, BNP and CRP were associated with the outcome (Table 2). In the stepwise selection procedure, BNP (hazard ratio [HR] per standard deviation 1.62 (95% confidence interval [CI] 1.41–1.85, P<0.0001) and CRP (HR 1.25, 95% CI 1.07–1.45, P=0.004) met the inclusion criterion (Supplementary Table 1). The biomarker selection was similar in the subsample of individuals with urinary albumin-to-creatinine ratio (BNP HR 1.63, 95% CI 1.40–1.89, P<0.0001; CRP HR 1.31, 95% CI 1.11–1.55, P=0.002). The urinary albumin-to-creatinine ratio was not significantly associated with incident AF. Addition of BNP alone, CRP alone, or both simultaneously to the model containing the clinical risk factors increased the Chi-square statistic from 223 to 303, 229, and 310, respectively. Cumulative event rates according to tertiles of the biomarker score incorporating BNP and CRP revealed an increase in AF events with the highest AF incidence observed in the top biomarker score tertile (Figure 1).

Table 2.

Multivariable-adjusted proportional hazards regression models for atrial fibrillation examining each log-transformed biomarker separately

Variable	Hazard Ratio	95% Confidence Interval		P Value
Inflammation
C-reactive protein	1.25	1.07	1.46	0.004
Fibrinogen	1.09	0.94	1.26	0.26
Natriuretic peptides
B-type natriuretic peptide	1.62	1.42	1.86	<0.0001
N-terminal pro-atrial natriuretic peptide	1.50	1.28	1.75	<0.0001
Renin-angiotensin-aldosterone system
Aldosterone	1.05	0.92	1.19	0.50
Renin	0.89	0.77	1.02	0.08
Oxidative stress
Homocysteine	1.08	0.94	1.24	0.28
Thrombosis, endothelial function
D-dimer	1.11	0.92	1.32	0.28
Plasminogen-activator inhibitor 1	1.13	0.96	1.33	0.15
Microvascular damage
Urinary albumin-to-creatinine ratio*	1.09	0.93	1.28	0.29

Biomarker concentrations are natural log-transformed measures.

Hazard ratios are provided per one standard deviation increase in log-biomarker concentration.

Models are adjusted for age, sex, body mass index, systolic blood pressure, hypertension treatment, PR interval, auscultatory valvular heart disease, and heart failure.

^*Urinary albumin-to-creatinine ratio was available in n=2673.

Figure 1.

Covariate-Adjusted AF cumulative incidence curves for tertiles of the biomarker score (including BNP and CRP) in women (panel A) and men (panel B). Mean values of the biomarkers for each tertile of biomarker score were used in creating the cumulative incidence estimates.

When assessing the biomarkers in addition to the risk factors identified for the recently developed AF risk score, the risk information derived from BNP increased the C-statistic from 0.78 (95% CI 0.75–0.81) to 0.80 (95% CI 0.78–0.83) (Supplementary Table 2) and improved net reclassification (Figure 2a). The analysis method of net reclassification has been developed to assess the putative clinical utility of a novel risk factor. It is based on pre-specified risk categories. A clinically useful biomarker would help to optimize risk classification beyond the model including known risk factors. Ideally, the novel marker would re-classify all individuals with a future event into the high-risk category and all individuals without the outcome into the lowest risk category. Among those participants who developed AF in our sample, the inclusion of BNP concentrations resulted in 25 individuals being reclassified into higher risk categories (correct direction, green shaded cells), but 28 were inappropriately down-classified (red shaded cells). Conversely, among individuals who did not develop AF over 10 years of follow-up, BNP concentrations would have led to undesirable up-reclassification of risk in 217 individuals, whereas inclusion of BNP concentrations would have appropriately reclassified 444 individuals into lower risk categories. The net reclassification improvement, which in our case indicates the overall reclassification in the desirable direction, was 0.06 (95% CI −0.01–0.14, P=0.09); reclassification occurred predominantly in the intermediate risk group. A clinically less intuitive method to assess reclassification is the calculation of the integrated discrimination improvement which does not rely on pre-specified risk categories, but represents a continuous measure was 0.03 (95% CI 0.02–0.04, P<0.0001) with 34.9% relative improvement.

Figure 2.

Reclassification based on biomarkers.

An even newer metric to evaluate novel biomarkers is the reclassification-calibration test introduced by Cook et al.³¹ which also takes into account censored data. The chi-square statistic for the risk-score variables only model (23.58, P=0.0003) decreased to 8.78 with the addition of BNP, P=0.12, indicating better fit (for these lack of fit statistics a lower value indicates better fit and non-significance is desirable). CRP (net reclassification improvement 0.009, 95% CI −0.04–0.06, P=0.72; integrated discrimination improvement 0.005, 95% CI 0.0002–0.01, P=0.04, relative integrated discrimination improvement 5.9%) achieved only a very small improvement in reclassification-calibration (Figure 2b). The reclassification-calibration chi-square slightly decreased from 7.26, P=0.20 to 7.05 P=0.22 when CRP was added. Supplementary Figure 1 provides plots of the estimated risk from the models with and without the biomarkers in addition to the risk factors. To create these plots we computed risk of AF for each person at each event time (n=203), first from the model with only the clinical risk factors and then from the model with the addition of BNP and CRP. We then averaged each model-specific set of risk estimates by event status, resulting in the set of sample average predicted risks for AF events and nonevents seen in the plots. Adding the information derived from BNP and CRP leads to a greater separation of the event curves mostly through a modest increase in estimated incidence for the AF event group.

The reclassification when using the combination of both biomarkers in addition to the model comprising the clinical covariates only was clearly driven by BNP (net reclassification improvement 0.11, 95% CI 0.04–0.19, P=0.002; integrated discrimination improvement 0.04, 95% CI 0.02–0.05, P<0.0001, relative integrated discrimination improvement 39.1%) (Figure 2c).

The C-statistic did not change appreciably (0.81, 95% CI 0.78–0.84) and the calibration χ² statistic slightly increased when CRP was added to the model including BNP.

When using the risk algorithm for 10-year incidence of AF (n=203 events) the final stepwise selection resulted in a similar model incorporating BNP (P<0.0001) and CRP (P=0.003). The reclassification statistics for the variables combined in a ‘risk score’ may inflate the reclassification and discrimination statistics for biomarkers compared to assessing the risk factors separately. In our case net reclassification improvement for BNP was 0.08, P=0.04. The integrated measure was 0.04, P<0.0001, with 48.2% relative improvement. Further details on the results when using the risk score are provided in the Supplement.

Secondary Analyses

We did not observe statistically significant age or sex interactions for BNP or CRP for the final model (global P=0.38). When adjusting for interim development of heart murmur or heart failure, the coefficients and significance of the estimates for BNP and CRP did not change materially (Supplementary Table 4). Excluding individuals (n=17) with prevalent heart failure at baseline did not change the final model appreciably (data not shown). Using a more parsimonious model with the strongest risk factors age, sex, hypertension and heart failure showed discrimination statistics comparable to the model incorporating the risk factors from the Framingham risk score (Supplementary Table 5). Using the broader range of risk factors moderately increased calibration and fit of the model. The strongest correlations for biomarkers were observed between BNP and N-ANP (Pearson correlation coefficient r=0.66) and fibrinogen and D-dimer (r=0.45) (Supplementary Table 6). CRP and BNP had a low positive correlation (r=0.04, P=0.02).

After adjustment of the final model incorporating both BNP and CRP for echocardiographic measures (left atrial diameter and left ventricular wall thickness and fractional shortening), the association of BNP with AF remained robust (HR 1.52, 95% CI 1.28–1.81, P<0.0001). However, CRP was no longer significantly associated with incident AF (HR 1.10 95% CI 0.91–1.34, P=0.33).

Discussion

Principal Findings

In a prospective, middle-aged to elderly community-based cohort we examined the association of 10 biologically plausible biomarkers with incident AF over a median of 9.7 years. The neurohumoral marker BNP emerged as the strongest predictor of incident AF. When used in addition to a risk score for AF incidence, it improved discrimination and resulted in a substantive net reclassification improvement of 7.9%, and a relative integrated discrimination improvement of almost 50%, which remained strong (35%) even after accounting for potential inflation of the results. The inflammatory biomarker CRP also was statistically significantly associated with the outcome, but did not markedly improve risk prediction beyond BNP. We observed that the final models were not substantively altered by analyzing cardiac disease as time-dependent variables or incorporating echocardiographic features. Furthermore, we did not observe significant effect modification by sex or age in the models incorporating BNP and CRP.

BNP as an indicator of cardiac stress is a highly plausible candidate biomarker for AF risk. Manifest AF is accompanied by elevated natriuretic peptide concentrations,³² even in paroxysmal AF^33,34 and in the absence of overt heart failure.³² Intuitively, the prohormone fragment of atrial natriuretic peptide, which is predominantly expressed in the atria, might be the member of the natriuretic peptide family that should have strongest predictive power for incident AF. Both natriuretic peptides are elevated in AF patients,³⁵ and the atria may be a main source for BNP even in the absence of ventricular dysfunction.³⁶ Correlates of BNP concentrations are left atrial size and left ventricular ejection fraction.³⁷ Our data demonstrated that even accounting for potential intermediate mechanisms by adjusting for interim cardiac disease or echocardiographic measures of left atrial dimensions and systolic function, BNP retained its strength of association with AF. Thus, BNP seems to provide risk information for AF beyond noninvasively assessed cardiac structure and function.

Recent investigations over a shorter follow-up and with fewer AF cases, including a study by Framingham investigators examining natriuretic peptides in relation to multiple cardiovascular outcomes (68 AF cases), suggested an association of natriuretic peptides with incident AF.^15,38,39 We now demonstrate that BNP provides additional risk information compared to known strong clinical risk factors for AF and compared to multiple other biomarkers that have been related to AF. The net reclassification improvement and relative integrated discrimination improvement, which takes into account the number of variables in the basic model and the gain of information by the addition of the novel variable, support the strength of BNP in addition to the clinical risk factors.

Manifest AF is accompanied by systemic inflammatory activity and increased oxidative stress.^40,41 We confirmed prior investigations that related the inflammatory biomarker CRP to incident AF,⁴² but did not find an association with fibrinogen that reached statistical significance.¹⁴ The magnitude of association we observed for CRP in the present study was similar to that observed previously. However, CRP did not perform as well as BNP in improving risk classification and application of CRP as a risk indicator in clinical practice is unlikely to be resource effective. Even if CRP does not substantially improve risk prediction, the observed relation may help to elucidate underlying mechanisms of AF and to identify therapeutic targets. Pleiotropic effects of statins have been shown to decrease inflammatory activity – and anti-inflammatory treatment might be a rationale for AF prevention based on the consistent association of CRP with AF.⁴³ Prior literature relating homocysteine to AF suggested that homocysteine is an indicator of endothelial dysfunction and susceptibility to thromboembolic events in manifest AF.¹⁷ Data have remained inconsistent,^44,45 and after multivariable-adjustment, we did not identify a significant association between homocysteine and occurrence of AF.

Strengths and Limitations

Some limitations merit consideration. Inherent to the study design, we may have missed asymptomatic AF episodes. We cannot exclude the possibility that baseline BNP concentrations may have in part been influenced by clinically undetected paroxysmal AF. Furthermore, whether BNP concentrations will be useful to predict risk of AF burden [duration and number of episodes] is a useful future research endeavor.

Offspring participants are almost exclusively of European ancestry, which may limit the generalizability of our findings towards other races/ethnicities. The mean age of the sample at baseline was 58.4 years and the strength of association of the risk factors and biomarkers may differ in younger individuals or patients with lone AF. Conversely, the risk score was derived from an ambulatory, community-based sample. We acknowledge that the generalizability to a referral-based sample, with a higher prevalence of heart failure is uncertain. The risk score may need to be recalibrated if the prevalence of AF risk factors varies substantively from that observed in the present sample. We had a modest number of AF cases; hence, we cannot exclude the possibility that with more events, biomarkers with more modest effect sizes also would have been related to AF onset. On the other hand, a larger number of AF cases might also have led to a regression towards the mean of BNP concentrations in individuals who developed AF.

Our present results will need confirmation in independent, prospective samples. The utility of the determination of BNP needs to be demonstrated, and potential preventive interventions need to be tested. The benefit of prediction algorithms ultimately depend on the demonstration of improved outcomes, i.e. the reduction in incidence of AF. At present, no strong preventive measures for AF have been established. BNP is an attractive candidate biomarker, but an observational study design cannot prove a causal relation. However, a better understanding of the relation of BNP to incident AF might provide valuable insights into the pathophysiology and help to identify targets for intervention. In the Framingham cohort, the correlation of N-ANP with BNP was moderately high at 0.66 and after incorporation of both natriuretic peptides into the model, BNP emerged as the stronger biomarker. Of note, the mean BNP concentrations in individuals developing AF were higher compared to individuals free of manifest AF at follow-up, but fell within the clinical range of normal BNP concentrations. Mildly elevated BNP thus shows a higher susceptibility for incident AF, yet our data clearly demonstrate that the measurement of BNP alone is not sufficient for AF risk evaluation. Our results can only suggest that BNP, in addition to careful assessment of clinical risk factors, may be able to refine risk prediction.

A major concern with respect to the validity of our reclassification results and conclusions remains the arbitrary choice of cut-offs for risk categories because to date, no established risk prediction scheme has been implemented for AF. We used the same risk classes as in the original publication of the risk algorithm.¹² A different definition of cut-points may result in changes in the net reclassification. For this reason we also provide data on integrated discrimination improvement that is not dependent on specified risk categories. The results of both analyses showed the same direction with a borderline improvement after the addition of BNP to the baseline model.

Strengths of the current study are the well-characterized community-based sample with routine ascertainment of clinical risk factors and potential confounders, strict quality control of biomarker measurements, continuous collection of information on outcomes over a comparatively long follow-up time frame, and rigorous ascertainment of incident AF cases. The availability of routine echocardiographic measures at the same exam cycle allowed us to explore mechanistic questions as to whether the relation of BNP to AF was mediated solely through cardiac remodeling. A great advantage of the present investigation is the ability to explore a broad range of pathophysiologically distinct biomarkers and to directly compare them for their strength of association.

In conclusion, the neurohormone BNP and the inflammatory biomarker CRP revealed significant associations with outcome in multivariable-adjusted analyses. BNP was the strongest single biomarker in relation to AF occurrence and significantly improved risk prediction beyond a risk score based on known clinical risk factors.

Clinical Perspective.

The prevalence of atrial fibrillation (AF) is expected to increase due to an aging population, improved therapies and longer survival with heart disease. Many pathophysiological pathways have been examined in animal and human studies in context with AF. We report the prospective association of a broad panel of blood and urinary biomarkers representing inflammation (C-reactive protein [CRP], fibrinogen), neurohormonal activation (B-type natriuretic peptide [BNP], N-terminal pro-atrial natriuretic peptide), oxidative stress and endothelial dysfunction (homocysteine), renin-angiotensin-aldosterone system (renin, aldosterone), and thrombosis (D-dimer, plasminogen-activator inhibitor), and microvascular damage (urine albumin excretion) in a community-based cohort with long-term incidence of AF. A recently published risk score for long-term incidence of AF combines well-established clinical risk factors for AF such as age, sex, body mass index, hypertension, and cardiovascular disease, including valvular disease and heart failure. We tested the predictive value of the strongest biomarkers in addition to the clinical variables combined in the risk algorithm. The neurohormone BNP and the inflammatory biomarker CRP revealed significant associations with outcome in multivariable-adjusted analyses. BNP was the strongest single biomarker in relation to AF occurrence and significantly improved risk prediction based on the risk algorithm. Whether determination of BNP contributes to strategies to prevent AF has to be established in future studies. Our findings may also provide valuable insights into the pathophysiology of AF.

Abbreviations

AF

atrial fibrillation

BNP

B-type natriuretic peptide

CI

Confidence Interval

CRP

C-reactive protein, HR, Hazard ratio

N-ANP

N-terminal pro-atrial natriuretic peptide

PAI-1

plasminogen-activator inhibitor-1