WIXLIB

Iron markers and new onset heart failure in the community(433) or cerebral arteries (434), coronary artery bypass grafting or percutaneous transluminal coronary angioplasty, and other vascular interventionsas percutaneous transluminal angioplasty or bypass grafting of aorta andperipheral vessels.Participants with a new diagnosis of HF were identified using criteriadescribed in the Heart Failure Guidelines of the European Society of Cardiology and an endpoint adjudication committee ascertained the diagnosisof HF, as described elsewhere.16 Additionally, HF was classified as HF witha reduced (HFrEF) or preserved ejection fraction (HFpEF) based on theleft ventricular ejection fraction at the time of diagnosis. To acknowledgethe most recent trends in cut-off for HFrEF and HFpEF in accordance withthe most recent HF guidelines, we set the cut-off for HFpEF at 50%.17 Inall subjects developing new onset heart failure in the present study data onLVEF was available.Baseline continuous variables are expressed as means with standard deviation (SD) when normally distributed, as medians with interquartile range(IQR) when distribution is skewed, or as numbers and percentages whencategorical. Inter-group differences were tested using trend-analyses. Forfurther analyses, skewed variables were transformed to a 2-log scale toachieve a normal distribution.By design, the PREVEND study over-selected subjects with an elevatedUAE ( 10mg/L). To overcome this over-selection of subjects with UAE, astatistical weighted method was applied in our regression analyses, allowing our conclusions to be generalized to the general population.18,19 The association of ferritin with baseline variables was examined by bootstrappingthe linear regression model 1000 times. Variables selected 700 times wereassumed to be accurate and included in the multivariate model.Overall and sex-stratified adjusted annual incidence rates of new onsetHF, CV events, and all-cause mortality were assessed assuming a Poissondistribution. The test for trend was calculated using the Mantel-Haenszelmethod. To preserve ferritin (and hepcidin) as continuous variables, multivariable fractional polynomials were employed to determine the best fittingfunctional form of both markers for all outcome parameters. To account fordeath as a competing risk for new onset HF, a subject’s follow-up time wascensored at the time of death whenever death occurred before the onset ofHF. Hazard ratios are then interpreted as the covariate effects on the causespecific (sub)hazard function of the transition from heart failure-free tonew onset HF in a competing risks model with new onset HF and all-cause5Statistical analyses105

mortality as the two absorbing states.20 Crude analyses were consecutivelyadjusted for 1) age and sex; 2) a multivariate model consisting of the following established HF and CV risk factors: age, sex, BMI, current smoking,diabetes, hypertension, hypercholesterolemia, history of MI or stroke, LVH,renal function, and levels of hemoglobin, erythropoietin, UAE, and presenceof inflammation (hs-CRP 5 mg/L); and 3) time-varying development of CVevents. Similar analyses were done for incident CV events and all-causemortality. All reported probability values are two-tailed and a value of P 0.05 was used as the nominal level of statistical significance. Models andanalyses were performed using STATA software version 13.0 (StataCorp LP,College station, Texas, USA).ResultsBaseline characteristics of the 6386 subjects, stratified per ferritin quartiles,are displayed in Table 1. Mean age was 53.1 12.0 years and 50.7% of thecohort was female. Median ferritin concentration was 97 (47 - 173) ug/L andwas higher in men (142 ug/L) compared to women (61 ug/L; P 0.001). Irondeficiency, defined by ferritin levels 30ug/L, was present in 928 subjects(14.5%) and more frequently observed in women (24.6% vs. 4.2% in men;P 0.001). Although postmenopausal women had significantly higher ferritin levels in relation to their premenopausal counterparts (94 ug/L vs. 36ug/L; P 0.001), iron deficiency was still more frequently seen when compared to men (8.7% vs. 4.2% in men; P 0.001).Per increasing ferritin quartile, individuals were generally older, moreoften male, and more likely to have cardiovascular risk factors (e.g. diabetes,hypertension). Additionally, higher concentrations of hemoglobin, hs-CRP,TSAT and hepcidin were observed, whereas renal function worsened andEPO concentrations declined. Baseline ferritin quartiles were additionally stratified by sex (Table S1A&1B). Similar results were detected for bothsexes, with the exception of prior MI and the presence of either LBBB orLVH, which were more frequently observed in women.Clinical associates of ferritin levelsIn both sexes, serum ferritin remained positively correlated with hemoglobinand MCV, other markers of iron homeostasis and glucose concentrations.Whereas ferritin was strongly related to cardiovascular risk factors (BMI,cholesterol) in men, inflammation and decreasing erythropoietin levelswere correlated with ferritin levels in women (Table 2). In a bootstrapped106

Iron markers and new onset heart failure in the communitymodel for ferritin (per doubling), these variables were highly selected. Thepresence of LVH (a strong HF risk factor) or LBBB was associated with ferritin levels in women (but not in men) in univariable regression only (TableS2). The multivariable model in women had a better model fit compared tomen (Table 2). Similar cardiovascular risk factors related to ferritin levels inmen and women without prior CV disease and adjusted R2 was comparable(Table S3).Markers of iron homeostasis and new onset heart failure.After a median follow-up of 8.3 (7.8-8.9), years, 199 individuals (3.1%) werediagnosed with new onset HF, 456 (7.1%) experienced a CV event and 356(5.6%) died from all-causes. Compared to other markers of iron status, ferritin had the best accuracy in predicting incident HF and other events (FigureS1). The adjusted annual incidence of new onset HF rose per increasing fer-5Figure 1. Total annual incidence of new onset heart failure according to ferritin (1A) and hepcidin(1B) quartiles and stratified by sex.107

percholesterolemia2.9Type 2 diabetes30.4Smoking or quit 1 yearMIMedical history (%)2.25.8LVH according to Cornell (%)68.5 10.0Heart rate (b.p.m.)LBBB (%)126.3 18.818.726.7 4.350.753.1 12.020.919.115.85.00.92.029.73.41.068.9 9.4119.5 17.112.725.5 4.382.048.3 10.927 (1–48)638697 (1–1925)Q1TotalSystolic BP (mmHg) 30 (%)BMI (kg/m2)Female sex (%)Age (years)Clinical signsFerritin (ug/L), min-maxnVariablesTable 1. Baseline characteristics according to ferritin quartiles.7.227.824.16.31.12.532.84.82.368.4 10.3125.0 18.715.526.2 4.356.653.3 12.071 (49–97)1579Q25.034.426.67.80.63.131.07.12.268.4 10.0128.5 18.119.126.9 4.240.554.6 11.7130 3 10.3132 18.627.628.1 4.123.156.4 11.7257 (173–1925)1595Q4 0.001 0.001 0.001 0.0010.68 0.0010.19 0.0010.0010.022 0.001 0.001 0.001 0.001 0.001NANAP - valuefor trend

3.27.8 (5.7–12.4)7.9 (5.8–13.9)5.589.9 17.683.4 16.94.7 (4.4–5.2)5.4 1.01.3 (0.6–2.9)7.7 (5.9–10.1)25.0 8.316.0 5.22.6 (1.8–3.5)90.7 4.113.7 1.1Q258.5 (6.1–16.0)6.788.2 17.486.5 19.05.0 (4.5–5.4)5.5 1.01.4 (0.7–3.2)7.5 (5.7–9.7)25.8 8.416.1 5.13.8 (2.8–5.1)90.8 4.214.0 1.1Q310.0 (6.7–20.3)7.786.2 17.790.4 23.65.1 (4.6–5.6)5.6 1.11.8 (0.8–3.6)7.3 (5.5–9.7)28.2 9.617.2 5.55.6 (4.2–8.0)91.3 4.414.3 1.1Q4 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001P - valuefor trendContinuous variables are presented as mean standard deviation or as median (inter-quartile range) when non-normally distributed. Binary categorical variables are presented aspercentages. Abbreviations: BMI Body mass index, BP Blood pressure, eGFR Estimated glomerular filtration rate, hs-CRP high-sensitive C-reactive protein, LBBB Left bundlebranch block, LVH Left ventricular hypertrophy, MCV Mean corpuscular volume, MI Myocardial infarction, TSAT Transferrin saturation, UAE Urinary albumin excretion.8.5 (6.0–15.1)5.7UAE (mg/24h) 60 (%)94.1 17.1eGFR (ml/min/1.73m2)78.9 14.184.8 19.189.6 17.7Creatinine (umol/L)4.6 (4.3–5.1)4.8 (4.4–5.3)5.1 1.01.0 (0.4–2.3)8.6 (6.5–12.0)20.6 9.5Glucose (mmol/L)5.4 1.01.4 (0.6–3.0)Cholesterol (mmol/L)7.8 (5.9–10.3)24.8 9.4TSAT (%)Erythropoietin (IU/L)15.9 5.6hs-CRP (mg/L)1.1 (0.6–1.7)3.1 (1.7–4.9)Hepcidin (nmol/L)Serum iron (umol/L)14.1 6.388.9 5.313.7 1.290.4 4.612.9 1.2Q1Hemoglobin (g/dL)TotalMCV (fL)Laboratory measurementsVariablesTable 1. Baseline characteristics according to ferritin quartiles. (continued)Iron markers and new onset heart failure in the community109

Table 2. Multivariable regression beta-coefficients between cardiovasular risk factors and serumferritin in men and women (selected after bootstrapping).Variables (selected afterbootstrapping)Menstruation (yes vs. no)BMI (per kg/m2)Smoking (yes vs. no)Women(n 3241)Men(n 3145)StandardizedßTPvalue---0.0905.4-- 0.001-Standardized ßT 0.129 7.4-- 0.035 2.6Pvalue 0.0010.011Hemoglobin (per g/dL)0.1086.5 0.0010.0714.1 0.001MCV (per fL)0.0996.0 0.0010.1247.6 0.001Log 2 hepcidin (nmol/L)0.68038.6 0.0010.62530.4 0.001TSAT (per 5%)0.0945.1 0.0010.1315.7 0.001--- 0.031 2.0-0.0593.5--0.0524.4Log 2 erythropoietin(IU/L)Log 2 hs-CRP (mg/L))--Cholesterol (mmol/L)0.0442.50.011Log 2 glucose (mmol/L)0.0433.6 0.001R2 total model0.60R2 total model0.042 0.001 0.0010.74For abbreviations, see Table 1.ritin quartile (Figure 1). This trend only appeared to be significant in women(P 0.001). Crude and multivariable fractional polynomials revealed theassociation between ferritin levels and the development of new onset HFto be best described by a linear function. In subsequent Cox regressionanalyses (with all-cause mortality as a competing risk) increasing ferritinconcentrations were strongly associated with the development of newonset HF (P 0.001; Table 3) However, this association was lost in multivariable analyses. Interaction analyses showed a significant interaction betweensex, ferritin levels and new onset HF (Pint 0.032). When stratified by sex, thisassociation persisted only in women (P 0.001) and remained significantin all multivariable models. Adjusting for menstrual status did not alter theresults. No interaction between menstrual status, ferritin levels and development of new onset HF was found in women (P 0.59). Additional subgroupanalyses in women demonstrated increasing ferritin levels to amplify risk forincident HF in almost every subgroup, but were significantly more predictive in women without anemia, without microalbuminuria and with an eGFR 60 ml/min/1.73m2.(Figure S2).Since body iron stores are firmly regulated at the systemic level by hepcidin, we additionally explored the role of hepcidin and risk of new onset HF.Similar observations regarding adjusted annual HF incidence were made forhepcidin, both overall and in women (Figure 1). Likewise, a strong associa-110

Iron markers and new onset heart failure in the communityTable 3. Levels of ferritin and hepcidin and risk for incident heart failure with all-cause mortalityas a competing risk.Ferritin (per doubling)sHR95%CIHarrell’s CP-value1Total population (n 6386)Univariable1.481.30–1.690.64 0.001Model 11.241.05–1.460.830.011Model 21.060.89 – 1.260.880.45Men (n 3145)Univariable1.000.81–1.230.520.98Model 11.060.87–1.290.820.54Model 20.920.75–1.140.870.45 0.001Women (n 3241)Univariable2.071.63–2.610.74Model 11.571.20–2.050.830.001Model 1A1.691.26–2.260.79 0.0011.351.02–1.790.870.038Model 31.391.05–1.860.870.024HR95%CIHarrell’s CP-valueHepcidin (per doubling)5Model 2Total population (n 6386)2Univariable1.451.23–1.720.61 0.001Model 11.200.98–1.450.830.07Model 21.090.92–1.300.880.39Men (n 3145)Univariable0.960.79–1.160.540.76Model 11.000.81–1.250.820.94Model 0.75 0.001Model 11.591.17–2.170.840.003Model 1A1.681.19–2.370.790.003Model 21.441.03–2.050.870.039Model 31.601.09–2.370.880.018Women (n 3241)Abbreviations: CI Confidence interval, sHR Sub-hazard ratio.1P for interaction ferritin, sex and new onset heart failure 0.032.2P for interaction hepcidin, sex and new onset heart failure 0.045.Model 1 is adjusted for age (per 10 years) and sex (in total population).Model 1A is adjusted for menstrual state instead of age.Model 2 is adjusted for model 1 cardiovascular and heart failure risk factors: BMI, current smoking, the presenceof diabetes, hypertension, hypercholesterolemia, history of MI or stroke, LVH, renal function, and levels of hemoglobin, erythropoietin, hs-CRP and UAE.Model 3 is adjusted for model 2 time-varying incident cardiovascular events.111

112Figure 2. Sex stratified risk for new onset heart failure according to the joint classification of ferritin levels and tertiles of TSAT (A), hepcidin (B), hs-CRP (C),BMI (D), glucose (E) or systolic blood pressure (F).

Iron markers and new onset heart failure in the community5tion between levels of hepcidin and the development of new onset heartfailure was observed in women only (Pint 0.045). Removing potential outliers (highest and lowest 1%) did not change the results. In subjects withoutprior myocardial infarction and/or stroke, both ferritin and hepcidin levelsremained strongly associated with the development of HF in women (TableS4). Sensitivity analyses in postmenopausal women showed ferritin - andnot hepcidin - to remain predictive for new-onset HF (Table S5).To assess whether the association between body iron stores and risk ofnew onset HF were modified by other markers of iron homeostasis (TSAT,hepcidin), inflammation (hs-CRP) or the metabolic syndrome (BMI, glucose,systolic blood pressure), the joint association of ferritin levels with eitherone of these 6 variables and the development of HF was examined. In thesejoint analyses, ferritin levels remained independently associated with newonset HF in women in almost all stratified analyses.(Figure 2).Finally, 79 of the 199 subjects newly diagnosed with HF were women.Of these women, 37 developed HFrEF and 42 HFpEF. Unadjusted causespecific hazard analyses demonstrated serum ferritin to be associated witheither HFrEF or HFpEF, with no significant interaction between both entities(Table S6). Multivariable analyses showed that levels of ferritin remainedstrongly associated with an increased risk for HFpEF only. However, againno significant interaction was found.Markers of iron homeostasis and risk for cardiovascular diseaseand all-cause mortality.A higher total annual incidence of new CV events and all-cause mortalitywas observed per increasing ferritin and hepcidin quartile (Figure S3). Whenstratified by sex, these observations only persisted in women. Multivariablefractional polynomials again showed the best fitting functional form between levels of ferritin or hepcidin and new CV events to be linear. Cruderegression analyses showed increasing ferritin and hepcidin concentrationsto be both strongly predictive for new CV events and all-cause mortality inthe overall population and in women (both P 0.001). However, age- andmultivariable analyses were non-significant. No interaction between variables was demonstrated.DiscussionIn this contemporary, prospective, population-based cohort, increasingferritin levels amplify the risk for new onset HF in women, but not in men.113

This relationship is independent of established conventional CV and HF riskfactors and the development of CV events over time. These observationsprovide evidence that increased ferritin levels are an independent risk factorfor the development of HF in women and might play a direct or indirect rolein the pathogenic mechanism leading to HF.How do elevated iron stores relate to incident heart failure?Whereas iron deficiency has detrimental effects on outcome in establishedheart failure,17,21,22 the role of iron markers or markers of erythropoiesis in thedevelopment of HF has only been investigated to some extent.23,24 Previousliterature suggested a U-shaped relationship between iron stores and newonset HF.3,4.In the present study, we observed a linear relationship betweenincreasing ferritin levels and the development of HF. Excess body iron isknown to accumulate in organs such as the heart and lead to cardiomyopathy. The generation of reactive oxygen species (ROS) in mitochondriais believed to be the most important pathogenic pathway determiningcardiomyocyte damage. Production of ROS causes lipid peroxidation andDNA damage, leading to cell death, fibrosis, and eventually cardiac dysfunction. However, iron overload is only observed when the binding capacity oftransferrin is fully saturated and non-transferrin bound iron is formed. Interestingly, joint analyses showed an association between serum ferritin andnew onset HF in all TSAT tertiles, suggesting that another pathway linkingincreased ferritin levels with cardiomyocyte damage may also be involved.Iron-mediated cell damage does not only occur under conditions of ironoverload. It has been proposed that iron maldistribution among organs,tissues and cellular compartments can also attenuate cell integrity and celllife. Dysmetabolic hyperferritinemia, meaning elevated ferritin levels withnormal TSAT, is commonly observed in clinical practice and has been associated with insulin resistance, obesity, hypertension, and other manifestationsof the metabolic syndrome.25 During circumstances of subclinical systemicinflammation, iron efflux in organs may be halted due to increased hepcidinproduction and subsequent internalization and degradation of its receptor,iron exporter ferroportin.26 Therefore, elevated ferritin levels (and hepcidin)might be a reflection to a low-grade inflammatory response of anotherpathophysiological process (e.g. metabolic syndrome), causally responsiblefor the development of HF.27 Supporting this hypothesis is the fact that weobserved a posit

Iron is an essential micronutrient for a wide range of metabolic and bio-logical processes. Systemic iron homeostasis is orchestrated by hepcidin, a liver-derived peptide that has emerged as the key circulating regulator of iron absorption and tissue distribution.1 In response to low body iron