ESC HEART FAILURE ORIGINAL RESEARCH ARTICLE ESC Heart Failure 2020; 7: 2223–2229 Published online 26 June 2020 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/ehf2.12621 Growth differentiation factor 15 as mortality predictor in heart failure patients with non-reduced ejection fraction Ana Belen Mendez Fernandez1, Andreu Ferrero-Gregori1, Alvaro Garcia-Osuna2, Sonia Mirabet-Perez1, Maria Jose Pirla-Buxo1, Juan Cinca-Cuscullola1, Jordi Ordonez-Llanos2,3 and Eulàlia Roig Minguell1* 1 Department of Cardiology, CIBERCV, Hospital Santa Creu i Sant Pau, Barcelona, Spain; 2Department of Biochemistry, IIB-Sant Pau, Barcelona, Spain; 3Department of Biochemistry and Molecular Biology, Universitat Autònoma, Barcelona, Spain Abstract Aims The prognostic value of biomarkers in patients with heart failure (HF) and mid-range (HFmrEF) or preserved ejection fraction (HFpEF) has not been widely addressed. The aim of this study was to assess whether the prognostic value of growth differentiation factor 15 (GDF-15) is superior to that of N-terminal pro-brain natriuretic peptide (NT-proBNP) in patients with HFmrEF or HFpEF. Methods and results Heart failure patients with either HFpEF or HFmrEF were included in the study. During their first visit to the HF unit, serum samples were obtained and stored for later assessment of GDF-15 and NT-proBNP concentrations. Patients were followed up by the HF unit. The main endpoint was all-cause mortality. A total of 311 patients, 90 (29%) HFmrEF and 221 (71%) HFpEF, were included. Mean age was 72 ± 13 years, and 136 (44%) were women. No differences were found in GDF-15 or NT-proBNP concentrations between both HF groups. During a median follow-up of 15 months (Q1–Q3: 9–30 months), 98 patients (32%) died, most (71%) of cardiovascular causes. Patients who died had higher median concentrations of GDF-15 (4085 vs. 2270 ng/L, P < 0.0001) and NT-proBNP (1984 vs. 1095 ng/L, P < 0.0001). A Cox multivariable model identified New York Heart Association Functional Class III (P = 0.04), systolic blood pressure (P = 0.01), left atrial diameter (P = 0.03), age >65 years (P < 0.0001), and GDF-15 concentrations (P = 0.01) but not NT-proBNP as independent predictors of all- cause mortality. The area under the curve was 0.797 for the basic model including NT-proBNP, and the area under the curve comparing the overall model was 0.819, P = 0.016 (DeLong’s test). Integrated discrimination improvement index after the in- clusion of GDF-15 in the model with the mortality risk factors was 0.033; that is, the ability to predict death increased by 3.3% (P = 0.004). Net reclassification improvement was 0.548 (P < 0.001); that is, the capacity to improve the classification of the event (mortality) was 54.8%. GDF-15 concentrations were divided in tertiles (<1625, 1625–4330, and >4330 ng/L), and sur- vival curves were evaluated using the Kaplan–Meier technique. Patients in the highest tertile had the poorest 5 year survival, at 16%, whereas the lowest tertile had the best survival, of 78% (P < 0.001). Conclusions Growth differentiation factor 15 was superior to NT-proBNP for assessing prognosis in patients with HFpEF and HFmrEF. GDF-15 emerges as a strong, independent biomarker for identifying HFmrEF and HFpEF patients with worse prognosis. Keywords GDF-15; HFmrEF; HFpEF; Chronic heart failure; Biomarkers Received: 8 July 2019; Revised: 4 December 2019; Accepted: 3 January 2020 *Correspondence to: Eulàlia Roig Minguell, Department of Cardiology, CIBERCV, Hospital Santa Creu i Sant Pau, Carrer de Sant Quintí, 89, Barcelona 08041, Spain. Tel: +34 93 553 59 58; Fax: +34 93 553 74 89. Email: eroigminguell@gmail.com © 2020 The Authors. ESC Heart Failure published by John Wiley & Sons Ltd on behalf of the European Society of Cardiology This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. 2224 A.B. Mendez Fernandez et al. Introduction Methods Until recently, heart failure (HF) was classified according left Patients meeting HF diagnosis criteria according the Euro- ventricular ejection fraction (LVEF) in HF with reduced or pre- pean Society of Cardiology guidelines,2 referred to our HF served LVEF fraction, although a ‘grey zone’ existed between unit (HFU) between 2010 and 2015, were included in the both HF phenotypes.1 The 2016 update of the European Soci- study. Patients were referred after an episode of decompen- ety of Cardiology guidelines defined a new group, HF with sated HF, either after an emergency room visit for dyspnoea mid-range EF (HFmrEF), as the HF with an LVEF between or hospital admission for HF; in both cases, HF was the final 40% and 49% and relevant structural heart disease or ele- diagnosis, and all patients were treated with diuretics. All pa- vated concentrations of natriuretic peptides.2 Thus, HFmrEF tients underwent echocardiography and were classified ac- shares some characteristics with both HF with preserved EF cording to EF as HFrEF, HFmrEF, or HFpEF. Patients with (HFpEF) and HF with reduced EF (HFrEF) and defines a group reduced EF were excluded from the study. Of all patients in- with an uncertain prognosis.3–6 HF is a condition resulting cluded, 276 patients had elevated natriuretic peptides [N-ter- from several cardiac processes, such as ischaemia, necrosis, minal pro-brain natriuretic peptide (NT-proBNP) >300 ng/L], stretch, neurohormonal activation, volume overload, inflam- while the remaining 35 patients, despite lower concentra- mation, and oxidation, and all these processes have their re- tions of NT-proBNP, had relevant structural heart disease spective biomarkers. As mentioned, HFpEF and HFmrEF (nine had left ventricular hypertrophy, 17 left atrial enlarge- patients share some characteristics, including elevated natri- ment, and four implanted pacemaker, and five were in atrial uretic peptide concentrations. For this reason, some studies fibrillation). One hundred and fourteen patients had hyper- have analysed biomarkers other than natriuretic peptides tensive cardiomyopathy, 77 ischaemic heart disease, 25 hy- for improving the prognosis of patients with both HF pheno- pertrophic cardiomyopathy, and 83 valve disease. Other types,7 but results were inconclusive. The lack of clear differ- miscellaneous aetiologies were three patients with previous ences in the analysed biomarkers and the difficulty in using cancer treatment, two with non-compaction cardiomyopa- some of them in daily practice justifies the search for new thy, and five with atrial fibrillation, all of whom had mild left prognostic biomarkers for both HF phenotypes. ventricular dysfunction. The remaining two patients had con- Growth differentiation factor 15 (GDF-15), also known as strictive pericarditis. macrophage inhibitory cytokine-1 and non-steroidal anti- During the first visit to the HFU, signs and symptoms were inflammatory drug-activated gene-1, is a member of the registered and blood was drawn. Serum samples were ob- transforming growth factor-β cytokine superfamily with im- tained after centrifugation and stored at 80 °C until analy- munosuppressive, anti-apoptotic, and anti-inflammatory sis. NT-proBNP and GDF-15 were measured by properties. GDF-15 is weakly expressed in healthy human electrochemiluminescence immunoassays on a Cobas e601 tissues, with the exception of the placenta. GDF-15 expres- platform (Roche Diagnostics, Basel, Switzerland). The mea- sion is up-regulated by p53, which in turn is a factor surement ranges for NT-proBNP and GDF-15 were 5–35 000 responding to daily-life stressors, such as inflammation, and 400–20 000 ng/L, and precision was ≤3.5% and ≤4.9%, hypoxia, and oxidative stress.8 Given the plurality of causes respectively, according to the manufacturer. Values reaching that increase GDF-15, its circulating concentrations have the top of the concentration ranges were equivalent to the been related with an increased risk of all-cause, cardiovas- maximum detectable value (i.e. 35 000 ng/L for NT-proBNP cular, and non-cardiovascular mortality, cardiovascular and 20 000 ng/L for GDF-15). Patients were followed in the events such as myocardial infarction, HF hospitalization or HFU, and all-cause mortality was registered. The study was stroke, and major bleeding in atrial fibrillation in community conducted in accordance with the principles outlined in the studies and in patients with stable or acute coronary dis- Declaration of Helsinki. All participants received information ease..9 However, GDF-15 concentrations have been mea- about the study and signed informed consent. The study sured in these studies using different methods, the was approved by the centre’s ethics committee. designs of which promoted a heterogeneity in GDF-15 con- centrations that was found to be associated with adverse outcomes. Recently, a fully automatic Statistical analyses electrochemiluminescence immunoassay has been made available for clinical use, which will facilitate the measure- Categorical variables were expressed as frequencies and per- ment of large number of samples with better analytical pre- centages. Continuous variables were expressed as mean ± cision than associated with previous methods. standard deviation or median (Q1–Q3, first and third quar- The aim of our study was to assess the potential of GDF- tiles), according to their Gaussian or non-Gaussian distribu- 15, measured by a fully automated immunoassay, in tion. The statistical differences between groups were predicting all-cause mortality in a large group of HFmrEF compared using the χ 2 test for categorical variables and the and HFpEF patients. Student’s t-test or the Mann–Whitney U test according to ESC Heart Failure 2020; 7: 2223–2229 DOI: 10.1002/ehf2.12621 GDF-15: a new prognostic biomarker in HFpEF and HFmrEF 2225 the Gaussian or non-Gaussian distribution of the continuous haemoglobin, NT-proBNP, and GDF-15 concentrations. A variables. Survival analysis was performed using proportional backward elimination method was then used to identify inde- hazards regression models dividing GDF-15 into tertiles. The pendent predictors of all-cause mortality. To evaluate the po- proportionality of the risks was tested using Schoenfeld resid- tential value of measured biomarkers in a prediction model of uals. The survival curves were also evaluated with the mortality risk, patients with HFmrEF and HFpEF were consid- Kaplan–Meier technique for the values of GDF-15 (analysed ered together, given the similarities found between both by tertiles), and the differences in survival were tested with groups. The potential value of measured biomarkers in a pre- the log-rank test. Several variables with known power or diction model of mortality risk, which included the indepen- power determined in the univariate analysis to identify out- dent predictors of mortality found in our cohort, as a base comes were entered in the multivariate analysis. Variables risk multivariable model, in addition to NT-proBNP, was eval- were age, HF aetiology, New York Heart Association (NYHA) uated by several methods. The Grønnesby and Borgan functional class, systolic blood pressure, atrial fibrillation, left goodness-of-fit test was used to determine if the proposed atrial diameter, estimated glomerular filtration rate, model fitted the observed outcome with the expected Table 1 Main characteristics of all patients and according to left ventricular ejection fraction: mid-range (40–49%) and preserved (≥50%) All HFmrEF HFpEF n = 311 n = 90 n = 221 P-value Age (years) 72 ± 13 67 ± 14 73 ± 12 <0.001 Gender (male) 175 (56%) 66 (73%) 109 (49%) <0.001 Aetiology <0.001 Hypertensive cardiomyopathy 114 (37%) 25 (28%) 89 (40%) 0.038 Ischaemic heart disease 77 (25%) 38 (42%) 39 (18%) <0.001 HCM 25 (8%) 4 (4%) 21 (10%) 0.100 Valve disease 83 (27%) 18 (20%) 65 (29%) 0.089 Miscellaneous 12 (4%) 5 (6%) 7 (3%) 0.339 Hypertension 240 (77%) 65 (72%) 175 (79%) 0.185 Diabetes mellitus 114 (37%) 32 (36%) 82 (37%) 0.797 Dyslipidaemia 150 (48%) 44 (49%) 106 (48%) 0.882 Atrial fibrillation 152 (49%) 34 (38%) 118 (53%) 0.01 Weight (kg) 77 ± 17 78 ± 15 76 ± 17 0.439 2 BMI (kg/m ) 29 ± 5 28 ± 5 29 ± 6 0.175 Blood pressure (mmHg) Systolic 131 ± 21 132 ± 20 127 ± 20 0.03 Diastolic 75 ± 11 75 ± 11 75 ± 10 0.818 NYHA-FC II 188 (60%) 61 (68%) 127 (58%) 0.241 III 123 (39%) 29 (32%) 94 (43%) 2 eGFR (mL/min/1.73 m ) 59 ± 24 62 ± 25 58 ± 23 0.247 Haemoglobin (g/L) 127 ± 19 128 ± 21 127 ± 18 0.515 NT-proBNP (ng/L) 1346 1362 1315 0.844 (median Q1–Q3) (542–2651) (658–2968) (536–2632) GDF-15 (ng/L) 2822 2748 2822 0.520 (median Q1–Q3) (1631–4378) (1218–5253) (1695–4176) Echocardiography LVEF (%) 58 ± 12 44 ± 3 64 ± 9 <0.001 LVEDD (mm) 50 ± 8 55 ± 7 47 ± 7 <0.001 IVST (mm) 13 ± 3 12 ± 3 14 ± 4 0.01 LAD (mm) 50 ± 10 49 ± 8 51 ± 9 0.104 MR (Grade 3–4) 40 (13%) 10 (11%) 30 (14%) 0.564 msPAP ≥ 40 mmHg 120 (39%) 27 (43%) 93 (52%) 0.223 msPAP (mmHg) 40 ± 16 38 ± 16 40 ± 15 0.789 E/A 1.2 ± 0.7 1.3 ± 0.8 1.2 ± 0.6 0.422 Deceleration time (ms) 221 ± 76 221 ± 84 220 ± 72 0.168 Treatment ACEIs/ARBs 229 (74%) 77 (86%) 162 (69%) 0.002 Beta-blockers 201 (65%) 75 (83%) 126 (57%) <0.001 Loop diuretics 261 (84%) 68 (76%) 193 (87%) 0.01 MRA 110 (35%) 41 (46%) 69 (31%) 0.01 ACEIs, angiotensin-converting enzyme inhibitors; ARBs, angiotensin receptor blockers; BMI, body mass index; E/A, early (E) mitral inflow peak/atrial (A) filling peak ratio; eGFR, estimated glomerular filtration rate; GDF-15, growth differentiation factor 15; HCM, hypertrophic cardiomyopathy; HFmrEF, heart failure with mid-range ejection fraction; HFpEF, heart failure with preserved ejection fraction; IVST, inter- ventricular septum thickness; LAD, left atrial diameter; LVEDD, left ventricular end-diastolic diameter; LVEF, left ventricular ejection frac- tion; MR, mitral regurgitation; MRA, mineralocorticoid receptor antagonist; msPAP, mean estimated systolic pulmonary artery pressure; NT-proBNP, N-terminal pro-brain natriuretic peptide; NYHA-FC, New York Heart Association functional class. ESC Heart Failure 2020; 7: 2223–2229 DOI: 10.1002/ehf2.12621 2226 A.B. Mendez Fernandez et al. calibration capacity of the model. The concordance index (C- Table 3 Main characteristics of patients grouped according the oc- currence of all-cause mortality statistic) checked the discrimination capacity of the model. The improvement in the discrimination capacity of the model Alive Died P- n = 213 n = 98 value including the biomarkers was assessed by the integrated dis- crimination improvement index, which evaluated the changes Gender (male) 119 (44%) 56 (32%) 0.833 Aetiology in the prediction probabilities of mortality estimated. The net Hypertensive cardiomyopathy 78 (37%) 36 (37%) 0.984 reclassification improvement, category-free version, was Ischaemic 54 (25%) 23 (23%) 0.721 used to test changes in the prediction of estimated mortality HCM 19 (9%) 6 (6%) 0.399 Valve disease 54 (25%) 29 (30%) 0.432 implying a change from one category to another.10 Finally, Miscellaneous 8 (4%) 4 (4%) 1.000 the receiver operating characteristic curve of global mortality Blood pressure (mmHg) was calculated, comparing GDF-15 and NT-proBNP. A value of Systolic 134 ± 22 125 ± 19 0.002 Diastolic 76 ± 12 73 ± 11 0.076 P < 0.05 was considered statistically significant. Statistical Hypertension 158 (74%) 82 (84%) 0.081 software SPSS Version 24.0 (IBM Corp., Released 2016, IBM Diabetes mellitus 78 (37%) 36 (37%) 1.000 SPSS Statistics for Windows, Version 24.0, Armonk, NY) and Dyslipidaemia 104 (49%) 46 (47%) 0.807 Atrial fibrillation 88 (41%) 64 (65%) 0.0001 R Version 3.3.2 (R Foundation for Statistical Computing, Vi- NYHA-FC enna, Austria) were used to perform all statistical analyses. II 137 (64%) 51 (52%) 0.002 III 76 (36%) 47 (48%) 2 eGFR (mL/min/1.73 m ) 64 ± 23 49 ± 21 0.0001 Haemoglobin (g/L) 129 ± 18 123 ± 19 0.01 Results NT-proBNP (ng/L) 1095 1984 0.0001 (median Q1–Q3) (145–2188) (880–4852) GDF-15 (ng/L) 2270 4085 0.0001 Three hundred and eleven patients were included in the (median Q1–Q3) (1403–3596) (2554–6756) study, of whom 221 had HFpEF and the remaining 90 Echocardiography HFmrEF; their main characteristics are shown in Table 1. Pa- LVEF (%) 57 ± 12 60 ± 11 0.02 LVEDD (mm) 50 ± 8 49 ± 8 0.455 tients with HFmrEF were younger and had lower left ventric- IVST (mm) 13 ± 3 14 ± 4 0.626 ular EF, more dilated left ventricles, and a higher incidence of LAD (mm) 48 ± 8 52 ± 10 0.001 ischaemic heart disease than patients with preserved EF. Of MR (Grade 3–4) 24 (11%) 16 (16%) 0.197 msPAP ≥ 40 mmHg 73 (45%) 47 (60%) 0.03 note, concentrations of GDF-15 [2748 ng/L (HFmrEF) vs. msPAP (mmHg) 38 ± 15 43 ± 16 0.01 2822 ng/L (HFpEF)] and NT-proBNP [1362 ng/L (HFmrEF) vs. E/A 1.1 ± 0.7 1.6 ± 0.9 0.01 1315 ng/L (HFpEF)] did not differ statistically between both Deceleration time (ms) 225 ± 76 198 ± 69 0.734 Treatment groups of patients (Table 1). During follow-up (median 15 ACEIs/ARBs 160 (75%) 69 (70%) 0.407 months, Q1–Q3: 9–30 months), 98 (32%) patients died (71% Beta-blockers 149 (70%) 52 (53%) 0.004 of cardiovascular causes); the main cause of death was refrac- Loop diuretics 166 (78%) 95 (97%) 0.0001 MRA 65 (30%) 45 (46%) 0.008 tory HF (Table 2). Table 3 shows the comparison of clinical characteristics of patients who died during follow-up vs. sur- ACEIs, angiotensin-converting enzyme inhibitors; ARBs, angioten- vivors. Patients who died were older and had lower systolic sin receptor blockers; BMI, body mass index; E/A, early (E) mitral in- flow peak/atrial (A) filling peak ratio; eGFR, estimated glomerular blood pressure, a higher incidence of atrial fibrillation, worse filtration rate; GDF-15, growth differentiation factor 15; HCM, hy- NYHA functional class, and lower eGFR and haemoglobin than pertrophic cardiomyopathy; IVST, interventricular septum thick- those who survived. Furthermore, they had higher GDF-15 ness; LAD, left atrial diameter; LVEDD, left ventricular end- diastolic diameter; LVEF, left ventricular ejection fraction; MR, mi- and NT-proBNP concentrations than patients with an tral regurgitation; MRA, mineralocorticoid receptor antagonist; msPAP, mean estimated systolic pulmonary artery pressure; NT- proBNP, N-terminal pro-brain natriuretic peptide; NYHA-FC, New Table 2 Causes of death York Heart Association functional class. n (%) uneventful follow-up: GDF-15: 4085 (inter-quartile range: CV death Heart failure 59 (60%) 2554–6651) ng/L vs. 2254 (inter-quartile range: 1389–3562) Arrhythmia 6 (6.1%) ng/L (±2674 ng/L), P < 0.0001, and NT-proBNP: 3613 ng/L AMI 2 (2.0%) (±5629 ng/L) vs. 1835 ng/L (±3449 ng/L) expressed as median CVA 3 (3.0%) Non-CV death and standard deviation, P < 0.0001. Multivariable analyses Neoplasia 7 (7.3%) identified NYHA Functional Class III (P = 0.04), systolic blood Infection 6 (6.1%) pressure (P = 0.01), left atrial diameter (P = 0.03), age >65 Renal failure 4 (4.0%) Aortic aneurism 2 (2.4%) years (P < 0.0001), and GDF-15 concentrations (P = 0.01) Vascular surgery 2 (2.4%) but not NT-proBNP as independent predictors of all-cause Unknown 7 (7.3%) mortality (Table 4). The area under the curve was 0.797 for AMI, acute myocardial infarction; CV, cardiovascular; CVA, cerebro- the basic model including NT-proBNP, and the area under vascular accident. the curve comparing the whole model was 0.819, P = 0.016 ESC Heart Failure 2020; 7: 2223–2229 DOI: 10.1002/ehf2.12621 GDF-15: a new prognostic biomarker in HFpEF and HFmrEF 2227 Table 4 Multivariable analyses for all-cause mortality 95% confidence interval Hazard ratio Lower Upper Significance NYHA-FC III 1.645 1.013 2.670 0.044 SBP 0.983 0.971 0.996 0.011 LAD 1.020 1.002 1.038 0.031 Age 1.072 1.043 1.102 0.000 GDF-15 per 100 ng/L 1.008 1.003 1.013 0.01 NT-proBNP 1.000 1.000 1.000 0.328 per 100 ng/L GDF-15, growth differentiation factor 15; GDF-15 per 100 ng/L, risk per each 100 ng/L increase; LAD, left atrial diameter; NT-proBNP, N- terminal pro-brain natriuretic peptide; NT-proBNP per 100 ng/L, risk per each 100 ng/L increase; NYHA-FC, New York Heart Association functional class; SBP, systolic blood pressure. Grønnesby and Borgan test shows a good model calibration: P = 0.769. (DeLong’s test). Integrated discrimination improvement index worse prognosis. Moreover, reduced functional capacity, after the inclusion of GDF-15 in the model with the mortality low systolic blood pressure, higher left atrial diameter, older risk factors was 0.033; that is, the ability to predict death in- age, and higher GDF-15 concentration were the clinical vari- creased by 3.3% (P = 0.004). Net reclassification improvement ables identified as predictors of mortality. Thus, GDF-15 was was 0.548 (P < 0.001); that is, the capacity to improve the the only biomarker with prognostic value identified in these classification of the event (mortality) was 54.8%. GDF-15 con- patients. Natriuretic peptides are recommended in clinical centrations were divided in tertiles (<1625, 1625–4330, and practice for assessing the diagnosis of HF independently of >4330 ng/L) and the respective survival compared by LVEF.2,10 Moreover, they are also recommended for the as- Kaplan–Meier survival curves. Statistically significant differ- sessment of prognosis in patients with HFrEF. However, their ences in 5 year survival were found among the three groups. predictive value in patients with HFmrEF or HFpEF is less Patients in the highest tertile had the poorest 5 year survival, clear,11 and several factors can account for this controversy. at 16%, whereas in the lower tertile, survival was 78% (P < In the case of NT-proBNP, different cut-off values according 0.001); survival among patients in the intermediate tertile to patient age have been suggested, but this could compli- (39%) differed statistically from the lower (P = 0.006) and cate individual risk evaluation and stratification. Furthermore, higher (P < 0.001) tertiles (Figure 1). previous studies used different LVEF values to classify pa- tients with low or preserved EF. In fact, HFmrEF is a very het- erogeneous group of patients that can include patients with recovered EF after previous reduced ventricular function Discussion and patients with persistent mild deterioration. In a recent publication of the Swedish HF registry,12 which divided pa- Our results suggest that GDF-15 is a useful biomarker for tients according to the new HF classification, the prognostic identifying HF patients with non-reduced EF associated with value of elevated NT-proBNP was maintained independently of EF. Similarly, increased BNP concentrations, although Figure 1 Kaplan–Meier curves showing cumulative survival according to lower in patients with HFpEF than in patients with HFrEF, tertiles of growth differentiation factor 15 (GDF-15) concentrations. were also associated with worse prognosis.13 Despite these results, new biomarkers are now available,14 which can im- prove prognosis prediction in these patients. Previous studies have shown that GDF-15 is elevated in HFpEF or HFmrEF pa- tients to a similar degree as in HFrEF patients.15 Increased GDF-15 has been shown to be independently associated with exercise capacity impairment and poor quality of life,16 and its accuracy has been at least as good as that of NT-proBNP, all suggesting that the combination of both markers could im- prove diagnostic power. GDF-15 is secreted from the myocar- dium in response to various stimuli such as ischaemia, wall stress, or pressure overload.17 It is also secreted from fibro- blasts in response to stress stimuli.18 Thus, although the mechanisms are still unclear, both cardiomyocytes and heart fibroblasts could be the source of GDF-15 in patients with HF. Because cardiac hypertrophy and interstitial fibrosis are com- mon pathological features of HFpEF and HFmrEF, the ESC Heart Failure 2020; 7: 2223–2229 DOI: 10.1002/ehf2.12621 2228 A.B. Mendez Fernandez et al. prognostic value of GDF-15 found in these patients is not en- independent predictor of all-cause mortality in patients with tirely surprising. Furthermore, GDF-15 is also a marker of in- LVEF > 40%. Furthermore, in our cohort, GDF-15 was supe- flammation and metabolic syndrome frequently seen in rior to NT-proBNP in the prognostic assessment of patients patients with HF. The association of metabolic co-morbidities with HFpEF and HFmrEF. such as obesity and diabetes can induce coronary microvas- cular inflammation. These metabolic changes can increase Limitations cardiomyocyte stiffness because of limited availability of ni- tric oxide and induce fibrosis because of myocardial infiltra- The low number of patients recruited from a single centre tion by activated macrophages. Experimental studies have may be the main limitation of the study. However, it is one also demonstrated that GDF-15 null mice show exacerbated of the largest studies to report the use of GDF-15 for prog- cardiac hypertrophy in response to pressure overload, sug- nostic assessment, and internal validation was applied. De- gesting that GDF-15 has antihypertensive properties.17 How- spite the fact that patients with preserved or mid-range EF ever, no significant correlation between serum GDF-15 level were analysed together, there were no significant differences and left ventricular mass index has been found in a recent in biomarker concentrations between them. Furthermore, study.19 In a recent study of 70 patients with HFpEF, elevated the cause of HF in our study was heterogeneous, so extreme GDF-15 was associated with higher BNP values and identified care should be taken when interpreting these results. as an independent predictor of adverse cardiovascular out- comes.20 Similarly to our results, in the study by Lok et al.21 Clinical perspectives that included patients with advanced HF, GDF-15 was also su- perior to NT-proBNP in assessing prognosis. However, in that New biomarkers are needed to identify high-risk patients study, patients were selected according to NYHA Functional with HFmrEF and HFpEF to improve their prognosis. Higher Class III–IV, but no information was provided on their ventric- GDF-15 values are associated with higher mortality in pa- ular function status. Unlike other studies, we did not find any tients with HFpEF and HFmrEF. However, more studies are prognostic power for NT-proBNP in patients with HFpEF and necessary before the implementation of GDF-15 in clinical HFmrEF, perhaps because the clinical variables identified as practice can be recommended. Our preliminary results should predictors of worse prognosis are variables tightly correlated be corroborated in larger cohorts. with NT-proBNP concentrations, and consequently, the bio- marker may have lost statistical significance for predicting events. However, because GDF-15 is a marker of fibrosis and inflammation, it may play a more important role in the Acknowledgements pathophysiology of HFpEF and HFmrEF and consequently in the prognosis of these conditions. Recent studies have also We thank Roche Diagnostics GmbH, Mannheim, Germany, for suggested that a model of several biomarkers may provide providing the kits for GDF-15 measurement, supported by an higher prognostic information in HF patients,22 but these institutional research grant. We gratefully acknowledge the models are expensive and not widely available. Thus, GDF- assistance of Heather Hamilton with the grammar text 15 concentrations add prognostic value to clinical information correction and emerge as a new biomarker that may help to assess prognosis in patients with HF. In our study, we measured GDF-15 with a fully automatic immunoassay that showed re- markably low analytical imprecision. In previous studies, GDF- Conflict of interest 15 was mainly measured with radioimmunometric23 or None declared. enzyme-linked immunosorbent assays20; these analytical ap- proaches usually show higher imprecision than fully auto- mated methods. Indeed, the method we used is developed Funding on an immunoassay platform available in many clinical labo- ratories, thus facilitating the transferability of results among This study was supported by Fundació d’Investigació Sant Pau different studies. In conclusion, GDF-15 emerged as an (G-60136934). References 1. Lam CSP, Solomon SD. The middle child range ejection fraction (40–50%). Eur J 2. Ponikowski P, Voors AA, Anker SD, in heart failure: heart failure with mid- Heart Fail 2014; 16: 1049–1055. Bueno H, Cleland JGF, Coats AJS, Falk ESC Heart Failure 2020; 7: 2223–2229 DOI: 10.1002/ehf2.12621 GDF-15: a new prognostic biomarker in HFpEF and HFmrEF 2229 V, González-Juanatey JR, Harjola VP, Hasdai D, Hoes A, Kirchhof P, Knuuti J, normal ejection fraction. Eur J Heart Fail Jankowska EA, Jessup M, Linde C, Kolh P, McDonagh T, Moulin C, Popescu 2010; 12: 1309–1316. Nihoyannopoulos P, Parissis JT, Pieske BA, Reiner Z, Sechtem U, Sirnes PA, 17. Xu J, Kimball TR, Lorenz JN, Brown DA, B, Riley JP, Rosano GM, Ruilope LM, Tendera M, Torbicki A, Vahanian A, Bauskin AR, Klevitsky R, Hewett TE, Ruschitzka F, Rutten FH, van der Meer Windecker S, McDonagh T, Sechtem U, Breit SN, Molkentin JD. GDF15/MIC-1 P, Authors/Task Force Members., Docu- Bonet LA, Avraamides P, Ben Lamin functions as a protective and ment Reviewers. 2016 ESC guidelines HA, Brignole M, Coca A, Cowburn P, antihypertrophic factor released from for the diagnosis and treatment of acute Dargie H, Elliott P, Flachskampf FA, the myocardium in association with and chronic heart failure. Eur J Heart Guida GF, Hardman S, Iung B, Merkely SMAD protein activation. Circ Res 2006; Fail 2016; 18: 891–975. B, Mueller C, Nanas JN, Nielsen OW, 98: 342–350. 3. Altaie S, Khalife W. The prognosis of Orn S, Parissis JT, Ponikowski P, ESC 18. Akiyama M, Okano K, Fukada Y, Okano mid-range ejection fraction heart fail- Committee for Practice Guidelines. ESC T. Macrophage inhibitory cytokine MIC- ure: a systematic review and meta- guidelines for the diagnosis and treat- 1 is upregulated by short-wavelength analysis. ESC Heart Fail J 2018; 5: ment of acute and chronic heart failure light in cultured normal human dermal 1008–1016. 2012. Eur J Heart Fail 2012; 14: fibroblasts. FEBS Lett 2009; 583: 4. Lauritsen J, Gustafsson F, Abdulla J. 803–869. 933–937. Characteristics and long-term prognosis 11. Kang S-H, Park JJ, Choi D-J, Yoon C-H, 19. Gawor M, Śpiewak M, Kubik A, Wróbel of patients with heart failure and mid- Oh I-Y, Kang S-M, Yoo BS, Jeon ES, Kim A, Lutyńska A, Marczak M, Grzybowski range ejection fraction compared with JJ, Cho MC, Chae SC, Ryu KH, Oh BH, J. Circulating biomarkers of hypertrophy reduced and preserved ejection fraction: KorHF Registry. Prognostic value of NT- and fibrosis in patients with hypertro- a systematic review and meta-analysis. proBNP in heart failure with preserved phic cardiomyopathy assessed by cardiac ESC Heart Fail J 2018; 5: 685–694. versus reduced EF. Heart 2015; 101: magnetic resonance. Biomarkers 2018 5. Lopatin Y. Heart failure with mid-range 1881–1888. Nov; 23: 676–682. ejection fraction and how to treat it. 12. Savarese G, Orsini N, Hage C, Vedin O, 20. Izumiya Y, Hanatani S, Kimura Y, Card Fail Rev 2018; 4: 9–13. Cosentino F, Rosano GMC, Dahlström Takashio S, Yamamoto E, Kusaka H, 6. Ouwerkerk W, Voors AA, Zwinderman U, Lund LH. Utilizing NT-proBNP for eli- Tokitsu T, Rokutanda T, Araki S, Tsujita AH. Factors influencing the predictive gibility and enrichment in trials in K, Tanaka T, Yamamuro M, Kojima S, power of models for predicting mortality HFpEF, HFmrEF, and HFrEF. JACC Heart Tayama S, Kaikita K, Hokimoto S, and/or heart failure hospitalization in Fail 2018; 6: 246–256. Ogawa H. Growth differentiation patients with heart failure. JACC Heart 13. Maisel A, Mueller C, Adams K, Anker factor-15 is a useful prognostic marker Fail 2014; 5: 429–436. SD, Aspromonte N, Cleland JGF, Co- in patients with heart failure with pre- 7. Moliner P, Lupón J, Barallat J, de hen-Solal A, Dahlstrom U, DeMaria A, served ejection fraction. Can J Cardiol Antonio M, Domingo M, Núñez J, Za- Di Somma S, Filippatos GS. State of the 2014; 30: 338–344. mora E, Galán A, Santesmases J, Pastor art: using natriuretic peptide levels in 21. Lok DJ, Klip IT, Lok SI, Bruggink-André C, Bayes-Genis A. Bio-profiling and bio- clinical practice. Eur J Heart Fail 2008; de la Porte PW, Badings E, van prognostication of chronic heart failure 10: 824–839. Wijngaarden J, Voors AA, de Boer RA, with mid-range ejection fraction. Int J 14. Cleland JGF, Taylor J, Freemantle N, van Veldhuisen DJ, van der Meer P. In- Cardiol 2018; 257: 188–192. Goode KM, Rigby AS, Tendera M. Rela- cremental prognostic power of novel 8. Vousden KH, Lane DP. p53 in health and tionship between plasma concentrations biomarkers (growth-differentiation disease. Nat Rev Mol Cell Biol 2007; 8: of N-terminal pro brain natriuretic pep- factor-15, high-sensitivity C-reactive pro- 275–283. tide and the characteristics and outcome tein, galectin-3, and high-sensitivity 9. Wollert KC, Kempf T, Wallentin L. of patients with a clinical diagnosis of di- troponin-T) in patients with advanced Growth differentiation factor 15 as a astolic heart failure: a report from the chronic heart failure. Am J Cardiol biomarker in cardiovascular disease. PEP-CHF study. Eur J Heart Fail 2012; 2013; 112: 831–837. Clin Chem 2017; 63: 140–151. 14: 487–494. 22. Jackson CE, Haig C, Welsh P, Dalzell JR, 10. McMurray JJV, Adamopoulos S, Anker 15. Chan MMY, Santhanakrishnan R, Chong Tsorlalis IK, McConnachie A, Preiss D, SD, Auricchio A, Böhm M, Dickstein K, JPC, Chen Z, Tai BC, Liew OW, Ng TP, Anker SD, Sattar N, Petrie MC, Gardner Falk V, Filippatos G, Fonseca C, Gomez- Ling LH, Sim D, Leong KT, Yeo PS, Ong RS, McMurray J. The incremental prog- Sanchez MA, Jaarsma T, Køber L, Lip HY, Jaufeerally F, Wong RC, Chai P, nostic and clinical value of multiple GY, Maggioni AP, Parkhomenko A, Low AF, Richards AM, Lam CS. Growth novel biomarkers in heart failure. Eur J Pieske BM, Popescu BA, Rønnevik PK, differentiation factor 15 in heart failure Heart Fail 2016; 18: 1491–1498. Rutten FH, Schwitter J, Seferovic P, with preserved vs. reduced ejection frac- 23. Baessler A, Strack C, Rousseva E, Wag- Stepinska J, Trindade PT, Voors AA, tion. Eur J Heart Fail 2016; 18: 81–88. ner F, Bruxmeier J, Schmiedel M, Zannad F, Zeiher A, Task Force for the 16. Stahrenberg R, Edelmann F, Mende M, Riegger G, Lahmann C, Loew T, Schmitz Diagnosis and Treatment of Acute and Kockskämper A, Düngen H-D, Lüers C, G, Fischer M. Growth-differentiation Chronic Heart Failure 2012 of the Euro- Binder L, Herrmann-Lingen C, Gelbrich factor-15 improves reclassification for pean Society of Cardiology, Bax JJ, G, Hasenfuss G, Pieske B, Wachter R. the diagnosis of heart failure with nor- Baumgartner H, Ceconi C, Dean V, The novel biomarker growth differentia- mal ejection fraction in morbid obesity. Deaton C, Fagard R, Funck-Brentano C, tion factor 15 in heart failure with Eur J Heart Fail 2012; 14: 1240–1248. ESC Heart Failure 2020; 7: 2223–2229 DOI: 10.1002/ehf2.12621