Article Text

Download PDFPDF
Original article
Diagnostic performance and reference values of novel biomarkers of paediatric heart failure
  1. Jakob A Hauser1,2,3,
  2. Svitlana Demyanets4,
  3. Krisztina Rusai1,
  4. Clara Goritschan1,
  5. Michael Weber5,
  6. Dilveer Panesar2,3,
  7. Lisa Rindler1,
  8. Andrew M Taylor2,3,
  9. Rodrig Marculescu4,
  10. Michael Burch3,
  11. Johann Wojta6,7,
  12. Ina Michel-Behnke1
  1. 1Department of Paediatrics and Adolescent Medicine, Division of Paediatric Cardiology, Medical University of Vienna, Vienna, Austria
  2. 2University College London, Institute of Cardiovascular Science, Centre for Cardiovascular Imaging, London, UK
  3. 3Cardiorespiratory Division, Great Ormond Street Hospital for Children, London, UK
  4. 4Department of Laboratory Medicine, Medical University of Vienna, Vienna, Austria
  5. 5Department of Biomedical Imaging and Image-Guided Therapy, Medical University of Vienna, Vienna, Austria
  6. 6Department of Internal Medicine II, Division of Cardiology, Medical University of Vienna, Vienna, Austria
  7. 7Medical University of Vienna, Core Facilities, Vienna, Austria
  1. Correspondence to Dr Jakob Hauser, University College London, Centre for Cardiovascular Imaging, 30 Guilford Street, London WC1N 1EH, UK; jakob.hauser{at}


Objective Biomarkers play a pivotal role in heart failure (HF) management. Reference values and insights from studies in adults cannot be extrapolated to the paediatric population due to important differences in pathophysiology and compensatory reserve. We assessed the diagnostic utility of four novel biomarkers in paediatric HF.

Methods Midregional (MR) pro-atrial natriuretic peptide (proANP), soluble ST2 (sST2), growth differentiation factor-15 (GDF-15), MR-pro-adrenomedullin (proADM) and N-terminal pro-B natriuretic peptide (NT-proBNP) were measured in 114 patients and 89 controls. HF was defined as the presence of HF symptoms and/or abnormal systolic ventricular function. Receiver-operating characteristics were plotted, and the area under the curve (AUC) was measured. This was repeated for subgroups with cardiomyopathy and congenital heart disease (CHD). Ventricular systolic function was measured by magnetic resonance or echocardiography. Reference values were calculated according to the current guidelines.

Results The AUC for diagnosing HF was 0.76 for MR-proANP (CI 0.70 to 0.84) and 0.82 for NT-proBNP (CI 0.75 to 0.88). These parameters performed similarly in the subgroups with CHD and cardiomyopathy. By contrast, MR-proADM, GDF-15 and sST2 performed poorly. When used in conjunction with NT-proBNP, no parameter added significantly to its diagnostic accuracy. NT-proBNP, MR-proANP, GDF-15 and sST2 could accurately discriminate between patients with preserved and patients with poor functional status. In a subset of patients with dilated cardiomyopathy, NT-proBNP, MR-proANP, MR-proADM and GDF-15 were associated with poor LV function.

Conclusions MR-proANP could accurately detect HF in children and adolescents. Its diagnostic performance was comparable with that of NT-proBNP, regardless of the underlying condition. Reference values are presented.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Paediatric heart failure (HF) differs substantially from its adult counterpart. Contrary to the latter, paediatric HF is caused predominantly by the large and heterogeneous group of congenital heart diseases (CHD) and by cardiomyopathy (CMP). Advances in CHD management have led to a rise in prevalence throughout all age groups.1 These specifics of paediatric HF present several problems. Symptoms and history are often ambiguous in young children and can be confounded by comorbidities. Functional assessments can be challenging in very small or uncooperative children, in complex CHD or in right ventricular (RV) disease, where echocardiography is unreliable. Measuring biomarkers in the blood can facilitate HF management as they provide objective information on disease severity, prognosis and treatment response.2–6

Novel biomarkers typically reflect different components of the complex HF pathophysiology such as fibrotic remodelling, myocardial stretch or inflammation, and their complementary use has been shown to greatly enhance their prognostic and diagnostic power.4 ,5 ,7–10 In adult cardiology, such parameters are usually studied and validated in relatively homogeneous populations of aged and multimorbid patients with left-sided HF of predominantly ischaemic aetiology.4 ,7–9 Findings from adults cannot be extrapolated to children as they fail to reflect important differences in pathophysiology and compensatory reserve inherent to this group. We therefore sought to assess four novel biomarkers in paediatric patients with HF and to provide reference values from normal children.


Study design and population

This study was conducted as a prospective trial in two tertiary centres in Austria and the UK. Patients aged 0–24 years were recruited from a cardiac ward and an HF outpatient clinic between September 2013 and July 2015 (n=114). This included a subset of patients with dilated CMP (DCM) undergoing elective cardiac MR (CMR) imaging (n=25). HF was defined as presence of HF symptoms and abnormal ventricular systolic function or, in clinically compensated patients without symptoms (ie, functional class (FC) I; see below), history of structural heart disease and abnormal ventricular systolic function on echocardiography or CMR. Abnormal systolic ventricular function was defined as fractional shortening (FS) <28% or ejection fraction (EF) <55% for the left ventricle (LV) and EF <50% or a Z-score >2 for tricuspid annular positive systolic excursion (TAPSE) for the RV.11 Patients with prior surgery for CHD were excluded if normal cardiac physiology had been completely restored. Reference values were obtained from children without heart disease undergoing phlebotomy prior to an elective procedure (n=89). Exclusion criteria were known renal, pulmonary, autoimmune or malignant comorbidity, haemodynamic instability and/or failure to provide informed consent. Written, informed consent was obtained from all patients and their parents prior to inclusion. The study was conducted in conformity with the Declaration of Helsinki, and ethical approval was obtained from both institutional review boards.

Functional classification

Disease severity was graded using the Ross classification.12

Biomarker sampling and analysis

Blood was collected using standard collection techniques on the day of admission in inpatients, or on the day of clinic visit in outpatients. Plasma and serum samples were spun and frozen on the day of collection and stored at −80°C for batch analysis. Soluble ST2 (sST2; Presage ST2 Assay, Critical Diagnostics, San Diego, California, USA) and growth differentiation factor 15 (GDF-15; human GDF-15 Quantikine ELISA Kit, R&D Systems, Minneapolis, Minnesota, USA) were measured in patient sera using a specific enzyme-linked immunosorbent assay (ELISA). The measurement range was 3.125–200 ng/mL for the Presage ST2 Assay (based on a 50-fold dilution of patient samples) and 23.40–1500 pg/mL for the human GDF-15 Quantikine ELISA Kit (based on a fourfold dilution). An automated immunofluorescent assay (KRYPTOR System, BRAHMS AG, Hennigsdorf/Berlin, Germany) was used to determine midregional pro-adrenomedullin (MR-proADM) from EDTA-plasma, as well as midregional pro-atrial natriuretic peptide (MR-proANP) processed from serum. Measurement ranges were 2.1–10 000.0 pmol/L for MR-proANP and 0.05–100 nmol/L for MR-proADM. N-terminal pro-B natriuretic peptide (NT-proBNP) was assessed using an Elecsys immunoassay on a cobas 8000 system (Roche Diagnostics, Mannheim, Germany). Laboratory measurements were performed by investigators who were blinded to patient history.

Imaging protocol

Using echocardiography, LV diameters and volumes were measured in end-diastole and end-systole as described (Vivid E9; GE Health Care, Fairfield, Connecticut, USA).13 CMR was performed on a 1.5 Tesla system, using one spine coil and one body matrix coil (Avanto, Siemens AG, Erlangen, Germany). A vectorcardiogram was used for cardiac gating. Ventricular volumes were measured using radial k-t SENSE imaging during free breathing and quantified by manual, slice-wise tracing of the endocardial border in systole and diastole by inclusion of papillary muscles and trabeculae in the myocardium (OsiriX; Pixmeo Sàrl, Bernex, Switzerland). EF was calculated, and Z-scores for LV volumes and diameters were obtained from published reference data.13–15

Statistical analysis

All statistical analyses were performed by a statistician (MW), using the pROC package for R (R Foundation for Statistical Computing, Vienna, Austria)16 and SPSS Statistics software V.23.0 (IBM, Armonk, New York, USA). Reference intervals were calculated according to the Clinical and Laboratory Standards Institute (CLSI) guideline C28-A3 using MedCalc V.12.5 (MedCalc Software bvba, Ostend, Belgium).17 Due to the non-normal distribution and the number of reference values of less than 120, the recommended ‘robust method’ of calculation was applied for all parameters. Metric and normally distributed data are presented as mean±SD. In case of skewed data distribution, medians are presented with their IQR. Due to data skewedness in the reference population, associations between biomarker levels and age or functional data were assessed using Spearman's rank correlation. This population was subsequently divided into four age groups, and the variance between the biomarker levels of these groups was assessed by Kruskal–Wallis test. In order to assess the association between biomarker levels and LV size and systolic function in patients with DCM, a binary variable was created, whereby LV dysfunction was defined as EF <55% or FS <28% and LV dilatation as a Z-score >2 for LV end-diastolic volume (LVEDV). The associations between biomarkers and these factors were determined by logistic regression in this subgroup. To examine the accuracy of each blood test to detect HF, a receiver-operating characteristic (ROC) was plotted and the area under the curve (AUC) quantified. From a bootstrap sample of n=10 000, 95% CIs were obtained. This procedure was subsequently repeated in subgroups of patients with CHD and CMP separately. In order to assess the incremental value of using novel parameters in conjunction with NT-proBNP, a stepwise multiple logistic regression model was first built. When a statistically significant association with HF was present, the resulting value from this model was used to repeat the ROC analysis, and the diagnostic accuracy of combinations of biomarkers was compared with that of NT-proBNP as a stand-alone parameter according to the method proposed by de Long et al.18 A significance level of p≤0.05 was defined for all statistical operations.


Study population

A total of 203 subjects were enrolled for the study (95 female; median age 7.5 years; IQR 2.4–13.0; range 5 days–24 years). This included 114 patients with HF (59 female; median age 5.9 years; IQR 2.1–12.5) and 89 controls (36 female; median age 8.9 years; IQR 2.7–14.5). No significant differences in sex or age distribution were found between the groups. Clinical data are listed in table 1.

Table 1

Study population characteristics

Diagnostic accuracy of novel biomarkers in the general study population

There was a significant difference between groups for NT-proBNP and MR-proANP (table 2). Similar trends for sST2 and GDF-15 did not reach statistical significance. ROC analysis was undertaken for MR-proANP, sST2, GDF-15 and NT-proBNP in 114 patients and 89 controls (figure 1A). Five datasets were excluded due to incomplete data. NT-proBNP and MR-proANP showed very good accuracy, whereas the remaining parameters did not perform sufficiently (table 3). Using an NT-proBNP cut-off value of 94 pg/mL yielded a sensitivity of 80% and a specificity of 71% with a positive predictive value (PPV) of 78% and a negative predictive value (NPV) of 74%. For MR-proANP, a 48 pmol/L cut-off resulted in a 75% sensitivity and a specificity of 70% (PPV 76% and NPV 68%). Due to missing data, adding MR-proADM to the head-to-head comparison reduced the number of valid cases to 76 in the HF group and 51 controls (table 4). ROC analysis showed poor accuracy for MR-proADM, whereas the outcome for the other parameters did not change significantly (figure 1B). When comparing their ability to discriminate high from low FC (ie, I and II vs III and IV), the AUC for NT-proBNP was 0.95 (CI 0.90 to 0.99, p<0.001), 0.93 for MR-proANP (CI 0.84 to 0.99, p<0.001), 0.89 for GDF-15 (CI 0.76 to 0.98, p<0.001) and 0.76 for sST2 (CI 0.57 to 0.92, p=0.001).

Table 2

Biomarker results in patients, controls and subgroups

Table 3

Receiver-operating analysis for heart failure biomarkers (without MR-proADM)

Table 4

Receiver-operating analysis for HF biomarkers (with MR-proADM)

Figure 1

Receiver-operating characteristic of novel biomarkers for the diagnosis of heart failure (HF) in the overall population alone (A and B) and combined (C); diagnosis of HF in patients with congenital heart disease (CHD) alone (D and E) and combined (F); diagnosis of HF in patients with cardiomyopathy (CMP) alone (G and H) and combined. GDF-15, growth differentiating factor 15; MR-proADM, midregional pro-adrenomedullin; MR-proANP, midregional pro-atrial natriuretic peptide; NT-proBNP, N-terminal pro-B natriuretic peptide; sST2, soluble ST2.

Diagnostic accuracy in CHD

Biomarker levels and age did not differ significantly between patients with univentricular and biventricular physiology (table 5). NT-proBNP and MR-proANP displayed good accuracy in detecting HF, whereas the other parameters performed poorly (table 3; figure 1D). Using an NT-proBNP cut-off value of 94 pg/mL yielded a sensitivity of 82% and a specificity of 70%, giving a PPV of 68% and a NPV of 83%. For MR-proANP, a 48 pmol/L cut-off resulted in a 75% sensitivity and a specificity of 71% (PPV 67% and NPV 78%). Adding MR-proADM to the analysis reduced the number of valid cases to 38 in the HF group and 51 controls (table 4). This revealed an unsatisfactory diagnostic power for MR-proADM, while that of the other parameters did not change significantly (figure 1E).

Table 5

Biomarker data and surgical history of patients with CHD

Diagnostic accuracy in CMP

In the CMP subgroup, ROC analysis showed good performance for NT-proBNP and MR-proANP (table 3, figure 1G), but not for the remaining parameters. The addition of MR-proADM reduced the case number to 38 patients and showed poor accuracy for MR-proADM, whereas the performance of the other parameters did not change significantly (table 4, figure 1H). Using an NT-proBNP cut-off value of 134 pg/mL yielded a sensitivity of 77% and a specificity of 82% (PPV 67% and NPV 88%). For MR-proANP, a 51 pmol/L cut-off resulted in 74% sensitivity and a specificity of 74% (PPV 58% and NPV 86%).

In 25 patients with DCM, with available CMR data, significant correlation of LV EF was found with NT-proBNP (r=−0.59, p<0.01), MR-proANP (r=−0.67, p<0.001), GDF-15 (r=−0.62, p<0.001) and MR-proADM (r=–0.45, p<0.05), whereas correlation with sST2 failed to reach statistical significance (r=−0.29, p=0.167). Significant correlation with LVEDV was found for NT-proBNP (r=0.50, p<0.05) and MR-proANP (r=0.46, p<0.05). In the regression model of all 38 patients with DCM, with available imaging data, only NT-proBNP was predictive of LV dilatation (B=0.003, CI 0.002 to 0.005, p<0.05). This was independent of age. Moreover, age-adjusted NT-proBNP and sST2 were predictive of LV dysfunction (NT-proBNP: B=0.001, CI 0.0001 to 0.002, p<0.05; sST2: B=0.003; CI 0.0001 to 0.006, p<0.05). When removed from the model, higher age remained an independent predictor for poor LV function (B=0.10, CI 0.014 to 0.19, p<0.05).

Combined use of HF biomarkers

In the regression model, the combinations of NT-proBNP and GDF-15, MR-proANP and GDF-15 as well as that of all three parameters were good predictors of HF. However, no combination yielded a significant increment in AUC over NT-proBNP as a stand-alone parameter (table 3; figure 1C, F, I).

Reference values and association with sex and age

Reference values are presented in table 6. Significant correlation with age was found for sST2 (r=0.278, p<0.01) and MR-proANP (r=−0.221, p<0.05). Moreover, there were trends for GDF-15 (r=− 0.192, p=0.072) and NT-proBNP (r= −0.205, p=0.055). NT-proBNP and MR-proANP showed peaks in the first year, whereas GDF-15 and MR-proADM were highest in children aged 6 years and younger (table 7). By contrast, sST2 was highest in the age group of 12 years and older. Females had significantly higher levels of sST2 and MR-proANP, though for the latter, this difference was only minor. No differences between sexes were found for the other blood parameters.

Table 6

Reference values from healthy children

Table 7

Values from normal children and association with age and sex


The present study is the first to demonstrate the diagnostic utility of MR-proANP in the diagnosis of paediatric HF, irrespective of the underlying cause. By contrast, MR-proADM, GDF-15 and sST2 did not perform sufficiently, and even when used in conjunction with the reference standard, NT-proBNP, none of the novel parameters added significantly to its diagnostic properties.

Only two studies have previously addressed the potential role of MR-proANP in paediatric cardiovascular disease.3 ,19 In adults, it was shown to be superior to NT-proBNP in the diagnosis and prognostic stratification of HF in elderly and overweight patients, where NT-proBNP is less reliable.4 ,7 In our study, the performance of NT-proBNP was slightly superior to that of MR-proANP, and our findings underpin previous works that demonstrated good diagnostic accuracy in a broad spectrum of cardiovascular abnormalities of the paediatric age.2 ,6 Increases in MR-proANP and NT-proBNP are attributed to closely related physiological mechanisms, and the causes underlying the disparity in their diagnostic power can only be speculated about. MR-proANP is typically elevated as a consequence of raised atrial wall stretch, whereas NT-proBNP increases predominantly in response to ventricular wall stretch.20 Moreover, cytokines have been found to influence NT-proBNP expression and secretion, while that of MR-proANP appears to be unaffected.21 Age-specific differences in filling pressures, atrial wall stretch and cytokine activity could account for the differences in the diagnostic performance of these two parameters observed in our study.

Adult HF is predominantly caused by ischaemic heart disease, a condition typically associated with a chronic state of systemic inflammation and an endpoint of fibrotic remodelling of the myocardium.5 In children, where CHD is the leading cause, the importance and prevalence of such fibrotic processes in different lesions are still poorly understood. This may partly explain the lack of diagnostic performance observed for sST2, one of the pivotal developments in HF biomarkers. Its clinical utility is well established both as a stand-alone parameter and in combination with NT-proBNP, in acute and chronic HF as well as in CAD.8–10 ,22 ,23 In children, it was found to be elevated in the setting of acute Kawasaki disease, where it was associated with poor diastolic function. However, sST2 was measured in the context of acute inflammation and coronary artery disease (CAD), which limits inferences for paediatric HF and underpins that inflammation is an important factor in its biology.24

In adults with CHD, GDF-15 was recently shown to be independently associated with elevated pulmonary pressures, poor functional status and cardiac dysfunction.25 ,26 By contrast, paediatric data are sparse, particularly in patients with CMP. In a mixed population of both children and adults with Fontan circulation, one group found elevated GDF-15 levels to be associated with poor functional status and impaired haemodynamic function. However, their findings cannot be uncritically transferred to the general paediatric HF population, which is predominantly biventricular in physiology.27 Similar data from Fontan patients were published for MR-proADM.28 Our findings are partly in keeping with the work of Norozi et al25 who found GDF-15 to be associated with low FC and poor function in patients with CHD. However, their patients were predominantly adults, and no effort was undertaken to provide details on comorbidities that potentially affect GDF-15 levels (eg, renal disease). Moreover, the authors excluded CMP from the analysis and did not provide reference data from healthy controls. The ability of GDF-15 to distinguish between patients with HF and controls was poor in our population, even though an association with impaired cardiac function was observed in the DCM group, as it was with MR-proADM. GDF-15 and MR-proADM can be elevated in various conditions and are expressed by different tissues. In cardiomyocytes, GDF-15 is secreted in the presence of abnormal loading conditions and metabolic stress, such as ischaemia, whereas MR-proADM is produced in response to myocardial stretch among other causes.5 Notably, in our study, levels of both parameters were significantly lower than previously described in adults.7 ,9 Age-related disparities in compensatory mechanisms, medication, comorbidity, cardiac loading conditions and underlying pathology are conceivable explanations for this phenomenon.

While paediatric reference values have been published recently for GDF-15 and sST2, such normal data for MR-proANP and MR-proADM outside of the neonatal period are largely unavailable.19 ,29 ,30 In our reference population, MR-proANP was significantly higher in the first year of life, in keeping with the previous studies on B-type natriuretic peptides.6 Our finding that sST2 levels were higher in females and in older children is inconsistent with the recently published data from a larger cohort.23 ,30 As our reference data are overall consistent with the published work, this finding is likely to be incidental in nature.


The inhomogeneity of our cohort in terms of underlying aetiology as well as the relatively small proportion of patients with severe HF may have obfuscated some of the statistical associations present in specific conditions and did not allow for a meaningful analysis of the associations between biomarkers and cardiac function. For this reason, we performed a subgroup analysis in a homogeneous group of patients with DCM. Moreover, a longitudinal analysis would have been problematic to interpret, as outcome trajectories vary substantially between different diseases in the paediatric age group. Further research is required to investigate whether and how reliably these novel biomarkers can predict outcome or treatment response in a more narrowly defined population. Finally, larger case numbers are needed to generate reliable reference intervals for specific age groups, and in particular, our sample size may have been underpowered for MR-proADM.


MR-proANP can accurately detect HF in children and adolescents with CHD and CMP, even if well compensated. Its diagnostic performance was comparable with that of NT-proBNP, regardless of the underlying condition. Further research is warranted to evaluate the diagnostic and prognostic significance of these tests in specific paediatric cardiac diseases.

Key messages

What is already known on this subject?

  • In heart failure (HF), blood-based biomarkers provide objective information on diagnosis, prognostic stratification and treatment response. However, validation data and reference values of novel parameters are only available for the adult population, where HF is typically of ischaemic cause and affects the left ventricle. Findings from adult studies cannot be transferred uncritically to the paediatric age group, where important differences in pathophysiology must be considered.

What might this study add?

  • The present study shows, for the first time, that midregional (MR) pro-atrial natriuretic peptide can be useful for the diagnosis of HF in children and adolescents. Its accuracy was independent of the underlying HF aetiology and was comparable with that of N-terminal pro-B natriuretic peptide, the reference standard. By contrast, our findings do not support sufficient diagnostic power for MR-pro-adrenomedullin, soluble ST2 and growth differentiation factor-15. Reference values from normal children are presented.

How might this impact on clinical practice?

  • Measuring biomarkers can aid the detection of HF, grade disease severity and evaluate response to treatment in patients where symptoms are equivocal, which is especially useful in very young children or patients with complex lesions, where imaging studies can be difficult to accomplish accurately. The reference ranges presented in this study will be of use for future research studies and clinical practice.



  • Contributors JAH: Acquisition of funds, study design, data collection and analysis, draft, critical review and submission of manuscript. SD and KR: Study design, data collection and analysis, critical review and approval of manuscript. CG: Data collection and analysis, critical review and approval of manuscript. MW and RM: Statistical analysis, critical review and approval of manuscript. DP, LR, MB and JW: Data collection, critical review and approval of manuscript.AMT: Acquisition of funds, data collection, critical review and approval of manuscript.IM-B: Acquisition of funds, study design, data collection, critical review and submission of manuscript.

  • Funding This work was supported by the Austrian Society of Cardiology (personal grant to JAH with funds to MW) and the European Commission (funds to JAH and DP) as part of the MD-PAEDIGREE research project (contract no. 600932).

  • Competing interests None declared.

  • Patient consent Obtained.

  • Ethics approval Medical University of Vienna; NRES Committee London Hampstead.

  • Provenance and peer review Not commissioned; externally peer reviewed.