A 3-Biomarker 2-Point-Based Risk Stratification Strategy in Acute Heart Failure

Introduction and Objectives: Most multi-biomarker strategies in acute heart failure (HF) have only measured biomarkers in a single-point time. This study aimed to evaluate the prognostic yielding of NT-proBNP, hsTnT, Cys-C, hs-CRP, GDF15, and GAL-3 in HF patients both at admission and discharge. Methods: We included 830 patients enrolled consecutively in a prospective multicenter registry. Primary outcome was 12-month mortality. The gain in the C-index, calibration, net reclassification improvement (NRI), and integrated discrimination improvement (IDI) was calculated after adding each individual biomarker value or their combination on top of the best clinical model developed in this study (C-index 0.752, 0.715–0.789) and also on top of 4 currently used scores (MAGGIC, GWTG-HF, Redin-SCORE, BCN-bioHF). Results: After 12-month, death occurred in 154 (18.5%) cases. On top of the best clinical model, the addition of NT-proBNP, hs-CRP, and GDF-15 above the respective cutoff point at admission and discharge and their delta during compensation improved the C-index to 0.782 (0.747–0.817), IDI by 5% (p < 0.001), and NRI by 57% (p < 0.001) for 12-month mortality. A 4-risk grading categories for 12-month mortality (11.7, 19.2, 26.7, and 39.4%, respectively; p < 0.001) were obtained using combination of these biomarkers. Conclusion: A model including NT-proBNP, hs-CRP, and GDF-15 measured at admission and discharge afforded a mortality risk prediction greater than our clinical model and also better than the most currently used scores. In addition, this 3-biomarker panel defined 4-risk categories for 12-month mortality.

Introduction and Objectives: Most multi-biomarker strategies in acute heart failure (HF) have only measured biomarkers in a single-point time. This study aimed to evaluate the prognostic yielding of NT-proBNP, hsTnT, Cys-C, hs-CRP, GDF15, and GAL-3 in HF patients both at admission and discharge.
Methods: We included 830 patients enrolled consecutively in a prospective multicenter registry. Primary outcome was 12-month mortality. The gain in the C-index, calibration, net reclassification improvement (NRI), and integrated discrimination improvement (IDI) was calculated after adding each individual biomarker value or their combination on top of the best clinical model developed in this study (C-index 0.752, 0.715-0.789) and also on top of 4 currently used scores (MAGGIC, GWTG-HF, Redin-SCORE, BCN-bioHF).

INTRODUCTION
The prognostic stratification of patients with acute heart failure (HF) is essential to establish an appropriate personalized followup plan. Cardiac biomarkers have improved the predictive models of HF patients beyond the already well-established clinical risk predictors such as functional class or physical examination (Levy et al., 2006).
More than 10 years ago, Braunwald provided a comprehensive review of the biomarkers related to the different pathophysiological substrates involved in HF (Braunwald, 2008), and remarked the need to identify the biomarkers with independent predictive value in large prospective cohorts of patients. Until now, a substantial number of studies have assessed the prognostic capacity of panels of 3 or more biomarkers in acute HF (Ishii et al., 2002;van Kimmenade et al., 2006;Januzzi et al., 2007;Rehman et al., 2008;Manzano-Fernández et al., 2009Niizeki et al., 2009;Zairis et al., 2010;Pascual-Figal et al., 2011;Shah et al., 2012;Bjurman et al., 2013;Lassus et al., 2013;Lok et al., 2013;Richter et al., 2013;Srinivas et al., 2014;Demissei et al., 2016Demissei et al., , 2017aHerrero-Puente et al., 2017;Tromp et al., 2017), but only in few of them the biomarkers were analyzed at both hospital admission and discharge (Demissei et al., 2016(Demissei et al., , 2017a. A single-point measurement would not allow to evaluate the width of the pathophysiological changes occurring during the clinical compensation and, moreover, might limit the predictive capacity of the biomarkers. Although the predictive capacity of the single-point measurement can be improved by increasing the number of the biomarkers in the panel, it is theoretically possible that a substantial improvement could be alternatively attained using only few of them, but measured at both admission and discharge. The prognostic yielding of sequential measurements of a single biomarker (van Vark et al., 2017) or a series of them (Demissei et al., 2017a,b) in patients with HF was evaluated in post hoc analysis of clinical trials.
However, the data derived from clinical trials might not entirely reflect the daily real practice.
Therefore, we aimed to analyze the performance of a multibiomarker panel covering distinct pathophysiological axes in HF measured both at admission and at hospital discharge, in a nationwide cohort of patients with acute HF (REDINSCOR II registry). We have selected the N-terminal pro-B-type natriuretic peptide (NT-proBNP) as marker of neurohormonal activation and myocyte stretch, high-sensitive T troponin (hs-TnT) linked to myocyte injury, cystatin-C (Cys-C) as indicative of renal dysfunction, growth differentiation factor 15 (GDF-15) and galectin-3 (GAL-3) as markers of matrix remodeling, and high sensitive C reactive protein (hs-CRP) as marker of inflammation.

Study Population
This is a subanalysis including 830 patients discharged alive with available biomarker data both at admission and discharge from the REDINSCOR II study. This is a multicenter, prospective nationwide registry, which enrolled consecutively patients from 18 secondary and tertiary hospitals since October 2013 to December 2014. Inclusion criteria were: (i) age older than 18 years; (ii) acute HF as the main cause for admission; and (iii) hospitalization ≥ 24 h in the Cardiology Department. Exclusion criteria were: (i) HF episode secondary to ST-segment elevation acute coronary syndrome; (ii) end-stage disease with a life expectancy < 1 year; and (iii) any condition that would prevent an appropriate follow-up. HF was diagnosed in accordance with current HF guidelines (McMurray et al., 2012). The study complies with the Declaration of Helsinki and the protocol was approved by the Ethics Committees of each participating center. All patients gave written informed consent.

Study Variables
Data were collected using specifically designed web forms and quality controls were done monthly. The following clinical variables were gathered at study inclusion and before discharge: demographic and previous clinical data, case history and physical examination, chest x-rays, ECG, echocardiography, laboratory blood tests, and pharmacological and non-pharmacological treatment (Appendix 1). Standard criteria were used to define the clinical variables. Left ventricular ejection fraction (LVEF) was categorized according to the recent HF European guidelines .

Biomarker Panel
Blood samples were obtained by venipuncture within the first 24 h of admission and thereafter at hospital discharge. The samples were centrifuged at 2,500 g for 15 min. Serum and plasma aliquots of 0.5 mL were separated and frozen at -80 • C until analysis; all samples of the same individual were processed in the same batch. Biomarker concentrations were measured at a core laboratory (Biochemistry Department, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain). We measured the serum levels of NT-proBNP, hs-cTnT, and GDF-15 by electrochemiluminescence immunoassays, and cystatin-C and hs-CRP by particle-enhanced turbidimetric immunoassays using reagents from Roche Diagnostics (Basel, Switzerland). Galectin-3 was also measured in serum using an enzyme-linked fluorescent immunoassay (BioMérieux, Marcy-l'Étoile, France). The imprecision of all assays was similar or even lower than that reported by manufacturers.

Follow-Up and Outcomes
In addition to the specific clinical follow-up needed by the patient, the vital status was also checked at 12 months after discharge. We used either telephone interviews or clinical records of hospitals, primary care, or institutional death registries. The primary outcome was all-cause mortality at 12-month after discharge. The secondary outcomes were cardiovascular mortality, HF mortality, and readmission for HF at the same time period. The reported events were reviewed by an ad hoc committee.

Statistical Analysis
Continuous variables are expressed as mean (standard deviation) or as median (interquartile range) whenever appropriate. Differences in continuous variables were tested by the analysis of variance (ANOVA), Student's t-test, or Wilcoxon signed rank test for independent samples. Categorical variables were presented as frequency and percentage. Differences in the categorical variables were assessed by the χ 2 test or by Fisher's exact test. A two-sided p-value < 0.05 was considered statistically significant. Missing data were imputed using the "MICE" package in R (Multivariate Imputation by Chained Equations) whenever necessary (m = 1). All the analyses were performed using R (v. 3.2) and STATA (v. 13.1).
Firstly, we developed the best clinical model to predict the occurrence of the primary endpoint using a multivariable Cox regression analysis. Clinical meaningful variables and those showing a p-value < 0.1 in the univariate analysis were thereafter included in the multivariate model. A backward stepwise method was used to identify independent predictors with a p-value < 0.05 as inclusion or deletion criteria. This model was finally composed of variables at admission (number of HF episodes during the last year, previous stroke, systolic blood pressure, presence of right HF signs, significant mitral valve regurgitation, hyponatremia, and body mass index), and variables at discharge (persisting HF signs, heart rate, left bundle branch block, eGFR < 60 ml/min/1.73 m 2 , and length of hospital stay). The discriminative ability of the model for all-cause mortality at 12-month after discharge assessed by the C-statistic index was 0.752 (95% CI 0.715-0.789).
On top of this clinical model, we then analyzed sequentially the added prognostic value of each individual biomarker and their combinations using the gain in the C-index, calibration (Grønnesby and Borgan, Brier score, Akaike and Bayesian criteria), integrated discrimination improvement (IDI), and net reclassification improvement (NRI). Moreover, we also analyzed the added prognostic value of each individual biomarker and their combinations on top of other well-validated clinical scores usually used in clinical practice as the MAGGIC (Pocock et al., 2013), GWTG-HF (Peterson et al., 2010), Redin-SCORE (Álvarez-García et al., 2015), and BCN bio-HF (Lupón et al., 2014). The ROC curve analysis was used to determine the optimal biomarker cut-off value to predict the primary endpoint employing the Youden criteria.

Clinical Characteristics of the Study Population
As shown in Table 1, most patients were elderly, male, Caucasian, had a previous history of HF (60%), and a high Charlson comorbidity index. According to LVEF at admission, 263 patients (32%) were classified as HFrEF, 207 (25%) as HFmrEF and 360 (43%) as HFpEF.

Biomarker Changes During Compensation of Acute Heart Failure Episode
As summarized in Table 2, the biomarkers linked to myocyte stress (NT-proBNP), inflammation (hs-CRP), and matrix remodeling (GDF-15) decreased significantly after the hospital stay whereas the percentage of change of GAL-3 and hs-TnT was negligible. The increase of biomarker reflecting renal damage (Cys-C) was lower than the expected by the biological variability. The Supplementary Table 1 summarizes the best cutoff points of each biomarker predicting the primary outcome according to the ROC analysis.

Clinical data at admission
Clinical profile of acute HF, n (%) Acutely decompensated chronic HF 595 (72) Pulmonary edema 115 (14) Right HF 34 (4) Others 86 (10) Heart rate, bpm, median (IQR) 85 (72-100) rise to the highest improvement in the C-index for 12-month mortality (0.782; 95% CI 0.747-0.817, p < 0.001). Of notice, the discrimination of this 3-biomarker model was better than that including the six biomarkers, and even better than that based only the biomarkers at discharge. Moreover, the 3-biomarkers model provided a huge reclassification of patients with and without increased risk reaching a statistically significant NRI of 56% for 12-month mortality. These scores were achieved with a correct calibration of the models (Supplementary Figure 1). Similarly, the addition of these 3 biomarkers on top of the MAGGIC, GWTG-HF, Redin-SCORE, and BCN bio-HF models was the best strategy in terms of gain of C-index and reclassification parameters. Table 3 summarizes the discrimination, calibration, IDI, and NRI parameters for the primary outcome given by the clinical models alone, in combination with the 6-biomarker model, and the 3-biomarker strategy. The discrimination capacity of the models for HF-mortality was also better than that for cardiovascular and overall mortality (Supplementary Figure 2). The 3-biomarker strategy also allowed to identify 4-risk categories for 12-month all-cause mortality: (1) low-risk group (529 patients) presenting either none or 1 elevated biomarker, (2) low-intermediate risk group (78 patients) presenting 2 or 3 elevated biomarkers at admission but none or 1 at discharge, (3) high-intermediate group (86 patients) presenting either none or 1 elevated biomarker at admission but 2 or 3 at discharge, and (4) high-risk group (137 patients) presenting 2 or 3 elevated biomarkers both at admission and discharge. As shown in the Figure 1, the 12-month mortality rates for these four categories was, respectively, 11.7, 19.2, 26.7, and 39.4% (p < 0.001 for the trend). Considering the low risk category as reference, the mortality risk-ratio was

Main Findings
Our study revealed that elevated concentrations of NT-proBNP, hs-CRP, and GDF-15 at hospital admission and discharge in patients with acute HF predicted 12-month mortality better than the best clinical model developed in our population and permitted to define 4 levels of risk. Moreover, the predictive capacity of this 3-biomarker panel was not increased by adding hs-TnT, cystatin C and Galectin-3 in the model and was also superior to the most the currently used scores.
Our study overcomes some of these limitations and emerges as the first observational, multicenter registry analyzing the capacity of a set of biomarkers double measured at hospital admission and discharge to predict relevant 1-year outcomes in a large group of patients with acute HF. We selected six biomarkers linked to the main processes involved in HF such as neurohormonal activation, myocyte stretch, injury or inflammation, myocardial remodeling and fibrosis, and impaired renal function involved in HF development. Among these 6 biomarkers, we found that the model including the NT-proBNP, hs-CRP, and GDF-15 was the best to predict 12-month mortality. Interestingly, these 3 biomarkers presented the largest magnitude of change from hospital admission to discharge suggesting that patients in whom the linked underlying mechanisms namely myocyte stretch, inflammation, and myocardial remodeling had not improved upon clinical compensation of the HF episode are at high risk of mortality. The percentage of change of the other 3 studied biomarkers hs-cTnT, GAL-3, and Cys-C at discharge was less than 5% and these biomarkers did not improve the discriminative risk capacity beyond that achieved by NT-proBNP, hs-CRP, and GDF-15. The lack of risk prediction of hs-cTnT in our study could deal with several causes. Elevated hs-TnT values in acute HF could be associated to ischemia, inflammation, oxidative stress or impaired renal function. However, all these alterations could also increase the 3 biomarkers, particularly GDF-15 and hs-CRP, already included in the model and, the prognostic role of hs-cTnT could be already covered (Kociol et al., 2010). In addition, we excluded in our study STsegment elevation acute coronary syndrome as a cause of HF hospitalization and hs-TnT is known to be a strong risk predictor in these patients.

Clinical Implications
A decrease in the plasma level of natriuretic peptides during the clinical compensation of a HF episode is associated with lower cardiovascular mortality and lower readmissions at 6 months (Savarese et al., 2014). However, a systematic recommendation on their use in clinical practice is not reflected in the current guidelines (Yancy et al., 2013;Ponikowski et al., 2016). Recently, a consensus document of the American Heart Association stated that the measurement of natriuretic peptides, cardiac troponin, and biomarkers of fibrosis at the time of presentation is useful and reasonable for establishing prognosis in patients with acutely decompensated HF (Chow et al., 2017). Our study contributes on this important issue by identifying the best combination of 3 out of 6 currently used biomarkers that are the most useful to predict 1-year mortality of patients after hospitalization for heart failure. Specially, the relevant IDI and NRI values by our 3biomarker model reinforce its role improving the ever-complex HF stratification process.

Study Limitations
This study includes 98% of Caucasian patients, thus our data might not be fully applicable to other ethnicities or countries. Considering that our study design necessarily required biomarker measurements available at hospital admission and discharge, we did not include patients lacking the discharge sample. Moreover, the size of the study sample did not allow analyzing the performance of the multi-biomarker strategies in subgroups of clinical interest. Therefore, external validation of the clinical model and the full model including biomarkers should be performed.

CONCLUSION
In a multicenter, prospective registry of patients with acute HF, we identified 3 out of 6 currently available biomarkers that afforded the highest discriminative power to predict 12-month mortality beyond the best clinical model and also above the currently used MAGGIC, GWTG-HF, Redin-SCORE, and BCN bio-HF scores. Moreover, this simple 3-biomarker panel permitted to define 4 predictive risk levels for 1-year mortality.
What is Known About the Topic?
• The prognostic stratification of patients with acute HF is essential to establish an appropriate personalized followup plan. • Cardiac biomarkers have improved the predictive models of HF patients beyond the already well-established clinical risk predictors. • Most multi-panel strategies in acute HF have only measured biomarkers in a 1-point time.
What Does This Study Add?
• We evaluate the prognostic role of 6 biomarkers at admission and discharge after HF admission. • Our study identifies a simple set of 3 biomarkers to predict prognosis of HF patients. • This panel permits to define 4 predictive risk levels for 12-month mortality.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

ETHICS STATEMENT
The protocol was approved by the Ethics Committees of each participating center. The patients/participants provided their written informed consent to participate in this study.

ACKNOWLEDGMENTS
We are indebted to Roche Diagnostics Intl. and BioMerieux for freely providing the reagents used for biomarker measurement. The companies had no further involvement in the different steps of the study.

SUPPLEMENTARY MATERIAL
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys.
Supplementary Figure 2 | Comparison of C-index between clinical, 6-biomarker, and 3-biomarker models for each outcome. Discrimination of the 3-biomarker model (red line) was better than that including only clinical variables (blue line), and even better than that considering all the biomarkers (green line). In addition, C-index for HF-mortality was also better than that for cardiovascular and overall mortality.