Prospective cohort.

This is study was conducted at Hospital Nacional Edgardo Rebagliati Martins, a national hospital in Lima, Peru.

We included adults diagnosed with acute non-ischemic HF during 2015. HF was defined upon when suggestive symptom criteria were met, NT-proBNP ≥125 pg/ml, and cardiac functional/structural alterations.4 We calculated a sample size of 167 cases in order to ensure a statistical power of 0.80.

Our exclusion criteria were pregnancy, hematologic diseases, blood transfusion, drugs that induce morphological changes in erythrocytes, cirrhosis, alcoholism, chemotherapy, major bleeding, and active infection. HF ischemic etiologies and other pro-inflammatory states, such myocarditis, and other inflammatory cardiomyopathies, are associated with increased inflammatory markers (RDW, IL-6, hs-CRP) and disease severity,14 therefore they are confounders of this association. We thus decided to exclude presenting a fine analysis with specific external validity.

The dependent variables were in-hospital mortality (yes or no) and disease severity. The latter was assessed with the Get With The Guidelines-HF (GWTG-HF),15 which is a severity score composed of systolic blood pressure (SBP), blood urea nitrogen (BUN), sodium, age, heart rate, race, and chronic obstructive pulmonary disease (COPD). The independent variables were RDW (%), hs-CRP (mg/l) and IL-6 (pg/ml). The covariables were collected at admission. They were age (years), gender (female or male), prior diseases (HF, diabetes, hyperlipidemia, chronic kidney disease, hypertension, COPD, and arrhythmia), type of HF (HF with preserved, mid-range or reduced ejection fraction)4 and number of hospital admissions. We also collected the vital signs (SBP, heart rate and respiratory rate), glucose, urea, BUN, and hemoglobin. We assessed the New York Heart Association (NYHA) Functional Classification (I-IV), the N-terminal pro-B-type natriuretic peptide, left ventricular ejection fraction (LVEF), and hospitalization duration (days).

Blood samples were used to determine a complete blood count, processed by the Sysmex XE-2100L brand blood cell, which carries out a daily and weekly calibration. We used the sequential immunometric method of enzyme solid-phase chemiluminescent immunoassay and the IMMULITE 2000 immunoassay analyzer (Siemens) to quantify IL-6 and hs-CRP.

We presented the qualitative variables in frequencies (absolute and relative), and the quantitative variables in mean (standard deviation, SD) or median (interquartile range (IQR)) according to its distribution, which was assessed with the histogram. To detect differences among patients who were dead or alive, we used the Chi-squared test, Fisher's exact test, Student's t-test, or Mann-Whitney U test according to distribution. Then, we used the Poisson regression model to determine the association between biomarkers and mortality. It was adjusted for significant factors according to the bivariate analysis and published literature. A p-value ≤0.05 indicated statistical significance. We excluded from the multivariate model the variables that presented collinearity. To assess the pattern of association between biomarkers and outcomes, we performed three hierarchical regression models per outcome. The first model was adjusted for age and gender; we added comorbidity confounders in the second model; and the third model was adjusted for all confounders. We reported the adjusted R2 of each model. We used the Youden index to identify the best mortality prognostic cut-off, additionally, we reported the receiver operating characteristic curves, area under the curve (AUC), sensitivity and specificity. Analyses were carried out in Stata v.14.0 (College Station, TX: StataCorp LP).

The Institutional Review Board of the Hospital Nacional Edgardo Rebagliati Martins approved this study protocol. The patients were identified with numeric codes in the database, and we did not collect any personal information. The costs of the tests were fully borne by the authors. We respected the ethical principles of the Declaration of Helsinki.

We assessed 167 patients with acute non-ischemic HF. Mean age was 72.61 (SD: 11.06), and male gender accounted for 59.28%. Median hospitalization duration was four days (IQR: 2-12). More than half reported an HF history (56.89%), moreover, hypertension was the most frequent comorbidity (60.48%). Regarding symptom classification, the majority of patients presented with NYHA II (40.12%) or III (43.11%), and less than half of patients had reduced ejection fraction (41.32%). RDW and IL-6 medians were 15% (IQR: 14.2-16.1) and 29 pg/ml (7.97-120.9), respectively. Hs-CRP mean was 48.60 mg/l (SD: 38.81) (Table 1).

Twenty-five patients (14.97%) died in-hospital. Compared to living patients, those who died presented significantly higher frequency or mean/median of hypertension (84% vs. 56.34%), COPD (72% vs. 9.15%), NYHA IV (36% vs. 9.86%), HF with reduced ejection fraction (100% vs. 30.99%), median urea (69 vs. 43.5), median NT Pro-BNP (32 444 vs. 2429.5), median RDW (17% vs. 14.5%), hs-CRP (96.62 vs. 40.14), and median IL-6 (386 vs. 19) (Table 2).

In univariate analysis, SBP, hemoglobin, and LVEF were associated with decreased risk of mortality, in contrast, hypertension, COPD, NYHA III-IV, reduced ejection fraction, heart rate, respiratory rate, urea, BUN, NT Pro-BNP, RDW, hs-CRP, IL-6, and number of admissions were associated with increased risk of mortality. After controlling for confounders, only SBP (RR: 0.90, 95 CI%: 0.81-0.99) was associated with decreased risk of mortality (Table 3).

Regarding the hierarchical regressions, in the first model all biomarkers were associated with increased mortality. This model explained the response variable at 32%. The second model explained the response variable at 48%. In this model, only IL-6 was associated with mortality (β: 0.0003, p-value: 0.054). In the third model, biomarkers did not show a significant association (Table 4). Similarly, we assessed a hierarchical regression between markers and severity. RDW and hs-CRP only had a significant effect in the first and second models (Table 5).

We determined the accuracy profile of biomarkers to predict mortality. The best cut-off for RDW, hs-CRP and IL-6 was 14.8 (sensitivity: 62.37%, specificity: 65.15%, AUC: 0.90), 68.7 (sensitivity: 46.53%, specificity: 80.3%, AUC: 0.90), 52.9 (sensitivity: 100%, specificity: 75.35%, AUC: 0.91), respectively (Supplementary Table 1, Appendix). The ROC curve of biomarkers is presented in Supplementary Figure 1 (Appendix).

The best cut-off of GWTG-HF for mortality was 52 (sensitivity: 92%, specificity: 90.85%, AUC: 0.97). We determined the accuracy profile of biomarkers to predict severity. The best cut-off of RDW, hs-CRP and IL-6 was 15.6 (sensitivity: 93.6%, specificity: 73.33%, AUC: 0.87), 65.4 (sensitivity: 87.2%, specificity: 82.50%, AUC: 0.86), 36.6 (sensitivity: 89%, specificity: 71.6%, AUC: 0.83), respectively (Supplementary Table 2).

We explored the role of three biomarkers in a cohort of HF patients. After adjusting for age, gender and comorbidity confounders, IL-6 resulted the only independent factor of in-hospital mortality, however its effect was not independent from other clinical confounders. The IL-6 and RDW presented the best accuracy for mortality and severity, respectively (Figure 1).

Figure 1

(0.15MB).

In our study, clinical confounders controlled the association between IL-6 and mortality, however its best cut-off for mortality prediction showed the best accuracy profile. IL-6 is a pleiotropic cytokine that increases in response to injury and activates immune cells and a signal protective response.14 Haugen et al.16 performed a case-control study, with the cases being the HF patients, and found that IL-6 was the only cytokine that predicted one-year mortality. Some characteristics may explain the differences with our results. First, the case-control study enables a better analysis; then, the group was mainly composed of octogenarian people, and the authors only included severe HF, additionally, authors did not state the confounders, but stated that they did not collect BNP data, which is a well-recognized marker.16 Another study suggested a potential association between IL-6 and mortality, however there were some details that reduce comparability: the population was composed of patients with HF and/or acute coronary syndrome, and the authors did not report the confounders of the multivariate model.17 IL-6 increases with aging,18 COPD,19 hypertension,20 among other diseases, so IL-6 rise may be mediated by the burden of comorbidity. Markousis-Mavrogenis et al.21 analyzed the data of 2516 patients with new-onset or worsening HF and reported an association between IL-6 and CV mortality through a model controlled by sociodemographic data, clinical data, and several comorbidities. In contrast, we added hemoglobin and NT pro-BNP, which might have better controlled the association since they are potential confounders.22,23 No previous studies have reported the accuracy of IL-6 in HF. Considering this context, IL-6 should not be used alone to predict mortality in HF.

In this study, hs-CRP was not an independent factor of mortality but of severity after controlling for age, gender and comorbidities. Nevertheless, other clinical confounders controlled its association. Previous studies have evidenced an association between CRP and mortality in HF with both reduced and preserved ejection fraction.24,25 This could be explained by the related physiopathological events, such as ventricular dysfunction.26 Although high CRP is linked to several chronic diseases,26 we controlled the potential comorbidity confounders, especially COPD, and hypertension. Nevertheless, the sensitivity of the best cut-off for hs-CRP was considerably lower, thus it could be limited in clinical practice. We did not find previous studies that reported the accuracy profile of hs-CRP to predict mortality among HF patients, but Aseri et al.27 reported high specificity and sensitivity in HF prediction among patients with myocardial infarction.

In our study, confounders controlled the association between RDW and mortality, however, it was shown to be highly correlated with disease severity. Konstantinos-Sotiropoulos et al.28 showed that the highest RDW quartile was associated with mortality in HF patients, however the model was only adjusted for age and gender. Not adjusting for potential confounders, such as hemoglobin and chronic diseases,29,30 may be a limitation. On the other hand, their association with ventricular dysfunction could explain the significant correlation with severity.31 In addition, a previous analysis in patients with HF with reduced ejection fraction showed that RDW was associated with lower global longitudinal strain, which estimates the myocardium contractibility.32 Kawasoe et al.33 demonstrated that the combined use of a similar RDW cut-off and BNP≥686 pg/ml was associated with mortality among HF patients. Other studies support that RDW adds a better prognosis capacity to NT pro-BNP.34,35 Nevertheless, there are no proposed potential direct effects of RDW on the myocardium.

We further studied the role of three biomarkers in predicting mortality and severity in HF patients. We do not recommend using these biomarkers alone but using them with other complex risk scores and clinical status. These tools are not expensive, indeed the RDW is a blood count component that can be routinely assessed, however, before applying it in clinical practice, we recommend carrying out cost-effectiveness studies in Peru.

The effect of biomarkers on mortality was controlled by clinical confounders, such as vital signs, NT Pro-BNP, and LVEF. However, IL-6 and RDW's effect on severity is independent from comorbidities, age, and gender. If we translated this evidence to a context of lack of laboratory and echocardiographic findings, RDW, an accessible tool, may be useful to classify the patient as severe. Nevertheless, in most situations there is not a lack of NT Pro-BNP or LVEF, consequently we prove that RDW and IL-6 use may not be the most pragmatic. Indeed, with the current evidence, it should not be used as criteria to modify HF therapy as LVEF is used. Although there is plausibility in their use, multi-center studies with a greater number of patients must be carried out to confirm our suggestion.

Our study presents several limitations. First, we only assessed three biomarkers, however currently there are other markers that could be investigated. While in previous studies mortality was assessed up to once a year, we only assessed in-hospital mortality. This could alter the interpretation of results since the effects of some biomarkers may dramatically change over time. For example, it has been suggested that initial IL-6 response protects heart tissue, but when it is chronically activated, it can lead to fibrotic disorders.14 Then, although our sample size ensured optimal statistical power, it was small in comparison to other papers. Finally, our results cannot be extrapolated to the whole Peruvian population since it is a single experience, moreover the external validity only applies to non-ischemic HF patients due to our inclusion criteria.