Comparison of Midregional Pro-Atrial and B-Type Natriuretic Peptides in Chronic Heart Failure

Influencing Factors, Detection of Left Ventricular Systolic Dysfunction, and Prediction of Death

Study sample

Patients were included from previous chronic HF studies conducted at our ambulatory heart failure unit at the Medical University of Vienna, Austria. Samples were collected between 1996 and 2003. Among these patients, whose residual plasma was frozen and stored at −80°C, we selected patients for this investigation according to the following criteria: history of chronic HF due to systolic dysfunction; documented left ventricular ejection fraction (LVEF) by echocardiography, radionuclide ventriculography, or contrast ventriculography at the time of blood sampling; documented history, demographic data, clinical status, standard laboratory parameters, and medication at the time of blood draw.

The protocol was approved by the institutional ethics committee.

End point

The end point analyzed was all-cause death, with patients receiving a ventricular assist device or heart transplantation censored at time of surgery. Mortality data were obtained from the Austrian Central Office of Civil Registration. End of follow-up was April 2006.

Measurement of natriuretic peptides

All blood samples were centrifuged within 1 h after collection; the resulting plasma was frozen at −80°C until analyzed in a blinded fashion for this study. Natriuretic peptides were determined using commercially available assays (MR-proANP: BRAHMS AG, Hennigsdorf, Germany [15]; NT-proBNP: Elecsys, Roche Diagnostics, Basel, Switzerland; BNP: Triage, Biosite, San Diego, California).

Statistical analysis

The GFR was estimated using the formula by Cockroft and Gault (18). For calculation and reporting, variables were transformed as follows: BMI in 5-point steps, GFR in 10 ml/min steps, and age in 10-year steps. The LVEF was ranked according to the following classes: ≥50%, 40% to 49%, 30% to 39%, and <30%.

Categorical data are presented as numbers (percent), continuous variables as mean ± SD. Due to right-skewed distributions, natriuretic peptides are presented as median and quartiles, and enter the models described below as their natural logarithms. Comparisons between groups were performed using chi-square tests for categorical data and the Student ttest for unpaired continuous data. Pearson correlation coefficients were calculated for each combination of natriuretic peptides.

The influences of age, sex, body mass index (BMI), LVEF, GFR, and ankle edema on the level of BNP, NT-proBNP, or MR-proANP were investigated using a multiple linear regression model. The results are presented by regression coefficients on the one hand and by proportions of explained variation (PEVs) on the other. Since the dependent variables of the regression models (the natriuretic peptides) were log-transformed, regression results are given as multiplicative factors, which are the exponentials of the estimated regression coefficients. Whereas regression coefficients show the direction and strength of a potential influence, PEVs allow for presentation and comparison of the proportion of the variability observed in the outcome that is explained by various independent variables, and thus their prognostic importance. Partial PEVs are given for each variable, thus adjusting for the respective remaining variables in the model. PEVs were calculated and compared with the method by Heinze and Schemper (19) using the conventional linear regression R²as measure of importance. For each natriuretic peptide, p values for all pairwise comparisons were corrected for multiple testing using the Shaffer method (20). To compare the influence of GFR on natriuretic peptide levels, the respective regression coefficients of a multivariate and multiple linear regression model were tested for equality.

The ability of the natriuretic peptides to detect an LVEF <40% was assessed using receiver-operating characteristic (ROC) curves and expressed by the area under curve (AUC) with 95% confidence intervals, which were compared by the method of DeLong et al. (21). A cutoff point was determined exploratively as the point on the ROC curve closest to the upper left corner. In order to investigate whether each natriuretic peptide's influence on LVEF depends on the GFR level, the interaction term of GFR with each natriuretic peptide was tested against 0 in multinomial logistic regression models with LVEF as an ordered dependent variable and sex, age, and BMI as adjustment variables in addition to GFR and the natriuretic peptides.

In order to assess and compare the prognostic values of different parameters, univariate and multiple Cox proportional hazard regression analyses were performed. To determine which of the natriuretic peptides were independent predictors of death, we calculated adjusted multiple Cox models for each natriuretic peptide separately. For confounding variables, we entered the following predictors of death in chronic HF as previously published in a systematic review (3): age, sex, GFR, the presence of diabetes mellitus, New York Heart Association (NYHA) functional class, LVEF, plasma sodium concentrations, BMI, systolic blood pressure, and the presence of ankle edema.

Finally, a full model with all 3 natriuretic peptides and all of the previously-listed predictors of death was calculated in order to determine which of the natriuretic peptides (if any) remained independent predictors of death.

To investigate whether each natriuretic peptide's influence on survival depends on the GFR level, the interaction term of GFR with each natriuretic peptide was tested against 0. The relative importance of the 3 natriuretic peptides to predict death was assessed by calculating the PEV. Testing for statistically significant differences between the PEVs of each natriuretic peptide was performed as previously described (19).

The reported p values are the results of 2-sided tests; values of p < 0.05 were considered to be statistically significant. All analyses were performed using SAS version 9.1 (SAS Institute Inc., Cary, North Carolina).

Patient characteristics

The study population included 797 patients with chronic HF comprising a wide range of ages (15 to 84 years), nutritional status (BMI: 16 to 52 kg/m²), renal function (GFR: 7 to 232 ml/min), and severity of HF as reflected by NYHA functional class, LVEF, and natriuretic peptide levels. Sex-specific baseline characteristics are presented in Table 1.

Factors influencing natriuretic peptide levels

Natriuretic peptides were highly correlated with one another, with correlation coefficients (r) of 0.82 (MR-proANP with NT-proBNP), 0.87 (MR-proANP with BNP), and 0.86 (NT-proBNP with BNP; p < 0.0001).

Multiple linear regression analysis showed that LVEF, GFR, and the presence of ankle edema were independent predictors of the 3 natriuretic peptides, whereas age and BMI were not. Sex was an independent predictor of MR-proANP and BNP, but not NT-proBNP (Table 2).When adjusted for the other 5 variables, a 1-class decline in LVEF (≥50%, 40% to 49%, 30% to 39%, and <30%) corresponded to 39%, 48%, and 78% increases in MR-proANP, NT-BNP, and BNP, respectively; a 10 ml/min decline in GFR corresponded to 15%, 18%, and 15% increases, respectively; and patients with ankle edema had 77%, 110%, and 149% higher levels of MR-proANP, NT-proBNP, and BNP, respectively, than patients without.

Partial (after adjustment for the respective other 5 confounding variables) PEVs showed that LVEF, GFR, and the presence of ankle edema had significantly more influence on each natriuretic peptide than age, BMI, and sex (corrected p < 0.05 for each comparison) (Fig. 1).Consistently, the combination of only LVEF, GFR, and the presence of ankle edema explained 38%, 33%, and 28% of the variation of MR-proANP, BNP, and NT-proBNP, respectively, whereas adding age, BMI, and sex did not significantly increase the explained variation of the natriuretic peptides (MR-proANP: 39%, BNP: 33%, NT-proBNP: 28%). No difference in the influence of GFR on levels of the 3 natriuretic peptide levels was found.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Factors Influencing Natriuretic Peptides

Relative influence of left ventricular ejection fraction (LVEF), glomerular filtration rate (GFR), the presence of ankle edema (edema), age, sex, and body mass index (BMI) on levels of midregional pro-atrial natriuretic peptide (MR-proANP) (A), B-type natriuretic peptide (BNP) (B), and amino-terminal pro–B-type natriuretic peptide (NT-proBNP) (C), as expressed by the partial proportion of explained variation. LVEF, GFR, and edema had a significantly stronger influence than age, sex, and BMI on levels of all 3 natriuretic peptides.

Detection of severity of left ventricular systolic dysfunction

ROC curves (Fig. 2)of natriuretic peptides to detect an LVEF <40% showed that MR-proANP and BNP had a higher accuracy than NT-proBNP (p < 0.0001 for each comparison), whereas no difference was found between MR-proANP and BNP (p = 0.7411). Respective optimized cutoff points with sensitivity and specificity are presented in Table 3.A dependence on GFR level of each natriuretic peptide's influence on LVEF could not be detected.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

ROC Curves

Receiver-operator characteristic (ROC) curves for detection of an LVEF <40% by MR-proANP (red line), BNP (green line), and NT-proBNP (blue line). Abbreviations as in Figure 1.

Prediction of death

Median (lower, upper quartile) follow-up time was 68 (55, 77) months. A total of 492 patients (62%) died, 49 (6%) received a ventricular assist device, and 25 (3%) received a heart transplantation as their first event.

All 3 natriuretic peptides were significant predictors of death, even after adjustment for the pre-defined confounders (Table 4).Every doubling of MR-proANP, NT-proBNP, and BNP levels increased the risk of death by 54%, 27%, and 23%, respectively. When all 3 natriuretic peptides were entered in the model, only MR-proANP remained an independent predictor of death, as did age, sex, NYHA functional class, LVEF, and systolic blood pressure, but not the BNPs or GFR. The partial PEVs to predict death were significantly higher for MR-proANP (4.91%) than for NT-proBNP (3.12%) and BNP (2.78%) (Fig. 3).The influence of each natriuretic peptide on death was not dependent on GFR.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Prediction of Death

Relative importance of MR-proANP, NT-proBNP, and BNP to predict death as expressed by the partial proportion of explained variation. MR-proANP was a significantly stronger predictor of death than NT-proBNP and BNP. Abbreviations as in Figure 1.

Similar results were obtained for the prediction of cardiovascular death (91% of all-cause death, data not shown).

Factors influencing natriuretic peptide levels

Previous works have recognized a variety of noncardiac factors influencing natriuretic peptide levels, including age, sex, BMI, and renal function (22–24). In our study, all 3 natriuretic peptides were to a higher degree dependent on LVEF, GFR, and the presence of ankle edema, whereas age, BMI, and sex contributed only marginally to explain natriuretic peptide levels. This supports the use of natriuretic peptides as biomarkers for the severity of HF, since a decline in LVEF, fluid retention, and HF-associated deterioration in renal function (the so-called cardiorenal syndrome) are all expressions of the progression of HF.

Age, sex, and BMI have been described as factors influencing natriuretic peptide levels in subjects without HF, so it has been suggested that reference values should be adjusted to these confounders (25). Although this might be relevant in diagnosing unknown HF (26), our data suggest that the influence of age, sex, and BMI is negligible in patients with known chronic HF when other, more disease-related factors such as ventricular function, volume status, and renal function become relevant.

Estimation of the severity of systolic dysfunction by natriuretic peptides

MR-proANP, NT-proBNP, and BNP performed well in detecting a reduced LVEF (<40%, AUCs from 0.73 to 0.80). The AUCs are on the same order of magnitude as in previous reports evaluating the performance to detect an LVEF <40% in patients with known cardiac disease (14) but lower than in studies investigating the usage of natriuretic peptides for the diagnosis of unknown heart failure. A potential explanation for this finding is that HF treatment by reducing wall stress confounds the relation between LVEF and natriuretic peptides found in untreated patients.

NT-proBNP performed slightly worse than the other 2 peptides, a difference that reached statistical significance due to the large patient population but can be considered of minor clinical relevance. Data on a differential influence of renal function on natriuretic peptides, which might explain the diagnostic inferiority of NT-proBNP compared with BNP, are conflicting (16,17). In our study, GFR had a similar influence on the 3 natriuretic peptides, and consistently, the ability of each natriuretic peptide to detect an LVEF <40% was not dependent on GFR. Therefore, it seems that in a chronic HF population: 1) there is no need for threshold adaption of natriuretic peptides according to GFR; and 2) even in the presence of renal dysfunction, there is sustained, similar diagnostic performance in all 3 natriuretic peptides.

A previous study found that NT-proANP had a significantly lower diagnostic accuracy (AUC: 0.65) than NT-proBNP (AUC: 0.79) and BNP (AUC: 0.83) in patients with left ventricular dysfunction (14). This is in accordance with other investigations showing that the inverse relation of LVEF with BNP was stronger than with ANP or NT-proANP (12,13). In contrast, another study indicated a better detection of reduced LVEF by MR-proANP than by NT-proBNP (27). We believe that differences in the analytical method used are the most probable reasons for the comparatively high diagnostic power of MR-proANP. Conventional NT-proANP assays use an antibody against the N-terminal region of proANP1-98 combined with a second antibody against either the midregion (28), or C-terminal region (29). However, under certain conditions, the amino-terminal region might be minimally accessible for antibody binding (6,30), because proANP1-98 can be subject to further fragmentation (5,31). We used a conceptually different sandwich immunoassay for MR-proANP (amino acids 53 to 90) (15), which has shown more biochemical stability and might therefore explain the improved diagnostic accuracy of MR-proANP (12–14).

Whereas NT-proBNP has been previously described as a relatively stable molecule in vitro (32), it is subject to truncation in vivo (5,6) to an even higher degree than NT-proANP, and especially in NT-proBNP, the smallest fragments can only be detected by assays directed at the central part of the molecule, leading to an underestimation of up to 40% with conventional assays (5).

Prediction of death

There are sparse data comparing the prognostic power of peptides of the ANP and the BNP families. Tsutamoto et al. (8) found BNP more useful than ANP to predict mortality in chronic HF, which can be attributed to the longer plasma half-life of BNP (4,33). Gegenhuber et al. (28) found that MR-proANP and BNP were comparable predictors of 1-year survival in patients hospitalized for acute HF. A similar prognostic power of NT-proANP and NT-proBNP to predict mortality was also found after myocardial infarction (29) and in community-based populations without HF (34).

The reason for the prognostic superiority of MR-proANP might again be explained by the higher biological stability of the molecule and the usage of an assay directed to the midregion of the molecule. Furthermore, it has been shown that NT-proANP (compared with NT-proBNP and BNP) levels are very robust against variability due to physical activity (11). GFR, however, as shown in our study, does not seem to be a major confounder of the prognostic power of natriuretic peptides and therefore does not account for any differences in the prediction of death.

Study limitations

A recent study has shown that the NT-proBNP assay we used markedly underestimates NT-proBNP values, whereas the BNP assay we used measured both BNP and proBNP (7). Therefore, we must assume that the comparative diagnostic and prognostic performance of natriuretic peptides reflect assay rather than analyte properties, so that our conclusions can only refer to the assays used in the present study and not to other assays or the natriuretic peptides as analytes themselves. To answer the question of which analyte is better, more sensitive and specific assays for these molecules are needed.

The study population consisted of patients from previous prospective studies, of which 100 patients had been randomized to beta-blocker or placebo for approximately 13 months. Given the high total patient number and the high observation time of the present study, we estimate the inherent bias as minimal.

Since MR-proANP is a new marker, no data exist on the long-term in vitro stability at −80°C. However, extensive in vitro stability testing at various conditions (15) and the high diagnostic and prognostic accuracy indirectly imply a negligible noise due to in vitro instability.

Our results cannot be extrapolated to a patient population without a history of systolic HF. Especially the diagnostic accuracies of natriuretic peptides to detect an LVEF <40% are likely to be lower in a less selected HF population.