Methods and material

Data collection

Data collection has been described in detail previously.10 11 Briefly, in 1995–6 screening was carried out using all records (n = 1168, patient age 65–82 years) of patients with symptoms and/or signs associated with heart failure who had contacted the primary healthcare centre in Kinda municipality, a mainly rural population of 10 300 inhabitants in south-eastern Sweden. All individuals, in whom heart failure could not be ruled out by carefully scrutinising (n = 548) the patient records, were invited to participate in the study. Individuals with atrial fibrillation were not invited.10 Of the 548 individuals, 510 accepted (participation rate 93%). Blood samples were obtained in 474 individuals (87% of the initial 548 individuals). Complete echocardiography was not possible in 59 individuals. Heart failure was defined as asymptomatic left ventricular systolic dysfunction (left ventricular systolic dysfunction (LVEF <0.40) or symptoms on heart failure (NYHA II-IV) and cardiac dysfunction (LVEF <0.40 or diastolic dysfunction (pseudonormalisation or restrictive filling pattern) or both).

Echocardiography

Doppler echocardiographic examinations (Accuson XP-128c), evaluating systolic as well as diastolic functions, were performed with the patient in the supine left position. Values for systolic function, expressed as LVEF, were determined semiquantitatively, being categorised into four classes with interclass limits of 30%, 40% and 50%. Normal systolic function was defined as LVEF >50%. Severe systolic dysfunction was defined as LVEF <30%. Diastolic dysfunction was defined as a reduced ratio of peak early diastolic filling velocity in relation to peak filling velocity at atrial contraction (E/A ratio).12 Pseudonormal filling pattern was identified by abnormal pulmonary venous flow pattern. Analysis of both parameters was age-adjusted. Diastolic function was categorised as follows: normal, abnormal filling pattern, pseudonormal filling pattern and restrictive filling pattern. Left atrial (LA) volume was obtained by M-mode.12 Valvular disease was quantified as described previously.11

Morbidity

History of diabetes, ischaemic heart disease, chronic obstructive pulmonary disease and hypertension was obtained as described previously.10 11

ProBNP and BNP analysis kits

Oslo

We quantified proBNP using a radioimmunoassay utilising primary antiserum raised against synthetic proBNP 1–21 conjugated in the C-terminus to bovine albumin.13 Inter-immunoassay variation was 14.6% at 219 pg/ml.

Roche (Elecsys)

Plasma concentrations of proBNP were analysed by electrochemiluminescence (ECLIA) on the Elecsys 2010.14 This assay uses two epitopes—one in the 39–50 region, and another in the C-terminus (1–21 region) of proBNP 1–76. Total variation (intra-assay and inter-assay) coefficient is reported (measured to be between 2.9% (2.1%) and 6.1% (4.8%) in the high and low range, respectively (at our laboratory).

Copenhagen

Plasma concentrations of proBNP were measured by a processing independent immunoassay described in detail recently.15 16 This assay uses one epitope: 1–10. Before incubation, plasma was treated with trypsin, which cleaved both NT-proBNP (fragment 1–76) and proBNP (fragment 1–108) into a uniform proBNP 1–21 fragment. Inter-immunoassay variation was 20% at 135 pg/ml and 10% at 591 pg/ml.

Shionogi

Plasma concentrations of BNP were measured by a non-extraction immunoradiometric immunoassay (Shionogi).16 Inter-immunoassay variation was 9.3% at 52 pg/ml and 5.4% at 193 pg/ml.

eGFR

eGFR was calculated by the four-component modification of diet in renal disease (MDRD) equation incorporating age, race, sex and serum creatinine concentration17: eGFR  =  186 × (serum creatinine (mg/dl))^−1.154 × (age (years))^−0.203. For women the product of the equation has to be multiplied by a correction factor of 0.742.17

Statistics

Patients characteristics, grouped according to absence or presence of eGFR <60 ml/min/1.73 m² ( = chronic kidney disease, CKD), are presented as percentage for dichotomous variables and means (median) and ranges for continuous variables. Baseline characteristics were compared with the use of χ² test for discrete variables and unpaired t tests (parametric) and Mann-Whitney U test (non-parametric) for continuous variables, as appropriate. ProBNP and BNP concentrations according to quartiles of eGFR were evaluated by one-way ANOVA analysis and Bonferroni correction (post hoc).

The crude effects of eGFR and creatinine on proBNP and BNP concentrations were evaluated by univariate analyses and the adjusted effects are (after adjustment for the confounders) presented in table 1. All linear covariates were evaluated by model control (linearity, variance homogeneity (Gaussian distribution around the regression line) and lack of interaction). Log-transformation was done where indicated. eGFR was log-transformed because of variance inhomogeneity. All logarithms are proportional (log_y(x)  =  log(x)/log(y)) and log_1.1 was chosen for eGFR and log_1.3 for creatinine to show the effects of 10% increases or decreases in eGFR and a 30% increase in creatinine on log(proBNP) and log(BNP) instead of a 2.7-fold increase (log) or 10-fold increase (log₁₀). Log_1.1 and log_1.3 were chosen over log and log₁₀ because these logarithms display the effect of clinically relevant changes in the parameter (10% and 30%) in a more direct fashion. Formal tests for interaction between the covariates in the final model were applied. In addition, eGFR was dichotomised (plus or minus 60 ml/min/1.73 m²) (chronic kidney disease) to see whether the effects of the other variables interacted with renal dysfunction. Logistic regression analysis was applied to evaluate whether the four peptides could discriminate between the presence or absence of LVEF <40 and presence or absence of heart failure by groups of presence or absence of CKD. Suggested cut-off values at a sensitivity at 90% are presented in table 2.

The effects of parameters estimates on log₁₀(proBNP) and log₁₀(BNP) were transformed back to percentage changes as follows: β =  log₁₀(proBNP₂/BNP₂) − log₁₀(proBNP₁/BNP₁)  =  log₁₀(proBNP₂/BNP₂/proBNP₁/BNP₁) ⇔ 10^β =  proBNP₂/BNP₂/proBNP1/BNP₁. Percentage changes were then calculated as: ((proBNP₂/BNP₂/proBNP1/BNP₁)−1) × 100% =  (10^β −1) × 100%.

Logistic regression analysis (receiving operating characteristics (ROC) curves) was used to evaluate whether the diagnostic sensitivity and specificity for left ventricular systolic dysfunction (LVSD) of the four peptides differed by groups of CKD.

A p value <0.05 (two-sided) was considered significant. Analyses were made using Statistical Analysis Software (SAS 9.1).

Results

Patient characteristics according to absence or presence of chronic kidney disease are presented in table 1. Patients with chronic kidney disease were older (p<0.001), more frequently male (p<0.001), had lower concentrations of haemoglobin (p<0.001), more frequently diabetes (p = 0.022), higher body mass indices (p = 0.033), higher concentrations of BNP (p<0.001) and proBNP (p<0.001 for all) and slightly smaller right ventricles (p<0.001).

The relation between the different proBNP immunoassays is presented in figure 1A–C. The relations are linear and highly significant, but do not intersect (0,0) and β is not equal to 1, indicating that the different immunoassays quantify different amounts of proBNP.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

(A–C) Scatter plots illustrating the relation between plasma concentrations of proBNP measured with the three different methods.

The crude and adjusted effects of eGFR on proBNP and BNP concentrations are presented in table 3. In multivariate models the effects of eGFR on the different proBNP and BNP concentrations are within the same range and confidence intervals overlap. The crude and adjusted effects calculated in percentage are presented in figure 2. The adjusted effect of a 10% decrease in eGFR on proBNP and BNP concentrations varies from 4–13%. We did not observe any interaction between eGFR and presence or absence of LVEF <40, or eGFR and presence or absence heart failure and proBNP or BNP levels (p values not shown).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Bars (error bars are 95% confidence interval) illustrating the crude and adjusted effects of eGFR on proBNP and BNP concentrations.

Logistic regression analyses are presented in table 2. All four peptides could discriminate between presence or absence LVEF <0.40 and presence or absence heart failure independently of absence or presence of CKD.

Discussion

The novel findings of this study are (1) that the adjusted effects of eGFR on proBNP measured with different immunoassays are low and within the same range, (2) that the crude effect of eGFR on proBNP concentrations is reduced substantially after adjustment for important confounders, and (3) that the adjusted effects of eGFR on proBNP and BNP are similar.

Recently, it has been suggested that proBNP and BNP should be evaluated in relation to renal function.18 Our analyses support this point of view, but also suggest that the effect of eGFR on proBNP or BNP concentrations is modest and partly confounded. This finding is supported by recently published experimental data from 165 hypertensive patients.7 This should be kept in mind in clinical practice, before renal function is offered as an explanation for increased proBNP or BNP concentrations.

It is noteworthy that the effect of eGFR on proBNP concentrations measured with three different immunoassays is similar (table 3) even though plasma concentrations are different (fig 1 A–C). The fragments detected by each immunoassay seem therefore to be cleared by the kidneys to a similar extent. This notion is supported by different physiological experiments with proBNP measured with three different immunoassays.4 5 6 7 Molecular heterogeneity of proBNP19 and differences in the immunoassays rather than different renal clearance seems therefore to explain different concentrations of plasma concentrations (fig 1 A–C). Finally, in adjusted analyses the effect of eGFR on proBNP and BNP concentrations seems similar (table 3 and fig 2) in accordance with the physiological experiments mentioned above, but in contrast to analyses from the Dallas Heart Study,20 which may be explained by different populations studied. Our analyses are therefore supported by experimental data, but the observed associations do not separate the potential different mechanisms for intra-renal clearance of proBNP and BNP: endothelial degradation, glomerular filtration and intra-tubular degradation.

In our logistic regression analyses we observed that the ability of the four assays to discriminate between absence and presence of LVEF <40 and heart failure did not differ in groups by presence or absence of CKD (table 2). Therefore, plasma concentrations of proBNP measured with three immunoassays and BNP (Shionogi) are associated with eGFR to a similar extent and can discriminate between absence and presence of heart failure to the same extent in patients with presence or absence of CKD and suspected heart failure. All peptides can therefore be used in patients with CKD if the values are evaluated in the clinical context, but it should be noted that the cut-off limits at a sensitivity at 90% only changes marginally and that the specificity decreases (higher false positive rate) in patients with CKD (table 2). The following considerations should be taken into account when a BNP or proBNP assay is selected for clinical use: (1) has the assay been evaluated in patients with similar demographic data—for example, CKD or atrial fibrillation?21 The existence of a single universal cut-off limit for heart failure is debatable, and the limits offered by the American and European Guidelines for heart failure are different.22 23 (2) Which assay is the department of clinical chemistry familiar with? Within-day and between-day analytical variation should be avoided and impression goals may differ between the assays.24 (3) Price? To keep down public healthcare expenses. (4) Optimal evaluation of the patients prognosis? The levels may differ from assay to assay (table 1 and fig 1A–C). The Roche assay is, from this point of view, superior to the other assays since it has been evaluated in large European cohorts with suspected25 and chronic heart failure.26 27 28

Some methodological considerations of this study should be discussed. Our analyses are retrospective and multiple testing may produce our results. An association between renal function and proBNP and BNP has, however, been observed several times, making this explanation unlikely. The generalisability of our results may also be limited. Our cohort consists primarily of elderly patients presenting with symptoms associated with heart failure without atrial fibrillation, and with a low to moderate frequency of left ventricular systolic dysfunction (13%), angina pectoris (25%) and severe chronic kidney disease (class IV-V) (15%). Our results should therefore be extrapolated with caution to other patient categories, but our results are in accordance with data published on patients with chronic heart failure and proBNP (Roche)2 and in accordance with experimental data from hypertensive subjects.7 We used E/A-ratio and retrograde lung vein flow and not E/e′12 as an estimate for diastolic function and LA diameter rather than LA volume index29 as an estimate for left atrial size. It may be argued that these measures are less accurate in determining LV diastolic function and that we, therefore, have overestimated the effects of eGFR on proBNP and BNP concentrations (residual confounding).30 eGFR underestimates true GFR,31 which may have resulted in misclassification of chronic kidney disease (table 1) and, in turn, biased the linear effect of renal function on proBNP and BNP concentrations. This is a cross-sectional study and we have measured the inter-individual relation between eGFR and proBNP and BNP concentrations. In theory, the intra-individual relation may be different. The strength of our data is a high participation rate (87%) (minimal selection bias) and the fact that we have adjusted for known clinical confounders and echocardiographic measures reflecting intracardiac pressures (minimal unmeasured confounding),32 making it likely that we have observed the true effect of eGFR on proBNP and BNP concentrations. When concentrations of proBNP measured with different immunoassays are compared, problems with scales may arise owing to different levels of measured plasma concentrations. We have avoided this phenomenon by using relative changes instead of absolute differences in the statistical models. Finally, proBNP measured by the Copenhagen immunoassay was measured on frozen samples (stored at −70°C and thawed once) some years after the study was performed, which may explain the deviation of data in the low range.

In conclusion, the effects of eGFR on proBNP concentrations measured with three different immunoassays and BNP are modest and within the same range. The effect of renal function on proBNP and BNP concentrations is reduced substantially after adjustment for important echocardiographic measures and clinical variables. These findings should be kept in mind before renal function is offered as an explanation for increased proBNP or BNP concentrations in elderly people with suspected heart failure.