Association of Copeptin and N-Terminal proBNP Concentrations With Risk of Cardiovascular Death in Older Patients With Symptoms of Heart Failure FREE

A central part in evaluation of elderly patients with symptoms of heart failure is to identify simple tools that can aid the clinician in identifying high-risk and low-risk patients. Combining a biomarker produced locally in the myocardium with a marker produced centrally in the body may be useful in patients with symptoms of heart failure. Studies have consequently tried to establish the clinical use of different markers in the circulation.

One such established marker is B-type natriuretic peptide^1- 2 and the N-terminal fragment of its precursor (NT-proBNP).^3- 4 Vasopressin, a noncardiac plasma marker of cardiovascular disease, is released from the neurohypophysis in response to changes in plasma osmolality and is involved in osmoregulation and cardiovascular homeostasis. The plasma concentration of vasopressin increases in patients with heart failure and is associated with left ventricular dysfunction.⁵ However, vasopressin as such is not a useful plasma marker, because it rapidly degrades in the circulation. Instead, copeptin, the C-terminal fragment of provasopressin, has emerged as a promising surrogate target for measurement of vasopressin concentration and also seems useful in cardiovascular disease.^6- 9

The primary diagnosis of heart failure remains a major challenge, and some reports have suggested that up to 50% of patients may go unrecognized.^10- 11 Measurement of copeptin concentrations has been tested mainly in patients with severe infection,¹² and some reports have suggested that copeptin measurement may have a role in patients with cardiovascular disease.^13- 14 However, to our knowledge, the value of measurement of copeptin concentration in elderly patients with symptoms of heart failure and the association with clinical outcomes has not yet been examined. We therefore evaluated the association of combined measurement of plasma copeptin and NT-proBNP concentrations with mortality in an elderly primary care population with symptoms of heart failure.

The primary design of the study has been published previously.¹⁵ Briefly, elderly patients with symptoms associated with heart failure at a primary health care center were evaluated. The study population all came from one center in southeast Sweden, and all patients were included between January 2 and December 4, 1996. All patients aged 65 to 87 years with symptoms and/or signs of heart failure were evaluated. Data from patient records and the official registry were assessed until December 31, 2009.

Symptoms and/or signs of heart failure included shortness of breath, peripheral edema, and/or fatigue. For each patient, a new patient record was started and clinical examination was performed, including New York Heart Association (NYHA) functional class, Doppler echocardiography, and standard blood analyses. Heart failure was diagnosed as a combination of symptoms and/or signs and an ejection fraction less than 40% on echocardiography.^16- 17 Mortality information was obtained from the Swedish National Board of Health and Welfare, which registers all deaths of Swedish citizens. Written informed consent was obtained from all patients, and the study was approved by the regional ethical review board in Linköping.

Doppler echocardiography examinations (Acuson XP-128c; Siemens AG, Munich, Germany) were performed with the patient in the supine left position. Values for systolic function,¹⁸ expressed as left ventricular ejection fraction, were categorized into 4 classes with interclass limits of 30%, 40%, and 50%.^19- 20 Normal systolic function was defined as left ventricular ejection fraction of 50% or greater. Severely impaired systolic function was defined as ejection fraction less than 30%.²¹

All blood samples were obtained while the patients were at rest in a supine position. The blood samples were collected in plastic vials containing EDTA. The vials were placed on ice before chilled centrifugation at 3000 g and then frozen at −70°C. No sample was thawed more than twice.

Concentrations of proBNP 1-76 (NT-proBNP) were measured on the Elecsys 2010 platform (Roche Diagnostics, Mannheim, Germany). The total coefficient of variation was 4.8% at 26 pmol/L and 2.1% at 503 pmol/L (n = 70). Plasma concentrations of copeptin were measured on the Kryptor Compact platform (BRAHMS, Hennigsdorf, Germany). The interassay coefficients of variation were less than 15% at 20 pmol/L, less than 13% for 20 to 50 pmol/L, and less than 8 pmol/L for concentrations greater than 50 pmol/L, according to manufacturer; assay validation has been reported previously.²²

Descriptive data are presented as percentages or means with SDs. Analyses were calculated using an unpaired 2-sided t test for continuous variables; the χ² test was used for discrete variables. Patients were stratified into quartiles of copeptin and NT-proBNP concentration, providing better sensitivity than the other approach sometimes used, ie, classifying values as above or below the median.

Evaluation of correlation was analyzed using the Pearson product-moment correlation coefficient. Cox proportional hazard regression analyses were used to analyze the risk of both all-cause and cardiovascular mortality during the follow-up period. Kaplan-Meier analyses were used to illustrate actual all-cause and cardiovascular mortality. The assumption of proportionality was tested, because long follow-up was used. Because the assumption was not fulfilled, we chose to present mortality data in 2 steps: as short-term mortality with a follow-up time of 5 years and as long-term mortality with a follow-up time of up to 13 years. To test the independent prognostic information of copeptin concentration in addition to NT-proBNP concentration in a multivariate analysis, a likelihood ratio test was also performed. Censored patients were those still alive at the end of the study period or those who died of reasons other than the variable analyzed, eg, cardiovascular mortality. Completers comprised patients with all-cause or cardiovascular mortality during the study period.

To evaluate the possible prognostic value of adding biomarkers in a multivariate analysis, a weighted variable was produced based on the β value in the multivariate Cox proportional hazard regression analysis. A receiver operating characteristic (ROC) curve analysis was performed using the method of DeLong et al,²³ with weighted variables. A ROC analysis was also performed testing the ability to identify risk of cardiovascular mortality using a 3-step model. Step 1 included clinical variables only (symptoms [fatigue, dyspnea, previous or present ischemic heart disease, diabetes, NYHA functional class III, male sex, age 70-75 years or >75 years]; signs [rales, jugular distension, peripheral edema]). Step 2 included clinical variables plus measurement of NT-proBNP concentrations. Step 3 included clinical variables plus measurement of NT-proBNP and copeptin concentrations, in 3 different ROC curves. The models were constructed using the β coefficients from a Cox regression multiplied by the value of the individual variable.

We also analyzed copeptin levels as a continuous variable and examined the association between a 10-pmol/L increase in copeptin concentration and mortality.

Considering the number of events in the cohort (226 vs 146), we had approximately 80% power (α = .05) for delivering a hazard ratio of 1.5 and 1.7 (for all-cause mortality and cardiovascular mortality, respectively, when comparing the fourth quartiles of the 2 biomarkers).

P < .05 was considered statistically significant, based on a 2-sided evaluation. All data were analyzed using Statistica version 10 (Statsoft Inc, Tulsa, Oklahoma).

Of 548 invited elderly patients with symptoms associated with heart failure according to patient records, 510 agreed to participate in the study (participation rate, 93%). Thirty-eight patients decided not to participate because of transportation difficulties, severe illness, mental insufficiency, or incapacity. Another 36 patients did not allow blood sampling, and echocardiograms from 4 patients were not of acceptable quality. Thus, the final patient population consisted of 470 patients. The mean age of the study population was 73 years, with an equivalent distribution of men and women (245 vs 225, respectively) (Table 1).

Study population data were assessed for 13 years (median, 4725 [range, 242-5112] days), with December 31, 2009, as the last follow-up date. Patients who survived during the follow-up period had a median observation period of 4923 (range, 4773-5112) days; those who did not survive had a median observation period of 2723 (range, 242-5018) days. During follow-up there were 226 deaths from all causes (48%), including 146 cardiovascular deaths (31%). The distribution of the different diagnoses of mortality is presented in eTable 1. No patients were lost to follow-up.

During the 13-year follow-up period, all patients were examined regarding all-cause and cardiovascular mortality. The mortality distribution across the copeptin quartiles ranged between 26.5% (first quartile) to 46.6% (fourth quartile) for cardiovascular mortality and between 38.5% (first quartile) to 69.5% (fourth quartile) for all-cause mortality. The corresponding distribution for NT-proBNP was 15.9% (first quartile) to 56.9% (fourth quartile) for cardiovascular mortality and between 28.3% (first quartile) to 75.9% (fourth quartile) for all-cause mortality (Table 2). We have chosen to present mortality data in 2 steps: as short-term and as long-term mortality.

In multivariate models comparing the second, third, and fourth quartiles against the first quartile of the biomarkers (Table 3), concentrations of copeptin and NT-proBNP were associated with long-term all-cause mortality (up to 13 years of follow-up), both separately (models 1 and 2) and in combination (model 3). The independent association of copeptin with mortality was confirmed using the likelihood ratio test (P = .02). Similar results were obtained in multivariate models examining cardiovascular mortality (Table 4), with a similar log likelihood trend test result for copeptin in the presence of NT-proBNP (model 3) (P = .048).

In a ROC analysis of cardiovascular mortality using the 10-year follow-up period, the area under the curve (AUC) increased from 0.70 to 0.74 (95% confidence interval [CI], 0.68-0.79) by adding NT-proBNP concentration to clinical examination variables, and increased to 0.76 (95% CI, 0.71-0.82) by adding copeptin concentration to clinical examination variables and NT-proBNP concentration. A significant difference in the AUC between the clinical variables alone and the addition of NT-proBNP concentrations plus those for copeptin was observed (difference, 0.06 [95% CI, 0.02-0.10]; P = .002). By only using NT-proBNP and applying the fourth quartile as cutoff value, 66 of 116 patients (56.9%) were correctly identified as at risk for cardiovascular mortality, whereas when copeptin was added (fourth quartile) the corresponding value increased to 36 of 52 patients (69.2%).

In a multivariate model using the fourth quartiles of the 2 markers at the same time vs quartiles 1 through 3 together with the clinical variables, the markers displayed prognostic information even when analyzed in the same model (all-cause mortality: hazard ratio [HR] for fourth quartile of copeptin, 1.70 [95% CI, 1.25-2.31] and HR for fourth quartile of NT-proBNP, 2.17 [95% CI, 1.60-2.94]; cardiovascular mortality: HR for fourth quartile of copeptin, 1.68 [95% CI, 1.15-2.45] and HR for fourth quartile of NT-proBNP, 2.93 [95% CI, 2.02-4.24]) (Table 5).

The prognostic strength of copeptin measurement in other diseases was evaluated in a univariate Cox proportional hazard regression (eTable 2). Measurement of copeptin concentration obtained the best prognostic information from heart failure and myocardial infarction. Among the noncardiovascular diseases, only infection was statistically significant in terms of prognostic information. In regard to short-term mortality (up to 5 years of follow-up), a slightly higher hazard ratio of copeptin and NT-BNP for all-cause and cardiovascular mortality was identified in a multivariate analysis (eTable 3).

Comparison of extreme values should be interpreted with caution. Recognizing this, Kaplan-Meier analyses of the 4 quartiles of copeptin and NT-proBNP concentrations, respectively, are shown in eFigure 1 and eFigure 2.

The Kaplan-Meier analysis of all-cause mortality (eFigure 1) shows the all-cause mortality associated with different combinations of copeptin and NT-proBNP, from a group with low plasma concentrations of both markers (group 1, with 63.7% survival) to a group with a combination of high plasma concentrations of both markers (group 4, with 16.5% survival). Prognostic information obtained by the markers was greater when both were combined. In absolute terms, 44 of 52 patients (84.6%) were correctly identified as at high risk in the group with both markers in the highest quartiles, compared with 99 of 287 patients (34.5%) with all-cause mortality and concentrations of both markers in the first quartile.

The Kaplan-Meier analysis of cardiovascular mortality (eFigure 2) similarly shows cardiovascular mortality associated with different combinations of the markers, from a group with lowest concentrations of both markers (group 1, with 74.6% survival) to a group with high concentrations of both markers (group 4, with 23.0% survival). In absolute terms, 36 of 52 patients (69.2%) were correctly identified as at high risk in the group with both markers in the highest quartiles, compared with 61 of 287 (21.3%) with cardiovascular mortality and concentrations of both markers in the first quartile.

In the multivariate analysis, for every increase of 10 pmol/L in copeptin levels, there was a corresponding increase in all-cause mortality of 1.21 (95% CI, 1.10-1.34) and in cardiovascular mortality of 1.22 (95% CI, 1.08-1.37) (Table 3, model 3).

The absolute risk reduction calculation for all-cause mortality using the markers in combination indicated that 6.8% more patients were correctly identified as at risk compared with measurement of only NT-proBNP concentration over a 13-year follow-up period. If the follow-up period was restricted to 10 years, an 8.4% increase in patients correctly identified as at risk was noted. In the same calculation for cardiovascular mortality over 13 years of follow-up, 12.3% more patients could be correctly identified as at risk. If the follow-up period was restricted to 10 years, 19.6% more patients were identified by the combination of the 2 markers.

The use of 2 plasma biomarkers, one well known in the management of heart failure (NT-proBNP) and the other a new marker of pituitary vasopressin release (copeptin), were evaluated during a follow-up time of 13 years. From our combined results, additive prognostic information was demonstrated for both all-cause and cardiovascular mortality.

Even after a 13-year follow-up period, copeptin concentration possessed prognostic information concerning mortality risk. Both NT-proBNP and copeptin provided independent prognostic information during a follow-up period of 13 years. To our knowledge, this has not been demonstrated in a corresponding population with such a lengthy follow-up period. A problem of using very long follow-up periods is that other diseases without connection to the markers, such as malignancies, will gain statistical influence.

In the ROC analyses, significant additive information was gained by adding one marker to the other in addition to clinical variables (AUC, 0.70-0.76). The use of 2 biomarkers alone or in combination has been reported by others.^8- 9,24 Apart from the study by Khan et al,⁹ the AUC values are consistent with ours (0.67-0.81vs 0.70-0.76 for the current study). The study by Khan et al used a slightly different methodological approach that might have produced higher AUC values (0.75-0.84). However, the ROC method is not informative in terms of when a specific death occurs, because it does not weigh the influence of whether death occurs early or late during the follow-up period, something that the Cox proportional hazard regression analysis does.

Because the assumption of proportionality was not fulfilled, we presented data in 2 steps; one for short-term and one for long-term mortality. This was done because the long-term follow-up in this study was unusually long, with a greater influence of other variables on mortality. In the evaluation of prognostic information of the markers, the second, third, and fourth quartiles of peptide concentrations were entered in a multivariate analysis of well-recognized clinical variables (Tables 3 and 4). From these analyses, copeptin exhibited independent prognostic information even when analyzed together with NT-proBNP. In addition, the prognostic information from the highest quartiles of the plasma markers analyses were evaluated for both short-term and long-term mortality (Table 5 and eTable 3). Because this comprises an evaluation of patients with the highest plasma concentrations, the results should be interpreted with caution. However, the combination of both markers in the highest quartiles resulted in increased risk for all-cause and cardiovascular mortality. This could also be seen in the short-term mortality table (eTable 3), in which both copeptin and NT-proBNP exhibited a slightly higher hazard ratio for both all-cause and cardiovascular mortality.

In a study by Neuhold et al,²⁴ the population was significantly younger (mean age, 57 years), whereas our patient cohort (mean age, 73 years) better conforms to the average age of those with heart failure in Sweden. Also, the population presented by Neuhold et al represents a population with more severe disease compared with our study population, as can be seen from the distribution of NYHA functional classes and also from the copeptin concentrations. Because our population consisted of a mixture of patients with heart failure in whom only approximately one-third had an ischemic background and none had overt ischemic symptoms at study inclusion, this may more accurately represent the spectrum of a population with heart failure that clinicians face in primary care. Moreover, Neuhold et al used the median plasma concentration as a cutoff. We chose a different strategy using the fourth quartiles, because we have found this strategy less sensitive to other factors that might influence the relatively low concentrations. The follow-up periods of other published studies were often between 1 and 3 years, whereas our study had a follow-up period of 13 years.

The relation between vasopressin and heart failure is not fully understood. Vasopressin is secreted from the neurohypophysis because of increased osmolality following reduced cardiac output and activation of baroreceptors in the carotid sinus.²⁵ Nevertheless, heart failure is associated with low osmolality, while plasma concentrations of vasopressin are increased in patients with heart failure. Accordingly, it may be that nonosmotic stimuli are involved.²⁶ As a vasoconstriction hormone, vasopressin also is involved in the adverse cardiac remodeling process via vasopressin 1a receptors. Stimulation of this receptor leads to increased myocyte protein synthesis and promotes development of myocardial hypertrophy, decreased contractility, and increased cardiac afterload.²⁷ Studies have demonstrated that vasopressin stimulates cardiac fibroblasts, resulting in increased myocardial fibrosis.²⁸ Thus, vasopressin plays an important pathophysiological role in patients with heart failure, because increased release of vasopressin is associated with both increased preload and increased filling pressures as well as increased afterload. Clinical observations have confirmed that plasma concentrations of vasopressin are increased in patients with severe heart failure (NYHA class III-IV).⁶

The objective of this study was to apply markers in a patient group commonly encountered in primary care, ie, elderly patients who often present with other diseases, making interpretation of symptoms difficult. The original design of our cohort study did not allow us to assess diagnostic elements of biomarker measurement. Instead, we focused solely on the prognostic information of the markers when applied in a primary care population. These data, together with our findings of the prognostic information provided by measurement of copeptin concentrations in elderly patients with symptoms of heart failure, suggest that vasopressin may be a potential target for therapeutic intervention.