Receiver operating characteristic curve analysis of amino-terminal pro–brain natriuretic peptide (NT-proBNP) testing for the estimation of 1-year mortality in dyspneic patients. The area under the receiver operating characteristic curve was 0.76 (P<.001).
Rate of mortality of dyspneic patients in the emergency department by 1 year, expressed as a function of amino-terminal pro–brain natriuretic peptide (NT-proBNP) in deciles. A strong threshold effect is noted at the seventh NT-proBNP decile, corresponding to an NT-proBNP concentration in excess of 972 pg/mL. P<.001 for trend across groups.
Age-adjusted Kaplan-Meier survival curves demonstrating the rates of mortality by 1 year associated with an elevated amino-terminal pro–brain natriuretic peptide (NT-proBNP) concentration at emergency department presentation with dyspnea. This risk was observed in those both with (A) (P = .001) and without (B) (P<.001) acute congestive heart failure at presentation.
Januzzi JL, Sakhuja R, O’Donoghue M, Baggish AL, Anwaruddin S, Chae CU, Cameron R, Krauser DG, Tung R, Camargo CA, Lloyd-Jones DM. Utility of Amino-Terminal Pro–Brain Natriuretic Peptide Testing for Prediction of 1-Year Mortality in Patients With Dyspnea Treated in the Emergency Department. Arch Intern Med. 2006;166(3):315-320. doi:10.1001/archinte.166.3.315
Copyright 2006 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2006
Amino-terminal pro–brain natriuretic peptide (NT-proBNP) is useful for diagnosis and triage of patients with dyspnea, but its role for predicting outcomes in such patients remains undefined.
A total of 599 breathless patients treated in the emergency department were prospectively enrolled, and a sample of blood was obtained for NT-proBNP measurements. After 1 year, the vital status of each patient was ascertained, and the association between NT-proBNP values at presentation and mortality was assessed.
At 1 year, 91 patients (15.2%) had died. Median NT-proBNP concentrations at presentation among decedents were significantly higher than those of survivors (3277 vs 299 pg/mL; P<.001). The optimal NT-proBNP cut point for predicting 1-year mortality was 986 pg/mL. In a multivariable model, an NT-proBNP concentration greater than 986 pg/mL at presentation was the single strongest predictor of death at 1 year (hazard ratio [HR], 2.88; 95% confidence interval, 1.64-5.06; P<.001), independent of a diagnosis of heart failure. Other factors associated with death included age (by decade; HR, 1.20), heart rate (by decile; HR, 1.13), urea nitrogen level (by decile; HR, 1.20), systolic blood pressure less than 100 mm Hg (HR, 1.94), heart murmur (HR, 1.92), and New York Heart Association classification (HR, 1.38 for each increase in class). The NT-proBNP concentration alone had an area under the receiver operating characteristic curve (AUC) of 0.76 for predicting mortality; the other significant covariates combined had an AUC of 0.80. The final model for predicting death, combining NT-proBNP with other covariates associated with mortality, had a superior AUC of 0.82.
In addition to assisting in emergency department diagnosis and triage, NT-proBNP concentrations at presentation are strongly predictive of 1-year mortality in dyspneic patients.
B-type natriuretic peptide testing is useful for the diagnostic evaluation of patients with dyspnea and suspected acute destabilized congestive heart failure (CHF).1- 3 Including both the B-type natriuretic peptide and its profragment, amino-terminal pro–brain natriuretic peptide (NT-proBNP), this class of cardiac biomarkers is useful not only for diagnosis or exclusion of CHF but also for stratification of long-term risk of mortality in community-based populations without CHF4 and those with chronic CHF.5- 9 It is also useful to predict prognosis in individuals with non-CHF states, such as coronary artery disease,10- 19 pulmonary embolism (PE),20,21 and critical illness,22 and following cardiac transplantation.23 However, the role of the B-type natriuretic peptides for defining the longer-term prognosis of patients who present with dyspnea to the emergency department (ED)—the most common indication for B-type natriuretic peptide measurement—remains undefined.
In the primary results of the ProBNP Investigation of Dyspnea in the Emergency Department (PRIDE) Study, we reported the value of NT-proBNP testing for the diagnostic evaluation of patients with dyspnea and suspected acute CHF.1 To further extend the usefulness of NT-proBNP measurement, we evaluated the association of NT-proBNP concentrations at the time of ED presentation with dyspnea and vital status of study participants at 1 year.
The institutional review board approved all investigational procedures involved in this study. The design and results of the PRIDE Study were recently reported.1 Briefly, 599 patients who presented to an urban ED with complaints of dyspnea were enrolled in a blinded study that evaluated the role of NT-proBNP testing for the diagnosis of acute CHF. At 60 days following enrollment, a follow-up telephone call was made to the participants and/or their physicians to review subsequent medical course following presentation. Study physicians (blinded to the results of NT-proBNP testing) assigned a final diagnosis for each patient using all clinical information available from the time of presentation through the 60-day follow-up period, including the results of telephone contact with each participant, as previously described.1 The diagnosis of acute CHF was based on clinical records and results of direct contact with patients and their caregivers.
Of the 599 patients in the study, 209 (35%) were judged to have acute CHF, whereas 390 were not. Of those who did not have acute CHF, 150 had exacerbation of obstructive airway disease, 64 had pneumonia, 31 had acute coronary syndrome (ACS), 19 had acute PE, and 10 had acute bronchitis. In an additional 116 participants, various other diagnoses were recorded, including anxiety attacks, allergic reactions, and ascites.1
We performed NT-proBNP analysis using a commercially available immunoassay (Elecsys proBNP; Roche Diagnostics, Indianapolis, Ind) on an Elecsys 1010 analyzer (Roche Diagnostics), using established methods. This assay is reported to have less than 0.001% cross-reactivity with B-type natriuretic peptide.
In the PRIDE Study, it was determined that the optimal strategy for identifying the presence of acute CHF using NT-proBNP measurement was an age-stratified approach, with a diagnostic threshold of 450 pg/mL for patients younger than 50 years and 900 pg/mL for patients 50 years or older. The manufacturer's recommended cut points of 125 and 450 pg/mL (for ages <75 or ≥75 years) for excluding CHF had a strong negative predictive value (NPV) of 100%, whereas a more simplified, age-independent cut point of 300 pg/mL had an NPV of 99.7%.
For the purposes of the present study, vital status at 1 year was ascertained by study physicians via telephone interview of the primary care physician or cardiologist for each study participant and/or review of hospital medical records. Physicians obtaining follow-up data were blinded to the results of cardiac biomarker testing. Of the 599 study participants, follow-up was complete in 595 (99.3%). For the purposes of the study, those lost to follow-up were assumed to be alive.
Receiver operating characteristic (ROC) analyses were performed in an effort to assess the relationship between NT-proBNP concentration and 1-year mortality. Once an optimal cut point for predicting mortality was identified, sensitivity, specificity, positive predictive value (PPV), and NPV were generated. The ROC analyses were performed with Analyze-It software (Leeds, England).
Comparisons of baseline clinical characteristics between survivors and nonsurvivors were performed using χ2 or Fisher exact tests for categorical data and t tests or Wilcoxon rank sum tests for continuous data, as appropriate. The NT-proBNP concentrations are expressed as medians and interquartile ranges.
To identify independent predictors of death at 1 year following presentation with dyspnea, age-adjusted Cox proportional hazards regression models were first performed to examine clinical variables of interest, with 1-year mortality as the dependent variable. All covariates associated with 1-year mortality with age-adjusted P<.05 were potentially eligible for inclusion in the final multivariable Cox model. The final multivariable model was selected using stepwise Cox regression with P<.05 as the cutoff for retention in the model. The results of the final multivariable model are presented. The variables in the final multivariable model were assessed in pairs for all possible first-order interactions and found to have none. Kaplan-Meier survival curves were plotted, and groups were compared using the log-rank test. Statistical analyses were performed with the use of Stata software, version 8SE (Stata Corp, College Station, Tex). A 2-sided P<.05 was considered statistically significant.
Of the original cohort of 599 dyspneic study participants, 91 (15.2%) died within 1 year of enrollment. All but 5 participants (94.6%) who died at 1 year were admitted to the hospital at index presentation (compared with an overall 75.8% admission rate in the PRIDE Study), but of these 91 decedents, only 14 (15.4%) died during index hospitalization.
Of those patients who died by 1 year, 55 (60%) had acute CHF at presentation, whereas 36 (40%) did not. The discharge diagnoses of the 36 patients without acute CHF who died by 1 year included pulmonary thromboembolism (8.3%), chest pain or ACS (11.1%), bacterial sepsis (11.1%), carcinoma (11.1%), pneumonia (16.6%), and exacerbation of obstructive airway disease (22.2%), whereas other single diagnoses accounted for the cause of death in the remaining 7 patients (19.4%). Of the 36 patients without acute CHF who died by 1 year, 5 (13.8%) had had prior CHF.
Given the high percentage of decedents with CHF, it is not surprising that compared with those surviving, those who died were older and more likely to have prior structural heart disease (including cardiomyopathy or prior CHF), to have more prevalent symptoms or signs consistent with CHF, to be using medications for CHF management such as diuretics or digoxin, and to have physical examination and radiographic results consistent with CHF (Table 1). Laboratory values of patients who died by 1 year were notable for significantly worse renal function.
A total of 139 patients (23.2%) underwent echocardiography as the standard of care at a mean of 51 hours following presentation. In this selected patient population, there were no significant differences in echocardiographic results between those dying and those surviving at 1 year (data not shown), with the exception of a relationship between ejection fraction and mortality.
Among those who died by 1 year, the median NT-proBNP concentration at presentation was 3277 pg/mL (interquartile range, 1086-9868 pg/mL), which was significantly higher than the NT-proBNP concentration in those surviving at 1 year, which was 299 pg/mL (interquartile range, 71-1807; P<.001 for difference).
The ROC analyses were performed to assess the ability of NT-proBNP to predict mortality at 1 year in all patients. The NT-proBNP concentration was both sensitive and specific, reflected in the area under the ROC curve (AUC) of 0.76 (Figure 1; P<.001). The cut point that yielded 90% sensitivity was 120 pg/mL, which had a PPV of 20.1% and an NPV of 94.6%. The cut point that yielded 90% specificity was 6899 pg/mL, which had a PPV of 41.2% and an NPV of 88.7%. The cut point that offered the best balance of sensitivity and specificity for predicting 1-year mortality among the entire cohort of dyspneic patients in the PRIDE Study was 986 pg/mL, which was 79% sensitive and 68% specific, with a PPV of 31% and an NPV of 95%.
Following stratification by NT-proBNP deciles, the percentage of patients who died by 1 year was assessed (Figure 2). For participants with levels below the seventh decile of NT-proBNP (representing a cutoff value of 972 pg/mL), mortality rates were low; however, at this point, a threshold effect for mortality was observed. Above this threshold, there was a marked increase in mortality (P<.001 for trend), which was particularly great for participants with the highest levels of NT-proBNP at presentation.
Results of the multivariable model predicting 1-year mortality are given in Table 2. An NT-proBNP concentration greater than 986 pg/mL was the single strongest predictor of mortality at 1 year (hazard ratio [HR], 2.88; 95% confidence interval [CI], 1.64-5.06; P<.001). Among the factors in the model predictive of death at 1 year were age, heart rate, blood urea nitrogen level, systolic blood pressure less than 100 mm Hg, heart murmur on examination, and New York Heart Association classification at presentation. Neither hospitalization at index presentation nor intervening clinical events during index hospitalization (such as ACS) influenced the powerful prognostic impact of an elevated NT-proBNP concentration at presentation.
As a single variable, an NT-proBNP concentration greater than 986 pg/mL alone had an AUC of 0.76 for predicting death; combining each of the other significant covariates without NT-proBNP results yielded a model with an AUC of 0.80. The final model combining NT-proBNP results with other covariates yielded a superior AUC (0.82); this final model had a likelihood ratio χ27 of 104.89, yielding a highly statistically significant P value (<.001) for the fit of the model. The pseudo R2 for this model was 0.21.
In a secondary analysis, we sought to determine whether any echocardiographic variables were significant predictors of 1-year mortality in addition to the covariates given in Table 2. In this analysis, restricted to the 139 participants who underwent echocardiography during the index hospitalization (of whom 19% died by 1 year), an NT-proBNP concentration greater than 986 pg/mL (HR, 7.12; 95% CI, 1.67-30.4; P = .008), blood urea nitrogen decile (HR, 1.29; 95% CI, 1.01-1.66; P = .04), and ejection fraction decile (HR, 1.03; 95% CI, 1.00-1.06; P = .04) were all significant predictors of 1-year mortality, with systolic blood pressure less than 100 mm Hg of borderline significance (HR, 2.66; 95% CI, 0.89-7.91; P = .08).
The 1-year crude mortality rates for patients with an NT-proBNP concentration greater than 986 pg/mL were strikingly higher than those of participants with NT-proBNP concentrations of 986 pg/mL or less, in patients both without acute CHF (30.2% vs 5.9%; P<.001) and those with acute CHF, in whom the mortality rate of those with concentrations below 986 pg/mL was 0% at 1 year (vs 29.9% for those with concentrations above 986 pg/mL; P<.001).
Notably, among those 53 patients without acute CHF whose NT-proBNP concentration was above 986 pg/mL, 18 (34%) had had prior CHF. As noted herein, of these 18 patients, 5 had died by 1 year, accounting for only 13.8% of decedents without acute CHF but with elevated NT-proBNP concentrations.
Age-adjusted 1-year survival curves are shown in Figure 3, stratified by NT-proBNP concentration for those with and without CHF at ED presentation. As depicted, increased mortality risk associated with high NT-proBNP concentration was observed in participants both with and without acute CHF. The risk of mortality appeared early following presentation and remained constant throughout the entire year of follow-up, as demonstrated by the continuously diverging event rates in patients both with and without acute CHF at presentation. Notably, no patients with acute destabilized CHF and an NT-proBNP concentration of 986 pg/mL or less died by 1 year.
In our cohort of patients who presented with dyspnea to the ED, we observed that an elevated NT-proBNP concentration at presentation was strongly predictive of death by 1 year, whereas lower concentrations of NT-proBNP had a high NPV for excluding risk of death by 1 year. Although natriuretic peptides have been shown to predict prognosis in healthy individuals,4 those with chronic CHF,5- 9 and patients with a wide variety of medical conditions other than CHF,10- 17,19,24,25 the role of natriuretic peptide testing for predicting long-term prognosis in patients who present with dyspnea to the ED is unknown. Since evaluation of the dyspneic patient represents the area of heaviest use of natriuretic peptide measurement, demonstration of an association between natriuretic peptide concentrations and long-term risk of mortality in patients with undifferentiated dyspnea is thus particularly important. To our knowledge, such an association has not previously been reported. Above an NT-proBNP cut point value (986 pg/mL), comparable to the optimal cut points in the PRIDE Study for the diagnosis of acute CHF, the risk of mortality at 1 year began to rise significantly. This risk was observed independent of a diagnosis of acute CHF. Our findings thus establish the prognostic utility of natriuretic peptide testing for the dyspneic patient; not only is natriuretic peptide testing useful for the diagnosis,1,3 triage,2 and management26 of patients with dyspnea, but also the results of the same blood sample used for diagnosis in the ED may now be applied to risk-stratify patients with respect to likelihood of death in the ensuing year.
The previous experience examining the association between natriuretic peptide values and outcomes among dyspneic patients has largely been restricted to studies with shorter follow-up and included outcomes other than mortality2,27 or examined groups of patients consisting solely of those with acute CHF,28,29 in whom NT-proBNP concentrations at presentation did not predict death.28,29 Our study benefited from a large sample size and inclusion of dyspneic patients without acute CHF, and notably, our follow-up period extends to a year, which is considerably longer than prior studies of B-type natriuretic peptide in dyspneic patients. In this context, we demonstrate a remarkably robust association between presenting NT-proBNP levels and the risk of death in the year following presentation.
A recent analysis by Fonarow et al30 of patients with acute CHF demonstrated the important association between abnormal renal function, low systolic blood pressure, and risk of death. However, that study examined only patients with CHF, and natriuretic peptide results were not included in their analysis. Although our study confirms their findings, it also establishes the independent and additive importance of NT-proBNP for predicting death in dyspneic patients, regardless of diagnosis of CHF at presentation. In our sample, the AUC was 0.64 for a model that contained the same covariates described by Fonarow and colleagues (blood urea nitrogen level ≥43 mg/dL [15.35 mmol/L] and systolic blood pressure <115 mm Hg), compared with an AUC of 0.82 (P<.001 for differences in models) when an NT-proBNP concentration greater than 986 pg/mL was added to the analysis.
The risk of death by 1 year associated with an elevated NT-proBNP was not simply related to the unfavorable prognosis connected to a diagnosis of acute or chronic CHF. Whereas patients with acute CHF accounted for 60% of patients who died by 1 year, 40% of those who died by 1 year had diagnoses other than acute CHF, and the prognostic utility of an elevated NT-proBNP concentration was useful in these patients as well. Considering that NT-proBNP concentration is expected to be elevated in those with prior heart failure, we examined the event rates in patients with chronic CHF who presented with dyspnea from another cause in the PRIDE Study, and those patients accounted for only 13.8% of the deaths in the non–acute CHF group. Thus, some other diagnosis was operative in most patients without acute CHF who died. The importance of NT-proBNP for the prediction of death in patients without CHF likely reflects the value of this marker for predicting mortality among dyspneic patients with diagnoses other than CHF, such as coronary artery disease10,13,17,24,25 or PE,20,21 which may present in a manner similar to acute CHF. It is also worthwhile to point out the value of NT-proBNP for the exclusion of risk of mortality by 1 year, since an NT-proBNP concentration of 986 pg/mL or less was valuable for identifying an extremely low risk of mortality. In this sample, not a single participant with acute CHF with an NT-proBNP concentration of 986 pg/mL or less at presentation had died at 1 year.
A major strength of our study is that outcomes in the trial were not confounded by knowledge of natriuretic peptide concentrations in our study participants because the treating physicians were blinded to the results of NT-proBNP testing, and routine natriuretic peptide testing was not performed at our institution at the time of the trial or the subsequent follow-up period. Our study is limited in that we have results only of NT-proBNP testing from presentation and were not able to evaluate the risk of mortality with respect to serial measurements of NT-proBNP, as has been previously described for patients with acute CHF28,29 or ACS.12 However, the value of routine serial natriuretic peptide measurements for the diagnosis or management of the dyspneic patient has not been described; such a strategy remains much less established than the value of natriuretic peptide testing at presentation for diagnosis and triage purposes, 2 uses clearly established in several trials.1- 3,26 Also, although the prognostic impact of an elevated NT-proBNP concentration appeared to be independent of a diagnosis of acute (or prior) CHF in breathless patients, the relatively small number of deaths in the non-CHF group is a limitation.
In summary, we demonstrate NT-proBNP testing to be a valuable prognostic tool for the identification of a high risk of mortality by 1 year following presentation to the ED with dyspnea with or without CHF. Because evaluation of the dyspneic patient represents the most common indication for natriuretic peptide testing, our results are of particular importance. In light of the value of NT-proBNP testing for the diagnosis,1,3 triage,2 management,26 and now long-term prognostic evaluation of dyspneic patients, we recommend that routine NT-proBNP testing for dyspneic patients in the ED be strongly considered. For those with elevated NT-proBNP concentrations, a diagnostic workup for the deleterious diagnoses known to elevate concentrations of NT-proBNP (notably including acute CHF, ACS, and PE) with appropriate diagnosis-specific therapeutic efforts is advisable.
Correspondence: James L. Januzzi, Jr, MD, Massachusetts General Hospital, Yawkey 5984, 55 Fruit St, Boston, MA 02114 (JJanuzzi@Partners.org).
Accepted for Publication: September 7, 2005.
Financial Disclosure: None.
Funding/Support: The PRIDE Study was supported by a grant from Roche Diagnostics Inc, Indianapolis, Ind.
Role of the Sponsor: The sponsor had no involvement in the data collection, data analysis or interpretation, or drafting of the manuscript.