How well does B-type natriuretic peptide predict death and cardiac events in patients with heart failure: systematic reviewBMJ 2005; 330 doi: https://doi.org/10.1136/bmj.330.7492.625 (Published 17 March 2005) Cite this as: BMJ 2005;330:625
- Jenny A Doust (), senior research fellow1,
- Eva Pietrzak, senior research officer2,
- Annette Dobson, professor of biostatistics2,
- Paul Glasziou, director3
- 1 Centre for General Practice, School of Medicine, University of Queensland, Herston Road, Herston, Qld 4006, Australia
- 2 School of Population Health, University of Queensland
- 3 Centre for Evidence-Based Medicine, Department of Primary Health Care, University of Oxford, Oxford
- Correspondence: J Doust
- Accepted 19 May 2005
Objective To assess how well B-type natriuretic peptide (BNP) predicts prognosis in patients with heart failure.
Design Systematic review of studies assessing BNP for prognosis in patients with heart failure or asymptomatic patients.
Data sources Electronic searches of Medline and Embase from January 1994 to March 2004 and reference lists of included studies.
Study selection and data extraction We included all studies that estimated the relation between BNP measurement and the risk of death, cardiac death, sudden death, or cardiovascular event in patients with heart failure or asymptomatic patients, including initial values and changes in values in response to treatment. Multivariable models that included both BNP and left ventricular ejection fraction as predictors were used to compare the prognostic value of each variable. Two reviewers independently selected studies and extracted data.
Data synthesis 19 studies used BNP to estimate the relative risk of death or cardiovascular events in heart failure patients and five studies in asymptomatic patients. In heart failure patients, each 100 pg/ml increase was associated with a 35% increase in the relative risk of death. BNP was used in 35 multivariable models of prognosis. In nine of the models, it was the only variable to reach significance—that is, other variables contained no prognostic information beyond that of BNP. Even allowing for the scale of the variables, it seems to be a strong indicator of risk.
Conclusion Although systematic reviews of prognostic studies have inherent difficulties, including the possibility of publication bias, the results of the studies in this review show that BNP is a strong prognostic indicator for both asymptomatic patients and for patients with heart failure at all stages of disease.
The clinical assessment of heart failure is notoriously difficult; it is difficult to determine which patients have heart failure and, once the diagnosis is established, to predict which patients are at risk of death or further cardiovascular events. Many studies have tried to determine which factors increase mortality and morbidity in patients with heart failure across a variety of clinical settings. Factors that have been shown to be predictors of mortality are increasing age, a history of diabetes mellitus or renal dysfunction, higher functional disability measures such as New York Heart Association class, lower left ventricular ejection fraction, lower sodium concentrations, lower body mass index, lower blood pressure, the presence of ankle oedema, and lower quality of life scores.1–4 However, none of these is a strong predictor, and so intense interest has emerged in the predictive value of B-type natriuretic peptide (BNP).
The natriuretic peptides are released by the heart in response to myocardial tension and increased intravascular volume and provide accurate tests for the diagnosis of heart failure compared with echocardiography or expert clinical consensus.5 In most countries, it is not currently standard clinical practice to measure these peptides to determine prognosis in patients with heart failure. Our aim in this study was to review systematically the literature to determine how well BNP or its precursor form, N-terminal pro-brain natriuretic peptide (NT-proBNP), predict mortality and morbidity in patients with heart failure, and to determine if this varied with the clinical setting or severity of heart failure. We also wanted to compare BNP with other traditional prognostic indicators, such as left ventricular ejection fraction, New York Heart Association class, serum sodium concentrations, age, history of diabetes mellitus, peak oxygen uptake (VO2), or a scoring system used to estimate the risk of death in patients awaiting heart transplantation, the heart failure survival score.4
We searched Medline and Embase from January 1994 to March 2004 for all studies of the prognostic value of BNP in patients with heart failure, including all stages of heart failure, all clinical settings, and all lengths of follow-up, with no restriction on the language of publication. We also included studies that had estimated the relation between BNP values and prognosis in “asymptomatic” patients. We excluded all studies conducted in patients with recent myocardial infarction because of the likely instability in the relation between BNP concentration and prognosis at this time. We also excluded studies that did not include a clear clinical end point, such as death, hospital admission, or further cardiovascular event. The search strategy included 17 MeSH or text word terms for the condition “heart failure” and five MeSH terms for the diagnostic test “natriuretic peptides.” The full strategy (see bmj.com) retrieved 861 citations. We subsequently checked the reference lists of primary studies and review articles identified by the search for further relevant studies.
Two reviewers (JAD, EP) checked the lists of abstracts and then the full papers for eligible studies and extracted data independently. Where they disagreed on inclusion or exclusion of a study or data extraction, the differences were resolved by consensus or by discussion with a third reviewer. Where possible, data were extracted from multivariable regression models of prognosis.
We assessed the quality of the included studies by determining how patients were selected for the study (in particular, whether the study was a prospective and consecutive cohort of patients), if follow-up of patients was complete and sufficiently long, and if the ascertainment of the end points was blinded and objective.6 We assessed the representativeness of each of the included studies by determining the clinical setting, the spectrum of the patients included in each study, the method for diagnosing heart failure, and the age of the patients. We also extracted data on study size and number of outcomes, the method for measuring BNP, the type of statistical model used, and the way in which BNP was modelled in the studies.
The most common form of analysis for prognostic studies is the Cox proportional hazards model. Such models measure the hazard ratio—the relative effect of a predictive factor on an outcome—by assuming that this relation is constant over time. To combine the data from as many studies as possible, we assumed that where the outcome is relatively rare, the relative risk or odds ratio approximates the hazard ratio. For the outcome of death, we planned to combine estimates of the hazard ratio, odds ratio, or relative risk from studies by using comparable measures of BNP using the “meta” command of Stata, version 7.0 (Stata Corporation, Texas USA, 2001). This command also tests for the presence of heterogeneity.
From the 861 citations, we identified 32 studies that assessed if BNP predicts death or cardiac events in patients with heart failure or in asymptomatic patients, either via estimating a relative measure of risk such as a hazard ratio, or by measuring the statistical significance of the BNP in a multivariable model of prognosis.7–38 We identified 19 studies that assessed the relative risk of death or cardiac events with rises in BNP in patients with heart failure and five studies in asymptomatic patients.26–30 Fourteen studies used BNP or NT-proBNP to predict the relative risk of death or cardiac death in heart failure patients (six used a continuous measure of BNP,7–12 six used a dichotomised measure,13–17 and four used a change in BNP over time8 13 19 20). Eleven studies used BNP or NT-proBNP to predict the risk of a cardiovascular event, most commonly death or hospital admission (three used a continuous measure of BNP,12 21 22 five used a dichotomised measure,13 14 16 16 17 23 and four used a change in BNP over time13 14 20 25). Tables 1, 2, 3, 4, 5, 6 show the results of each of these groups of studies.
In most studies, the primary outcome of interest was either death or cardiac death. These are reasonably objective end points, but it is difficult to assess from the study reports how completely patients in the studies were followed up and how completely outcomes were ascertained in each study. Three studies reported that some patients in the study were lost to follow-up; the remainder either reported complete follow-up or the calculations imply complete follow-up. A possibility exists of the selective reporting of outcomes or the biased reporting of only models with significant results.
The studies were conducted in various clinical settings and used various BNP tests. Although BNP and NT-proBNP seem to have skewed distributions, most of the models included BNP as either a continuous variable linearly related to the outcome or used a discrete cut-off point rather than a logarithmic transformation of the variable.
We combined the results of four of the five studies that estimated the relative risk of all cause mortality by using a continuous measure of BNP in a random effects model.7 8 10 11 We excluded the study by Bettencourt et al because the published report did not provide results to sufficient accuracy to enable us to estimate a plausible hazard ratio. Pooling the other four studies gives an estimate of the relative risk of death per 100 pg/ml of 35% (95% confidence interval 22% to 49%, heterogeneity χ = 6.3, df = 3, P = 0.096). Including the one study that used a continuous measure to estimate the relative risk of cardiac death12 in the pooled estimate (again excluding the study by Bettencourt et al) gives a similar result of 37% (22% to 54%, heterogeneity χ2 = 10.2, df = 4, P = 0.037).
The studies that used dichotomous measures showed considerable variation in results, possibly because of the differences in the cut-offs used and because several of the studies estimated the relative risk of BNP to predict mortality or cardiovascular events unadjusted for other risk factors. They show, however, a consistently increased risk of either death or cardiovascular events with raised concetrations of BNP (tables 2 and 5). The pooled estimate from the studies using a continuous measure was consistent with the results seen of the largest study using a dichotomised measure—that is, a study of a subset of patients (4305 patients) from the valsartan heart failure trial (Val-HeFT) trial.13 This study showed in patients with BNP concentrations > 97 pg/ml a hazard ratio of death of 2.10 (1.79 to 2.42).
Patients whose BNP values fail to fall in response to treatment seem to be at particularly high risk of death or a cardiovascular event (tables 3 and 6). Models that included both initial measurements and measurements after treatment showed that the values after stabilisation on treatment were more significant predictors of death and further events than baseline values.8 13 19 20 24 25
BNP and NT-proBNP also predict mortality and cardiovascular events in asymptomatic patients (tables 7 and 8). Again, the studies used various methods for measuring the relation between BNP and mortality or cardiovascular events. The two largest studies used relatively low cut-off points (≥ 17.9 pg/ml in the study by McDonagh et al, or ≥ 20.0 pg/ml in men and ≥ 23.3 pg/ml in women in the study by Wang et al).26 27 We could not assess from the data in the studies in this review whether the mortality risk associated with BNP is continuous or there is a threshold effect, but even using these relatively low cut-off levels of BNP, the relative risk of death doubled during the follow up periods of four and five years.
Comparison of BNP with other prognostic markers
Thirty five multivariable models included BNP or NT-proBNP to predict survival, cardiac death, readmission, or cardiac events; these included some models that did not estimate the relative risk or hazard ratio.31–38 In 23 of the 35 multivariable models, BNP or NT-proBNP had the smallest P value. In nine of the 35 models, BNP or NT-proBNP was the only predictor that reached significance; other prognostic markers contained no information beyond that provided by BNP.7 16 19 20 22 23 30 31 37 Many clinical features that have been shown to be associated with increased mortality, such as New York Heart Association class, serum creatinine concentration, lower systolic blood pressure, and higher heart rate1 no longer reached significance in models that included BNP. In two models, BNP or NT-proBNP was not a significant predictor, and in both cases N-terminal pro-atrial natriuretic peptide (N-proANP) reached significance.8 35 N-proANP also excluded BNP and vice versa in the model developed by Wang et al26 but did not reach significance in 10 other models that included BNP.
Assessing the relative strength of prognostic markers on a continuous scale is difficult because of differences in the scale of each marker. We therefore estimated standardised hazard ratios (see bmj.com). Although theoretically this allows a better comparison between BNP and left ventricular ejection fraction as predictors, the results were quite inconsistent between studies (table 9). Another way to compare the predictive value of prognostic markers in heart failure is to compare the area under a receiver operating characteristic (ROC) curve for each variable, as this method also removes the scaling of the variable. We found only one study (n = 142) that estimated the predictive ability of factors for all cause mortality in advanced heart failure by using ROC curves.16 The areas under the ROC curve were 0.738 for NT-proBNP, 0.640 for left ventricular ejection fraction, 0.650 for peak oxygen uptake (VO2), and 0.654 for the heart failure survival score, indicating that NT-proBNP has the greatest predictive value.
The strength of prognostic variables in models may also be confounded by decisions on treatment. For example, patients with low left ventricular ejection fractions may be treated more aggressively by clinicians, thereby diluting some of the prognostic value of left ventricular ejection fraction. However, BNP remained a significant predictor of prognosis, even in models in which treatment was included as a variable7 13 27 30 32 36 38 BNP may also add to the prognostic information of left ventricular ejection fractions. In the cohort of participants in the 1992 multinational monitoring of trends and determinants in cardiovascular disease (MONICA) risk factor survey in Glasgow, four year mortality from all causes was determined for patients with and without left ventricular dysfunction (defined as left ventricular ejection fraction ≤ 40% and > 40%) and raised and normal concentrations of BNP (defined as ≥ 17.9 pg/ml and < 17.9 pg/ml).27 The risk of mortality for the group with raised BNP alone was 7%; with reduced left ventricular ejection fractions alone, 8%; and with the two factors combined, 17%, indicating an apparently additive risk (table 9).
Comparison of BNP with NT-proBNP
BNP was directly compared with NT-proBNP in only one model. In the multivariable analysis, both log BNP and log NT-proBNP reached significance in univariate analysis, but only log BNP remained significant in the multivariable analysis.30
BNP was a consistently significant prognostic indicator in patients diagnosed with heart failure and in asymptomatic patients in the studies under review. The prognostic information seems to be at least additive with that of left ventricular ejection fraction, and BNP should be used to assess prognosis in patients with heart failure.
Defining heart failure
If BNP predicts prognosis, including in patients not diagnosed with heart failure, it raises important questions concerning the way that heart failure is defined and diagnosed. In most recent trials of treatment and in studies of diagnostic accuracy, the reference standard for the diagnosis of heart failure has been systolic function as measured by left ventricular ejection fraction. This is despite the fact that it is recognised that 20-50% of patients with heart failure have preserved systolic function.39 Currently, no criteria are agreed for how to categorise patients with “diastolic dysfunction.” BNP is a strong indicator of cardiac risk and may therefore be a better way of identifying the cohort of patients who would benefit from treatment. This hypothesis could be tested by a trial of heart failure treatment in patients with discordant results for BNP guided diagnosis compared with standard echocardiographic or clinical diagnosis. This raises further questions. It would not be difficult to enrol patients in a trial of treatment who have a raised BNP measurement but normal left ventricular function. Is it also possible that patients with a low left ventricular ejection fraction but a normal BNP do not benefit from treatment?
Cut-off values for BNP
The question also arises of what should be considered a “normal” value of BNP. The risk of death and cardiovascular events seems to rise with even small values of BNP. In the studies in asymptomatic patients by Wang et al and McDonagh et al,26 27 the relative risk of death and cardiovascular events was doubled at values well below those currently considered diagnostic for heart failure, at 80-100 pg/ml.17 At what measurement might the benefits of treatment be effective and cost effective?
Monitoring heart failure
The fact that patients with a raised BNP value after treatment, whether in hospital or as outpatients, were at high risk of a further event also implies that BNP may be useful to monitor treatment response and guide decisions on further treatment. Two small trials have proposed that using BNP to guide treatment results in fewer cardiac events than traditional clinical assessment,40 41 but these results are preliminary and need confirmation in larger clinical trials.
What is already known on this topic
Factors shown to be predictors of mortality in heart failure are increasing age, a history of diabetes mellitus or renal dysfunction, higher New York Heart Association class, lower left ventricular ejection fraction, lower sodium concentrations, lower body mass index, lower blood pressure, the presence of ankle oedema, and lower quality of life scores
The clinical assessment of prognosis in heart failure is difficult, however, and none of the above factors are strong predictors of survival or cardiovascular events
What this study adds
B-type natriuretic peptide is a strong prognostic indicator for patients with heart failure at all stages of disease and seems to be a better predictor of survival than many traditional prognostic indicators, such as New York Heart Association class, serum creatinine, and possibly left ventricular ejection fraction
The relative risk of death increases by about 35% for each 100 pg/ml increase in BNP in patients with heart failure patients
Raised BNP values also predict survival in patients not known to have heart failure, with the risk doubled in patients with a BNP value > 20 pg/ml
Despite the abundance of studies, this review has several limitations. In part this is because systematic reviews of prognostic studies are hampered by the standard of reporting of the original studies. Finding all prognostic studies is difficult as they are not tagged as such in Medline, and finding negative studies—that is, studies where the variable was considered but did not reach significance—is particularly difficult. Many of the studies did not report on features that would ensure objective and unbiased estimates of prognostic indicators. In addition, the true impact on prognosis may be less than estimated from these studies because studies that did not show a significant effect have possibly not been published.
BNP is a powerful prognostic indicator for patients with heart failure at all stages of disease. Both initial values and values after starting treatment are important indicators of disease severity.
The full search strategy is on bmj.com
Contributors The idea for this study arose from a previous review of diagnostic accuracy studies. JAD designed the study, and JAD and EP assessed the studies for inclusion and extracted data. PPG and AD provided advice on the statistical analysis and interpretation of the studies. JAD, PPG, and AD drafted the paper. JAD is the guarantor.
Funding source National Health and Medical Research Council programme grant 211205: Screening and Test Evaluation Programme (STEP) grant.
Competing interests None declared.