NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Balion C, Don-Wauchope A, Hill S, et al. Use of Natriuretic Peptide Measurement in the Management of Heart Failure [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2013 Nov. (Comparative Effectiveness Reviews, No. 126.)

Cover of Use of Natriuretic Peptide Measurement in the Management of Heart Failure

Use of Natriuretic Peptide Measurement in the Management of Heart Failure [Internet].

Show details


A comparative effectiveness review (CER) was undertaken to assess the state of the evidence for diagnosis, prognosis, treatment, and biological variation of B-type natriuretic peptide (BNP) and N-terminal proBNP (NT-proBNP) in patients with heart failure (HF). HF is a major concern for health care systems because of its chronic nature and resource implications. BNP and NT-proBNP have emerged as promising markers for HF diagnosis, prognosis, and treatment; use of these markers has been recommended in guidelines.400

The search strategy for this CER uncovered a very large volume of literature and the inclusion/exclusion criteria ensured the selection of the most relevant evidence for each of the seven Key Questions (KQs). Given the complexity of these questions and the volume of literature, we partitioned the discussion to reflect the four major areas evaluated in this review. Issues relevant to diagnosis, prognosis, treatment, and biological variation are detailed below in the context of the relevant KQ.

Key Question 1. In patients presenting to the emergency department or urgent care facilities with signs or symptoms suggestive of heart failure (HF)

  1. What is the test performance of BNP and NT-proBNP for HF?
  2. What are the optimal decision cutpoints for BNP and NT-proBNP to diagnose and exclude HF?
  3. What determinants affect the test performance of BNP and NT-proBNP (e.g., age, gender, comorbidity)?

Overview: Key Question 1

There were 51 publications that met the criteria for KQ1 and examined BNP,3,72-121 and 39 articles that met the criteria for KQ1 and examined NT-proBNP.1,2,26,88,108-122,124-143 In patients with signs and symptoms suggestive of HF presenting to an emergency department or urgent care center, measurement of BNP or NT-proBNP is a useful tool to rule out HF as a cause of the symptoms. Irrespective of the cutpoint chosen, which could be the lowest in each study, the manufacturers' suggested cutpoint, or the optimal cutpoint selected by a study's authors, the sensitivity is high and the negative likelihood ratio (LR-) is low. On the other hand, both BNP and NT-proBNP displayed lesser ability to rule in HF as to the cause of patients' symptoms.

The selection of an “optimal” cutpoint was evaluated in order to rule out and rule in HF in this population. Low cutpoints, either the lowest cutpoint reported, or the manufacturers' suggested cutpoint, resulted in high sensitivity and low LR-. To evaluate the rule-in capability of the tests, higher cutpoints proposed by the studies were examined. For BNP, 100 pg/mL is suggested by all manufacturers as the diagnostic cutpoint. All BNP studies that presented diagnostic performance data examined this cutpoint. This cutpoint provides excellent rule-out capability and moderate rule-in capability. For NT-proBNP, attempts to increase the value of these tests to rule in HF by using an optimal cutpoint (often set as the best combination of sensitivity and specificity) resulted in an increase in specificity and LR+, with a small loss of sensitivity and LR-. There was no agreement among the studies as to which optimal cutpoint(s) to choose. One study2 reported on a consensus amongst four studies where the analysis was pooled for 1,256 patients in 3 continents. They reported an age stratified “rule-in” strategy of 450, 900, and 800 pg/mL for ages <50, 50 to75, and >75 respectively, and an age independent “rule-out” cutpoint of 300 pg/mL. The European Society of Cardiology guidelines391 recommends a rule out cutpoint of 300 pg/mL, and further investigation (echocardiogram) above this.

BNP concentrations increase with age. Three101,111,119 of four studies examining diagnostic performance propose increased cutpoints with age, but no consensus was reached. NT-proBNP concentrations also increase with age. Three studies2,138,141 proposed consistent cutpoints of 450 pg/mL for patients <50 years, 900 pg/mL for patients 50 to 74 years, and 1,800 pg/mL for patients ≥75years.

Both BNP and NT-proBNP concentrations increase as renal function (as measured by estimated glomerular filtration rate (eGFR)) decreases. Four authors109,113,114,120 suggested increasing the diagnostic threshold with declining renal function, but the studies differ in the proposed cutpoints. For NT-proBNP, one author113 suggested increased cutpoints for patients with reduced renal function.

Not enough evidence exists to make firm conclusions with respect to the effects of sex, ethnicity, BMI, or the presence of diabetes on the diagnostic performance of BNP or NT-proBNP.

Applicability in Diagnostic Studies

The diagnosis of HF in patients presenting to emergency departments is difficult.401 The differential diagnosis for patients presenting with the chief complaint of dyspnea is large, including cardiac causes, pulmonary causes, combined cardiac and pulmonary causes, and neither cardiac nor pulmonary causes.401

In KQ1 of this review, the focus was on studies that enrolled patients presenting to the emergency department with the clinical symptoms of HF as the chief complaint, regardless of comorbidities, to create a summary of the evidence with maximum generalizability. Studies that required the presence of a specific disease or condition as a criterion for enrollment were excluded.

For BNP, we present data on the common cutpoint of 100 pg/mL as proposed by all manufacturers of FDA-approved BNP assays. This should provide users of the test with robust information on the applicability of the test to patients in the emergency department with appropriate symptoms. For NT-proBNP, few studies commented on the diagnostic performance of the test using the manufacturers' recommended cutpoints of 125 pg/mL for those less than 75 years and 450 pg/mL for those older. Researchers proposed various cutpoints based on age. This lack of uniformity for NT-proBNP suggests clinicians should apply the findings of this report cautiously to their practices in emergency departments and urgent care centers.

Conclusions for Diagnostic Studies

Diagnostic Studies From Emergency Settings

For patients presenting to emergency departments or urgent care settings with signs and symptoms suggestive of HF, BNP and NT-proBNP have good diagnostic performance to rule out, but lesser performance to rule in, the diagnosis of HF compared with the reference standard of overall global assessment of the patient's medical record. The strength of evidence (SOE) was as high for sensitivity and moderate for specificity for both BNP and NT-proBNP at all cutpoints examined. Nevertheless, we rated the overall SOE as high. Further studies are unlikely to change the conclusions presented here. Comorbidities, including age, renal function, and BMI (BMI for BNP only) have important effects on the performance of these tests. There is, however, no agreement amongst the studies regarding the appropriate cutpoints that should be applied, dependent on the test, age and renal function of the patient.

Key Question 2. In patients presenting to a primary care physician with risk factors, signs, or symptoms suggestive of HF

  1. What is the test performance of BNP and NT-proBNP for HF?
  2. What are the optimal decision cutpoints for BNP and NT-proBNP to diagnose and exclude HF?
  3. What determinants affect the test performance of BNP and NT-proBNP (e.g., age, gender, comorbidity)?

Overview: Key Question 2

There were 12 articles that met the criteria for KQ2 that examined BNP,148-159 and 20 articles that met the criteria for KQ2 examining NT-proBNP.154,156-159,161-175

In primary care settings, patients often present with risk factors but have mild or no obvious symptoms of HF. Thus, diagnosis can be challenging. BNP or NT-proBNP tests are often used with these patients as the first step in the diagnostic algorithm. Those with low BNP or NT-proBNP values can be safely ruled out, whereas those with increased values can be diagnosed directly, or referred for further confirmatory testing.

This review indicates that BNP and NT-proBNP are useful diagnostic tools to identify patients with HF in primary care settings. The results obtained from this review are in agreement with a recent systematic review using individual patient data meta-analysis where both BNP and NT-proBNP had high sensitivities for diagnosis of HF.402 When separating the sensitivities of the studies into the optimum cutpoint as defined by the authors of included studies, the lowest cutpoint, or the manufacturers' cutpoint all provided similar pooled sensitivities. However, the pooled specificities for diagnosis of HF were substantially lower.

In the case of BNP, studies that reported results for the manufacturers' suggested cutpoint of 100 pg/mL were pooled, since this is likely the cutpoint that the majority of laboratories would use. The study by Barrios et al.153 had a substantially lower sensitivity and a high specificity for identifying patients with HF. Predominantly elderly patients were enrolled in this study and HF was defined according to the Framingham criteria. Sixty percent of patients had diastolic dysfunction and only 2.8 percent had a reduced left ventricular ejection fraction (LVEF). The authors suggested that the reduced sensitivity for diagnosis of HF found in this study, relative to the other studies, is due to the high proportion of diastolic HF.

Only two studies159,168 looked at the manufacturers' suggested cutpoints for NT-proBNP. The sensitivities were somewhat different; however, the specificities were similar. Gustafsson et al.168 used an LVEF of <40 percent to identify patients with HF, while Christenson et al.159 used cardiologist adjudication, including an LVEF <40 percent as well as other signs, symptoms, and other objective markers. This may account for the lower sensitivity in the Christenson report.

When the effect of various determinants on BNP and NT-proBNP were examined, we found that values for both peptides increased with age and declining renal function, and decreased as BMI increased.

A single study looked at the age effect on BNP and demonstrated that a higher cutpoint is required in patients greater than 65 years to maintain an optimal sensitivity compared with patients less than 65 years.158 A similar age-related increase in NT-proBNP is seen in the same study, with higher cutpoint required to maintain an optimal sensitivity.158 A pooled analysis performed by Hildebrandt et al. showed similar results by demonstrating that higher cutpoints are required to maintain equivalent diagnostic accuracy as age increases.403

In terms of sex, two studies investigated the effect on BNP. Both Fuat et al.156and Park et al.158 did not identify any significant effects. Five studies156,158,162,166,170 examined the effect of sex on NT-proBNP, and although the authors identified different optimal cutpoints for males and females, no clear conclusions could be drawn regarding optimal cutpoints.

The effect of BMI on BNP and NT-proBNP was investigated by several studies. Most studies showed a negative correlation of BMI with BNP or NT-proBNP, with decreasing sensitivities for diagnosing HF. However, no BMI-specific cutpoints were suggested in the included articles.

Decreased renal function, measured by creatinine clearance (concentration <60 mL/min), was determined by Park et al.158 to increase the levels of both BNP and NT-proBNP; however, the effect was more significant with NT-proBNP. The differential effect is likely due to the fact that NT-proBNP is cleared by the kidneys,404 while BNP is not.405

Applicability in Diagnostic Studies

In primary care settings the majority of patients do not present to general practitioners with obvious serious symptoms of HF. Many of the patients may present with limited symptoms or subclinical disease. Identification of patients at risk of developing HF or those with subclinical or limited symptoms is critical, as there are effective treatments for HF and in undiagnosed patients, the condition will progress without treatment, increasing the cost to the health care system and decreasing the quality of life of the patient.

BNP, using either the optimal or manufacturers' suggested cutpoint, is effective at identifying patients at risk of HF or patients with few or no symptoms of HF. NT-proBNP is effective at identifying patients at risk of HF using the optimal cutpoint; however, limited evidence exists for using the manufacturers' suggested cutpoint. Goode et al.173 performed a cost-benefit analysis of using NT-proBNP to identify patients at high-risk of developing HF. In their population, 7.5 percent had undiagnosed left ventricular systolic dysfunction and use of NT-proBNP was effective for identifying patients at risk and provided a significant cost benefit.

Conclusions for Diagnostic Studies

Diagnostic Studies From Primary Care Settings

Both BNP and NT-proBNP have good diagnostic performance in primary care settings for identifying patients who are either at risk of developing HF, or have fewer symptoms and/or less severe signs suggestive of HF. Using the manufacturers' suggested cutpoint, BNP can effectively be used to rule out the presence of HF in primary care settings. In the case of NT-proBNP, limited evidence is available to determine if the manufacturers' suggested cutpoint is as effective. We rated the SOE for sensitivity as high and specificity as moderate. We rated the overall SOE as high. Further studies are unlikely to change the conclusion presented here.

Limitations of the Review of Diagnostic Studies in KQ1 and KQ2

This review examined the evidence for the use of the BNP and NT-proBNP in the diagnosis of HF, without examining this test in combination with other diagnostic tools. The effect of BNP and NT-proBNP as part of “test panels” or in combination with other diagnostic algorithms was not investigated.

The effect of heterogeneity among the studies on the overall estimates of diagnostic performance was not investigated. Mastandrea et al.406 examined factors that can contribute to heterogeneity of meta-analyses of studies using BNP and NT-proBNP. He examined 98 samples from 67 studies (52 samples/41 studies of BNP, 46 samples/24 studies of NT-proBNP) and found that disease severity, disease prevalence, and the reference test were factors that contributed to heterogeneity for BNP. Whereas disease severity is an intrinsic factor in the pathology of the disease, the disease prevalence and the reference test were considered to be true elements of interference. For NT-proBNP, Mastandrea et al. were unable to identify factors contributing to heterogeneity.

One study86 for BNP used the echocardiogram as the sole criterion for the reference test in the diagnosis of HF. All others used a combination of signs, symptoms, and objective criteria (e.g., X-ray, electrocardiogram, echocardiogram) and diagnostic scorecards (e.g., Framingham, Boston, National Health And Nutritional Examination Survey (NHANES)). Similarly, for NT-proBNP, one study137 used echocardiogram as the sole diagnostic criterion. All others used the same global criteria as BNP. The lack of a single “gold standard” for the diagnosis of HF necessitates the use of the clinical diagnosis.

Future Research Recommendations in Diagnostic Studies in KQ1 and KQ2

  1. More studies are needed to determine the effect of age on the diagnostic cutpoints, especially for NT-proBNP. Common cutpoints that can be used in all situations would increase the applicability of this test.
  2. More studies are needed to determine the effect of declining renal function on the diagnostic performance of both BNP and NT-proBNP, and to establish cutpoints in situations of reduced renal function.
  3. More studies are needed to determine the effect of sex, ethnicity, and BMI on BNP and NT-proBNP concentrations and ultimately on the cutpoints for diagnosis.
  4. There is a need to examine the evidence for the value of BNP and NT-proBNP in multi-marker panels for the diagnosis of HF.
  5. A more detailed study of the effects of heterogeneity amongst the studies would allow a clearer understanding of the effects of various confounders, including comorbidities.

Key Question 3. In heart failure populations, is BNP or NT-proBNP measured at admission, discharge, or change between admission and discharge an independent predictor of morbidity and mortality outcomes?

Overview: Key Question 3

Overview of Issues in Studies Evaluating Decompensated Heart Failure Subjects

The majority of studies were not designed with the primary objective to evaluate the prognostic ability of BNP /NT-proBNP nor were higher level model validation computations undertaken. Almost all of the studies with large sample cohorts were designed for another purpose (usually intervention assessment) and were primarily aimed at “predictor finding” analyses showing some association between BNP/NT-proBNP and the outcomes of interest.

Seventy-nine studies evaluated levels of BNP (n=38), NT-proBNP (n=35), or both (n=6) as predictors of mortality and morbidity outcomes in subjects with decompensated HF, ranging over time intervals from 14 days to over 6 years. When considering single outcomes, most publications (n=55) evaluated mortality outcomes, predominately all-cause; morbidity outcomes were inconsistently defined and assessed as endpoints less frequently (n=8). The majority of studies assessing single outcomes, evaluated admission BNP levels with fewer studies evaluating serial measurements (while hospitalized), change from admission levels, or discharge levels prior to leaving the hospital as potential prognostic factors. Composite outcomes were reported as frequently as all-cause mortality outcomes and within these, all-cause mortality and morbidity were most frequently assessed. Studies with composite outcomes had relatively equal numbers of studies assessing admission and discharge or change levels as predictors.

In general, higher levels of admission BNP and NT-proBNP incurred greater risk for the outcomes of mortality, morbidity, or a combination of both. A decrease in BNP levels was also predictive of decreased rates of mortality and morbidity. The range of thresholds for high or higher levels was markedly varied across studies. Similarly, for the studies evaluating pre-hospital discharge BNP/NT-proBNP levels as a predictor, or a change relative to baseline, the thresholds or percent change varied markedly across studies. Comparison of BNP study results relative to NT-proBNP levels were limited to six studies and were inconsistent across studies; the findings of these studies would not indicate superiority of one test relative to the other.

When considering threats to internal validity of the studies evaluating levels in patients with decompensated HF as a whole, many studies were rated as problematic for establishing the validity and reliability of the methods used to ascertain the outcome. Similarly, a minimum of four key confounder domains identified in the 2006 report for the Agency for Healthcare Research and Quality (AHRQ)20 (age, sex, BMI, and renal function) were established a priori as confounders that the clinical experts judged to be important, and therefore studies were downgraded if they did not include or consider these covariates in their analyses. Many studies did not consider all of these factors concurrently. Finally, when applying the Hayden58 criteria to assess appropriate statistical analyses, our evaluations were relatively less stringent than those proposed elsewhere407 and, as such, most studies rated well; however, problems with reporting sufficient information to replicate the statistical analyses were noted across these studies. This issue decreases the confidence in the approaches that these studies used to estimate the prognostic strength of BNP and NT-proBNP. This group of studies is at high risk of bias for validity of outcome measurement and for confounding; however, considering all other criteria within the Hayden checklist, the overall risk of bias was judged as moderate because of the uncertainty with these two criteria.

An important factor influencing the interpretation of the study findings is the length of followup. Study findings were presented as a function of intervals for followup and in the context of decompensated HF patients, this was short term (up to 31 days, 2 to 3 months) and longer term (6 to 11 months, 12 to 23 months, 24 months and greater). We observed the fewest number of studies for the shortest (up to 31 days) and longest time intervals (24 months or greater); within these studies the levels of BNP used, the thresholds for determining high and low risk, and the prognostic models differed. As such, the consistency of the direction of effect and the magnitude varied. The most frequently evaluated interval was the medium range time interval (6 to 12 months), and these studies consistently showed that BNP or NT-proBNP concentrations are independent predictors of all-cause and cardiovascular mortality, some morbidity outcomes, and composite outcomes. This was shown across studies despite the variations in the factors included within the statistical models. These factors included: different cutpoints when used as a dichotomous data, other potential prognostic factors in the statistical models, and the time intervals. It would be important for the clinical community to reach a consensus on what are the most clinically relevant short-term time intervals for predicting specific outcomes; these intervals could reflect optimal timepoints when additional or different interventions may assist in minimizing risk of morbidity and mortality (both for the shorter and longer term) following an acute episode of decompensation. Conversely, it may be equally important to provide a rationale for the longest interval that would be meaningful for clinicians to expect that BNP/NT-proBNP levels from admission or discharge of a current episode are relevant.

The challenge with these differing study factors is in interpreting the magnitude of the predictive values across studies. As noted previously, with differing prognostic models, it is problematic to assume a hazard ratio (HR) equal to two in one study is in fact comparable to that same estimate from another study. Within the decompensated HF studies there was the added problem of when the BNP/NT-proBNP levels were measured. Levels measured at admission would suggest that the subjects had not had significant intervention to manage the acute episode. Serial measurements during the course of hospitalization reflects a short-term response (or lack of response) to treatment that was commenced following admission. Pre-discharge values reflect that the patient is considered to be sufficiently stable that hospitalization is no longer required; it also reflects a degree of response to treatment. From a methodological perspective, treatment intervention associated with the decompensation episode is a confounder (associated with changing BNP/NT-proBNP levels and with the outcomes of mortality and morbidity). The timing of receiving treatment relative to when the BNP/NT-proBNP levels were measured is important to consider when interpreting the magnitude for risk.

Overview: Populations With Chronic Stable Heart Failure

Fifteen publications evaluating BNP levels, 88 publications for NT-proBNP, and one study evaluating both assays considered these tests as predictors of mortality and morbidity in patients with chronic stable HF. For BNP levels in patients with chronic stable HF, there is an association between BNP and the outcome of all-cause mortality. The other mortality outcomes (i.e., cardiac and sudden cardiac) demonstrated less convincing association, which did not remain statistically significant in all of the reviewed studies after multivariable adjustment. The importance of BNP as an independent predictor appears to depend on severity of the HF and possibly the length of followup. Severity is suggested as an important factor. A study that selected New York Heart Association (NHYA) level III or IV subjects, found a significant HR for BNP >1,000 pg/mL,261 while three other studeis that used more general HF populations did not find a significant relationship to all-cause mortality at 24 months.268,269,271 The studies that extended beyond 24 months in more general HF populations also found a significant relationship to all-cause mortality.274,275 The other mortality outcomes (i.e., other than all-cause mortality) were less frequently reported and thus consistency in the findings is not generalizable to this group.

The outcome of hospitalization for HF also demonstrated an association with BNP using a natural log (ln) transformed BNP (lnBNP), but this was only reported in one study.274

The composite outcome of all-cause mortality and cardiovascular morbidity demonstrated a significant independent association for BNP with the outcomes selected by the investigators. This was consistent for six of the seven papers in this subsection. The HRs reported here were often a little higher than the ones for all-cause mortality alone.

The use of cutpoints for determining risk is problematic considering the range of cutpoints reported in this review: 250 pg/mL to 1,000 pg/mL for BNP in all-cause mortality and 55 pg/mL to 590 pg/mL for BNP in the combination of all-cause mortality and cardiovascular morbidity. Most often the studies determined the cutpoint from their own population using receiver operating characteristic (ROC) analysis, median, or mean values. Predetermined cutpoints are required for any study aiming to assess the prognostic ability of a test used in a dichotomous fashion. Similar comments would apply to tertiles, quartiles, or quintiles and these values should be selected based on previous studies rather than determined in the study population. Cutpoints are attractive to the clinician because they are easy to remember, but they are likely to lose valuable information from the continuous variable. The use of log transformed BNP seems to hold as much predictive value as that not transformed; an alternative to a predetermined cutpoint could be lnBNP.

The negative association of BMI with BNP has been demonstrated in the paper by Horwich et al.,262 as well as in a paper that was excluded from this review because the authors did not use BNP to diagnose or prognose HF.408 Studies should include either BMI or another measure of body fat, such as waist circumference or waist-to-hip ratio, in their variables. Other variables such as age, sex, and renal function are included in the papers reviewed; these are also known to have strong associations with BNP. Measured parameters, such as LVEF and the NYHA stages, also have strong associations with BNP and should be included in predictive models to prove that BNP holds independent predictive ability. In addition, common factors used in the prediction of cardiovascular disease (CVD) outcome such as hypertension, diabetes, total cholesterol to HDL-cholesterol ratio, and smoking, should be included in predictive models as these have been shown to be associated with mortality from CVD and should thus be accounted for in all-cause mortality and cardiovascular specific mortality assessment.

While the independent association with all-cause mortality and hospitalization for HF is suggested, it is not always found. The applicability of these findings to patient care is not demonstrated in the papers reviewed, as there are no transferable common cutpoints and there is no risk stratification model that has been studied that uses BNP in the risk score. Some of these findings will be discussed under KQ4 where the direct comparison between other prognostic markers is considered in more detail.

Eighty-eight publications evaluated NT-proBNP levels as predictors of mortality and morbidity in patients with chronic stable HF. Overall, the evidence consistently supports the trend that NT-proBNP is an independent predictor of mortality and morbidity outcomes in people with chronic stable HF. The applicability of the aforementioned results rests largely in middle-aged or elderly males. The included studies did not explore whether the prognostic effects of NT-proBNP would differ by age, sex, or time period. Also, the studies did not suggest a single cutpoint to optimize the prognostic ability of the peptide. In general, the studies were problematic with respect to measuring the outcome and including a predefined set of confounders.

The largest number of studies, and the strongest evidence, concerns the outcome of all-cause mortality. Fifty-two publications included all-cause mortality as an outcome and all of the point estimated measures of association, whether statistically significant or not, indicated positive associations between NT-proBNP and all-cause mortality. This conclusion applies across all periods of followup, from 12 months to 44 months.

For cardiovascular mortality, the evidence in 17 publications also suggests a positive association with NT-proBNP. However, this conclusion largely applies to studies with followups that are longer than 24 months.

Twelve studies examined the prognostic value of NT-proBNP for morbidity in persons with stable HF. Overall, higher NT-proBNP levels were shown to be associated with greater hospitalization in eight studies. Twenty-six publications evaluated composite outcomes and showed that NT-proBNP is an independent predictor; the results also suggest that higher levels of NT-proBNP predict greater numbers of composite events.

Overview: Populations With Heart Failure Following Cardiac Surgery

There were eight studies that evaluated BNP/NT-proBNP levels in HF patients who underwent cardiac surgery. Five studies evaluated the effect of resynchronization therapy on BNP levels (n=3) and NT-proBNP levels (n=2) and one study evaluated the effect of cardiac resynchronization defibrillator therapy on BNP. Both assays were shown to be independent predictors of all-cause and cardiovascular mortality and morbidity. The remaining three studies evaluated surgical interventions of intracoronary infusion of bone marrow-derived mononuclear progenitor cells, noncardiac surgery (e.g., abdominal, orthopedic), and peritoneal dialysis. All showed that BNP or NT-proBNP were independent predictors of all-cause and cardiovascular mortality and morbidity, with the exception of the peritoneal dialysis study.

General Issues With Prognosis Studies Evaluating BNP and NT-proBNP as Predictors of Mortality and Morbidity

This systematic review netted a large number of studies (198 publications) and would have been larger still, had the criteria included studies using non-FDA approved BNP/NT-proBNP assays. Despite this large study base, consistent issues with research methodology were observed. These issues, with respect to definitions of HF populations, selection of cutpoints for determining high risk groups, defining and validating outcomes, study design, and statistical modeling approaches, are detailed below.

Defining the Heart Failure Population: Classification Systems for Heart Failure Are Problematic for Establishing Levels of Prognostic Risk

One of the important issues in evaluating any potential prognostic factor in patients with HF is the current classification system for this cardiovascular disorder. HF is considered to be a syndrome rather than a primary diagnosis.15 HF has many different causes and variations in clinical features and exists with a number of comorbidities. In this systematic review, all definitions of HF (i.e., as provided by the study authors) were considered; however, it could not be certain that the patients within the studies were clearly patients with HF or were similar across studies, and it was therefore assumed that findings could be compared across studies with respect to this clinical syndrome classification. This assumption does, however, reflect clinical practice and thus this limitation does not negate the findings. It would, however, be helpful if investigators defined explicitly which categories of HF were included in their populations.

A division among the studies was established to distinguish those patients who were recruited with acute episodes and those who were stable but chronic. This is an important clinical division as the required clinical response is often different in these two settings.12 It was assumed that the level of acuity was adequately categorized by the site of recruitment and that patients who were recruited from emergency or hospital admissions were acute and likely decompensated. Patients recruited from outpatient settings were assumed to be stable and chronic. It would be helpful if authors defined the acuity of their subjects in the methods or results of the study. The case has been made that there is inconsistency in defining the subtypes of acute HF, decompensated HF, or exacerbation of HF.16 The European Society of Cardiology divides acute HF syndromes into six clinical profiles (worsening or decompensated chronic, pulmonary edema, hypertensive HF, cardiogenic shock, isolated right HF, acute coronary syndrome, and HF).13 The American College of Cardiology Foundation (ACCF)/American Heart Association(AHA) has a four stage classification system for acute HF.10 It is not clear that the eligibility criteria of studies included in the acute decompensated category of this review made these distinctions; nor is it clear which of these definitions or subgroups may likely influence the predictive ability of BNP/NT-proBNP for the outcomes of interest.

Defining the Heart Failure Population: Influence of Comorbid Conditions

As patients age, the incidence and prevalence of HF increases409 as do the comorbid conditions of patients. Comorbidity was not consistently considered within the prognostic models, and the degree to which such conditions can confound the estimates of the predictive ability of BNP/NT-proBNP levels needs to be considered appropriately in the analysis of the study.

BNP/NT-proBNP Transformations in Statistical Models and Selection of Thresholds or Cutpoints

When undertaking statistical computations for outcomes that are dichotomous, logistic regression is undertaken and study authors must decide whether to model BNP/NT-proBNP as a continuous or categorical covariate. BNP and NT-proBNP are continuous measures, and typically the distributions are heavily skewed. When they are included as continuous variables, it is recommended that markers that are skewed should be log transformed to “normalize” the distribution in subsequent computations.48 In the presence of such skewing, if the distribution of the BNP or NT-proBNP marker is not transformed, then there is a great risk that results will be misleading. It was observed that the minority of studies log transformed the BNP or NT-proBNP distribution. The practical implication is that one must transform the results back to the previous scale and as such, the HR estimate as reported is not intuitively understood. It is recognized that some study authors may be reluctant to log transform the BNP or NT-proBNP data because of issues with interpretation (which would require a back translation of the log HR). However, it is necessary that the assumptions used in logistic regression are not violated. A tool that calculated risk based on the log transformed test result would be a simple practical way for clinicians to use BNP/NT-proBNP in the clinical setting. An alternative approach is to categorize the BNP or NT-proBNP covariate, typically into quartiles. This option is preferred when the relationship between the BNP or NT-proBNP and the outcome is nonlinear;48 if a continuous covariate were used in this instance, then error is introduced in the estimate of predictive strength. However, if a linear relationship exists between BNP and NT-proBNP, then not analyzing this covariate as a continuous variable will decrease the ability for the model to accurately evaluate the prognostic value. In general, the justification for either approach was not always well reported, which serves to decrease our confidence in the magnitude of the HR.

Another challenge with interpreting results from statistical models was the widely varying thresholds to determine who was or was not at greater risk for future adverse events. Many studies provided a rationale for selecting cutpoints (typically based on ROC analysis or use of mean, median, or tertiles); however, this choice of threshold may in effect select the point producing the largest difference in outcome between categories. If this is the case, then the models would likely overestimate the predictive ability of BNP/NT-proBNP. Finally, interpretation of estimates of predictive strength are problematic from a pragmatic perspective. It is not clear what thresholds to suggest to clinicians, because most studies have overlapping cutpoints. This essentially makes these tests of little use in the clinical setting.

Unspecified Interventions for Patients With Heart Failure in Prognosis Studies

Although the intervention is not often described in prognostic studies, from a methodological perspective it can be considered an important confounder, particularly if patients receive different treatments based on perceived prognostic risks. Interventions were not always well described in the majority of studies reviewed, and it is not clear to what extent diverse treatments have comparable effects on BNP/NT-proBNP concentrations. Often the results of both treatment groups were put together for the purposes of the secondary paper that described prognosis and it was difficult to work from the primary paper which group may have influenced the results. Although in theory the effect of interventions may be less important than other intrinsic prognostic factors (e.g., age, sex, disease stage), it is entirely possible that these studies are at risk of bias for confounding by indication (a variant of selection bias in observational studies).410 Typically, in observational studies, the indication for treatment or the way in which treatment is administered to subjects is poorly reported. Thus if patients differ at baseline with respect to perceived prognostic risk, then either these patients will not receive adequate treatment or will receive more aggressive or different treatment. This bias can result in over- or underestimation of the predictive ability of the factor of interest. Additionally, if an explanatory variable representing treatment is included in the model, then a clear definition (standardized and reproducible description) would be required.411

Selection and Definition of Other Prognostic Factors Within the Prognostic Models

It is important to clearly define all variables included in the prognostic risk models. Within this systematic review, the definitions of prognostic factors included in the predictive models were generally not clearly defined to the level that would allow reproducibility or facilitate comparison across models. Difficulties arise when common and accepted predictors are operationalized differently across studies, particularly those that dichotomize or categorize continuous variable (e.g., age and BMI). Additionally, reporting standards with respect to how factors were selected and included in models were inconsistently reported. Hayden et al.412 present some convincing arguments that much of the prognostic research lacks explicit theoretical frameworks to establish the potential relationship among variables within prognostic models. This would imply the need to hypothesize the potential for intermediary or mediating pathways among prognostic factors. This may involve the use of multilevel or structural equation modeling the aim of which is to evaluate the strength of relationships among the variables.

Study Designs and Phased Hierarchical Approach To Establishing Predictive Value of BNP and NT-proBNP

Several attempts have been made to develop frameworks for establishing sequential or hierarchical phases of prognostic research in order to establish convincing evidence of the value of a predictive marker (prognostic indicator). Table 45 shows four such attempts, with one framework specifically developed for cardiovascular markers.44 Appendices E & F detail the explanation for these phases of development for prognostic research. These frameworks, showing a phased sequential approach to prognostic research, can be paralleled to grading systems for the SOE with respect to credible validation of predictive strength of BNP or NT-proBNP concentrations.

Table 45. Frameworks for sequential development of prediction models that assess the contribution of potential prognostic factors.

Table 45

Frameworks for sequential development of prediction models that assess the contribution of potential prognostic factors.

In our judgment, irrespective of the prognostic model used, the majority of BNP and NT-proBNP studies reviewed within this evidence synthesis, fall into the earliest phases of prognostic study development. At the lowest level of prediction, prognosis studies are designed to identify potential associations of the factors of interest and are termed “exploration”412 or “predictor finding studies”.43 From 198 studies eligible for KQ3, only 41 undertook statistical procedures related to discrimination, calibration, or reclassification of risk; from these, 15 did not report the results of these computations. As such, we would classify the majority of studies in KQ3 as having the aim of establishing or exploring the independent contribution of BNP/NT-proBNP, but these studies did not attempt to evaluate the predictive performance of the model and therefore, represent the early phases of multivariable prognostic research (predictor variable studies). Clearly, this reflects that, as a whole, the evidence for prognostic ability of BNP/NT-proBNP evaluated within this systematic review is based on early and less convincing statistical evidence for predictive strength.

Ideally, prognostic studies would employ prospective cohort or RCT designs.411 In addition to the study design, establishing the predictive value of a marker can be considered to be phased or hierarchical in nature (see Table 45). Specifically, a six-phase model has been proposed for the development and evaluation of cardiovascular risk markers.44 In this systematic review, the majority of studies can only be viewed as meeting the earliest phases of development, irrespective of the particular framework used; most studies were aimed at establishing that BNP and NT-proBNP were independent predictors but did not seek to establish incremental value (relative to base model and other markers) or attempt validation (internal or external) of the predictive model.

Although prospective designs are ideal, we observed that retrospective cohort designs were frequently used in the eligible prognosis studies; retrospective designs may contain bias or omit information critical to the subsequent model used to establish the relative importance of predictors. Additionally, some of the prospective studies were not originally designed to establish the prognostic predictive strength of BNP/NT-proBNP but were secondary analyses from intervention trials, which may also be prone to the same issues as observational retrospective studies. Studies were not restricted by their design type in this review. A few studies addressed the more advanced phases of the evaluation of BNP and NT-proBNP as predictors, attempting internal or external validation (see KQ4 and validation of models).251,353,357 This review found very few studies that addressed the impact of the prediction models on clinical practice (final phases). Although this represents a significant gap in the literature, it is problematic to undertake such studies unless there is clear evidence from high quality predictive models that BNP/NT-proBNP are important predictors of the outcomes of interest.

Development of Statistical Models To Establish Predictive Strength

The multivariable nature of prognostic research can pose some challenges with respect to estimating adequate sample sizes.411 The issue of sample size is particularly important when one considers the number of explanatory variables within statistical models (model development or validation) used to predict HR relative to the number of outcome events. The rule of thumb is that there should be a minimum of 10 events for every prognostic factor included within the multivariate model;411,413 this suggests that some studies included in this review did not have adequate sample sizes with respect to the statistical analyses related to the number of prognostic factors. Conversely, because of the limited sample sizes some studies may have been limited in the number of possible confounders or covariates to include in their prognostic models. The result of this is that the HR will be overestimated. The studies eligible for this review undertook multivariate or multivariable analyses. However, it was difficult to assess the validity of these computations because of the lack of detail in the reporting of the computation methods. Had we evaluated the studies for adequate reporting criteria for multivariate analyses, we suspect that the studies would not have performed well. Additionally, the evaluation of statistical models for use within patient care should take into account the intended purpose of the model. The purpose of prognostic models may be more complex than those of other clinical aims (e.g., diagnostic accuracy). Although multivariable models can predict future events, the issue of discrimination (accurate classification of those with or without the outcome or disease), calibration (estimating probabilities or predictive values for future risk), and reclassification methods are key aspects that need to be taken into account.45,46 Similarly, there is a need to identify the intended aims of the study with respect to the prognostic factor. We have described the phased nature of prognostic research in Table 45. In this systematic review, the majority of studies did not specify the main aim of the research in the context of these frameworks and, as such, we surmised their aim based on the statistical analyses that were attempted and presented. We also note that many of the included studies did not specify that the primary purpose of the study was to evaluate BNP/NT-proBNP; in these studies, BNP/NT-proBNP was one of many predictor variables that were being evaluated.

Some studies in KQ3 could be classified as developmental studies, undertaking discrimination and calibration statistics to establish the model performance. These were the studies that we then included for KQ4, as they provided some information about the incremental added value of BNP/NT-proBNP. Some of the studies in KQ4 provided validation of the model, using internal validation approaches; in this review, only two studies251,375 attempted external validation. This systematic review identified very few impact studies176,414 that attempted to evaluate the clinical impact of the prognostic model on decisionmaking and patient outcomes. Future research studies also need to move toward developing impact studies.

Future research should consider undertaking consensus exercises to establish a minimum set of prognostic factors to be consistently evaluated (or potentially included) in the base statistical models in these prognostic studies. In the best case scenario, the base model contains prognostic factors that have already been established. Unfortunately, this is not clear or consistent in the literature we evaluated. This makes comparison across studies or evaluation of incremental value of adding BNP or NT-proBNP problematic. In this systematic review, we established a priori a minimum set of confounders that were felt to be important for this population and these included age, sex, BMI (or some other metric of body mass), and any measure of renal function which we used to assess risk of bias criteria for confounding; the rationale was based primarily on theoretical biological grounds but none have been definitely established.

Defining Outcomes in Prognostic Studies of BNP and NT-proBNP

The use of composite outcomes is prevalent in the prognosis literature dealing with CVDs. Approximately one half of the studies in the decompensated and stable BNP and NT-proBNP studies eligible in this review used composite outcomes and about one third reported combined outcomes only. The interpretation of these combined outcomes is problematic for clinicians and for patients and could result in misinterpretation of study findings. Although composite endpoints are common in cardiovascular studies because they are used by clinicians or because they increase the event rates and assist in statistical analyses, they can be misleading as the combined outcomes have widely varying importance to patients. Clearly, mortality and morbidity are likely to be valued differently by patients; similarly, even combined outcomes within one category (i.e., morbidity: hospital re-admission combined with reduced quality of life) can be valued differently by patients. For example, patients might place higher value on improved quality of life rather than hospital-free survival. In addition, mixing of a hard outcome, such as cardiac death, with a soft outcome, such as clinical symptoms of angina or HF, is not ideal, as the soft outcomes are more subjective.44 There are also data to suggest that clinicians may overestimate the impact of treatments on preventing adverse events that matter most to patients when considering composite outcomes.415 The events that are often combined within composite endpoints tend to have widely differing frequencies and therefore, different relative risk reductions.416

In the context of prognosis or establishing BNP/NT-proBNP as predictors of composite outcomes, the interpretation for patients and clinicians can be equally challenging. If composite outcomes are to be presented, we recommend that they be presented in conjunction with noncomposite outcomes. Further, study authors should justify why they are combining outcomes (i.e., with similar biological factors and hence similar frequencies or risk). Alternatively, a suitable combined cardiovascular outcome could be defined by cardiology societies. When large variation among the individual components of combined outcomes exist, likely the best choice is to avoid combined outcomes.416 Even if combined estimates were to be used in studies, there is a need for consistency in how these are combined. For example, consider the composite endpoint of cardiovascular death and re-admission to hospital. It is not clear how these events are counted within the same patient, where re-admissions can occur in more than one instance for the same patient. It is not clear if the combined outcome considers these events once per subject or as multiple events per subject; this is further compounded by the use of “and” in some studies and “or” in other studies. Greater clarity in this would be helpful.

Applicability in Prognosis Studies

When one considers the applicability of the BNP and NT-proBNP findings to clinical situations, note that the majority of papers pertained to populations aged 60 years or older, although we could not find specific evidence to suggest that the predictive value of BNP or NT-proBNP varies by the age of the study population. The majority of studies included samples whose composition was over 50 percent, and sometimes over 80 percent, male. Thus, we cannot conclude that the results are equally applicable to males and females.

In these articles we reported on the variety of cutpoints used for developing the prognostic models. It is not clear if these thresholds are truly generalizable because there is such wide variation in practice.

Limitations of This Review for Prognosis Studies in Both Decompensated and Chronic Stable Heart Failure Populations

In studies with decompensated HF patients, it was necessary to assume that the level of acuity was adequately categorized by the studies and so any study that recruited subjects from emergency or hospital admissions was classified as being acute; conversely, subjects not recruited from these settings were considered to be non-acute or stable and chronic. We contacted seven authors to clarify the acuity levels of their studies. From these, five replied but two did not. A judgment call was then made to classify all seven as chronic stable populations. In general, most studies did not provide sample size calculations for either the decompensated or chronic stable HF populations. This is particularly important when one considers the number of explanatory variables within the statistical modeling (model development or validation). Studies were not restricted to those that used appropriate statistical methods (or reported these adequately). However, studies with univariate analyses (including univariate ROC analyses) alone were excluded; for studies that reported univariate and multivariable or multivariate analyses, only the latter were reported and considered in our review synthesis.

We also found a few studies that reported negative BNP and NT-proBNP results, but these studies were most often reporting primarily on alternative markers. The potential bias for not reporting negative BNP and NT-proBNP association is very high and may suggest the risk of publication bias and selective outcome reporting bias. It is expected that publication bias may be particularly problematic for prognostic studies that employ nonrandomized or observational study designs, especially retrospective analyses of existing databases.417 We did not formally assess publication bias for prognosis studies using statistical computations such as funnel plots. Currently, no registry for protocols of prognostic prediction studies exists. As such, it is difficult to assess the potential for selective outcome reporting and the Hayden criteria do not address this specific bias.

Conclusions for Prognosis Studies

The findings demonstrate that there is an association between BNP and NT-proBNP predominately for the outcomes of all-cause mortality and composite outcomes in both decompensated and stable populations. The other mortality outcomes (cardiac and sudden cardiac) demonstrated a less convincing association in chronic stable populations, and were less often evaluated in populations with decompensated HF. In studies with decompensated HF patients, admission and discharge levels and change from admission were all shown to be predictors. The majority of studies were characterized as early phases of prognostic research attempting to establish the independent association of BNP or NT-proBNP with the outcomes of interest. Far fewer studies attempted to undertake model validation methods either in internal or external samples. Very few studies evaluated the impact of using BNP or NT-proBNP on clinical decisionmaking or cost-benefit analyses. Six studies evaluated the prognostic ability of BNP/NT-proBNP in patients undergoing resynchronization therapy and were shown to be independent predictors of all-cause and cardiovascular mortality and morbidity.

The conclusions regarding the evidence must be considered in light of the risk of bias. Many of the papers did adjust for multiple confounders and most included the important covariates of age and sex in the regression models. Our moderate risk of bias rating for this domain can thus be considered more of a caution than a reason to impugn the results. The same could be said of the moderate risk of bias that was assigned to the domain for the measurement of outcomes.

We do not believe that the potential for a moderate risk of bias in these two domains mitigates the overall conclusion that BNP and NT-proBNP are independent predictors of mortality and morbidity outcomes in persons with decompensated and chronic stable HF. However, it is difficult to provide useful, clinically applicable information from these data because there are neither established cutpoints nor simple means of interpreting the test in the different clinical situations.

Future Research Recommendations for Prognosis Studies in Decompensated and Chronic Stable HF Populations

A number of recommendations for future research in assessing the prognostic strength of BNP/NT-proBNP in decompensated acute and chronic stable HF patients are listed.


  1. Include more women and subjects of different races when assessing the predictive value of BNP/NT-proBNP in both decompensated and chronic stable HF patients. Reporting the racial composition of study participants would also be important.
  2. Evaluate the impact of different age tertiles on the predictive value of BNP and NT-proBNP.
  3. Identify clearly if the study subjects are acutely ill (decompensated) or chronic and stable HF patients; this should be specified irrespective of the setting in which treatment is administered.
  4. Improve clarity (better reporting) with regard to the different classifications of decompensated HF subjects. This will minimize misclassification of subjects, improve comparability across studies, and assess potential differences in risk prediction for the HF disease subgroups (that may vary with the different disease taxonomy categorizations).

Intervention (Measurement and Analysis of BNP/NT-proBNP)

  1. For studies of decompensated HF patients, greater clarity in reporting when BNP/NT-proBNP levels were measured relative to the commencement of treatment (e.g., BNP levels were taken within 2 hours of admission prior to pharmacological treatment, etc.).
  2. Report if BNP/NT-proBNP levels were normally distributed and if skewed, the method of adjustment (e.g., log transformation) for subsequent inclusion in the prognostic model.
  3. Consider assessing the same sets of cutpoints in different age groups to examine whether the predictive value of BNP or NT-proBNP changes with age.
  4. Consider prognostic study design to include predetermined cutpoints (based on the literature).
  5. For populations with decompensated HF, there is the need for studies to consistently evaluate potential differences between admission and discharge levels of BNP/NT-proBNP with respect to their predictive ability for both short-term and long-term outcomes.
  6. Future research should adhere to transparent and reproducible methods when defining and selecting all prognostic factors included within the model.407

Study Design

  1. Aim to increase the number of studies that employ prospective designs with a primary aim to establish developmental and external validation models (prospective second-phase studies). This review showed a large number of retrospective studies not primarily designed to assess BNP/NT-proBNP as an independent predictor; there is a need to move away from these retrospective designs.
  2. Increase the number of studies designed to assess the impact (including cost-effectiveness) of BNP/NP-proBNP that demonstrate how decisionmaking and patient outcomes are affected.
  3. Provide a sample size calculation. Consider the number of potential predictors relative to the number of events to prevent overestimation of the predictive ability of BNP/NT-proBNP or other markers (as the number of predictors is larger than the number of outcome events).

Comparators/Covariates in Prognostic Model

  1. Adherence to transparent and reproducible methods when defining and selecting all prognostic factors included within the model.407
  2. Consensus on using a minimum standard set of covariates to account for potential confounding; age, sex, BMI, and renal function is what was suggested by the clinical experts in this systematic review.
  3. Comorbidities are important confounders and attempts should be made to assess and report these within study subjects and possibly adjust for these in the prognostic model.
  4. Clarification of method used to adjust for age, BMI (i.e., another measure of body mass such as waist circumference or waist-to-hip ratio) in the predictive model.

Statistical Prognostic Models

  1. Adherence to reporting standards407 that allow for adequate assessment of the validity of the methods undertaken to develop the predictive model and estimate the prognostic risk. All covariates placed into the model and tested should be reported.
  2. Do not limit statistical analyses to univariate methods (even for ROC analyses). The assumption that BNP/NT-proBNP levels are not mediated by other prognostic factors or that time does not change their predictive ability412 is problematic.


  1. Consensus on defining key outcomes is needed. Outcome assessment should be standardized, both in terms of the types of outcomes investigated and the ways in which these outcomes are defined and measured. This standardization will improve the uniformity of research in this domain and enhance the comparability of results across different studies. The outcomes should be predefined and the investigators should only report on the predefined outcomes,
  2. Report negative as well as positive findings from multivariate or multivariable analyses (even if negative findings are shown). Most authors will run all the possible variables through logistic regression but only report those that demonstrate a significant relationship.
  3. Report findings of single outcomes when composite outcomes are reported.


  1. For subjects with decompensated HF, consensus on what are the most clinically relevant time intervals (shorter and longer term) for predicting outcome.

Key Question 4. In HF populations, does BNP measured at admission, discharge, or change between admission and discharge add incremental predictive information to established risk factors for morbidity and mortality outcomes?

Overview: Key Question 4

From the 198 publications that evaluated prognosis in KQ3, we examined a subset of 41 studies specifying that their intent was to assess incremental value. From these 41 studies, 17 were not extracted as they did not provide data2,247 or included BNP in the base prognostic model,106,196,210,212,273 in the NT-proBNP predictive model,282,303,316,339,343,348,352,362,375 or both assays were included in the model.217 In all these circumstances, the incremental value could not be extracted.

Incremental Value of BNP and NT-proBNP in Patients With Decompensated Heart Failure

Seven publications evaluated incremental value of BNP/NT-proBNP in decompensated HF subjects for admission BNP3,187,193,198,205 and admission NT-proBNP.251,256 Within the BNP publications incremental value was consistently shown to predict all-cause mortality for short-term (3 and 6 months) and longer-term (9 and 12 months). Two studies compared the incremental value of BNP to other cardiac markers (carbohydrate antigen125 (CA125),205 C-reactive protein (CRP),193 and cardiac troponin-T (cTnT)193) and did not show superiority. Within the two NT-proBNP publications, both studies251,256 showed incremental value at 22 months and 6.8 years for predicting all-cause mortality. In those studies that considered other cardiac markers and all-cause mortality, the highest incremental predictive value was achieved when BNP/NT-proBNP was combined with these other markers. Only two studies evaluated predicting cardiovascular mortality in the short term (31 days) and longer term (9 months) and showed BNP did add incremental value; NT-proBNP studies did not evaluate cardiovascular mortality.

Only mortality related outcomes were evaluated in these studies and none evaluated outcomes of morbidity or composite outcomes. All studies evaluated admission BNP levels and none evaluated discharge or change in BNP/NT-proBNP levels. Future research in patients with decompensated HF should endeavor to evaluate incremental predictive value for morbidity and composite outcomes and also to evaluate BNP/NT-proBNP levels at discharge from acute care centers or change relative to baseline but before discharge.

The majority of studies were predictor finding or developmental with respect to phased development of prognostic validation. None of the BNP publications included in KQ4 undertook internal or external model validation computations. Only one of the NT-proBNP studies251 evaluated incremental value and presented internal model validation computations. Future research in the incremental value of BNP/NT-proBNP should endeavor to undertake internal and external validation computations consistently to better assess the role of these assays.

Overall, despite the differences in the base models, cutpoints, and lengths of followup, evidence from lower hierarchical statistical approaches and early phase prognostic development studies suggest that BNP or NT-proBNP adds incremental predictive value in patients with decompensated HF.

Incremental Value of BNP and NT-proBNP in Patients With Stable HF

No eligible studies evaluated the incremental value of adding BNP in patients with stable chronic HF. Fifteen publications283,286,301,306,309,320,329,340,344,349,353,357,360,373,376 evaluating chronic stable HF patients considered the prognostic value of NT-proBNP.

When considering all-cause mortality, all but one study344 showed incremental value of adding NT-proBNP to the base models. The findings from four publications (with relatively large sample sizes) show consistent trend for incremental value to predict all-cause mortality at approximately 2 years. Similarly, four publications that evaluated the incremental value predicting mortality at 30 and 34 months were consistent in showing the added value of NT-proBNP. When incremental predictive value of NT-proBNP is compared to midregional pro-atrial natriuretic peptide (MR-proANP), hs-cTnT, or ST2, the relative contribution appears similar but the greatest increment was shown when NT-proBNP is combined with the other markers. When considering cardiovascular mortality, three studies consistently reported on the incremental value of NT-proBNP in patients with stable chronic HF for predicting from 12 to 24 months. Six publications that evaluated five different composite outcomes that combined mortality and morbidity events all suggest that NT-proBNP adds incremental value in predicting these outcomes from 22 to 37 months.

All but two publications, which evaluated the same cohort, undertook validation approaches, and the remaining studies were predictor finding or developmental with respect to phased development of prognostic research. Overall, despite the differences in the base models, cutpoints, and lengths of followup, these studies consistently show that NT-proBNP adds incremental predictive value for predicting mortality, morbidity and composite outcomes in patients with stable HF.

Applicability Issues in Prognosis Studies Evaluating Incremental Value

When considering the applicability of the BNP and NT-proBNP studies for KQ4, note that they do not differ from those that were identified for KQ3. Studies for KQ4 were derived from those eligible for KQ3; however, a much smaller pool of studies is considered. Of particular note is that the base models (covariates included), cutpoints, and lengths of followup varied widely across studies; it is not clear how these might impact applicability. Time intervals were heterogeneous for both studies of decompensated HF (from 31 days to 6.8 years) and stable chronic HF (from 12 to 37 months), making comparisons across studies problematic.

Conclusions for Prognosis Studies Adding Incremental Value

There is limited but consistent evidence that BNP or NT-proBNP adds incremental value for patients with decompensated HF for all-cause mortality and cardiovascular mortality in the short (3 and 6 months) and longer term (22 months to 6.8 years); outcomes of morbidity or composite outcomes have not been evaluated. There were no studies assessing the incremental value of BNP in populations with stable chronic HF. There is a consistent trend showing that NT-proBNP adds incremental value to predicting outcomes of all-cause mortality, cardiovascular mortality, and composite outcomes from 1 to 3 years when considered with other prognostic factors. Clinical utility of using multi-factor prognostic scoring need to be designed and evaluated before this becomes an established clinical tool.

Future Research Recommendations for Adding Incremental Value

  1. There is a need to evaluate outcomes of morbidity and composite outcomes in subjects with decompensated HF respect to the incremental value of BNP and NT-proBNP.
  2. There is a need to evaluate BNP in stable chronic populations with respect to incremental predictive value using appropriate computations.
  3. There is a need to move to higher level hierarchical approaches (internal and external validation) when selecting statistical evaluations (i.e., reclassification methods), as well as designing impact studies.
  4. Future research recommendations for KQ3 are also applicable for KQ4 for both decompensated and chronic stable populations.

Key Question 5. Is BNP or NT-proBNP measured in the community setting an independent predictor of morbidity and mortality outcomes in general populations?

Overview: Key Question 5

The use of markers to predict adverse outcomes in the general population has become fairly well established in the field of cardiology, especially with the advent of risk stratification tables for predicting CVD outcomes using variable such as age, sex, smoking, diabetes, systolic blood pressure, total cholesterol, and high-density lipoprotein (HDL)-cholesterol. These scoring systems have limitations and unfortunately are based on combined mortality and morbidity outcomes. The use of BNP or NT-proBNP in a community setting to add to these prediction scores would be valuable. The findings demonstrate clearly that an association exists between NT-proBNP and the outcomes of morbidity (HF and atrial fibrillation (AF)), as well as mortality (all-cause, cardiovascular, and sudden cardiac). No studies reported on the use of BNP in a community setting.

The adjusted HR demonstrates the log-linear relationship between baseline NT-proBNP and cardiovascular death, as well as all-cause mortality, taking into consideration age, sex, BMI, and renal function. The loss of independence in the prediction of cardiovascular death when baseline CVD is documented requires further assessment. The loss of independence may be a result of the smaller number of events (92) compared with 220 events in the whole population.379

For outcomes that are associated with cardiac disease (incident HF and AF), there appears to be a log linear relationship between NT-proBNP and the outcome, taking into consideration age, sex, BMI, and renal function. In addition, NT-proBNP seems to perform well, even when adjusted for other conventional risk markers and some of the more recently investigated biomarkers.

The prediction of AF became nonsignificant when all the other factors were used in a backward elimination adjustment.377 This suggests that when all the factors are considered, NT-proBNP may not provide independent prediction of future AF. It should be noted that this reference did measure another natriuretic peptide (MR-proANP) that showed significance in the model.

Applicability for Prognostic Studies From the General Population

While the association is clear, the directness of these findings to patient care is not demonstrated well in the papers reviewed. The statistical approaches to considering discrimination of prediction risk, Harrels c-statistic, the integrated discrimination of improvement (IDI), and net reclassification improvement (NRI), were evaluated in a number of papers.377,379-381 All of these demonstrated statistical benefit in including NT-proBNP in the prediction models (using other traditional risk factors) for incident HF,377 all-cause mortality,379 cardiovascular death,380,381 and combined cardiovascular outcomes.381 In these studies, the addition of NT-proBNP made a significant change to the c-statistic when added to conventional risk markers (similar but not identical in the papers).377,379-381 The reclassification data in these papers are presented using the best fit models that include NT-proBNP along with other biomarkers.377,380 One paper reported IDI and NRI.377

To translate this into clinical practice will require the development of specific risk calculators that take into consideration the confounders for NT-proBNP/BNP (renal function and BMI) and any other established risk markers (age, diabetes, hypertension, total cholesterol, HDL-cholesterol, smoking, and high sensitivity C-reactive protein). Such models will require testing in population cohorts before the use of NT-proBNP/BNP can be validated for use as a prognostic marker in community settings. These studies will have to demonstrate that measurement of NT-proBNP/BNP and any other biomarkers will clearly add to the predictive power of the risk calculation and change patient outcomes (all-cause mortality or cardiovascular mortality). In addition, to demonstrate economic benefit, the impact on actual outcomes is essential for the public to understand the benefit of the test in addition to all the other measurements that are usually required.

The term general population was strictly applied to this review. One study418 had selected subjects based on urine albumin excretion but claimed that they weighted their participants to model a general population. A companion paper419 was excluded because of the exclusion criteria reported in the study (type 1 diabetes mellitus) and the fact that they selected subjects based on urinary albumin excretion. This study had recruited 8,592 individuals and had data for 7,819 available for analysis.418 It is interesting to note that the reclassification statistics (HR, Harrell c-statistic, and IDI) reported in this paper largely confirm the findings of other reports and suggest that the weighting applied in the paper is a reasonable simulation of a general population.418 For all-cause mortality, HR=1.28 (95% CI, 1.11 to 1.47); Harrel c-statistic 0.84 (95% CI, 0.83 to 0.86); IDI 1.86 (95% CI, 1.26 to 2.45), are similar to the summarized results in Appendix L Table L-1. Similarly, for cardiovascular mortality, HR=1.40 (95% CI, 1.03 to 1.87); Harrel c-statistic 0.92 (95% CI, 0.89 to 0.95); IDI 2.06 (95% CI, 1.10 to 3.02); and cardiovascular events (HR=1.23 (95% CI, 1.11 to 1.38); Harrel c-statistic 0.83 (95% CI, 0.81 to 0.85); IDI 1.07 (95% CI, 0.73 to 1.40) are comparable.

Conclusions for Prognosis in Studies From the General Population

The findings demonstrate clearly that there is an association between NT-proBNP and the outcomes of morbidity (HF and AF), as well as mortality (all-cause, cardiovascular, and sudden cardiac) in the general population. The use of discrimination of risk statistics has shown that NT-proBNP adds statistical significance to the models of risk prediction. The development of a risk model for direct comparison against a standard risk model has not yet been reported.

Future Research Recommendations for Prognosis in Studies From the General Population

Future research should develop specific risk calculators that take into consideration the confounders and any other established risk markers. BNP has not been evaluated in the general population. The findings of the other sections of this report suggest that these would be similar and thus we would recommend that future studies consider measuring both NT-proBNP and BNP in the general population.

Such models will require testing in population cohorts before the use of NT-proBNP or BNP can be validated for use as a prognostic marker in community settings. It would also be helpful to have studies designed to help us understand which parameters in cardiac and renal function can be changed based on NT-proBNP or BNP measurement to improve clinical outcome.

Key Question 6. In patients with HF, does BNP-assisted therapy or intensified therapy, compared with usual care, improve outcomes?

Overview: Key Question 6

This systematic review question on BNP-guided therapy falls under the overarching question of how best to manage patients with HF. There were nine RCTs that addressed this question. Variation in study design, patient selection, baseline characteristics of patients, therapy goals, BNP/NT-proBNP cutpoint, outcome types, and how they were reported limited the option of performing any meta-analyses to derive summary estimates. Four studies reported at least one outcome that was better in the BNP/NT-proBNP group compared with the usual care group.4,53,386,388

The studies were carried out primarily in settings of cardiologists, which may attenuate the advantage of using BNP/NT-proBNP. Patients who are seen by a cardiologist will likely get less benefit from BNP-guided therapy, compared with those who are seen by a community physician who does not have the same expertise. Studies may also have been underpowered as few provided sample size calculations. In two studies, the followup time was only 3 months.384,387 All but two studies384,386 were done in multiple sites, but randomization was still patient-based. Maisel420 suggests that randomization should be based on site rather than on patient as this can reduce the “learning biases” in single-center randomizations.

The type of patients selected in these studies varied as there were different inclusion and exclusion criteria used. These studies were also limited to patients with systolic HF, as preserved ventricular function was only considered in the “Can Pro-brain-natriuretic peptide guided therapy of chronic heart failure IMprove heart fAilure morbidity and mortality?” (PRIMA) trial.385 The severity (or disease burden) of patients enrolled is therefore inconsistent across studies. Some studies specifically chose patients who were recently diagnosed with HF and therefore early in their time-point of the syndrome. There was a broad spectrum of patients with HF, including the very elderly and those with multiple comorbidities. Therapy at baseline was also variable. For example, NT-proBNP Testing to Guide Heart Failure Therapy in the Outpatient Setting (PROTECT) patients were receiving optimal therapy as 99 percent of the patients received angiotensin converting enzyme I (ACE-I) or angiotensin receptor blockers (ARB) and 94 percent of these patients had the recommended dose. Beta-blockers were taken by 99 percent of the patients and 59 percent of these were taking the recommended dose. Similarly, a high percentage of patients in the Trial of Intensified vs. standard Medical therapy in Elderly patients with Congestive Heart Failure (TIME-CHF) were receiving the recommended HF therapy.53 Strategies for Tailoring Advanced Heart Failure Regimens in the outpatient setting (STARBRITE) optimized therapy before the start of the trial.

The goals of therapy for the BNP/NT-proBNP group compared with the usual care group were different mainly in the target concentration set. A higher target means concentration did not need to decrease too much and were therefore, less likely to change symptoms or outcomes. A lower target runs the risk of adverse events outweighing the benefits. There was no consistency in trials with lower compared with higher BNP/NT-proBNP concentrations. In a subanalysis of data from the PRIMA study,385 patients who achieved their target concentration did better than those who did not. This provides some support to using individualized target concentrations rather than population-based targets. Furthermore, the application of biological variation data (see Results KQ7), specifically the reference change value (RCV), may be enhanced when therapy is altered (e.g., titration of medications or addition of medications). We know that patients can vary widely between serial BNP/NT-proBNP measurements, some to a larger degree and others to a smaller degree, but sporadic increases could also occur. A predefined BNP/NT-proBNP cutpoint that is seen with the most stable patients with HF (e.g., <200 pg/mL for BNP and <1,000 pg/mL for NT-proBNP) may be a reasonable choice.3 The frequency of measurements is another aspect that has not been assessed in BNP/NT-proBNP therapy studies. Another consideration is whether BNP/NT-proBNP measured using point-of-care devices (e.g., Triage BNP), has a higher analytical variation and therefore contributes to a higher RCV, and is less sensitive to detecting a change in HF status. There are also different forms of BNP/NT-proBNP that may vary depending on worsening symptoms and other comorbidities and the assays may measure these species differently. Making patients and caregivers aware of the BNP or NT-proBNP test result may encourage patients to stay on treatment but evidence is limited. Two studies used this approach, one with a positive outcome386 and one with a negative outcome.389

The aggressiveness of therapy among the studies appeared to vary, but this was difficult to assess as not all studies reported drug titrations in the same way. The timing was not always reported, nor the change in dose or when additional medications were given. A structured approach would be difficult, as patient care is individualized, but the data need to be captured to compare interventions. The recommendation for therapy suggested by Maisel420 is to establish predefined treatment goals, at least to recommended guideline doses, and to use clinical judgment to individualize medications according to the patient's response. That is, mirror what is normally done. In the BNP/NT-proBNP group, Maisel suggests to increase followups and increase doses as long as there are no adverse events (e.g., decreased blood pressure or worsening kidney function). Also, have additional followups if the condition is worsening. Furthermore, ensure there is documentation that the clinician has responded to an elevated BNP/NT-proBNP concentration for the BNP/NT-proBNP-guided group. Another suggestion is to enhance data collected from these studies to consider measuring other biomarkers that reflect HF pathology, including more heart-specific and renal biomarkers. A multi-marker panel may offer greater value than a single marker in guiding therapy by adding greater precision to the estimate of pathology.

A successful BNP/NT-proBNP-guided therapy study is one in which hospital admissions are reduced, clinicians and physicians adhere to HF therapy guidelines, renal function is preserved, and quality of life is improved.420 All studies captured information on hospital events and most measured kidney function, but only four had quality of life data. No studies reported on how well physicians followed therapy guidelines.

There were six studies that used composite endpoints, but because the combination of outcomes were different it was difficult to compare studies. There was no relationship between the number of individual endpoints within the composite and overall effect. Combining endpoints into a composite helps to reduce the number of patients required to achieve adequate power. However, it can also obscure the component in the composite that had the most events causing a misinterpretation of the positive or negative outcome achieved. For example, the PROTECT trial386 had the combined outcome of cardiovascular death, HF hospitalizations, acute coronary syndrome, cerebral ischemia, significant ventricular arrhythmias, and worsening HF. However, the only difference between the two treatment arms was for the individual endpoints of HF hospitalizations and worsening HF. Mortality was no different between treatment arms, and only two studies4,53 that included this endpoint in the composite found a difference (lower in the BNP/NT-proBNP arm) and happened to be the two of three studies with the longest followup. Endpoints such as mortality would occur less frequently and therefore there are fewer events to capture in shorter trials, but these trials can achieve sufficient power by recording more frequent events like hospitalizations. In addition, in trials where adverse events were collected, BNP/NT-proBNP-guided therapy differed between treatment groups. This finding suggests that clinicians used other information in addition to the BNP/NT-proBNP results to make decisions on therapy.

Five studies reported negative results, three (Beck-da-Silva,384 SIGNAL-HF,5 STARBRITE387) had short followups (3 to 9 months), which would have limited the number of outcomes that would have occurred over a longer period of time. In the other two studies, one385 only required a 10 percent reduction in BNP/NT-proBNP from baseline, and in the other study389 patients had the most type of medications, 35 percent of which were taking an ARB. Studies have shown that ARB use decreases mortality, and in one study cardiovascular mortality was decreased in patients with HF and reduced LVEF.

Data interpreted based on age may also be important. In the TIME-CHF study,421 younger patients (≤75 years) benefited more than older patients (>75 years), but there was no difference between these age groups in the Use of Peptides in Tailoring hEart failure Project (UPSTEP).389 Younger patients may seem to do better, but this may depend on how care is given, as older patients need a more careful, gradual approach.

One limitation to this systematic review was the exclusion of two trials, the first trial assessing BNP/NT-proBNP-guided therapy in 2000,422 and a more recent study in 2010 done by the same research group.423 They were not included because the method for NT-proBNP measurement is not a commercially available one, but an in-house method. The data from these trials would have strengthened the results of this systematic review but not altered the conclusions. Also, meta-analyses were not performed because of the heterogeneity among the studies, and therefore no quantitative summary estimates could be made. Two previously published studies did conduct meta-analysis and reported reduced mortality in the BNP/NT-proBNP guided group.35,424

Applicability for BNP-Guided Therapy

Understanding the usefulness of BNP or NT-proBNP measurement in the assessment of HF status will allow for better management of patients with HF. It may or may not be useful. If it is useful, it would essentially serve as a barometer for disease improvement or deterioration. Currently, the data from the studies that have evaluated BNP or NT-proBNP for this purpose are inconclusive.

Conclusions for Intervention Studies

Over the last 10 years, few studies have been undertaken to assess whether BNP/NT-proBNP-guided therapy has benefits over usual care. The conclusions from these studies are varied in part because of the difference in study design and outcomes. Differences among studies provide greater understanding on how BNP/NT-proBNP-guided therapy can be used, regardless of whether trials succeeded or failed. The SOE for all-cause mortality was low.

Future Research Recommendations for Intervention Studies

The data reported from the nine studies evaluating the utility of BNP or NT-proBNP for guiding therapy in patients with HF provides a rich basis of information to draw upon to design further RCTs. Based on the information gathered, future trials should consider the following design features:

  1. Therapy optimized at baseline according to clinical guidelines.
  2. BNP or NT-proBNP target near the median value for patients with stable HF.
  3. Consider using the RCV when considering a change in therapy.
  4. Followup of 2 years or more.
  5. Include all relevant endpoints: cardiovascular mortality, total mortality, days alive and not hospitalized for HF, number of HF hospitalizations, number of HF events not requiring hospitalization, surrogate measures of renal function (e.g., creatinine) and ischemia (e.g., troponin), number of patients who have achieved target BNP/NT-pro-BNP concentration, and number of patients who have achieved recommended medication doses. Also, include as part of medication information the number of patients who are taking additional medications or doses above the recommended amounts. Quality of life questionnaires would be of additional value.
  6. Provide sample size calculations to demonstrate adequate study power for the outcomes selected.
  7. Consider age in the statistical analyses to determine how age affects outcome (treatment effect).
  8. Consider regression analyses to test for interactions between intervention and characteristics such as age, sex, NHYA class, and disease.
  9. Provide confidence intervals for all statistical measures to allow meta-analyses to be performed as recommended by the CONSORT Statement.425
  10. Consider evaluating other biomarkers in establish a panel that can be used to assess disease improvement or deterioration.

Key Question 7. What is the biological variation of BNP and NT-proBNP in patients with HFand without HF?

Overview: Key Question 7

It is important to know biological variation for BNP and NT-proBNP in order to be able to effectively use these measurements for managing patients with HF. Specifically, what constitutes a significant change in serial measurements or RCV? In other words, this information provides knowledge about the reproducibility of the test result in patients with no change in clinical status, deterioration, or improvement. This systematic review found six studies that contained biological variation data in patients with stable HF. The requirement for stable HF was made so as to eliminate variation from individuals who were not optimized on medical therapy and thus could have a change in their HF status or who had experienced a recent event such as hospitalization or myocardial infarction. The value in doing this is to be able to apply the biological variation data to the group of patients where they would be used as biological variation maybe different in other patient groups. From this systematic review, the two studies where healthy individuals were evaluated, the RCV values were higher than those in the studies of patients with stable HF. However, this difference may also reflect the difference in age as the healthy groups were younger than the HF groups. The age dependence of within-individual variation is known for other analytes, that is, lower variation compared to younger individuals.426

Within-individual variation was similar for BNP (median=25%) compared with NT-proBNP (median=20%), but lower in short measurement intervals (hours, days) compared to longer measurement intervals (weeks, year). Although the circulating half-life of BNP is much shorter (21 min) compared with NT-proBNP (60 to 120 min), this did not seem to affect the biological variation values for within-individual (CVi) values by much.427 Another factor to consider when interpreting the CVi values is that they are calculated from the difference in variance between total variation (CVt) and analytical coefficient of variation (CVa). Thus, a lower CVa will provide a more accurate and higher CVi. The highest CVa values were obtained from the point-of-care instrument (Triage BNP) and correspondingly resulted in higher RCV values. Reduction of CVa is possible by using automated instruments and measuring samples in duplicate.

Accuracy of biological variation estimates is a function of study design, including the selection of participants, preanalytical factors such as participant preparation (e.g., fasting, posture, and stress), and time of collection (to minimize diurnal variation; NT-proBNP, and more so BNP, increase during the day and stabilize in the afternoon). Further precision can be gained by increasing the number of samples collected within the measurement interval (study time frame), number of replicates for each sample (e.g., duplicate), and statistical methods. The number of replicates becomes more important when variation (analytical or biological) is high. In the study by Schou et al.37 the number of determinations of a sample on the biological variation estimates was explored, with small changes seen between single and double determinations. This was explained by the very low analytical variation for both BNP and NT-proBNP.

Most studies included in this systematic review considered at least some known pre-analytical factors and tried to minimize or address them. However, the determinants of within-person biological variation have not been well explored; more is known about between-person variation, such as sex, age, exercise, and comorbidity.428 The biological variations are likely due to subclinical changes in hemodynamics, hormonal regulation, clearance, and perhaps even differences in the type of circulating forms of BNP, as well as whether the measurement method detects them.427

Calculations for biological variations should also consider the distribution of the data. It is well known that the distribution of NT-proBNP data is skewed to the right and log transformation of data is appropriate for statistical analysis. The reason for this skewness is not known but may indicate the population is heterogeneous or nonbiological variation factors are present. If Gaussian distribution is assumed, then all data (99.7%) will fall within ±3 SD of the mean. Therefore, in an RCV calculation, the CVi cannot be greater than 33.3% without including negative values. There is a linear relationship between CVi and NT-proBNP concentration, but after log transformation, CVi is reduced and the association with concentration is removed. Shou38 examined the difference in CVi values using year-to-year NT-proBNP normal and log data and found the mean CVi to drop from 35 percent to 5.4 percent. The log CVi suggests the variation in NT-proBNP to be fairly stable. However, monitoring on a log scale is difficult because it carries a risk that small changes reflecting a true biological change are missed. Therefore, biological variation data should be interpreted on a non-log scale.

No meta-analysis could be done to compute summary estimates for CVi or RCV as confidence limits were not provided for variance data in any study. Recently, Roraas429 described how experimental design greatly influences the confidence interval and reliability of the biological variation estimate.

The index of individuality (IOI) for BNP and NT-proBNP was between 0.03 and 0.14, which is lower than any of the common biochemistry analytes.430 For example, the IOI for creatinine is 0.24 and for cholesterol it is 0.33. This means patients are not like each other and reference intervals or decision limits are not as useful. A low IOI (<0.48) is considered to reflect strong individuality, which in turn indicates that an individual patient should be assessed with respect to his or her individual hormonal level. In contrast, a high IOI (>1.4) indicates this patient should be assessed with respect to population-derived reference intervals (or decision points). In practice, serial monitoring of patients using the RCV provides the best assessment of change. However, this information is rarely provided on laboratory reports to assist clinicians in interpreting test results.

Applicability Issues in Biological Variation

The applicability of the RCV values calculated from patients with stable HF is to assess instability in patients with HF. Although the inclusion criteria of patients with stable HF varied among studies, some stricter than others, this did not seem to influence the RCV values by a large degree. The time frame of collection for the biological variation data seemed to influence the RCV. The within-hour and within-day values were much lower, yet there was no discernible difference beyond this time period (up to 2 years). Interestingly, the RCV values for BNP were about double those for NT-proBNP. This information, in addition to the shorter half-live of BNP (minutes) compared to NT-proBNP (hours), raises the possibility that NT-proBNP may have an advantage over BNP to detect the same clinical change. Since NT-proBNP has a longer half-life it can be regarded as an averaging effect of the biologically active BNP. An analogy to BNP and NT-proBNP in HF could be drawn from fructosamine and glycated hemoglobin (HbA1c) in diabetes. Both tests measure glycation but fructosamine has a higher RCV and shorter half-life compared to HbA1c (10.2% and 2 to 3 weeks compared to 7.6% and 8 to 12 weeks, respectively).431 Current practice recommends HbA1c for monitoring diabetic control because it correlates better with diabetic complications compared to fructosamine.

Conclusions for Biological Variation

The data on biological variation for BNP and NT-proBNP offer insight into the changes that can be expected in patients with stable HF and in healthy individuals. The difference in serial results, expressed as RCV, was higher for BNP compared with NT-proBNP. Furthermore, the IOI for BNP and NT-proBNP was very low, thereby highlighting the individuality of this hormone and suggesting serial measurements need to be interpreted carefully.

Future Research Recommendations for Biological Variation Studies

  1. Additional studies would provide supporting evidence of the biological variation parameters. These studies should be designed to capture sources of biological variation determinants by multivariable regression analysis requiring large sample sizes. These analyses may also provide clues as to why the data distributions for BNP and NT-proBNP are right-skewed.
  2. Preanalytical and analytical variation should be minimized by collection of samples in the early morning when BNP and NT-proBNP are at their nadir, increasing the frequency of collection and duplicating determinations to increase accuracy of the measure.
  3. Statistics used should be clearly described, include all biological variation components, and provide confidence intervals to show reliability and allow meta-analyses to be done.
PubReader format: click here to try


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (9.6M)

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...