This review assessed the diagnostic accuracy of brain natriuretic peptide (BNP) for the diagnosis of heart failure. The authors concluded that BNP is an accurate marker of heart failure and a cut-off of 15 pmol/L achieves high sensitivity. Although the review was well conducted, this conclusion appears over-optimistic given the evidence presented and should be viewed with caution.

To assess the diagnostic accuracy of brain natriuretic peptide (BNP) for the diagnosis of heart failure and to compare it with the diagnostic accuracy of atrial natriuretic peptide (ANP).

MEDLINE and EMBASE were searched from January 1994 to December 2002; the search terms were not reported but are available from the authors. The reference lists of retrieved primary studies and reviews were checked for further studies.

Study designs of evaluations included in the review

Diagnostic accuracy studies were eligible for inclusion. Case control studies were excluded. Studies examining the association between natiuretic peptides and heart failure were excluded.

Specific interventions included in the review

Studies that measured natriuretic peptide levels were eligible for inclusion. The included studies measured BNP or N-terminal proBNP; some also measured ANP or N-terminal ANP. The studies used a variety of diagnostic cut-off points. Accuracy data were extracted from included studies for the cut-off closest to 15 pmol/L (range: 5.2 to 41.3).

Reference standard test against which the new test was compared

To be included in the review, the studies were required to use a reference standard for heart failure. However, further inclusion criteria were not specified. The included studies used a variety of reference standards which the review categorised into the following groups: left ventricular ejection fraction (LVEF) less than 30%; LVEF less than 40%; LVEF less than 45 to 55%; clinical diagnosis; diastolic failure; systolic or diastolic failure.

Participants included in the review

There were no inclusion criteria for participants. The included studies involved a variety of settings and participant groups, ranging from the general population aged over 45 years to patients diagnosed with heart failure.

Outcomes assessed in the review

Studies that reported results allowing the construction of a 2x2 table were eligible for inclusion.

How were decisions on the relevance of primary studies made?

The authors did not state how the papers were selected for the review, or how many reviewers performed the selection.

The following methodological quality criteria were assessed: patients were a consecutive series or random sample; index and reference tests were assessed independently and blinded to the other test result; all patients received both tests; the methods for performing both tests were described; characteristics of the study population were described; no time delay between the two tests. Two reviewers independently assessed study quality. Any disagreements were resolved by consensus or by consulting a third reviewer.

Two reviewers independently extracted the data. Any disagreements were resolved by consensus or by consulting a third reviewer. Data were extracted to construct a 2x2 table for each cut-off point reported. The diagnostic odds ratio (DOR) was calculated for each cut-off, and the average of these was calculated for each study. The statistical methods were described in full in the report.

How were the studies combined?

Studies of the diagnostic accuracy of BNP were combined using a DerSimonian and Laird random-effects model to generate pooled estimates of the DOR with 95% confidence intervals (CIs) for each reference standard. An unweighted least-squares regression model was used to test for a relationship between DOR and cut-off point. Positive and negative likelihood ratios (LRs) were calculated where the studies had used similar cut-offs. Data from studies comparing the diagnostic accuracy of BNP and ANP were used to generate summary receiver operating characteristic curves for each analyte; their comparative performance was then assessed by examining the difference between the estimated areas under the curves (AUCs) divided by the variance of AUC.

How were differences between studies investigated?

Statistical heterogeneity between studies used to generate pooled DORs was assessed using the chi-squared test. Subgroup analyses were performed to pool studies using similar diagnostic thresholds and to compare studies carried out in different clinical settings.

The review included 20 studies, involving 9,702 participants in total.

Study quality was generally high: each of the criteria was satisfied by between 14 and 20 of the studies.

The diagnostic accuracy of BNP compared with each reference standard was as follows.

LVEF less than 40% (8 studies): the pooled DOR was 11.6 (95% CI: 8.4, 16.1) and there was no evidence of heterogeneity. Pooling only those studies that used a cut-off for BNP of between 14 and 19 pmol/L (5 studies) gave an estimated positive LR of 4.1 (95% CI: 2.6, 6.6) and a negative LR of 0.35 (95% CI: 0.17, 0.72).

LVEF 45 to 55% (7 studies): the pooled DOR was 5.6 (95% CI: 3.7, 8.5) but there was statistically significant heterogeneity (P<0.01).

Clinical criteria (7 studies): the pooled DOR was 30.9 (95% CI: 27.0, 35.4) and there was no evidence of heterogeneity. The largest study in this group had a cut-off for BNP of 14.4 pmol/L with a positive LR of 2.6 (95% CI: 2.3, 2.8) and a negative LR of 0.05 (95% CI: 0.03, 0.07).

Diastolic failure (3 studies): the pooled DOR 28.3 (95% CI: 2.66, 300.5), but there was significant heterogeneity (P<0.001).

Systolic or diastolic failure (2 studies): the pooled DOR was 37.7 (95% CI: 5.9, 237.2) but there was significant heterogeneity (P=0.02).

There was no significant difference between the pooled DORs of studies carried out in general practice or community settings and those performed in hospital.

The 3 studies comparing the diagnostic accuracy of BNP with N-terminal ANP found BNP to be marginally more accurate as a diagnostic marker of heart failure (P=0.048).

BNP is an accurate marker of heart failure, although there was considerable unexplained variation in estimates of diagnostic accuracy. The use of a cut-off value of 15 pmol/L achieves high sensitivity for heart failure in patients in whom disease is suspected, and BNP levels below this can be used to rule out heart failure. BNP has greater accuracy when compared with clinical criteria than when compared with LVEF alone.

The review question was clear although the inclusion criteria were fairly broad; this means that the included studies encompassed a wide range of participants, diagnostic thresholds and reference standards. The search for primary studies was limited to searches of two databases and the scanning of reference lists, and there was no attempt to locate unpublished studies. It is therefore possible that relevant studies were missed. Two independent reviewers performed the data extraction and quality assessment, which should have minimised the introduction of bias or error at these stages of the review process. Appropriate techniques were used to analyse and pool study results, statistical heterogeneity was formally assessed, and clinical heterogeneity was investigated in terms of study setting. The authors' conclusions appear over-optimistic given the evidence presented, and the conclusion regarding a BNP threshold of 15 pmol/L should be treated with caution as few of the included studies used this value as a cut-off.

Practice: The authors stated that the sources of variation in the diagnostic accuracy of BNP for heart failure should be understood before it is recommended for routine clinical use.

Research: The authors stated that further research is needed to define the role of BNP in clinical management.

This is a critical abstract of a systematic review that meets the criteria for inclusion on DARE. Each critical abstract contains a brief summary of the review methods, results and conclusions followed by a detailed critical assessment on the reliability of the review and the conclusions drawn.