Limited value of brain natriuretic peptide as a prognostic marker in acute heart failure--a meta-analysis.

Brain natriuretic peptide (BNP) is secreted by cardiomyocytes in response to excessive stretching. BNP plasma levels are useful for diagnosis in patients with dyspnea [1] predicting heart failure better than clinical assessment alone [2]. Therefore, determination of BNP when heart failure is suspected has become standard of care [3]. However, for patients with established heart failure who undergo an exacerbation, it is unknown whether BNP determination adds any clinical value. Although studies have shown an association between greater BNP levels and adverse events [4,5], the additive value of its determination on top of known predictors for heart failure prognosis has not been tested. We performed a systematic review and meta-analysis of all studies inwhich BNP levels were used for prognosis in patients admitted with decompensated heart failure in order to assess the predictive value of an above median BNP level for all cause mortality, and to determine the discriminative value of elevated BNP levels. Two trained investigators (A.B., A.A.) independently performed a systematic search for studies using PubMed and The Cochrane Collaboration CENTRAL database from January 1990 to December 2009. Proceedings from the American College of Cardiology, American Heart Association, the European Society of Cardiology and the Heart Failure Society of America annual meeting were also searched for the prior 3 years. Potentially relevant studieswere recovered as completemanuscripts and assessed for compliance to inclusion and exclusion criteria. Cross examination of the reference lists was performed to collect additional relevant studies. Inclusion criteria were: a) determination of BNP values within the first 24 h of hospitalization, b) subjects admitted with decompensated heart failure, c) studies that had reported all cause mortality, and d) enrolling N10 subjects. Studies that employed NT pro-BNP were excluded. When uncertain about published data, the principal investigator was contacted by correspondence at least two times. Eventually, in case of unobtainable data, the study was excluded. All cause mortality was chosen as the endpoint of interest. A fixedeffect model, computing relative risks (RRs) with 95% confidence