Forty-nine studies were included in the review (n=23,886). Study quality was generally described as adequate, although delays of up to a year between the index test and reference standard were reported in two screening studies.
Detection of heart failure in symptomatic patients (27 studies, n=7,062):
There was significant heterogeneity between the 15 studies (p<0.001, I2=79%) of BNP. Meta-regression showed that the DOR was decreased by a factor of 2.0 (95% CI 1.0 to 4.2) for each additional decade of age for studies of BNP. There was no significant effect of sex balance, in-patient and out-patient status, width of distribution of age and the use of a prespecified threshold. The results of nine studies of BNP in patients with a mean age of 80 years or less showed a summary DOR of 27.7 (95% CI 21.6 to 35.6). There was no evidence of heterogeneity (p=0.34, I2=5%).
There was significant heterogeneity between the nine studies of NT-proBNP (p<0.001, I2=68%). Meta-regression showed that the DOR was decreased by a factor of 2.5 (95% CI 1.7 to 3.7) for each additional decade of age for studies of BNP. After removal of one poor-quality study, the results studies of NT-proBNP in patients with a mean age of 80 years or less showed a summary DOR of 37 (95% CI 26.6 to 51.6). There was no evidence of heterogeneity (p=0.32, I2=12%).
Community screening studies (13 studies, n=14,590):
Studies that assessed BNP and used a severe definition of left ventricular systolic dysfunction (ejection fraction ≤40%) showed a summary DOR of 19.9 (95% CI 12.5 to 31.9). There was some evidence of heterogeneity (p=0.14). Meta-regression showed no significant effect of age, quality or age distribution on the DOR. Studies that assessed two levels of left ventricular systolic dysfunction severity consistently showed higher DOR for severe disease than for mild disease, which suggested that the assay was more accurate for more severe disease. There were only three studies of NT-proBNP and these showed a summary DOR of 9.3 (95% CI 4.7 to 19.0).
Mixed samples (eight studies, n=2,234):
Results of these studies varied widely and generalisability was unclear.
Comparison of assays (seven studies):
Accuracy of BNP was better than that of NT-proBNP in five studies; in two studies accuracy of NT-proBNP was greater. The pooled ratio of odds ratios showed that BNP was significantly more accurate than NT-proBNP (OR 1.77, 95% CI 1.06 to 2.95).