Adjusting for clinical covariates improves the ability of B-type natriuretic peptide to distinguish cardiac from non-cardiac dyspnoea: a sub-study of HEARD-IT

We sought to create a model that adjusts B-type natriuretic peptide (BNP) for specific covariates to better distinguish cardiac from non-cardiac dyspnoea. HEARD-IT was a multicentre, prospective study of the diagnostic utility of acoustic cardiography in the emergency department. Dyspnoeic patients more than 40 years were eligible. Two cardiologists independently adjudicated the HF outcome. Using logistic regression, a model adjusting BNP for pertinent covariates was developed (n = 740). The mean age was 66 +/- 13 years. Age, gender, ethnicity, body mass index, blood urea nitrogen, and creatinine affected BNP levels independently of HF. The model adjusting BNP for these covariates improved the area under receiver operator characteristic curve for HF compared with BNP alone (0.948, 95% CI 0.934-0.963 vs. 0.937, 95% CI 0.920-0.954; P = 0.004). Net reclassification improvement, a novel metric of model performance, was 3.5% for those without HF (P = 0.05) compared with conventional, unadjusted BNP cut-offs. Thirteen of 116 (11%) patients without HF, but with unadjusted BNP values >= 100 pg/mL, were correctly reclassified as not having HF with the adjusted BNP model. Adjusting BNP for important covariates may improve its ability to distinguish cardiac from non-cardiac dyspnoea.