The final diagnosis was dyspnea due to CHF in 47% of patients, dyspnea due to noncardiac causes in 5% of patients with known left ventricular dysfunction, and dyspnea not due to CHF in 49%. While median BNP levels were significantly different between each group, there was a large degree of overlap (median BNP levels [±SD], 675 ± 450 pg/mL for dyspnea due to CHF, 346 ± 390 pg/mL for dyspnea not due to CHF in patients with known left ventricular dysfunction, and 110 ± 225 pg/mL for dyspnea due to noncardiac causes). The BNP levels were more accurate than clinical predictors, such as physical examination and Framingham and NHANES congestive heart failure scores, for diagnosing heart failure. A cutoff value of 100 mg/mL provided the highest accuracy, with a sensitivity and specificity of 90% and 76%. Given positive and negative likelihood ratios of 3.75 and 0.13 at this cutoff, it is clear that clinical judgment must still play an important role in the diagnosis of CHF, as interpretation of the BNP result remains dependent on the physician's estimate of pretest probability. For example, with a pretest probability of 50% for CHF, a BNP level greater than 100 pg/mL would increase the likelihood of CHF to 79%, while a level less than 100 pg/mL would reduce the likelihood that dyspnea was due to CHF to 12%. At very low values, BNP level is useful for excluding the diagnosis of acute CHF as the cause of dyspnea, with a negative predictive value for BNP level less than 50 pg/mL of 96%. The authors did not provide specificity data for BNP cutoff levels greater than 150 pg/mL, but it is reasonable to infer from these data that markedly elevated BNP levels do indicate heart failure.