illustrates the results of ROC curve analysis in which the BNP level was employed to identify CSR-CSA associated with CHF, with an area under the ROC curve of 0.79 (95% CI, 0.668 to 0.913; p < 0.001). Table 2
shows the sensitivity, the specificity, and the positive and negative predictive values for several BNP cutoff values (ie, 80, 116.25, and 152 pg/mL) selected from the ROC curve. The best result was obtained with the cutoff value of 116.25, with a sensitivity of 62%, a specificity of 92%, and an accuracy of 83%. Covariance analysis supported the differences between the two groups of patients in terms of their BNP levels (p = 0.001) after adjusting for age, gender, diabetes, renal insufficiency (ie, creatinine level > 2 mg/dL), and LVEF. In stepwise multiple regression analysis, the 30.5% of variability in BNP levels was predicted by a model that included AHI and DI, with the remaining independent variables (sex, age, LVEF, presence of diabetes, and renal insufficiency) excluded from the model because of a lack of statistical significance.