Figure 4

Receiver operating characteristics (ROC) for 2 to 5 protein panels for identification of normal patients with 95% specificity for detection of coronary artery disease (CAD) patients. The ROC curves are derived from 4 separate panels optimized to detect 101 normal patients (true positives in this figure) at highest specificity while maintaining a sensitivity of 95% for patients with CAD (138 samples). The ROC curves are obtained by iteratively testing each biomarker panel for classification of a randomly excluded portion (20%) of the dataset. The areas under the curve (AUC) were comparable as indicated in the curves for two proteins (osteopontin (OPN) and resistin: AUC = 0.839), three proteins (OPN, resistin, apolioprotein B100 (APO-B100): AUC = 0.845), four proteins (OPN, resistin, matrix metalloproteinase 7 (MMP7) and interferon γ (IFNγ): AUC = 0.839) and five proteins (OPN, resistin, MMP7, IFNγ and C-reactive protein (CRP): AUC = 0.827). The predicted specificity for detection of normal patients at 95% sensitivity for CAD patients was two proteins = 50%, three proteins = 52%, four proteins = 63% and five proteins = 64%.