Table 2.
Evaluation of Biomarkers and Biomarker Profiles Using the Area Under the ROC Curve Method
| CSF | Plasma | CSF & Plasma | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| EV | PeV-A3 | EV | PeV-A3 | EV | PeV-A3 | ||||||
| TNF-α | 0.88 | Disc2 | 0.866 | Fractalkine | 0.854 | Disc1 | 0.952 | Disc2 | 0.924 | Disc1 | 1 |
| MIP-1α | 0.861 | IL-1β | 0.726 | GM-CSF | 0.774 | IFN-α2 | 0.901 | Disc1 | 0.912 | Disc2 | 0.558 |
| IFNg | 0.86 | MIP-1α | 0.725 | IL-15 | 0.74 | IL-15 | 0.898 | ||||
| Disc1 | 0.85 | IL-6 | 0.717 | Disc2 | 0.72 | MCP-1 | 0.898 | ||||
| IL-6 | 0.844 | IFNg | 0.686 | IL-17A | 0.718 | IP -10 | 0.891 | ||||
| IL-10 | 0.842 | TNF-α | 0.674 | IL-10 | 0.716 | IL-1Ra | 0.881 | ||||
| IL-1Ra | 0.841 | IL-15 | 0.659 | IFN-α2 | 0.712 | Fractalkine | 0.874 | ||||
| IL-1β | 0.839 | GM-CSF | 0.639 | IL-2 | 0.676 | IL-8 | 0.755 | ||||
| IL-12p40 | 0.835 | IP-10 | 0.632 | MIP-1α | 0.666 | IL-12p40 | 0.753 | ||||
| GM-CSF | 0.826 | MCP-1 | 0.623 | IL-6 | 0.66 | IL-10 | 0.728 | ||||
| IL-8 | 0.812 | IL-10 | 0.61 | IL-12p40 | 0.65 | RANTES | 0.663 | ||||
| Fractalkine | 0.809 | IL-8 | 0.604 | Disc1 | 0.64 | GM-CSF | 0.655 | ||||
| IFN-α2 | 0.8 | Disc1 | 0.601 | MCP1 | 0.632 | TNF-α | 0.609 | ||||
| IL-4 | 0.774 | RANTES | 0.593 | IL-8 | 0.612 | IL-13 | 0.595 | ||||
| IL-13 | 0.765 | IL-17A | 0.548 | TNF-α | 0.604 | IL-6 | 0.595 | ||||
| IP-10 | 0.749 | IL-13 | 0.543 | IL-1Ra | 0.6 | IL-5 | 0.571 | ||||
| Disc2 | 0.702 | IL-12p40 | 0.541 | IP-10 | 0.564 | IL-4 | 0.568 | ||||
| IL-2 | 0.69 | IL-1Ra | 0.54 | IL-4 | 0.548 | MIP-1α | 0.556 | ||||
| MCP-1 | 0.682 | IL-5 | 0.531 | IL-13 | 0.544 | IL-17A | 0.524 | ||||
| IL-15 | 0.679 | IL-2 | 0.527 | IL-5 | 0.54 | IL-2 | 0.519 | ||||
| IL-17A | 0.673 | Fractalkine | 0.525 | IFNg | 0.528 | IL-1β | 0.51 | ||||
| RANTES | 0.649 | IFN-α2 | 0.516 | IL-1β | 0.526 | Disc2 | 0.507 | ||||
| IL-5 | 0.556 | IL-4 | 0.516 | RANTES | 0.524 | IFNg | 0.503 | ||||
Analogous to principal components in PCA, discriminant analysis computes discriminants. Both principal components and discriminant functions are eigenvectors that can be viewed as artificial variables comprised of contributions from observed variables. In a multivariate problem, data points are plotted in multidimensional space with as many axes as variables. To transform this into a simpler 2- or 3-dimensional presentation, variables are combined into eigenvectors ranked by the amount of variance they explain. In PCA, the direction of the eigenvector that explains the most variance (ie, first component) is selected so that it explains the maximum amount of variance that can be explained by 1 vector (ie, maximizing the amount of variance explained). In discriminant analysis, the eigenvector that explains the most variance (first discriminant function) is selected to maximize group separation. As these eigenvectors are linear combinations of observed variables, they may be used as a way to combine variables and test them using the AUC method. Here, we selected discriminant functions (Disc1 and Disc2) as they are likely to be superior to principal components given the way they were computed. The heat map corresponds to a range of AUC values from the lowest (dark red) to highest (dark blue).
Abbreviations: AUC, area under the curve; CSF, cerebrospinal fluid; Disc1/Disc2, first/second discriminant function; EV, enterovirus; GM-CSF, granulocyte macrophage–colony-stimulating factor; IFN, interferon; IL, interleukin; IP, interferon-γ-inducible protein; MCP, monocyte chemoattractant protein; PCA, principal component analysis; PeV, parechovirus type A3; TNF, tumor necrosis factor.