Figure 2.

Separation of Delirium Versus Non-Delirium Using the 32 Significant Differential Proteins by Three Unique Linear Analysis Methods. (A). Heat map derived from the machine-learning Hierarchical Cluster Analysis (HCA) algorithm reflecting relative minimum and maximum expression levels for each given protein among patients as determined by proteomics analysis results. Comparisons were made between all patients with delirium and non-delirium control subjects. p < 0.05 comparing delirium to matched non-delirium controls for the 32 differential proteins is shown. In the HCA heat map, the color bar represents the display range, [−3, 3]; for the standardized signal values mapped to the red–green continuum, red denotes upregulation and green denotes downregulation. Protein names are represented by the gene symbol, and full names are presented in Table 2. CNT = control non-delirium samples; DEL = delirium samples. * indicates an incorrect classification: this means a CNT sample clusters with the DEL samples or vice versa. (B). Principal components analysis (PCA) separation of patients with delirium versus non-delirium control subjects into two dimensions with support vector machines (SVMs). The separating line is based on a linear kernel using the 32 protein CSF panel derived from proteomics analysis. The first principal component accounts for 22.48% of the variance, and the second principal component for 12.69% of the variance. PCA demonstrates that the derived proteomics data contain a significant component separating delirium cases from matched control samples (43/48 = 89.6%). Red circles denote delirium samples. Blue triangles represent non-delirium controls. (C). Receiver operating characteristic (ROC) curve analysis based on the 32 protein delirium signature. Our CSF delirium signature discriminates delirium subjects from non-delirium control subjects with an area under the curve (AUC) = 0.91 [0.80–0.97]. The 95% confidence interval for the AUC is shown in brackets.