Figure 2

New Classification and Regression Tree to Predict 28-Day Mortality. The classification tree consists of four biomarker-based decision rules with five terminal daughter nodes. The tree incorporates three of five PERSEVERE biomarkers: interleukin-8 (IL8), heat shock protein 70 kDa 1B (HSPA1B), and chemokine ligand 3 (CCL3). Each node denotes the number of subjects in the node, the serum concentration of a given biomarker determining the branch point (pg/mL), and both the total number and accompanying rate of survivors and non-survivors. Terminal nodes 1, 2, and 4 are low risk nodes. Terminal node 3 is higher risk, with terminal node 5 being the highest risk node. The AUROC for this model was 0.87.