Figure 1. Classification tree resulting from the Classification and Regression Tree (CART) analysis including the nine prognostically most relevant proteins in our study cohort (i.e. EGFR, AURKA, VEGFA, PTGS2, SLC2A1, KCNQ1, CEA, MMP9 and HIF1α).

The optimal prediction of three-year survival was obtained by a classification tree including AURKA, PTGS2 and MMP9 expression. Class A contained patients with low AURKA expression, class B patients with high AURKA, low PTGS2 and low MMP9 expression, class C patients with high AURKA, low PTGS2 and high MMP9 expression and class D patients with high AURKA and high PTGS2 expression. Based on the HRRs of the individual classes, classes A and B were grouped and classes C and D were grouped, resulting in classes I and II, respectively. Excluded from analysis were patients of whom no data was available on survival status or three-year survival and with unknown expression of AURKA, PTGS2 and MMP9. HRR: hazard rate ratio, p: p-value as determined by Cox regression analysis.