Fig. 4

Predictive ability of FN1 and VTN ELISA data concerning recurrence-free survival. A AUCs for the Hamburg cohort based on FN1 and VTN levels determined using MS (n = 78). Shown are median AUCs of 50-fold cross-validation (grey) of our model using FN1 and VTN (protein) plus PSA plus Gleason score (Bx; orange) versus PSA alone (light blue) and PSA plus Bx (dark blue). B Boxplots of our model for the Hamburg cohort with protein levels determined by MS (PSA/Bx + protein) compared to PSA alone and PSA plus Bx. Each box indicates min, 25%-quantile, median (black line), 75%-quantile, max, mean (black cross), and std (gray bar). Statistics: paired t-test, corrected for multiple testing with the Benjamini and Hochberg method [57]. C Prediction of 5-year biochemical recurrence-free survival for the validation ProCOC (n = 263) with our model based on the protein signature determined by ELISA (FN1, VTN, PSA, and Bx, orange) versus PSA alone (light blue), PSA plus Bx (dark blue), and NCCN alone (green). Statistics: DeLong test for ROCs. D Kaplan–Meier plots for recurrence-free survival of ProCOC (n = 263) stratified based on PSA (light blue), PSA plus Bx (dark blue), NCCN alone (green), or our score (orange lines). Statistics: Likelihood ratio test