Recurrence-free survival analysis in the full cohort of HCC patients (N = 44) using gene expression in nontumoral samples. Expression of 25,073 genes was used as a variable in Cox proportional hazard model. The power of recurrence-free survival prediction for the top-ranked (based on Cox score) genes was assessed using the Nearest Template Prediction (NTP) algorithm and cross-validated with a leave-one-out procedure. (A) Log-rank P-value and error rate of prediction for up to 150 top-ranked genes. (B) Recurrence-free survival curves based of the prediction made using 14 top-ranked genes. (C) Heat map of the 14 gene-based predictors. Red and blue indicate higher or lower than median gene expression, respectively.