Multivariable proportional hazards regression using forward variable selection was used to select the best model to predict overall survival.

C = Cytoplasm; N = Nucleus. All biomarkers that were significant predictors of overall survival in univariate models from Tables 2, 3, 4 were categorized in two groups, high and low, using the median as a cutoff point. Gleason score was categorized into two groups: ≥8 and <8 and the two dichotomous variables were combined to generate four groups: Gleason ≥8 and Biomarker High, Gleason ≥8 and Biomarker Low, Gleason <8 and Biomarker High, and Gleason <8 and Biomarker Low. ‘Biomarker High’ indicates biomarker values greater than the median of the (continuous) biomarker.