Should I use the mean concordance index from cross validation, or aggregate and build only one cox proportional hazards model?

Question

I have been conducting survival analysis on cancer samples using 10 fold cross validation. For each training fold, I\'m feature selecting and then during testing I\'m running a