I have seen that Frank Harrell's rms package does not offer a CI for Somers Dxy (and subsequently the c-statistic/AUROC).
I am trying to look at a method with good performance and a plan to implement this properly, probably through R.
I know an optimism corrected AUC already involves a bootstrap, so would the best way to be to have a script run the optimism corrected AUC from a model, save it, and repeat the bootstrapping/correction of AUC with a loop of say B=500 times? Then just take the 2.5 and 97.5 quantiles?
What might be some issues with this? I don't think I have enough bootstrap/resampling expertise to foresee what problems might ensue.
1) is the above method reasonable? 2) what would be another method? 3) what issues or limitations are pertinent to numbers 1 and 2?