It looks like a good paper. An important reference from Brad Efron was omitted; I think he was one of the first to develop a fix for the dependence on the censoring distribution. I wish I had the reference handy.
While the c-index is a good measure of predictive discrimination for a single model, we need to quit using it to compare models. It’s not sensitive enough for that purpose. I would prefer to emphasize pseudo adjusted R-square for all purposes.
2 Likes