Explained variation for relative survival models

[aolszewski]

Hi, I am interested in studying a relative survival model (using the user-written command -stpm2- available from SSC).
After fitting a model, I would like to get some characteristics conceptually related to concordance and/or explained variation known from standard proportional hazard models. I was wondering if anyone is aware of such measures. I have not found any information about using some equivalent of the c-statistic or Somers' D rank statistic in this setting. For explained variation, it is technically possible to obtain the variants of R2 statistics (such as those generated by the user-written -str2d- or -str2ph- commands), but I wonder if they are interpretable and reliable, since in a relative survival model there is no "observed outcome" that is recorded in the dataset outside of the knowledge of the expected mortality rate. Perhaps such measures just do not make intrinsic sense but I wonder if some equivalent could still be used to describe how much of variation in outcome is explained by the variables in the model.
I would be grateful for any insight,
Adam Olszewski

[Veronica Leoni]

Good Afternoon Adam,

I see that this a pretty old post. Did you find an answer to your question?

I have run survival analysis and I used the str2ph package too. My adj.R^2 is around 0.3240 and I really cannot find any benchmark to interpret the goodness of fit.

Thank you

Regards

VL