Compare (with p-values) Harrell C's statistics after the "estat concordance" command in STATA

[Panagiotis Antiochos]

Dear all,

This is a recurrent question. After a Cox prediction model, we have the post estimation option of estat concordance that calculates the Harrell C for the model. After that, we add a new predictor to the previous model and re-run estat concordance which calculates the new Harrell C for the new model after the inclusion of the new predictor.

Is there a way to compare those 2 Harrell 's C statistics, calculated after the estat concordance commands?

I am aware of the excellent paper of Roger Newson, (ref.: The Stata Journal (2010) 10, Number 3, pp. 339–358. Comparing the predictive powers of survival models using Harrell’s C or Somers’ D) who perfectly explains the problem of doing such a thing and proposes the somersd package which i have been using with success.

However, apart from the somersd, is there a simpler (less robust i guess) method to compare the exact numbers that are calculated after the estat concordance command? It seems like in SAS for example there is a simple way to do that with no need to split the dataset into a training and validation set as with the somersd package in STATA. Is there anything similar in STATA?

Thank you very much for any help.

*************************************************

Panagiotis Antiochos, M.D.

Post-doctoral Research Fellow

Cardiac Magnetic Resonance Laboratory

Brigham and Women's Hospital

75 Francis Street, Boston, MA 02115

Phone: 617-938-7156

Email: [email protected]

*************************************************

[Rich Goldstein]

I am out of town and don't feel like searching at this point; however, I note that I have previously published in Statalist, citations for articles that show that the Wald p-value on the new variable is the p-value for the change in c-statistics; the articles appeared in the same issue of Statistics in Medicine, and one is by Pepe and colleagues and the other by Begg and colleagues (if I remember correctly, Begg is not the first author of his article); thus, you should be able to find these citations by searching in the forum - but maybe I have misunderstood your question???