Comparisons of ROC curves with covariates

[Piotr Lewczuk]

Good afternoon,
I'm comparing the areas under the ROC curves of two diagnostic tests (x1 and x2; both give continuous values (concentrations) as the results) on a cohort of 100 subjects divided into Controls (Status=0) and Positives (Status=1). None of the two tests is "golden standard".
The first, preliminary analysis (roccomp), suggests that the AUC of x2 is highly significantly larger, which makes sense to me. Now I'd like to fine-tune the analyses by adjusting for the age and the sex of the subjects (Controls are younger and have more females, which, theoretically, may influence how accurate the two tests are).
Would somebody be so kind to suggest, how I should do it? I guess some kind of rocreg would be of use?
Thank you very much in advance!
Piotr Lewczuk

[Carlo Lazzaro]

Piotr:
does the example #4 under -rocreg- entry in Stata .pdf manual meet you needs?

[Piotr Lewczuk]

Dear Carlo,
Indeed! I think this is what I was looking for. Just realized I should not specify "nobootstrap" to get the ROC curves comparisons.
Thank you very much!
Regards,
Piotr

[Piotr Lewczuk]

Good afternoon, again.
Carlo, once again - thanks!
After some additional search, I came across this paper:

http://www.stata-journal.com/sjpdf.h...iclenum=st0023

and I'm wondering if someone could explain me the pros and cons of the two approaches. For me, the way suggested by Dr. Cleves (in the paper above) is very straightforward compared to rocreg in the manual, but perhaps I'm missing something?
Thank you in advance.
Regards,
P. Lewczuk