Conditional ROC curve analysis?

[Giulia Ferrannini]

Hi!
I was asked by a reviewer to produce a ROC curve for my data, testing the performance of a continuous classifier in predicting a disease. I have matched data with 805 cases and 805 controls matched for sex, age and living area.
Is there a way to produce a conditional ROC curve in STATA?

Thank you!

[Rich Goldstein]

I am not all sure I understand what you are asking but a general procedure is to estimate your model, obtain predicted probabilities and obtain the ROC curve based on that; if you are using -logistic- or -logit-, a direct way to do this is through the post-estimate -lroc- command but you can also use the various ROC commands; see

h lroc
h roc

[Joseph Coveney]

. . . testing the performance of a continuous classifier in predicting a disease. I have matched data with 805 cases and 805 controls matched for sex, age and living area.
Is there a way to produce a conditional ROC curve in STATA?

A method of estimating the AUC of the ROC in exactly your circumstances is given in this paper.

It looks as if it wouldn't be too difficult to implement the method in Stata by hand, but I don't know of any canned procedure (user-written command) for doing it.