ROC commands with multiple variables

[geraldine tran]

Hello,

I would like to calculate the AUC using one of the roc commands. My reference/outcome variable is diseases status (0/1), and the classifier variables are length (continuous variable), and hormone status (0/1).
Which roc command should I use? (i.e. roctab, roccomp, rocreg?)
Is there special syntax for categorical classifiers like hormone status (0/1)? For example, when using the logit command we place an "i." infront of categorical variables.

Thank you!
Geraldine

[Marcos Almeida]

You may type - help rocreg - and take a look at the example 1, where the text shows comparisons between roctab and rocreg.

[geraldine tran]

Hi Marcos, I've looked at both roctab and rocreg examples in help. It looks like they only use reference and one classifying variable (i.e. "roctab disease, rating" which would be similar to my reallife example, "roctab disease, length."

If I want to add more classifying variables, do I need to use logistic/logit, and the lroc?

The problem here is that lroc does not give confidence intervals.

Ideally, I could have AUC (of model including more than 1 classifying variable of which some are continuous and some are categorical), 95% CI, and also a p-value showing difference from AUC of 0.5.

I am pretty new at STATA, so any help is appreciated.

Thank you,
Geraldine

[Rich Goldstein]

what you want to do is estimate your model using logistic, then get the predicted probabilities and use that as the "classvar" in your roctab command; see

help logistic postestimation

and scroll down to "predict" and click on it