ROC-Curve p value

[Ankit Bhardwaj]

I performed ROC curve analysis on a diagnostic test. While my results displayed the AUC, 95% CI, and standard error, they did not show the p-value. Can you advise me on this?

[Joseph Coveney]

. . . they did not show the p-value. Can you advise me on this?

You don't say what command you used, but from your description, I assume that it's roctab.

You could try something like the following. I use the auto dataset for illustration and test the area under the receiver operating characteristic curve for a car's price to discriminate its manufacturing origin different from chance level.

version 18.0

clear *

quietly sysuse auto

*
* Begin here
*
roctab foreign price

display in smcl as text "{it:P} = " ///
    as result %04.2f  2 * normal(-abs(r(area) - 0.5) / r(se))

// or (quick 'n' dirty)
ranksum price, by(foreign)

exit

I suppose that you could make an argument that it should be a one-sided test of a diagnostic test's ability to discriminate disease at better-than-chance level. I think that the rule of thumb typically calls for (much) better than an ROC AUC of 0.70 for a diagnostic test of any commercial interest.

[Ankit Bhardwaj]

I have noticed that in SPSS, the AUROC, 95% CI, and p-value are all presented in one table. Yes I have used 'roctab' command.
However, I was wondering if there is a command that can display these values in the same table in STATA as the AUROC, 95% CI, and standard error.
Thank you for the help