st: Re: Comparing AuROC for same predictor but different outcomes in single survey-weighted sample

[Anna DeBenedetti Rubinsky <adeben@gmail.com>]

Thank you, Dr. Dwamena, but I believe roccomp and rocreg only
facilitate the (more common) comparison of two AuROCs for different
predictors (e.g. screening instruments) of the same binary outcome,
and not the comparison of two AuROCs for the same predictor (e.g.
screening instrument) but different variants of the binary outcome, as
I wish to do.

Anna


On Thu, Apr 5, 2012 at 1:25 PM, Anna DeBenedetti Rubinsky
<adeben@gmail.com> wrote:
> We have data from a stratified random sample, with final weights to account
> for the sampling design and nonresponse.
> We would like to compare AuROCs for a single screening instrument based on
> different variants of the outcome (because diagnostic criteria for the
> disease of interest are changing).
>
> We have been able to estimate separate AuROCs and CIs using Roger Newson's
> somersd command, factoring in the sampling design (pweight=final weight;
> cluster=primary sampling unit)
> per the suggestion on
> http://www.stata.com/statalist/archive/2009-01/msg00689.html.
>
> However, I am not sure how to compare these AuROCs (from the same sample)
> and estimate the CI and p-value for the difference.
>
> The listserve has addressed how to compare AuROCs for two predictors of the
> same outcome (http://www.stata.com/statalist/archive/2012-01/msg00477.html),
> but I have not been able to find how to compare AuROCs of the same predictor
> for two variations of the outcome, especially given that the data is from a
> survey with a complex sampling design.
>
> Advice would be much appreciated.
> Thank you,
> Anna

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/