Re: st: using roctab when lower values of 'classvar' indicate higherrisk

[Gary Longton <glongton@fhcrc.org>]

Morrison Hodges wrote:

> The manual says that when using 'roctab', higher values of 'classvar' must
> indicate higher risk. I have data using a cardiac variable called ejection
> fraction (EF) as 'classvar', which varies from 0.00 to 1.00, with lower
> values indicating higher risk. The 'refvar' is 'status', with 0=alive and
> 1=dead. The lower the EF the more likely the patient is dead.
>
>  roctab status ef, graph
>
> produces a curve below the 45 degree line; I want the curve to be above the
> 45 degree line. What is a legitimate way to make the curve be above the 45
> degree line? Exchange 'refvar' values? Subtract 'classvar' values from 1?
> Something else? Thanks. 
It is only by *convention* that, in ROC methods, larger values of the test 
result are more indicative of disease (or, in this case, death).  You can 
legitimately satisfy this convention in this situation by redefining your test 
result as Y = -EF.

Alternatively exchanging the <refvar> values will give you a different and 
likely unwanted result: a plot of specificity vs. 1-sensitivity in the situation 
where it is believed that lower test values are more indicative of disease or 
other outcome?

- Gary

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/