R help - Aug 2005 - compare c-index of two logistic models using rcorrp.senc() of the Hmisc library

Dear R-help,

Would it be appropriate to do the following to
calculate a p-value for the difference between c-ind
of x1 and c-inx of x2 using the output from
rcorrp.senc()
> r<-rcorrp.senc(x1,x1,y)
> pValue<-1-pnorm((r[11]-r[12])/(r[2]/r[5])*1.96) 
Osman O. Al-Radi, MD, MSc, FRCSC
Chief Resident, Cardiac Surgery
University of Toronto, Canada

Osman Al-Radi wrote:
> Dear R-help,
> 
> Would it be appropriate to do the following to
> calculate a p-value for the difference between c-ind
> of x1 and c-inx of x2 using the output from
> rcorrp.senc()
> 
> 
>>r<-rcorrp.senc(x1,x1,y)
> 
> 
>>pValue<-1-pnorm((r[11]-r[12])/(r[2]/r[5])*1.96) 
> 
> 
> Osman O. Al-Radi, MD, MSc, FRCSC
> Chief Resident, Cardiac Surgery
> University of Toronto, Canada
Osman,

Because tests for differences in two ROC areas are not very powerful, 
rcorrp.cens changes the hypothesis to "are predictions from one method 
more concordant with the outcome than predictions from the other method, 
within paired predictions".  You can't get a difference in ROC areas 
from the U-statistic computed by rcorrp.cens.

Frank



-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University