R help - Nov 2009 - ImproveProb Function in Design Package

Hello

Thanks to Frank Harrell for the great Design package. I noted from Pencina's
Article (Statistics in Medicine Vol 27, pp157-172) that the result obtained for
the Net Reclassification Index depends upon categorical cut-offs for risk (eg in
the paper <6%, 6-20% and >20%). I wondered how the improveprob function
arrives at its cut-offs, and whether there was any function to change the
default cut-offs used.

Thanks for any help you can offer.

Yours sincerely

David McAllister


*****************************************************************
The information contained in this message may be confide...{{dropped:8}}

McAllister, David wrote:
> Hello
> 
> Thanks to Frank Harrell for the great Design package. I noted from
Pencina's Article (Statistics in Medicine Vol 27, pp157-172) that the result
obtained for the Net Reclassification Index depends upon categorical cut-offs
for risk (eg in the paper <6%, 6-20% and >20%). I wondered how the
improveprob function arrives at its cut-offs, and whether there was any function
to change the default cut-offs used.
> 
> Thanks for any help you can offer.
> 
> Yours sincerely
> 
> David McAllister
> 
Note that improveProb is in the Hmisc package.

I believe that almost any statistical method that requires arbitrary 
binning of data is problematic.  improveProb implements continuous 
measures as emphasized by Pencina et al.  Their NRI does not use cutoffs.

Frank

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