R help - Jul 2004 - Evaluating the Yield of Medical Tests

Hello,

I'm a biostatistician in Toronto. I would like to know if there is
anything in survival analysis developed in R for the method "Evaluating
the Yield of Medical Test" (JAMA. May 14,1982--Vol 247, No.18  Frank E.
Harrell, Jr,PhD; Robert M. Califf, MD; David B. Pryor, MD;Kerry L.Lee,
PhD; Robert A. Rosait,MD.)

Hope to hear from you and thanks

Lisa Wang, MSc
Project Organiser & Researcher
Cancer Informatics,
Ontario Cancer Institute/Princess Margaret Hospital, University Health
Network;
Email: lisawang at uhnres.utoronto.ca
Phone: 416 946 4501 ext. 5201
Fax: 416 946 4619

On Mon, 2004-07-19 at 14:37, Lisa Wang wrote:
> Hello,
> 
> I'm a biostatistician in Toronto. I would like to know if there is
> anything in survival analysis developed in R for the method
"Evaluating
> the Yield of Medical Test" (JAMA. May 14,1982--Vol 247, No.18  Frank
E.
> Harrell, Jr,PhD; Robert M. Califf, MD; David B. Pryor, MD;Kerry L.Lee,
> PhD; Robert A. Rosait,MD.)
> 
> Hope to hear from you and thanks

I do not have access to the full text of Frank's article, however I read
the brief abstract on Medline and cross-referenced the citation of the
article with content in Frank's book (Regression Modeling Strategies -
http://biostat.mc.vanderbilt.edu/twiki/bin/view/Main/RmS). 

Thus, I am going to take a (hopefully well considered) guess that what
you are looking for will be in the combination of the Hmisc and Design
packages, which Frank has kindly made available for R. These are
available for installation from CRAN.

More information on Hmisc and Design is available at:

http://biostat.mc.vanderbilt.edu/twiki/bin/view/Main/RS

which is the link to Frank's site at Vanderbilt. Looking at the authors'
names, Frank was at Duke when the cited article was written.

I suspect that Frank will reply (RSN) with the acceptance or rejection
of my guess, however...  ;-)

HTH,

Marc Schwartz

Marc Schwartz <MSchwartz at MedAnalytics.com> is correct.  The rcorr.cens 
function in the Hmisc package was written to do what is described in 
that article.  It also computes standard errors, and there is a function 
rcorrp.cens that in paired predictions can assess whether one predictor 
is "more concordant" with survival time than the other.  In the binary
Y
case rcorr.cens gives you the ordinary Somers Dxy and standard error, 
translatable to ROC area.

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