Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Survival analysis: finding best cut-off values


From   "Nick Cox" <n.j.cox@durham.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: Survival analysis: finding best cut-off values
Date   Tue, 6 Mar 2007 21:29:04 -0000

The practice of dividing good continuous
variables into categories is retrograde. 
See Frank Harrell's book on "Regression modeling
strategies" from Springer in 2001. 

Nick 
n.j.cox@durham.ac.uk 

Diego Bellavia
 
> I am performing a survival analysis on a dataset with many 
> variables. Multivariate cox proportional-hazard models 
> defined the best predictors (around 7 out of 270 variables). 
> I would like to give the readers some cut-off values 
> they can use in the clinical practice, so I divided the most 
> significant predictors in tertiles, create the dummy variables 
> and run Cox models for each variable (groups of dummy vars). 
> Doing so, I obtain significant/unsignificant tertiles and 
> Kaplan-Meyer graphs 
> stratified by tertiles. Thsi way works pretty well. But what 
> if I would like to find only one cut-off per variable ? 
> I thought to use ROC curves to define the best diagnostic 
> cut-offs and see if they are good also for prognosis, but 
> unfortunately not all the best
> predictors are so good also to discriminate groups of patients. 
> In conclusion my question is: there is a way to obtain the 
> best prognostic cut-off value using Cox models ? 

*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index