R help - Apr 2010 - C-index and Cox model

Dear all R users,

I am building a Cox PH model on a small dataset. I am wondering how to
measure the predictive power of my cox model? Normally the ROC curve or Gini
value are used in logistic regression model. Is there any similar
measurement suitable for Cox model?

Also if I use C-index statistic to measure the predictive power, is it a
time-dependent value (i.e. do I need to calculate it for each time period?)
or we can calculate it as a single value for the whole model ignoring the
time?

Thank you so much.

Bessy
-- 
View this message in context:
http://n4.nabble.com/C-index-and-Cox-model-tp1774155p1774155.html
Sent from the R help mailing list archive at Nabble.com.

Bessy wrote:
> Dear all R users,
> 
> I am building a Cox PH model on a small dataset. I am wondering how to
> measure the predictive power of my cox model? Normally the ROC curve or
Gini
> value are used in logistic regression model. Is there any similar
> measurement suitable for Cox model?
> 
> Also if I use C-index statistic to measure the predictive power, is it a
> time-dependent value (i.e. do I need to calculate it for each time period?)
> or we can calculate it as a single value for the whole model ignoring the
> time?
> 
> Thank you so much.
> 
> Bessy
install.packages('Hmisc')
?rcorr.cens

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

Bessy wrote:
> Dear all R users,
> 
> I am building a Cox PH model on a small dataset. I am wondering how to
> measure the predictive power of my cox model? Normally the ROC curve
or Gini
> value are used in logistic regression model. Is there any similar
> measurement suitable for Cox model?
> 
> Also if I use C-index statistic to measure the predictive power, is it
a
> time-dependent value (i.e. do I need to calculate it for each time
period?)
> or we can calculate it as a single value for the whole model ignoring
the
> time?
> 
> Thank you so much.
> 
> Bessy
Frank H added:
> install.packages('Hmisc')
> ?rcorr.cens
You can also use the survConcordance function, which is part of the
survival package.  (Why two functions?  I wrote this one before I was
aware of Frank's.)

Terry T