[R] percent inclusion in cph

Frank E Harrell Jr f.harrell at vanderbilt.edu
Tue Nov 13 13:49:13 CET 2007


sushi4u wrote:
> Dear R-help,
> 
> I have to calculate the percent inclusion of each variable in a bootstrap validation of a cox proportional hazards model(described in Sauerbrei and Schumacher, Stat Med 11:1093, 1992).
> 

This approach is not recommended.   Collinearities can ruin the result, 
and the selection frequencies will just replay what the original 
P-values tell you.  Further, there is no reason to do variable selection 
in your case, and selection will distort all statistical inferences. 
Just pre-specify a model, fit it, and stop.

Frank

> First I need to get a bootstrap sample from my dataset, which I did with the sample function. Then I tried to run a cph model and looked which covariates are significant. This I would repeat 200 times and at the end calculate how many percent the covariates were included. 
> 
> This is what I entered:
> boot1 <- sample(Dataset, 300, replace=T)
> cph1 <- cph(Surv(months,status) ~ cov1 + cov2 + cov3 + cov4, data=boot1)
> 
> Unfortunately, I get exactly the same results (coefficient, SE, p-value) then when I would fit a Cox model without drawing a bootstrap sample before.
> 
> How do I do it right? Or is there another way to calculate the percentage?
> 
> Sorry for my bad English. 
> 
> Thanks.
> 
> Dott. Mario Rossi
> University of Foggia, Italy
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
> 


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



More information about the R-help mailing list