R help - Feb 2011 - Forced inclusion of varaibles in validate command as well as step

Hello all

 

I am a very new R user

 

I am used to using STATA

 

My problem:

 

I want to build a Cox model and validate this.

 

I have a large number of clinical relevant factors and feel the need to
reduce these. Meanwhile I have some clinical variables I deem sufficiently
important to force into the model regardless of AIC or p value.

 

This is my present log over commands

************

 

library(rms)

library(survival)

library(Hmisc)

 

data1 <- read.table("optimism.csv", header=T, sep=",")

attach(data1)

 

coxmodel4 <- coxph(formula=Surv(OS,mors) ~
iAJCC2+iAJCC3+iPS2+iPS3++alder_diag+gender+vol_GTV+iforb2+iforb3+hem_LNL+ser
o_thromb+LDH_UNL+ALAT_UNL+BASP_UNL+sero_bili+resection_perf+sero_WBC,
data=data1, x=TRUE, y=TRUE,method=c("efron"))

coxmodel.streg<-step(coxmodel4)

 

I would like to "lock" iAJCC2 iAJCC3 and iPS2 + iPS3 regardless, but I
cannot seem to get the step function to accept this.

 

Further

 

Once I have the model I would like to validate it with the validate command

 

I am presently using this****************

 

fit <- cph(formula=Surv(OS,mors) ~
iAJCC2+iAJCC3+iPS2+iPS3+alder_diag+gender+vol_GTV+iforb2+iforb3+hem_LNL+sero
_thromb+LDH_UNL+ALAT_UNL+BASP_UNL+sero_bili+resection_perf+sero_WBC,
data=data1, x=TRUE, y=TRUE)

fit

validate(fit, method="boot", B=40,bw=TRUE, rule="p",
type="residual",
sls=0.15, aics=0, pr=TRUE)

 

Due to my small data set 153 patients with 130 events I have chosen to lift
the p limit from 5% to 15% as suggested by Steyerberg.

************************************************************************

 

I would appreciate any help with the lock term (also if it cannot be done) 

As I mentioned I am a bit of a rookie, and not too experienced as a
programmer (I am a MD after all)

However I am quite impressed with R so far since I have been trying to get
this far in STATA for a few weeks.

 

Sincerely

 

Jon Kroll Bjerregaard, MD. Dep of Oncology Odense University Hospital

 


	[[alternative HTML version deleted]]

Jon,

Version 3.3-0 of rms will be released within 2-3 days.  It has a new option
"force" for fastbw, validate, calibrate.  force is an integer vector
of the
parameter numbers to force into every model.  It is meant to work with
type='individual' and its performance with type='residual' needs
to be
studied (I doubt if it works as you want).  Suppose you have a Cox model
(which does not have an intercept to include in the sequential numbering for
force) like this:

f <- cph(S ~ sex + pol(age,2) + rcs(height,4))

and you want to force age into every model.  You would specify force=2:3. 
Typing coef(f) will show you the sequential parameters in the model. 
Someday I'll extend this to force='age' but that's not in the
current
version.  Note that fastbw always pools age and age^2 effects when judging
significance.

As always use stepwise methods at your own risk.  They are dangerous to your
health.

If you are using Linux I can send the new version by e-mail.

Frank

-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
-- 
View this message in context:
http://r.789695.n4.nabble.com/Forced-inclusion-of-varaibles-in-validate-command-as-well-as-step-tp3324901p3325922.html
Sent from the R help mailing list archive at Nabble.com.