[R] Interpreting the example given by Prof Frank Harrell in {Design} validate.cph

vikkiyft s067835 at alumni.cuhk.net
Mon Feb 21 05:46:53 CET 2011


Dear R-help,

I am having a problem with the interpretation of result from validate.cph in
the Design package.

My purpose is to fit a cox model and validate the Somer's Dxy. I used the
hypothetical data given in the help manual with modification to the cox
model fit. My research problem is very similar to this example.

This is the model without stratification: 
> library(Design) 
> f1 <- cph(S ~ age, x=TRUE, y=TRUE)
> coef(f)
       age 
0.0440095
> set.seed(1)
> validate(f1, B=10, dxy=T)
         index.orig      training          test      optimism
index.corrected  n
Dxy   -0.3376784858 -0.3287537760 -0.3376784858  0.0089247099    
-0.34660320 10
R2     0.0627722521  0.0636136044  0.0627722521  0.0008413523     
0.06193090 10
Slope  1.0000000000  1.0000000000  0.9896987441  0.0103012559     
0.98969874 10
D      0.0237993965  0.0239476118  0.0237993965  0.0001482153     
0.02365118 10
U     -0.0008208378 -0.0008141441  0.0007104737 -0.0015246178     
0.00070378 10
Q      0.0246202342  0.0247617559  0.0230889228  0.0016728331     
0.02294740 10
 
But if I fit a stratified cox model to the same data, the result becomes: 
> f2<- cph(S ~ age + strat(sex), x=TRUE, y=TRUE, surv=TRUE, time.inc=2) 
> coef(f)
       age 
0.04271953
> set.seed(1)
> validate(f2, dxy=TRUE, u=2, B=10)
         index.orig      training         test      optimism index.corrected 
n
Dxy    0.3514778665  0.3259011492 0.3044982080  0.0214029412    0.3300749254
10
R2     0.0622369082  0.0651967502 0.0622369082  0.0029598419    0.0592770663
10
Slope  1.0000000000  1.0000000000 0.9621830568  0.0378169432    0.9621830568
10
D      0.0257519780  0.0267239073 0.0257519780  0.0009719293    0.0247800486
10
U     -0.0009257142 -0.0009125388 0.0009102968 -0.0018228356    0.0008971213
10
Q      0.0266776922  0.0276364461 0.0248416812  0.0027947649    0.0238829273
10 

The coefficients are similar between the models, so I expect the results
would be somewhat similar, yet the two models give totally contrasting Dxy.
I reckon a negative Dxy value is normal in the sense that the survival time
and the prediction are concordant, but why does the result become discordant
when stratification is used? 
Is there something wrong or is there a sensible interpretation for this? 

This problem is very critical to me because the Design package is the only
one I can use for my purpose. Any advice is greatly appreciated. 


Best regards,
Vikki
-- 
View this message in context: http://r.789695.n4.nabble.com/Interpreting-the-example-given-by-Prof-Frank-Harrell-in-Design-validate-cph-tp3316820p3316820.html
Sent from the R help mailing list archive at Nabble.com.



More information about the R-help mailing list