Hello,

I am performing cross-validation on a number of survival models fitted with the coxph function and would like to use the likelihood ratio test on the left-out cases as the criteria for comparison of PCA, SPCA, ridge, lasso, etc.  Is there an easy way to do this?  I think predict.coxph might be the answer but stumped on the implementation.

For example, if I had

    fit <- coxph(Surv(time, status) ~ age + ph.ecog + strata(inst), lung[1:200,]) 
    > fit
    Call:
    coxph(formula = Surv(time, status) ~ age + ph.ecog + strata(inst), 
        data = lung[1:200, ])
    
    
              coef exp(coef) se(coef)    z       p
    age     0.0125      1.01   0.0106 1.18 0.24000
    ph.ecog 0.5371      1.71   0.1423 3.78 0.00016
    
    Likelihood ratio test=19.7  on 2 df, p=5.17e-05  n= 198, number of events= 156 
       (2 observations deleted due to missingness)




How to test likelihood of this model on lung[201:228,]   ?

Thanks for any advice,
	[[alternative HTML version deleted]]