Dear All:
I am fitting a cox model with a time dependent
covariate:
model.surro<-coxph(Surv(T1, T2,
survind)~(treat+S):strata(trial),data="">
where S is the time dependent covariate and
surro.time is the data set with the format needed to fit this time of models in
Splus. I am also fitting the model without the time dependent covariate and
using the original data set (surro):
model.original<-coxph(Surv(surv,
survind)~(treat):strata(trial),data="">
when I try to calculate the likelihood ratio test
to compare both models I get:
-2*(model.original$loglik[2]-model.surro$loglik[2])=
-25959
which is negative. I also tried to fit the model
without S using the surro.time data set
model.without.surro<-coxph(Surv(T1, T2,
survind)~(treat):strata(trial),data="">
and the likelihood ratio test becomes
then:
-2*(model.without.surro$loglik[2]-model.surro$loglik[2])=837.4
which is positive as one would expect. However the
estimates from model.without.surro and
model.original differ.
Does anybody could explain what is going on here
and how can I calculate the likelihood ratio statistics.
Kind Regards,