Dear all,
I didn't get any suggestion for my querry concerning the NLTM package so I
am re-posting it hoping that someone can give me any clue. I apologize for doing
so. Please find attached a copy of my previous email:
Dear R users,
I have a question concerning the nltm package. Before posting to the R list I
first contacted the author of the package twice but no succes. May be I've
got the wrong email! My question is about the object "surv" given in
the package "nltm". As explained, the object "surv"
represents the MLE estimates of the baseline survival function evaluated at the
estimated profile likelihood parameters. When I extracted this baseline
survival function using "surv" I was surprised by the fact that
almost all its values are close to 1. I thought maybe because It is a simulated
example. So I used the same example provided in the package using the PH family
in "nltm.model" and then I compared the estimated baseline survival
with the one provided by the usual coxph package. As a result, the parameters
theta are similar as expected BUT the estimated baseline survivals are
completely different. Am missing something here? I would be really very grateful
if anyone can let me know where I am wrong. The examples are as follows:
# fit a Cox PH Model using nltm package
data(melanoma)
fit.nltm <- nltm(Surv(time,status) ~ size + age, data=melanoma,
nlt.model="PH")
surv.nltm <- fit.nltm$surv
# fit a Cox PH model using coxph package
data(melanoma)
fit.coxph <- coxph(Surv(time,status) ~ size + age, data=melanoma)
surv.coxph <- summary(survfit(fit.coxph))[[1]]
The finite-dimension parameters theta in fit.nltm and fit.coxph are similar but
as you can see "surv.nltm" and "surv.coxph" are completey
different. Furthermore, "fit.nltm" have all values close to 1.
All the bests
Abderrahim
[[alternative HTML version deleted]]
