I have responded to this particular misconception so often I begin to grow grumpy about it
(not the particular fault of YH). The cumulative hazard function from
fit <- coxph( some model)
sfit <- survfit(fit, newdata= set of covariate values)
gives the survival curve and cumulative hazard for that particular set of
covariate values.
There is nothing special about a "baseline hazard". Any cumulative hazard is for some
particular set of covariate values, including "all values =0" which is what most textbooks
refer to as the baseline. The survfit routine will by default return one centered at the
means of the predictor variables, simply because there is less round off error if one
stays near the center. Any baseline is as good as any other.
cum haz at covariate values "z" = cumulative haz for values "x" * exp(beta *
(z-x))
Note that for a random effect, the survfit routine uses 0 as the centering
value.
Terry Therneau
On 11/05/2013 05:00 AM, r-help-requ...@r-project.org wrote:
Hi Dr. Therneau,
Yes, -log(sfit$surv) gives me the cumulative hazard but not the baseline
cumulative hazard. I know that Nielsen and Klein have SAS Macros to get
such estimates by using EM approach. I'm wondering if I can obtain the
baseline hazard estimates from coxph() for gamma frailty model since I
think coxph() is a very powerful function. I feel there may be some way
that I don't know.
Thanks,
YH
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
