[R-sig-eco] CI for stratified Cox PH model

[Bob O'Hara]

Is there any reason you can#'t use confint()? e.g.

  test1 <- list(time=c(4,3,1,1,2,2,3),
                    status=c(1,1,1,0,1,1,0),
                    x=c(0,2,1,1,1,0,0),
                    sex=c(0,0,0,0,1,1,1))
confint(coxph(Surv(time, status) ~ x + strata(sex), test1))

Another trick is to use predict(), with new data at all of hte levels 
you are interested in:

pred1 <- list(x=c(0:2, 0:2),
                    sex=c(0,0,0,1,1,1))
predict(coxph(Surv(time, status) ~ x + strata(sex), test1), 
newdata=pred1, type="lp" , se.fit=TRUE)

(it doesn't provide CIs, but mean +/-1.96s.e. should work OK for the 
linear predictor scale)

Bob

On 05/06/18 18:47, Bertolo, Andrea wrote:
> Hi everyone,
> 
> I have a doubt about the way to calculate 95% CI for coefficients in
> the stratified Cox proportional hazard models and your help is welcomed
> .
> 
> Say that I have a variable of interest Imi and a stratifying variable
> UV in a interaction model (since the interaction between Imi and UV is
> of interest for me and the interaction model has a better fit than the
> no-interaction model:
> 
> library(survival)
> model1 <- coxph(Surv(start,stop, Status.time)
>                     ~ Imi + Imi:UV + strata(UV) + cluster(ID),
>                       weights = NB_Event, data=Data.unfold)
> 
> 
> 
> Whereas it is pretty straightforward to calculate the coefficients (and
> associated HR) for each combination of Imi and UV, I am not sure about
> how to calculate the associated CI (note that, of course, I got the CI
> for the estimate of "Imi:UV" from the output of model1).
> 
> Is it correct to calculate separately a model for each UV level and use
> the CI for the Imi variable to get the CI for the two levels of UV (see
> below) ?
> 
> # 2 models (one per UV level)
> data.Low <- subset(Data.unfold,UV=="low")
> model2.1 <- coxph(Surv(start,stop, Status.time)
>                     ~ Imi + cluster(ID),
>                       weights =
> NB_Event, data=data.Low)
> 
> data.High <- subset(Data.unfold,UV=="high")
> model2.2 <-
> coxph(Surv(start,stop, Status.time)
>                     ~ Imi +
> cluster(ID),
>                       weights = NB_Event, data=data.High)
> 
> Alternatively, is there a way to get the CI directly from the output of the stratified model ?
> Many thanks
> Andrea Bertolo
> 
> 
> 
> Université du Québec à Trois-Rivières
> 3351, bd des Forges
> C.P.500, Trois-Rivières (Québec) Canada
> G9A 5H7
> 
> _______________________________________________
> R-sig-ecology mailing list
> R-sig-ecology using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
> 


-- 
Bob O'Hara
Institutt for matematiske fag
NTNU
7491 Trondheim
Norway

Mobile: +49 1515 888 5440
Homepage: http://www.ntnu.edu/employees/bob.ohara
Journal of Negative Results - EEB: www.jnr-eeb.org