[R] Problem with survfit when model has interaction with strata

[Sebastian Pölsterl]

Dear all,

I encountered a problem when running survfit using a coxph model that 
contains an interaction with a strata variable (see the attached 
example). I hope this is the right place to report problems.

 > source("~/survfit-example.R")
Error in scale.default(x2, center = xcenter, scale = FALSE) :
   length of 'center' must equal the number of columns of 'x'

 > traceback()
9: stop("length of 'center' must equal the number of columns of 'x'")
8: scale.default(x2, center = xcenter, scale = FALSE)
7: scale(x2, center = xcenter, scale = FALSE)
6: survfit.coxph(fit, newdata = newd)
5: survfit(fit, newdata = newd) at survfit-example.R#10
4: eval(expr, envir, enclos)
3: eval(ei, envir)
2: withVisible(eval(ei, envir))
1: source("~/survfit-example.R")

 > sessionInfo()
R version 3.2.1 (2015-06-18)
Platform: x86_64-unknown-linux-gnu (64-bit)
Running under: Fedora 22 (Twenty Two)

locale:
[1] C

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base

other attached packages:
[1] survival_2.38-3 foreign_0.8-65

loaded via a namespace (and not attached):
[1] tools_3.2.1   splines_3.2.1


I believe the problem is occurring on line 263 of survfit.coxph.R, where 
a model.matrix from newdata is created. The values passed to scale look 
like:

 > x2
   RXplacebo RXtreatment:strata(SEX)male RXplacebo:strata(SEX)male
1         0                           1                         0
2         0                           0                         0

 > xcenter
                 RXplacebo strata(SEX)male:RXplacebo
                 0.5000000                 0.2380952


Clearly, model.matrix is adding an additional (redundant) column that 
was omitted when fitting the original model. Since interactions with 
strata variables are a special cases in coxph, I believe similar care 
should be taken when creating the model matrix for newdata.

Best regards,
Sebastian Pölsterl

-------------- next part --------------
library(foreign)
library(survival)
data <- data.frame(read.spss("http://web1.sph.emory.edu/dkleinb/allDatasets/surv2datasets/anderson.sav"))

S <- with(data, Surv(SURVT, I(STATUS == "relapse")))
fit <- coxph(S ~ strata(SEX)*RX, data=data)
newd <- data.frame(RX=c("treatment", "treatment"), SEX=c("male", "female"))
sfit <- survfit(fit, newdata=newd)