Stratified cox regression in R using package namespace

Question

I'm doing stratified cox regression in R with the survival package:

cox <- coxph(response~strata(x), data=data)

This works fine, but when using the sampling package the strata function from the survival package is hidden and cannot be used directly.

However, calling it with the explicit namespace survival::strata does not create a stratified analysis:

library(survival)

# sample data
mort28day <- c(1,0,1,0,0,0,0,1,0,0,0,1,1,0,1,0,0,1,0,1,1,1,1,0,1,1,1,0,0,1)
daysurv <- c(4,29,24,29,29,29,29,19,29,29,29,3,9,29,15,29,29,11,29,5,13,20,22,29,16,21,9,29,29,15)
levo <- c(0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,0,0)
acbi30 <- data.frame(mort28day, daysurv, levo)

# create survival object
test.Surv <- Surv(acbi30$daysurv, acbi30$mort28day)

test.surv_fit_1 <- survfit(coxph(test.Surv~strata(levo), data=acbi30))
test.surv_fit_2 <- survfit(coxph(test.Surv~survival::strata(levo), data=acbi30))

So,

test.surv_fit_1$strata
# levo=0 levo=1 
#    12      3 

test.surv_fit_2$strata
# NULL

Why is this the case and how can I access this function when it is hidden by sampling::strata (apart from detaching the sampling package ;-)

I'm using R version 2.15.2

Thanks, DeltaKappa

Update:

As a workaround to using survival::strata together with the sampling package I did

library(survival)
library(sampling, pos="package:base")

which loads the sampling package with a low search priority right above the base package. Then I can do

coxph(test.Surv~strata(levo), data=acbi30)

while calling sampling::strata explicitly.

`coxph` looks for certain functions, including `strata`, in the supplied formula to identify as "special" terms. It calls `terms.formula` (via `terms`) with `special = c("strata", "cluster", "tt")`, and it looks like this fails to identify `survival::strata` as a special term. That missing special term then propagates throughout the rest of the calculations and leaves you with no strata. `terms.formula` uses internal code, and I don't have the nerve (or perhaps the know-how) for source-surfing right now. — BenBarnes, Jul 01 '13 at 19:01
I guessed, it would be something in that direction, since assigning `st <- strata(levo)` and using it in the formula `test.Surv~st` also didn't work... thanks for claryfing @BenBarnes — DeltaKappa, Jul 02 '13 at 08:10

score 0 · Answer 1 · answered Sep 19 '21 at 19:53

The proximate reason for failure was correctly described in the comments , but there is an incredibly simple fix that was not offered. As you pointed out the strata function is not exposed but the use of the triple-colon operator (:::) lets you get around that aspect:

test.surv_fit_1 <- survfit(coxph(test.Surv~strata(levo), data=acbi30))
test.surv_fit_2 <- survfit(coxph(test.Surv~survival:::strata(levo), data=acbi30))
test.surv_fit_1
#--------
Call: survfit(formula = coxph(test.Surv ~ strata(levo), data = acbi30))

        n events median 0.95LCL 0.95UCL
levo=0 25     13     24      19      NA
levo=1  5      2     NA      13      NA

test.surv_fit_2
#-------------
Call: survfit(formula = coxph(test.Surv ~ survival:::strata(levo), 
    data = acbi30))

      n events median 0.95LCL 0.95UCL
[1,] 30     15     24      19      NA

Whether this was the correct solution in the context of using the sampling package, I'm not able to render an opinion, but it would definitely circumvent the masking of the survival:::strata function.

Stratified cox regression in R using package namespace

1 Answers1