Fix error in plotting qgcomp.cox.boot objects when using variable transformations (e.g. splines) on covariates

qgcomp_boot_transform_plot_workaround.R

# scoping issue when using in-formula transformations
# example: spline basis functions
# problem: when including functions like `rms(x)` in a formula to put a spline
#   function on 'x', plotting a qgcomp.cox.boot object yields an error that 'x' can't be found
# solution: explicitly create the spline basis functions as variables in a data.frame. Example follows.

library(qgcomp)
library(rms)
library(survival)

# testing data
set.seed(50)
N=200
dat <- data.frame(time=(tmg <- pmin(.1,rweibull(N, 10, 0.1))), 
                  d=1.0*(tmg<0.1), x1=runif(N), x2=runif(N), z=runif(N))
expnms=paste0("x", 1:2)

# restricted cubic spline for Z - using the spline function in the formula
# using splines in the formula works for model fitting, but plotting it will fail
f_nlin1 = survival::Surv(time, d)~x1 + x2 + rcs(z, quantile(dat$z, c(.1, .5, .9), include.lowest=TRUE))
set.seed(12312)
(fit_nlin1 <- qgcomp.cox.boot(f_nlin1, expnms = expnms, data = dat, B=10, MCsize=200))
# plot(fit_nlin1) # error: Error in rcs(z) : object 'z' not found

# workaround: explicitly bring in the spline variables
zspl = unclass(rcs(dat$z, quantile(dat$z, c(.1, .5, .9), include.lowest=TRUE)))
colnames(zspl) <- paste0("zspl", 1:ncol(zspl)) # check the number of columns to be sure you have the formula correct
dat2 = cbind(dat, as.data.frame(zspl))

f_nlin2 = survival::Surv(time, d)~x1 + x2 + zspl1 + zspl2

# confirm that results are the same in a standard Cox model first
coxph(f_nlin1, data = dat2)
coxph(f_nlin2, data = dat2)

set.seed(12312)
(fit_nlin2 <- qgcomp.cox.boot(f_nlin2, expnms = expnms, data = dat2, B=10, MCsize=200))
plot(fit_lin) # this works