tests/test-interval-timeSplitter.R

In Greg: Regression Helper Functions

library(Greg)
library(dplyr)
data("melanoma", package = "boot")
melanoma <- mutate(melanoma,
                   status = factor(status,
                                   levels = 1:3,
                                   labels = c("melanoma-specific death", "alive", "other death")),
                   ulcer = factor(ulcer,
                                  levels = 0:1,
                                  labels = c("Absent", "Present")),
                   time = time / 365.25)

library(survival)
regular_model <- coxph(Surv(time, status == "melanoma-specific death") ~
                         age + sex + year + thickness + ulcer,
                       data = melanoma,
                       x = TRUE, y = TRUE)

spl_melanoma <- timeSplitter(melanoma,
                             by = .1,
                             event_var = "status",
                             event_start_status = "alive",
                             time_var = "time",
                             time_related_vars = c("age", "year"))

interval_model <- update(regular_model,
                         Surv(Start_time, Stop_time, status == "melanoma-specific death") ~ .,
                         data = spl_melanoma)

mismatch <- abs(sum(coef(interval_model) - coef(regular_model)))
if (mismatch > 10^-10) {
  stop("Failed to match interval with regular cox model.",
       " Total coefficient difference = ", mismatch,
       "\n Regular: ", paste(txtRound(coef(regular_model), 3),
                             collapse = ", "
       ),
       "\n Interval: ", paste(txtRound(coef(interval_model), 3),
                              collapse = ", "
       )
  )
}