Add IPCW weights to competing risks long-format data

Estimates the probability of remaining uncensored and appends it as column p_notcens to the dataset. Supports two estimation strategies: a Cox proportional hazards model (strat = "no") or non-parametric (KM) estimates within each level of the covariate (strat = "yes").

Usage

add_ipcw_weights_cr(
  data_long,
  covariate = "z1",
  strat = "no",
  new_data = NULL,
  by.start = TRUE
)

Arguments

data_long: A data frame in long format, as returned by wide_to_long_cr(). Must contain columns tstart, tstop, censor, and the covariate named by covariate.
covariate: Character string. Name of the covariate column used to model the censoring distribution. Default is "z1".
strat: Character. "no" (default) fits a single Cox model for the censoring distribution using covariate. "yes" estimates the censoring distribution non-parametrically within each stratum of covariate.
new_data: Optional data frame to which weights are applied. If NULL (default), weights are computed for data_long itself.
by.start: Logical. If TRUE (default), the weight is P(not censored by tstart). If FALSE, the weight is P(not censored during the interval from tstart to tstop).

Value

new_data (or data_long if new_data is NULL) with an additional column p_notcens.

Examples

set.seed(42)
dat <- sim_data_cr(n = 100, censoring = "baseline")
dat_long <- wide_to_long_cr(dat)
dat_long <- add_ipcw_weights_cr(dat_long, strat = "no")
summary(dat_long$p_notcens)
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
#>  0.0547  0.7641  0.9126  0.8413  0.9757  1.0000