This github repo contains the code used for the simulation study of the cheap subsampling confidence interval. To run the simulation study, one needs to use R with the targets package and run tar_make(). The code for the introduction is included in intro_cross_validation while the code for two simulation studies are included in simulation_study_with_cross_validation and simulation_study_without_cross_validation. In addition, one needs the rtmle package which, at the time of writing, are not on CRAN. The necessary packages can be installed as follows:

install.packages(c("targets","data.table", "ggplot2", "tarchetypes", "crew", "crew.cluster","devtools", "dplyr", "tidyr", "tibble", "gt", "ggpubr"))
devtools::install_github("tagteam/rtmle")

An example run for the simulation study can be run by running the following code in R:

# Load the necessary packages
library(data.table)
library(rtmle)

## Working directory
setwd(
  "~/phd/cheap_subsampling_simulation_study_repo/simulation_study_without_cross_validation/"
)

# Arguments
time_horizon <- 2
eta <- 0.632
sample_size <- 1000

# Load functions from simulation study
source("functions/rtmle_functions.R")
source("functions/simulation_functions.R")

# Generate some data
simulated_data <- simulate_data(n = sample_size, time_horizon = time_horizon)

# RTMLE initialization
x <- rtmle_init(
  intervals = time_horizon,
  name_id = "pnr",
  name_outcome = "Y",
  name_competing = NULL,
  name_censoring = "C",
  censored_levels = c(1, 0),
  censored_label = 0
)

# Add data
x <- add_baseline_data(x, data = simulated_data$baseline_data)
x$data$outcome_data <- simulated_data$outcome
x$data$timevar_data <- simulated_data$timevarying_covariates
x$data$timevar_data <- x$data$timevar_data[simulated_data$regimen, on = "pnr"]
x <- protocol(x, name = "Always_A", intervention = data.frame("A" = factor("1", levels = c("0", "1"))))
x <- prepare_data(x)

# Specify the target parameter
x <- target(
  x,
  name = "Outcome_risk",
  strategy = "additive",
  estimator = "tmle",
  protocols = "Always_A"
)
x <- model_formula(x)
x <- run_rtmle(x, learner = "learn_glm", time_horizon = 1:time_horizon)
x <- cheap_bootstrap(
  x,
  B = 5,
  time_horizon = 1:time_horizon,
  M = 0.632 * NROW(x$prepared_data)
)