SCL: Successive classification learning for estimating quantile optimal treatment regimes

This directory contains all scripts to reproduce the simulation and real-data results. The entry point is Run_all.R, which sequentially sources the individual result scripts. All the tables and figures in the main paper and the supplementary material will appear in the output/ folder after running the scripts.

Summary of the files

• required_packages.R: script to install all the necessary R packages to reproduce our results
• Run_all.R: the main script to run all the result scripts
• code_result/: folder containing all the individual result scripts for simulation and real-data analysis
  • Result_main_500.R produces main simulation results with sample size n=500 and covariates dimension d=2, i.e., Table 1 in the main paper, Tables S4–S8 in the supplementary material, and Figures S2–S7 in the supplementary material.
  • Result_main_diff_np.R produces simulation results with different (n, d) combinations, including (250, 2), (1000, 2), and (500, 10), i.e., Tables S1–S3 in the supplementary material.
  • Result_inconsistency.R produces simulation results on the inconsistency issue of Wang's method, i.e., Figure 1 in the main paper and Figure S1 in the supplementary material.
  • Result_alter_smoothing.R produces simulation results with alternative smoothing techniques, i.e., Tables S9 and S10 in the supplementary material.
  • Result_DR.R produces simulation results about the doubly robust property of our SCL method, i.e., Tables S11–S16 in the supplementary material.
  • Result_nonconvexity.R produces simulation results for nonconvexity issue of Wang's method stated in Section S1.1 in the supplementary material.
  • Result_realdata.R produces real-data analysis results on the AIDS study data, i.e., Tables 2 and 3 in the main paper.
  • Result_smooth_survival.R produces plot of the survival functions and the smooth survival functions, i.e., Figure 2 in the main paper.
  • Result_kernels.R produces simulation results with different kernel choices, i.e., Tables S17 and S18 in the supplementary material.
  • Result_survival.R produces simulation results for survival data, i.e., Tables S19 and Figure S8 in the supplementary material.
• code_functions/: folder containing all the functions used in the simulation and real-data analysis.
  • function_main.R: functions including our SCL method, the competing methods, and auxiliary functions
  • function_inconsistency.R: functions for the inconsistency issue of Wang's method
  • function_nonconvexity.R: functions for the nonconvexity issue of Wang's method
  • function_kernel.R: functions to test the robustness of our SCL method with different kernel choices
  • function_survival.R: functions to implement our SCL method for survival data
  • simulation_case1.R: functions to generate data for case 1 in the main paper and execute the competing methods
  • simulation_case2.R: functions to generate data for case 2 in the main paper and execute the competing methods
  • simulation_case3.R: functions to generate data for case 3 in the main paper and execute the competing methods
  • simulation_case1_dr.R: functions to test the doubly robust property of our SCL method in case 1
  • simulation_case2_dr.R: functions to test the doubly robust property of our SCL method in case 2
  • simulation_case3_dr.R: functions to test the doubly robust property of our SCL method in case 3
  • simulation_inconsistency.R: functions on the inconsistency issue of Wang's method
  • simulation_nonconvexity.R: functions on the nonconvexity issue of Wang's method
  • simulation_kernel.R: functions to generate data and test the robustness of our SCL method with different kernel choices
  • simulation_survival.R: functions to generate survival data and test our SCL method for survival data
  • simulation_alter_smoothing.R: functions to test the alternative smoothing techniques for discrete outcomes for our SCL method
  • realdata_value.R: functions to clean the AIDS study data (real data) and implement the cross-validated to evaluate competing methods in value function
  • realdata_DC.R: functions to clean the AIDS study data (real data) and implement the cross-validated to evaluate competing methods in decision concordance (DC)
• output/: folder to save all the tables and figures in the main paper and the supplementary material

Prerequisites

R version

R version 4.3.1.

Multi-core parallelization on a single machine

Number of cores used: 16.

Necessary packages

The necessary packages to reproduce our results are as follows. You can run the script required_packages.R to intall all these packages.

• cowplot_1.1.3
• doParallel_1.0.17
• dplyr_1.1.4
• foreach_1.5.2
• ggplot2_3.5.1
• ggtext_0.1.2
• glmnet_4.1.8
• grid_4.3.1
• knitr_1.43
• latex2exp_0.9.6
• MASS_7.3.60
• mpath_0.4.2.25
• parallel_4.3.1
• quantoptr_0.1.3
• quantreg_5.97
• speff2trial_1.0.5
• tidyr_1.3.1
• WeightSVM_1.7.11
• survival_3.5.5

How to replicate our simulation and real-data results

Execute: r source("./Run_all.R")

This will run:

• code_result/Result_main_500.R
• code_result/Result_main_diff_np.R
• code_result/Result_alter_smoothing.R
• code_result/Result_inconsistency.R
• code_result/Result_DR.R
• code_result/Result_nonconvexity.R
• code_result/Result_realdata.R
• code_result/Result_smooth_survival.R
• code_result/Result_kernels.R
• code_result/Result_survival.R

All the tables and figures will be saved in the output/ folder.