README

Introduction

This repository contains manuscript-relevant code for the paper titled “Time-to-Event Estimation with Unreliably Reported Events in Medicare Health Plan Payment” by Oana M. Enache and Sherri Rose, arxiv:2602.04092.

More specifically, this code:

• (In Section All of Us) Extracts co-occurring Hierarchical Condition Categories (from version 28 of the Medicare Advantage risk adjustment payment algorithm) of All of Us respondents

• (In Section Simulations on Sherlock) Runs 1000 replicates of simulations for manuscript

• (In Section Manuscript) Generates plots and manuscript.

Each of these have their own environment, which is specified in the corresponding section. In general, you can restore a project library from an renv lockfile by calling renv::restore(lockfile="path/to/lockfile"); more details are available in the renv package documentation. Code listed should be run in the order written unless otherwise specified. Output relevant to each section is also available, so only using/re-running parts of the code is also possible.

Repository Structure and Contents

• R/

  • data_simulation/

    • AllOfUs_platform/ : Code used on All of Us Researcher Workbench for baseline data generation

    • manuscript_simulation/ : Code to generate simulation data and estimates for manuscript

  • figures_and_tables/ : Summary statistics exported from All of Us Researcher Workbench and code to generate all figures in manuscript

• manuscript/: Code to generate manuscript and copies in PDF

• renv/: Additional local renv files to use for analyses besides on All of Us or Sherlock

• all_of_us_renv.lock : renv lockfile to restore environment for All of Us using R version 4.4.0 (current as of January 2026).

• renv.lock: renv lockfile to restore environment used for all analyses besides on All of Us or Sherlock

All of Us

All of this code needs to be run in the All of Us Researchers Workbench.

Setup

Obtain Registered Tier data access.

Restore the environment all_of_us_renv from all_of_us_renv.lock.

Code

All code is in R/data_simulation/AllOfUs_platform.

Run:

1. 01_make_hcc_to_AoU_concept_map.qmd

2. 02_save_cohort_subsets_to_csv.qmd

3. 03_aggregate_cohort.qmd

4. 04_basic_eda_and_define_hcc_sets.qmd

Output

Co-occurring HCCs with respondent counts is available in the upcoding package in the global variable cooccurring_v28_hcc_counts_allofus.

Descriptive tables from All of Us respondents are in R/figures_and_tables/all_of_us_summary_statistics.

All of Us-related summary tables are in Supplementary Information of the main manuscript.

Simulations on Sherlock

Sherlock is a high-performance computing cluster at Stanford University that uses Slurm for job scheduling and resource management. See Sherlock Documentation for further details.

Setup

As of December 2025, it does not work to use the local renv on Sherlock. So, you can get setup to run the slurm script as follows. This setup only needs to happen once.

On Sherlock, run the following commands in this order:

```         
# Launch interactive dev session with 4 CPUs

$ sh_dev -c 4

# Load the required modules

$ ml purge binutils libgit2
$ ml R/4.4.2
$ ml fribidi/1.0.12 libwebp/1.3.0 freetype/2.9.1 

# Launch R and install tidyverse

$ R
> install.packages("tidyverse", repos = "http://cran.us.r-project.org", Ncpus=4)
> install.packages("here")
> install.packages("furrr")
> install.packages("remotes")
> install.packages("gert", configure.vars = list(USE_SYSTEM_LIBGIT2 = 1))
> install.packages("devtools")
> library(devtools)
> devtools::install_github("oena/khsmisc") # installs Oana's fork of khmisc package
> devtools::install_github("StanfordHPDS/upcoding") # install upcoding package
```

Separately, you need to create a new SSH key for Sherlock (you only need to do the section “Generating a new SSH key”) and add it to your Github account, if you haven’t already. You also need to authorize this SSH key for use with Stanford’s single-sign on using the instructions here. Once this is set you should be able to clone this repo to your account.

Code

All simulations (1000 replicates) can be run by calling sbatch run_simulations.sh on Sherlock from the directory it’s in, R/data_simulation/manuscript_simulation; there is a seed that is modified slightly for each replicate. It typically takes 2-4 hours, and we recommend running it in your $SCRATCH folder as several large interim files are generated (and then deleted).

If you do use $SCRATCH, keep in mind that files are automatically deleted after 90 days; you may want to move your output files elsewhere like $HOME or $GROUP_HOME. Also note that your $HOME folder likely does not have enough space and that (per Stanford Research Computing) $GROUP_HOME is not recommended for persistent and heavy computational tasks; so, we do not recommend running your code in either of these spaces.

Output

There are two output folders generated by the simulation in R/data_simulation/manuscript_simulation : (1) output_files/ and (2)hcc_counts/ . You can also find a copy compressed version of each of these in R/data_simulation/manuscript_simulation.

Manuscript

Setup

Restore the renv renv.lock . This should occur automatically when you open the project.

Code

Generate all figures:

Rscript generate_simulation_figures.R . This will generate figures from the zipped R/data_simulation/manuscript_simulation/output_files.zip and R/data_simulation/manuscript_simulation/hcc_counts.zip; you do not need to unzip them yourself.

The manuscript is generated from the file:

manuscript/upcoding_metrics_manuscript.qmd

Output

A PDF of the manuscript can be found at manuscript/upcoding_metrics_manuscript.pdf. All figures generated are in manuscript/images/. The supporting information PDF and corresponding images are available in manuscript/supporting_information/supporting_information.pdf and manuscript/supporting_information/images/ respectively.