A teaching-grade, fully reproducible RWE project. Loads the public survival::colon adjuvant-chemotherapy cohort as a stand-in for an EHR-derived oncology dataset, augments it with a simulated tumor biomarker, and estimates biomarker-specific treatment effects on overall survival.

Modern oncology drug development is biomarker-driven. RWE studies that stratify by biomarker provide some of the most actionable insights for early-phase decision making — they tell us who benefits. This project demonstrates the canonical analytical pattern:

1. Load and clean a real-world cohort (here: survival::colon, public).
2. Attach (or merge) a biomarker.
3. Build descriptive Table 1 by treatment arm.
4. Estimate Kaplan–Meier curves overall and by biomarker stratum.
5. Estimate Cox PH HRs for each subgroup.
6. Test the treatment × biomarker interaction.
7. Visualize as a forest plot.
8. Discuss limitations a regulator would expect addressed.

The analytical helpers ship as an installable R package (rwbiomarker) so the report stays clean and the helpers are unit-tested and reusable.

rwbiomarker/
├── .devcontainer/devcontainer.json   # one-click reproducible env
├── .github/workflows/render.yml      # CI: tests + renders the report
├── DESCRIPTION                       # R package metadata
├── NAMESPACE                         # exported functions
├── R/
│   ├── rwbiomarker-package.R
│   ├── load.R                        # load_colon()
│   ├── biomarker.R                   # add_biomarker()
│   └── analyze.R                     # fit_subgroup_hr(), fit_interaction_cox()
├── man/                              # roxygen-generated help
├── tests/testthat/                   # unit tests
├── analysis/
│   ├── report.qmd                    # the Quarto walkthrough
│   ├── references.bib
│   └── _quarto.yml
├── Makefile                          # `make install | test | render`
├── LICENSE / LICENSE.md              # MIT
└── README.md

1. Install Docker and the Dev Containers extension.

2. Open this folder in VS Code.

3. F1 → Dev Containers: Reopen in Container.

4. Wait for the build (R 4.4 + Quarto + R packages).

5. From the integrated terminal:

   make test
   make render   # produces analysis/report.html

install.packages(c(
  "devtools", "survival", "survminer", "broom", "gtsummary",
  "tibble", "dplyr", "ggplot2", "knitr", "gridExtra"
))
devtools::install(".")

quarto render analysis/report.qmd
open analysis/report.html

• The mechanics of a biomarker-stratified survival analysis.
• Why a significant interaction matters more than significant subgroup-specific HRs.
• How to draw and interpret a forest plot of subgroup HRs.
• The difference between predictive and prognostic biomarkers (the interaction term distinguishes them).
• How to package an analysis for transparent regulatory review.

• Pinned R version via the rocker-org dev-container image (tidyverse:4.4).
• Pinned package set in DESCRIPTION (use renv::init() if you want a lockfile).
• The biomarker simulation uses a fixed seed.
• CI runs the tests and re-renders the report on every push.
• The rendered report can be published as a GitHub Pages site from the docs/ folder if Pages is enabled for this repository.

• Replace the simulated biomarker with a real one (e.g., from cgdsr / cBioPortal).
• Add competing-risk analyses (cmprsk / tidycmprsk).
• Add multiplicity-adjusted subgroup p-values (e.g., Hochberg).
• Run a sensitivity analysis under different biomarker prevalences and random seeds.

MIT — see LICENSE.md.

For education only. The simulated biomarker is not clinically meaningful and the analyses are not regulatory advice.