r4clinstats

Learn R for clinical statistics, interactively. A swirl-style, learn-by-doing course for clinicians and clinical researchers: type real code against real-feeling synthetic clinical data and get immediate, kind feedback. Lessons are learnr tutorials graded by gradethis.

All fifteen modules (0–14) are built and runnable (M0–M5 are the MVP).

What's here so far

• Module 0 — Why R for clinical data (narrative: scripts, reproducibility, your first line of R).
• Module 1 — Tidy data as design (narrative: tidy shape, wide vs long, design before collection).
• Module 2 — The building blocks of R (objects, value types, factors, logical vectors).
• Module 3 — Wrangling I (filter, count, arrange, select, the pipe |>), adapted from the prototype.
• Module 4 — Wrangling II (mutate, group_by + summarise).
• Module 5 — Tidy data in practice (pivot_wider/pivot_longer, left_join).
• Module 6 — Reading & cleaning data (read_csv/here, recoding, fixing types, missing values).
• Module 7 — Visualisation I (the ggplot2 grammar of graphics: scatter, box and bar plots).
• Module 8 — Visualisation II (trends over time, faceting, tidying a figure for a paper).
• Module 9 — Summary tables (the "Table 1" every clinical paper opens with, via finalfit).
• Module 10 — Describing data (histograms, mean/SD vs median/IQR, choosing between them).
• Module 11 — Comparing groups (t-test, chi-squared, tidy results with broom).
• Module 12 — Regression (lm, baseline adjustment, logistic glm, odds ratios).
• Module 13 — Survival analysis (censoring, Kaplan-Meier, log-rank, Cox hazard ratios).
• Module 14 — Reproducible reports (the capstone: the whole trial analysis as a Quarto report, built from a shipped template via start_report() and rendered for real).
• Each interactive module closes with a Practice section: open exercises (no blanks) with hints, solutions and grading.
• Datasets patients, labs, outcomes and linelist (synthetic, reproducible).
• learn(), check_setup(), use_local_library(), start_report().
• A test harness in tests/ that runs every tutorial solution and locks the datasets.

The curriculum is complete: foundations, real data, analysis, and reporting.

Build it from source (developer)

From the package root, in R:

source("data-raw/patients.R")   # writes data/patients.rda
source("data-raw/labs.R")       # writes data/labs.rda  (run after patients.R)
source("data-raw/outcomes.R")   # writes data/outcomes.rda  (run after patients.R)
source("data-raw/linelist.R")   # writes data/linelist.rda
devtools::document()            # regenerate NAMESPACE and man/ from the roxygen comments
devtools::install(dependencies = TRUE)   # install locally, with the course packages
devtools::test()                # run the harness

gradethis is not on CRAN. It installs from Posit's R-universe; check_setup() handles this, or install it directly:

install.packages("gradethis",
                 repos = c("https://rstudio.r-universe.dev", getOption("repos")))

The portal (this website)

The landing page, install guide and browser taster are a pkgdown site. Build it locally to preview:

install.packages("pkgdown")
pkgdown::build_site()

On every push to master, .github/workflows/pkgdown.yaml rebuilds the site and publishes it to GitHub Pages at https://apsmunro.github.io/r4clinstats/. Enable Pages once in the repository settings, serving from the gh-pages branch.

Get started (learner)

install.packages("r4clinstats",
                 repos = c("https://apsmunro.r-universe.dev",
                           CRAN = "https://cloud.r-project.org"),
                 dependencies = TRUE)

library(r4clinstats)
check_setup()     # confirm the install, and fix anything still missing
learn()           # list the lessons
learn("m0")       # start at the beginning

dependencies = TRUE matters: the packages the tutorials run on (the tidyverse, finalfit, broom and the rest) are in Suggests, which a plain install skips. They are in Suggests deliberately, so that a missing gradethis cannot stop library(r4clinstats) from loading and taking you to check_setup().

gradethis is fetched separately by check_setup(), from the rstudio R-universe. It is deliberately not named in the repos above: that universe also carries development builds of learnr and rmarkdown, and install.packages() takes the highest version it can see rather than the first repository listed, so naming it there would quietly put every learner on an unreleased build of the engine the lessons run on.

Installing from apsmunro.r-universe.dev rather than with remotes::install_github() keeps learners clear of GitHub's anonymous API limit of 60 requests an hour per network address, which a shared hospital or university connection exhausts quickly when a group installs together.

The full walkthrough is the Get started guide.

Licence

MIT. The bundled datasets are synthetic and contain no real patient data.