Stepped-Wedge Clinical Trial Analysis and Power Simulation

Provides reusable functions for aggregated cluster-period data, mixed-effects analysis, and simulation-based power and type I error evaluation in stepped-wedge cluster randomized trials. The design and mixed-effects analysis follow Hussey and Hughes (2007) . Intraclass correlations for binary outcomes are converted to logistic-normal random-intercept standard deviations following Eldridge, Ukoumunne and Carlin (2009) . Monte Carlo uncertainty in estimated power is summarized using the exact binomial interval of Clopper and Pearson (1934) . The simulation engine supports sequence-specific baseline risks, cluster random effects, direct intraclass-correlation specification, Monte Carlo uncertainty intervals, and model-fitting diagnostics. Applied physician and specialty helpers are retained for backward compatibility and for an example health-services workflow.

stepwedgepower

stepwedgepower provides a general simulation-based workflow for stepped-wedge cluster randomized trials with aggregated binary outcomes. Version 0.1.1 adds:

• generic cluster/sequence terminology,
• direct ICC specification for logistic random-intercept models,
• Monte Carlo standard errors and exact confidence intervals,
• convergence, fit-failure, and singular-fit diagnostics, and
• backward compatibility with the original physician/specialty interface.

Project background

This package is based on a PhD biostatistics rotation project on statistical methods for stepped wedge clinical trial designs. According to the rotation evaluation, the project involved:

• research of published literature,
• summarizing literature in slides/presentations, and
• building well-organized, well-documented R software.

The software was also used to provide sample size calculations for a study under development. In the permission email shown in the screenshots, Prof. Florin Vaida approved publishing the software to GitHub.

What was changed from the original script

The original file mixed together:

1. raw CSV imports,
2. one-off data cleaning,
3. model fitting,
4. ad hoc probability extraction, and
5. repeated simulation loops.

This package reorganizes those steps into exported functions:

• prepare_physician_data()
• summarize_by_specialty()
• fit_specialty_rate_model()
• estimate_specialty_rates()
• analyze_lpa_outcomes()
• simulate_stepwedge_trial()
• run_stepwedge_analysis()
• estimate_power()
• estimate_type1_error()

Installation

# install.packages("remotes")
remotes::install_github("AmandaLinLi/stepwedgepower")

Minimal workflow

library(stepwedgepower)

power_out <- estimate_power(
  n_simulations = 1000,
  treatment_or = 1.50,
  n_clusters_per_sequence = c(10, 10, 10, 10),
  sequence_names = paste0("Sequence ", 1:4),
  baseline_probs = c(0.05, 0.05, 0.05, 0.05),
  icc = 0.05,
  n_per_cluster_period = 20,
  seed = 2026
)

power_out
power_out$power
power_out$mcse
c(power_out$conf_low, power_out$conf_high)

Version 0.1.0 argument names remain available with deprecation warnings.

Example data

A small synthetic dataset is included for quick testing:

ex_dat <- read_example_physician_data()
head(ex_dat)

Repository structure

stepwedgepower/
  DESCRIPTION
  NAMESPACE
  R/
  man/
  inst/extdata/
  inst/scripts/
  tests/
  .github/workflows/

Notes

• The package is structured to be GitHub-ready.
• I assumed an MIT license for convenience; you can change that before publishing.
• The original external CSV files are not bundled here, so the package ships with synthetic example data only.
• Because the current environment does not have an R runtime, this package scaffold was prepared carefully but not executed with R CMD check inside the container.

Suggested next steps before publishing

1. Open the package locally in RStudio.
2. Run devtools::document().
3. Run devtools::check().
4. Replace YOUR_GITHUB_USERNAME in README.md.
5. Replace the maintainer email placeholder in DESCRIPTION.
6. Commit and push to GitHub.