Bayesian (and frequentist) re-analyses of published RCTs

Interactive apps

These Shiny apps provide interactive Bayesian (and frequentist) re-analyses of published randomized controlled trials. They are part of the Assessing clinical trial interpretability research project funded by Heart & Stroke Canada.

Enter the 2×2 event counts from any RCT, choose an effect measure (relative risk or risk difference), and explore how different prior assumptions shape posterior conclusions.

Static entry image.

Static entry image

Bayesian RCT Re-analysis (Bayes only)

A focused Bayesian app with four priors — Weak, Enthusiastic, Skeptical, and Empirical Bayes (van Zwet et al. 2021/2025) — on either the relative risk (RR) or risk difference (RD) scale.

Features

  • Three likelihood options: Normal (analytic), Binomial (Laplace), MCMC (brms)
  • RR and RD effect measures via tabbed prior panels (priors entered in natural units)
  • MCID-aware posterior summaries, predictive distributions, and forest plots
  • Prior vs Likelihood vs Posterior facet plots
  • CSV export of all summaries

Sample static output image.

Static output image

Source code on GitHub

Full Bayesian + Frequentist Re-analysis

Everything in the Bayesian app, plus frequentist calibration metrics and an optional Design-informed prior.

Additional features

  • Greenland S-values (vs RR = 1 and vs RR = MCID)
  • Gelman–Carlin Type-S and Type-M error analysis
  • Per-prior posterior calibration table (Type-S, Type-M, coverage, replication probability)
  • Empirical-Bayes mixture component decomposition (Bk, B̄)
  • Optional 5th Design-informed prior centred on the trial’s protocol target (RR mode)

Source code on GitHub

Running locally

Clone the shiny_light repository and run either app in RStudio or from the command line:

shiny::runApp("app_bayes_alone.r")
shiny::runApp("app_full.r")

Requires R ≥ 4.1 with shiny, ggplot2, dplyr, ggdist, tibble, bslib, and rlang. The MCMC backend additionally requires brms and cmdstanr (or rstan).