Efficacy and Safety Graphics for Clinical Review Packages

Builds submission-style efficacy and safety figures used in clinical review packages, with one simulator for each therapeutic area. Oncology figures include Kaplan-Meier curves, RECIST waterfalls, swimmer plots and spider plots. Cardiovascular, endocrine, immunology, infectious disease and neuroscience figures include time-to-event curves, mean change from baseline and responder rates. Shared safety figures include an adverse-event dot plot, an eDISH liver plot, a laboratory shift plot and an exposure plot. Reference lines follow RECIST 1.1 and the FDA drug-induced liver injury guidance. This is not an FDA product and does not guarantee regulatory acceptance.

trialfigs

Efficacy and safety graphics for a clinical review package, organised by therapeutic area.

This is an independent R package. It is not an FDA product and not a guarantee that a figure will be accepted. Reference lines follow RECIST 1.1 and the FDA drug-induced liver injury guidance. Data from the simulators are synthetic.

Suggested GitHub repository name: trialfigs. R package name: trialfigs.

Install

# Once the repository exists:
install.packages("remotes")
remotes::install_github("YOURUSER/trialfigs")
library(trialfigs)

From a local checkout:

install.packages(c("ggplot2", "survival"))
install.packages(".", repos = NULL, type = "source")

What to plot, by therapeutic area

ta_plot_catalog() returns this map. The function column is what the example scripts call.

Primary contrasts in a real submission should come from the model named in the SAP (Cox, MMRM, negative binomial). The intervals in plot_change are mean ± 1.96 SE, labelled as such, so they are not silently substituted for an MMRM LS-mean.

Real example

Data below are simulated. Do not describe them as trial results.

library(trialfigs)

onc <- simulate_oncology(n_per_arm = 80)

plot_km(
  onc$adsl,
  time = "os_months", event = "os_event", arm = "arm",
  title = "Overall survival",
  xlab = "Months", ylab = "Overall survival"
)

plot_waterfall(onc$adsl)
plot_swimmer(onc$adsl, max_n = 24)
plot_spider(onc$adtr)

forest <- rbind(
  cox_hr(onc$adsl, "os_months", "os_event", "arm", subgroup = "Overall"),
  cox_hr(onc$adsl[onc$adsl$sex == "F", ], "os_months", "os_event", "arm", subgroup = "Female"),
  cox_hr(onc$adsl[onc$adsl$sex == "M", ], "os_months", "os_event", "arm", subgroup = "Male")
)
plot_forest(forest, n = "n", title = "Overall survival by subgroup")

saf <- simulate_safety()
plot_ae(saf$adae, denom = saf$adsl, reference = "Placebo")
plot_edish(saf$adlb)
plot_exposure(saf$adsl)

Build the full gallery from the package root:

source("inst/examples/run_all.R")
# PNGs land in inst/examples/figures/

Repository layout

trialfigs/
  DESCRIPTION              package metadata
  NAMESPACE                exports
  LICENSE                  MIT
  README.md
  NEWS.md
  trialfigs.Rproj
  .Rbuildignore
  .gitignore
  .github/workflows/R-CMD-check.yaml
  R/theme.R                palette, theme, catalog
  R/simulate.R             one simulator per therapeutic area
  R/plots_efficacy.R       KM, waterfall, swimmer, spider, forest, change, responder
  R/plots_safety.R         AE, eDISH, lab shift, exposure
  man/trialfigs-package.Rd
  inst/examples/           runnable review-style scripts

Put it on GitHub

cd trialfigs
git init
git add .
git commit -m "Initial trialfigs package"
gh repo create trialfigs --public --source=. --remote=origin --push

Replace YOURUSER in DESCRIPTION and in the install snippet above.

Conventions worth keeping

• Population in the caption (ITT, safety, virologic), not only in a footnote nobody reads.
• Number at risk on every Kaplan–Meier. Censoring ticks on.
• RECIST lines only on tumor-burden plots, and only if the endpoint is RECIST.
• eDISH quadrant cases get a narrative. Alkaline phosphatase and an alternative cause still have to be checked before anyone writes “Hy’s law”.
• Subgroup forests are descriptive unless the subgroup was prespecified and the multiplicity plan says otherwise.
• Do not invent a p-value the SAP did not specify.

References the figures follow

• Eisenhauer EA et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009.
• FDA. Guidance for Industry. Drug-Induced Liver Injury: Premarketing Clinical Evaluation. 2009. eDISH and Hy’s law quadrant.
• PHUSE. Analysis and Displays Associated with Adverse Events, and the safety-topics white papers.
• ICH E3. Structure and Content of Clinical Study Reports.