README.md
In trialfigs: Efficacy and Safety Graphics for Clinical Review Packages

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.

# 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")

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

| Therapeutic area | Display reviewers expect | Function | | --- | --- | --- | | Oncology | Kaplan–Meier of overall survival and progression-free survival, with censoring ticks and number at risk | plot_km | | Oncology | Waterfall of best percent change in sum of diameters. Dashed lines at −30% and +20% (RECIST 1.1) | plot_waterfall | | Oncology | Swimmer plot: time on treatment, objective response, progression | plot_swimmer | | Oncology | Spider plot: individual tumor trajectories plus the arm mean | plot_spider | | Oncology | Objective response rate with a Wilson interval | plot_responder | | Oncology | Forest of hazard ratios by prespecified subgroup | plot_forest, cox_hr | | Cardiovascular | Kaplan–Meier of time to first MACE. State the competing-risk limitation if non-CV death is common | plot_km | | Cardiovascular | Forest of hazard ratios, overall and by subgroup | plot_forest | | Cardiovascular | Mean change in SBP or LDL-C with a 95% interval | plot_change | | Endocrinology | Mean change in HbA1c | plot_change | | Endocrinology | Responder rate, for example HbA1c < 7% at the landmark visit | plot_responder | | Immunology | Mean change in DAS28, PASI, or CDAI | plot_change | | Immunology | Binary responder (ACR20, PASI75, DAS28 improvement) | plot_responder | | Infectious disease | Mean change in log10 viral load | plot_change | | Infectious disease | Time to virologic clearance | plot_km | | Neuroscience | Mean change in MADRS, ADAS-Cog, or UPDRS | plot_change | | Neuroscience | Landmark responder, for example 50% MADRS reduction | plot_responder | | Safety, all areas | Adverse-event incidence dot plot with Wilson intervals, ordered by risk difference | plot_ae | | Safety, all areas | eDISH: peak ALT × ULN vs peak total bilirubin × ULN. Quadrant is ALT ≥ 3× and bilirubin ≥ 2× | plot_edish | | Safety, all areas | Baseline vs post-baseline maximum laboratory value | plot_lab_shift | | Safety, all areas | Exposure: treatment duration by arm | plot_exposure |

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.

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/

| Script | Area | Figures | | --- | --- | --- | | inst/examples/01_oncology.R | Oncology | OS, PFS, waterfall, swimmer, spider, forest, ORR | | inst/examples/02_cardiovascular.R | Cardiovascular | MACE, subgroup forest, SBP change | | inst/examples/03_endocrinology.R | Endocrinology | HbA1c change, HbA1c < 7% | | inst/examples/04_immunology.R | Immunology | DAS28 change, DAS28 responder | | inst/examples/05_infectious.R | Infectious disease | Viral-load change, time to clearance | | inst/examples/06_neuroscience.R | Neuroscience | MADRS change, 50% responder | | inst/examples/07_safety.R | Safety, all areas | AE dot plot, eDISH, ALT shift, exposure |

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

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.

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.

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.