cardinal is on CRAN

A few years ago {cardinal} started life as a quarto catalog of copy-paste template scripts — a cross-company effort (Boehringer Ingelheim, Moderna, Roche, Sanofi and friends) to open-source a harmonized set of the FDA safety tables everyone in clinical reporting keeps rebuilding from scratch.

It just grew up. cardinal is now on CRAN — installable, versioned, dependency-managed:

install.packages("cardinalfda")

Yes, the package is cardinalfda on CRAN (plain cardinal was already taken). Same initiative, same website, now a proper release you can pin in a lockfile and trust to still install next year.

What changed under the hood

If you looked at cardinal a while back, the tables were {rtables}. That’s no longer the engine. The catalog has moved to the {gtsummary} + {cards}/{cardx} + {crane} stack:

• {gtsummary} builds the table,
• {cards} / {cardx} build the ARD (Analysis Results Dataset) behind it,
• {crane} supplies the pharma-specific bits (e.g. tbl_listing() for listings).

And that middle point is the part I’m most excited about: every cardinal table now comes with an ARD. More on that below.

The anatomy of a cardinal template

The nicest thing about the new layout is that every template tells the same story — and it maps onto how a statistical programmer actually thinks. Each catalog page has the same four tabs: Table Preview → Setup → Build Table → Build ARD. Let’s walk FDA Table 2 (Baseline Demographic and Clinical Characteristics, Safety Population) through it.

1. The IG — the FDA Integrated Guide

cardinal’s templates are drawn from the FDA’s Standard Safety Tables and Figures Integrated Guide (the “IG”), and are indexed by its table numbers — fda-table_02 is IG Table 2. The IG is the specification: it says exactly what the table should contain — the rows, the summary statistics, the population, the footnotes. So there’s never any doubt about what the output is supposed to be.

2. The output snapshot — the preview

Every catalog page opens with a Table Preview — a snapshot of the target output, the “here’s what we’re building” picture. It’s the contract between the IG spec and the code.

FDA Table 2 preview

3. The actual code

Then the part programmers came for — the actual, runnable code. cardinal ships each IG template as a self-contained script, so nothing is a black box: the template gives you the layout, you supply your ADaM data. Here is the real fda-table_02 template, run against the public {pharmaverseadam} ADSL:

library(dplyr)

adsl <- pharmaverseadam::adsl |>
  # removing screen failure observations
  filter(TRT01A != "Screen Failure")

tbl <- adsl |>
  dplyr::filter(SAFFL == "Y") |>
  gtsummary::tbl_summary(
    by = dplyr::all_of("TRT01A"),
    include = dplyr::all_of(c("SEX", "AGE", "AGEGR1", "ETHNIC", "RACE")),
    type = gtsummary::all_continuous() ~ "continuous2",
    statistic = list(
      gtsummary::all_continuous() ~ c(
        "{mean} ({sd})",
        "{median} ({min}, {max})"
      ),
      gtsummary::all_categorical() ~ "{n} ({p}%)"
    ),
    label = list(AGEGR1 = "Age Group, Years")
  ) |>
  gtsummary::add_overall(last = TRUE, col_label = "**Total Population**  \nN = {N}") |>
  gtsummary::remove_footnote_header(columns = dplyr::everything())

4. The real output

Because the template is real code over real (public) data, the last step is just to print it — and that same table is what you’d drop into your report:

tbl

Characteristic	Placebo N = 86	Xanomeline High Dose N = 72	Xanomeline Low Dose N = 96	Total Population N = 254
Sex
F	53 (62%)	35 (49%)	55 (57%)	143 (56%)
M	33 (38%)	37 (51%)	41 (43%)	111 (44%)
Age
Mean (SD)	75 (9)	74 (8)	76 (8)	75 (8)
Median (Min, Max)	76 (52, 89)	76 (56, 88)	78 (51, 88)	77 (51, 89)
Age Group, Years
18-64	14 (16%)	11 (15%)	8 (8.3%)	33 (13%)
>64	72 (84%)	61 (85%)	88 (92%)	221 (87%)
Ethnicity
HISPANIC OR LATINO	3 (3.5%)	3 (4.2%)	6 (6.3%)	12 (4.7%)
NOT HISPANIC OR LATINO	83 (97%)	69 (96%)	90 (94%)	242 (95%)
Race
AMERICAN INDIAN OR ALASKA NATIVE	0 (0%)	1 (1.4%)	0 (0%)	1 (0.4%)
BLACK OR AFRICAN AMERICAN	8 (9.3%)	9 (13%)	6 (6.3%)	23 (9.1%)
WHITE	78 (91%)	62 (86%)	90 (94%)	230 (91%)

Prefer not to read the whole script? Every template is also runnable in one line, which sources the shipped script and hands you back the built objects:

env <- run_template("fda-table_02")
env$tbl # the gtsummary table
env$ard # the ARD (see below)

The bit I like most: an ARD falls out for free

An ARD (Analysis Results Dataset) is the tidy, machine-readable version of a table — every statistic as a row, before any formatting. It’s the emerging CDISC direction for analysis results, and it’s what makes results reusable, comparable and reviewable rather than locked inside a rendered table.

With the {cards}-based approach, you don’t build the ARD separately — it’s gathered straight from the table you already made:

ard <- gtsummary::gather_ard(tbl)
ard$tbl_summary |>
  dplyr::select(dplyr::any_of(
    c("group1", "group1_level", "variable", "stat_name", "stat")
  )) |>
  head(8)

# An ARD data frame: 8 × 5
  group1 group1_level variable stat_name   stat
  <chr>  <list>       <chr>    <chr>     <list>
1 TRT01A Placebo      SEX      n         53    
2 TRT01A Placebo      SEX      N         86    
3 TRT01A Placebo      SEX      p          0.616
4 TRT01A Placebo      SEX      n         33    
5 TRT01A Placebo      SEX      N         86    
6 TRT01A Placebo      SEX      p          0.384
7 TRT01A Placebo      RACE     n          0    
8 TRT01A Placebo      RACE     N         86    

Same numbers as the table, now as data you can filter, diff against a previous run, or feed into a meta-analysis. Table and ARD, from one template — that’s the direction cardinal is heading.

Try it, and come help

Install it, run a template, and if you spot a table your team builds differently, that’s an invitation — cardinal is a collaboration, and new templates are exactly the kind of contribution the initiative wants. The cardinal website has the full catalog, and the pharmaverse examples site is a great next stop for seeing these packages used end to end.

Last updated

2026-10-20 15:02:04.247126

Details

Source, Session info