03 — Pharmaverse TLG Production Practice and What "FDA Submission Level" Means

Research for open.csr. Everything below is either fetched from a cited URL or verified locally with Rscript against the installed packages (marked [verified locally]). Local session: R with pharmaverseadam 1.1.0, cards 0.6.1, cardx 0.2.5, gtsummary 2.3.0, gt 1.0.0, tfrmt 0.1.3, r2rtf 1.3.1, admiral, quarto.

1. What "FDA submission level" actually means for a TFL

There is no single "FDA-quality TFL" standard. In practice it decomposes into four independent layers, and open.csr needs an answer for each:

open.csr's stated principles (code-generated TFLs, live regeneration, saved iterations, end-to-end traceability, test evidence) map almost exactly onto layers B + D — which is where the open-source ecosystem is weakest and where a differentiated product exists. Layers A and C are already well served by existing packages and should be consumed, not rebuilt.

2. FDA Standard Safety Tables and Figures (ST&F) — the closest thing to an FDA-defined TFL set

CDER's Office of New Drugs published the Standard Safety Tables and Figures: Integrated Guide (ST&F IG), first released August 2022 via docket FDA-2022-N-1961-0046, with a current copy on fda.gov. It is what FDA reviewers build for NDA/BLA safety review, so it is the best available proxy for "what FDA wants to see."

• Docket PDF: https://downloads.regulations.gov/FDA-2022-N-1961-0046/attachment_1.pdf
• Docket landing: https://www.regulations.gov/document/FDA-2022-N-1961-0046
• FDA program page: https://www.fda.gov/drugs/development-resources/standard-safety-tables-and-figures-stfs
• Integrated Guide (fda.gov copy): https://www.fda.gov/media/187065/download
• Targeted Analysis Guides (TAGs): Kidney Injury https://www.fda.gov/media/187063/download · Muscle Injury https://www.fda.gov/media/187064/download

Key ST&F conventions worth copying:

• FDA Medical Queries (FMQ/OCMQ) — FDA's own standardized MedDRA PT groupings (narrow/broad), used instead of or alongside SMQs. Introduced publicly in 2022; see https://www.thefdalawblog.com/2022/09/fda-unveils-its-own-medical-queries-a-standardized-approach-for-grouping-meddra-preferred-terms-that-will-impact-nda-bla-safety-analyses-and-drug-labeling/
• Safety population, pooled analyses as the default population/scope label in titles.
• Threshold displays — e.g. "Common Adverse Events Occurring at ≥X% Frequency."
• Titles are fully self-describing: "Patients With Serious Adverse Events by System Organ Class and FDA Medical Query (Narrow), Safety Population, Pooled Analyses" (Table 10).

{cardinal} (pharmaverse) is the open-source implementation of the ST&F IG: https://pharmaverse.github.io/cardinal/ · https://github.com/pharmaverse/cardinal · catalog: https://pharmaverse.github.io/cardinal/quarto/index-catalog.html

Cardinal's catalog currently maps these ST&F tables (ID → title):

Cardinal is built on {cards} + {gtsummary} and explicitly aligns with the CDISC ARD/ARM effort — i.e. the same ARD-first architecture open.csr proposes. It is listed as "Under development / Upcoming" on the pharmaverse TLG page (https://pharmaverse.org/e2eclinical/tlg/), so it is a collaborator, not a competitor.

3. The pharmaverse TLG package landscape (with current CRAN versions)

Fetched from crandb (https://crandb.r-pkg.org/) on 2026-07-25:

pharmaverse's own recommendation split (https://pharmaverse.org/e2eclinical/tlg/): Recommended = rtables, chevron, pharmaRTF, Tplyr, gtsummary, tfrmt, tidytlg, cards, rlistings, ggsurvfit. Available = cardx, siera, tfrmtbuilder, eudract, tern, teal, gridify, docorator, clinify, autoslider, tidyCDISC.

Note the ecosystem is bifurcated: the NEST/Roche stack (rtables → tern → chevron → rlistings) and the ARD stack (cards → gtsummary/tfrmt). Both are pharmaverse-blessed. cards/gtsummary is the one aligned with CDISC ARS and the one Cardinal chose.

4. The TLG Catalog (insightsengineering/tlg-catalog)

• Site: https://insightsengineering.github.io/tlg-catalog/ · Repo: https://github.com/insightsengineering/tlg-catalog (open-sourced 2023, permissive license)
• Built with Quarto; book/ contains tables/, listings/, graphs/, appendix/.

Verified via GitHub API — table categories and entry counts (.qmd files):

ADA 5 · ECG 5 · adverse-events 15 · concomitant-medications 4 · deaths 1 · demography 1
disclosures 3 · disposition 3 · efficacy 15 · exposure 1 · lab-results 18
medical-history 1 · pharmacokinetic 9 · risk-management-plan 5 · safety 1 · vital-signs 2
= 89 table entries

Listings categories: ADA, ECG, adverse-events, concomitant-medications, development-safety-update-report, disposition, efficacy, exposure, lab-results, medical-history, pharmacokinetic, vital-signs. Graphs: efficacy, other, pharmacokinetic. (Public write-ups cite "over 220 standard table variants across 8 categories" once variants within each qmd are counted.)

{chevron} exposes the same catalog as callable functions with sponsor-style IDs (https://insightsengineering.github.io/chevron/). Selected IDs — this is the closest thing to a canonical "CSR TFL numbering scheme" in open source:

• AE: AET01 Overview of Deaths and AEs · AET01_AESI AESI summary · AET02 AEs by SOC and PT · AET03 AEs by greatest intensity · AET04 AEs by highest NCI-CTCAE grade · AET05 / AET05_ALL exposure-adjusted rates per patient-year · AET10 Most common (x%) AE PTs · AEL02 AE listing · AEL03 SAE listing · AEL01_NOLLT PT/verbatim glossary
• DM/DS/EX: DMT01 Demographics and Baseline Characteristics · DST01 Patient Disposition · EXT01 Exposure Summary
• LB: LBT01 results + change from baseline by visit · LBT04 abnormalities not present at baseline · LBT05 single/replicated marked abnormalities · LBT06 abnormalities by visit and baseline status · LBT14 shift to highest NCI-CTCAE grade · LBT15 shifts to grade 3–4 post-baseline
• VS/EG: VST01, VST02 · EGT01, EGT02, EGT03 (shift baseline vs min/max), EGT05_QTCAT
• CM/MH/DTH/PD: CMT01A, CMT02_PT, CML02A_GL · MHT01 · DTHT01 · PDT01, PDT02
• Efficacy: RSPT01 binary outcomes · TTET01 time-to-event · COXT01/COXT02 Cox models · FSTG01/FSTG02 subgroup forest · CFBT01 change from baseline
• RMP: RMPT01, RMPT03–RMPT06 (EU risk-management-plan exposure/extent tables)
• Graphs: KMG01 Kaplan-Meier · MNG01 mean plot

Chevron functions all take adam_db — a named list of ADaM datasets — as first argument. A useful API convention for open.csr's TFL Builder.

5. {pharmaverseadam} contents — verified locally

pharmaverseadam 1.1.0 ships 23 datasets [verified locally]:

adae adbcva_ophtha adce_vaccine adcm adeg adex adface_vaccine adis_vaccine adlb adlbhy
admh adoe_ophtha adpc adpp adppk adrs_onco adsl adsl_vaccine adtr_onco adtte_onco
advfq_ophtha advs advs_peds

Sizes and study IDs [verified locally]:

Provenance (https://pharmaverse.github.io/pharmaverseadam/): datasets are produced by a script that runs the templates shipped in admiral, admiralonco, admiralophtha, admiralvaccine, admiralpeds, admiralmetabolic, admiralneuro against pharmaversesdtm SDTM. Specs are kept in adams-specs.xlsx. Regenerated on new package releases and ad hoc when templates change. These are test data, not QC'd production datasets — the site directs users to the changelog to know which package versions produced a given snapshot.

Underlying study: CDISCPILOT01, the CDISC pilot (xanomeline transdermal in Alzheimer's). TRT01A levels [verified locally]: Placebo (86), Screen Failure (52), Xanomeline High Dose (72), Xanomeline Low Dose (96).

Gaps that matter for a CSR demo [verified locally]:

1. No primary-efficacy ADaM for CDISCPILOT01. There is no adadas / adqs / adcibc. The CDISC pilot's primary endpoint (ADAS-Cog(11) change from baseline at Week 24, LOCF) lives in the R Consortium pilot packages, not in pharmaverseadam. The only efficacy data is oncology-shaped (adtte_onco PARAMCDs = OS, PFS, RSD; adrs_onco; adtr_onco) — which is not the CDISCPILOT01 protocol.
2. Population flags are thin. adsl *FL variables are only DTHFL, SAFFL, DTH30FL, DTHA30FL, DTHB30FL — no ITTFL, EFFFL, COMPLFL, RANDFL. The R Consortium pilot ADSL has these. open.csr must either derive them or source ADSL from the pilot packages.
3. ARM/ACTARM include "Screen Failure" (52 subjects), so disposition/screening tables need explicit handling of the screen-failure column.
4. Structure is otherwise complete: adae has AEDECOD, AEBODSYS, AESEV, AESER, AEREL, AEOUT, TRTEMFL, AESDTH, ASTDT, AENDT, AEACN, AOCCIFL, SAFFL; adlb has PARAM, PARAMCD, AVAL, BASE, CHG, PCHG, ANRIND, BNRIND, AVISIT, AVISITN, ATOXGR, BTOXGR, ABLFL, ANL01FL (shift tables are fully derivable); advs PARAMCDs = BMI, BSA, DIABP, HEIGHT, MAP, PULSE, SYSBP, TEMP, WEIGHT; adeg PARAMCDs = EGINTP, HR, QT, QTCBR, QTCFR, QTLCR, RR, RRR [all verified locally].

6. The standard CSR TFL set, mapped to ICH E3

ICH E3 places outputs in §14 (in-text and end-of-text tables/figures) — conventionally 14.1 subject disposition & demographics, 14.2 efficacy, 14.3 safety — and §16.2 appendix listings (16.2.1 discontinued subjects, 16.2.2 protocol deviations, 16.2.3 excluded from efficacy, 16.2.4 demographic data, 16.2.5 compliance/drug concentration, 16.2.6 individual efficacy response, 16.2.7 AE listings, 16.2.8 individual lab listings).

A defensible minimum viable set for open.csr's Report Template Library, cross-referenced to chevron IDs / FDA ST&F numbers:

PHUSE's Standard Analyses and Code Sharing white papers are the reference for the analysis conventions inside each of these (treatment-emergent definition, EAIR/patient-years, descending-frequency sort, ≥X% thresholds, SOC/PT nesting): AE white paper at https://phuse.s3.eu-central-1.amazonaws.com/Deliverables/Standard+Analyses+and+Code+Sharing/Analyses+and+Displays+Associated+with+Adverse+Events+Focus+on+Adverse+Events+in+Phase+2-4+Clinical+Trials+and+Integrated+Summary.pdf (companion papers cover demographics/disposition/medications, measures of central tendency, outliers/shifts, hepatotoxicity, QT). Standard scripts index: https://github.com/phuse-org/phuse-scripts/wiki/Standard-Script-Index

7. Output conventions and the RTF/PDF path

r2rtf (Merck) — the de facto submission RTF engine

r2rtf 1.3.1. Verified locally, the complete export list is:

rtf_page, rtf_page_header, rtf_page_footer, rtf_title, rtf_subline, rtf_colheader,
rtf_body, rtf_footnote, rtf_source, rtf_figure, rtf_read_figure, rtf_read_png,
rtf_rich_text, rtf_encode, write_rtf, assemble_rtf, assemble_docx, utf8Tortf

That export list is the CSR display grammar. Verified argument sets:

• rtf_page(tbl, orientation, width, height, margin, nrow, border_first, border_last, border_color_first, border_color_last, col_width, use_color) — page size, landscape / portrait, margins, and rows per page (pagination) are first-class.
• rtf_title(tbl, title, subtitle, text_font, text_format, text_font_size, ...) — multi-line titles + subtitles.
• rtf_body(...) includes group_by, page_by, new_page, pageby_header, pageby_row, subline_by, last_row, plus per-cell borders, text_justification (incl. decimal alignment), indents, fonts, colors.
• rtf_footnote() / rtf_source() — footnotes and the "Source: ..." provenance line, with as_table control.
• assemble_rtf() / assemble_docx() — concatenate many outputs into one deliverable.

Docs https://merck.github.io/r2rtf/ · paper Wang, Ye, Anderson, Zhang (2020) PharmaSUG https://pharmasug.org/proceedings/2020/DV/PharmaSUG-2020-DV-198.pdf · customization https://pharmasug.org/proceedings/2021/EP/PharmaSUG-2021-EP-085.pdf

The canonical textbook is R for Clinical Study Reports and Submission (r4csr.org), whose chapters are a CSR TFL curriculum: Overview, Disposition, Analysis population, Baseline characteristics, Efficacy table, Efficacy figure, AE summary, Specific AE, Assemble TLFs — https://r4csr.org/tlf-overview.html. It frames deliverables as following "ICH E3 guidance and the FDA's PDF Specifications," while noting each organization defines its own TFL rules.

Other output paths

• gt → RTF: gt 1.0.0 exports as_rtf, as_word, as_latex, as_raw_html, as_gtable [verified locally]. gtsummary 2.3.0 exports as_gt, as_flex_table, as_hux_table, as_kable*, as_hux_xlsx — no direct as_rtf [verified locally]; the route is gtsummary → as_gt() → gt::as_rtf(). gt's RTF lacks the page/pagination/footnote grammar r2rtf has. See PharmaSUG QT-263, "R Tables via GT for Regulatory Submissions", https://pharmasug.org/proceedings/2023/QT/PharmaSUG-2023-QT-263.pdf
• pharmaRTF — last CRAN release 2021-09-28 (0.1.4). Still "recommended" by pharmaverse but a maintenance risk; prefer r2rtf.
• tfrmt (GSK) — display metadata as data: body_plan, frmt, frmt_combine, frmt_when, col_plan, col_style_plan, row_grp_plan, footnote_plan, page_plan, big_n_structure, span_structure, tfrmt_n_pct, tfrmt_sigdig, make_mock_data, print_mock_gt, tfrmt_to_json / json_to_tfrmt, layer_tfrmt [verified locally, 0.1.3]. The JSON round-trip plus mock-table generation is directly relevant to open.csr's "TFL definition is version-controlled source" principle. Companion Shiny app: https://gsk-biostatistics.github.io/tfrmtbuilder/ · paper: https://www.lexjansen.com/phuse-us/2024/sm/PAP_SM09.pdf
• Frameworks for headers/footers: gridify, docorator, clinify (pharmaverse "Available" tier) wrap tables/figures with the page furniture.

Precision and rounding

The near-universal SAP convention (confirmed across many public SAPs indexed by clinicaltrials.gov): min/max at collected precision; mean and median at one more decimal place than collected; SD and SE at two more (equivalently, one more than the mean); typically capped at 3 decimals; percentages to one decimal. tfrmt::tfrmt_sigdig() implements significant-digit plans declaratively; r2rtf supports decimal alignment via text_justification. Rounding must be half-up, not R's default banker's rounding — a classic R-vs-SAS discrepancy and a required test case.

Population/denominator conventions

Every safety display is "Safety Population" with N from SAFFL == "Y"; treatment columns from TRT01A/TRTA (actual) for safety and TRT01P/TRTP (planned) for efficacy; TEAE subsets by TRTEMFL == "Y"; "big N" header counts come from ADSL, not the analysis dataset — a distinction tfrmt::big_n_structure() and cards::ard_total_n() both encode.

8. ARD-first architecture: cards / cardx / gtsummary / CDISC ARS

CDISC Analysis Results Standard (ARS) v1.0 was published 19 April 2024: https://www.cdisc.org/standards/foundational/analysis-results-standard/analysis-results-standard-v1-0 Model entities: ReportingEvent, AnalysisSet, DataSubset, GroupingFactor, Analysis, AnalysisMethod, Operation, Output, OutputDisplay. Deliverables include a logical model, JSON/YAML schema, and Excel representation.

{cards} builds the ARD. Verified locally (cards 0.6.1):

cards::ard_continuous(pharmaverseadam::adsl, by = "TRT01A", variables = "AGE")
# columns: group1, group1_level, variable, context, stat_name, stat_label, stat,
#          fmt_fun, warning, error
#   group1 group1_level variable    context stat_name stat_label     stat fmt_fun
# 1 TRT01A      Placebo      AGE continuous         N          N       86       0
# 2 TRT01A      Placebo      AGE continuous      mean       Mean  75.2093       1
# 3 TRT01A      Placebo      AGE continuous        sd         SD 8.590167       1

The warning / error list-columns are the built-in quality-evidence channel — every statistic carries its own condition record. That is an unusually good fit for open.csr's "test evidence for every component" requirement.

cards generators [verified locally]: ard_continuous, ard_categorical, ard_dichotomous, ard_hierarchical, ard_hierarchical_count, ard_stack, ard_stack_hierarchical, ard_stack_hierarchical_count, ard_strata, ard_pairwise, ard_missing, ard_total_n, ard_attributes, ard_complex, ard_identity, ard_formals.

cardx adds ~40 model/test ARDs [verified locally]: ard_stats_t_test, ard_stats_anova, ard_stats_chisq_test, ard_stats_fisher_test, ard_survival_survfit, ard_survival_survdiff, ard_survival_survfit_diff, ard_incidence_rate, ard_emmeans_mean_difference, ard_regression, ard_smd_smd, ard_continuous_ci, ard_categorical_ci, etc. — i.e. KM medians, hazard ratios, LS-mean differences and exposure-adjusted incidence rates are all ARD-native.

gtsummary 2.3.0 consumes ARDs directly via tbl_ard_summary, tbl_ard_continuous, tbl_ard_wide_summary, tbl_ard_hierarchical, plus tbl_hierarchical, tbl_hierarchical_count (the AE-by-SOC/PT workhorses), tbl_survfit, tbl_split_by_rows, tbl_split_by_columns [verified locally]. gtsummary was refactored to use cards as its backend, so a table can emit its own ARD.

{siera} (Clymb Clinical, CRAN 0.5.6) closes the loop the other way: readARS() ingests ARS metadata (JSON or xlsx) and auto-generates the R scripts that produce the ARD. https://clymbclinical.github.io/siera/ · https://github.com/clymbclinical/siera

Reference talks: Sjoberg, Harnessing CDISC's Emerging ARD Standard with {cards} and {gtsummary} — https://www.danieldsjoberg.com/ARD-RinPharma-talk-2024/ and the 2025 CDISC COSA spotlight https://www.danieldsjoberg.com/CDISC-COSA-Spotlight-ARD-gtsummary-2025/

9. R Consortium FDA submission pilots 1–7 — what was submitted and what FDA said

Hub: https://rconsortium.github.io/submissions-wg/ · 2026 status post: https://r-consortium.org/posts/submissions-wg-2026/

Pilot 1 outputs (the canonical four): Demographics & baseline characteristics; primary efficacy — ADAS-Cog(11) change from baseline to Week 24 (LOCF); a secondary efficacy table; Kaplan-Meier plot of time to first dermatologic event. https://rconsortium.github.io/submissions-pilot1/

FDA feedback themes from Pilot 3 (https://r-consortium.org/posts/news-from-r-submissions-working-group-pilot-3/):

1. "Clear ADRG documentation on computing environment, package dependencies, and expected warnings."
2. "Clear documentation on data processing rules and statistical method implementation."
3. "Good statistical practice in confirmatory trials, such as avoiding the possibility of p-hacking."

Cross-pilot lessons (Appsilon synthesis, https://www.appsilon.com/post/r-in-fda-submissions):

• ADRGs for R submissions need substantially more detail than SAS equivalents, including step-by-step Windows installation instructions. "You can never have too much documentation."
• Reviewers are on Windows. "You better work on Windows."
• Reviewers prefer CRAN packages over sponsor-internal ones; a custom package "just opens up another risk." (Pilots still shipped one — pilot3utils_0.0.2.zip.)
• Reviewers were content to spend 5–10 min installing from an renv.lock, and in fact preferred compiling packages themselves over receiving binaries.
• Interactivity is acceptable; interactive inference is not.

10. eCTD packaging — the exact folder shape (verified from the pilot repos)

Verified via GitHub API, RConsortium/submissions-pilot3-adam-to-fda (65 paths):

m1/us/cover-letter.pdf
m1/us/report-tlf-pilot3.pdf
m1/us/response-FDA-IR-pilot3.pdf
m5/sap-cdiscpilot01.pdf
m5/datasets/rconsortiumpilot3/tabulations/sdtm/{ae,cm,dm,ds,ex,lb,mh,qs,...}.xpt
m5/datasets/rconsortiumpilot3/tabulations/sdtm/{define.xml, define.pdf, blankcrf.pdf}
m5/datasets/rconsortiumpilot3/analysis/adam/datasets/{adsl,adae,adlbc,adadas,adtte}.xpt
m5/datasets/rconsortiumpilot3/analysis/adam/datasets/{define.xml, define2-0-0.xsl, adrg.pdf}
m5/datasets/rconsortiumpilot3/analysis/adam/datasets/adam-pilot-3.xlsx     <- specs
m5/datasets/rconsortiumpilot3/analysis/adam/programs/{adsl,adae,adlbc,adadas,adtte}.r
m5/datasets/rconsortiumpilot3/analysis/adam/programs/{tlf-demographic,tlf-efficacy,
                                                      tlf-kmplot,tlf-primary}.r
m5/datasets/rconsortiumpilot3/analysis/adam/programs/pilot3utils_0.0.2.zip
m5/datasets/rconsortiumpilot3/analysis/adam/programs/renv-lock.txt

Pilot 1's package is the same shape with a wider ADaM set (adsl, adae, adadas, adcibc, adlbc, adlbcpv, adlbh, adlbhpv, adlbhy, adnpix, adtte, advs) and r0pkg.txt.

Notable conventions to copy: .r (not .R) lowercase program files, renv.lock renamed to renv-lock.txt (eCTD file-type restrictions), the internal package shipped as a .zip, adrg.pdf living next to the datasets, and define.xml accompanied by its XSL stylesheet. eCTD rules also cap total path length at 180 characters and require lowercase names with no spaces or special characters.

11. ADRG and analysis metadata expectations

• FDA's Study Data Technical Conformance Guide (current version v6.0, March 2025; Technical Specifications Document reissued Dec 2025) recommends an SDRG for clinical/nonclinical tabulation data and an ADRG for clinical analysis data, submitted as adrg.pdf. Landing page: https://www.fda.gov/industry/fda-data-standards-advisory-board/study-data-standards-resources (Direct PDF fetches from fda.gov/hhs.gov were 403/404 from this environment — retrieve manually before quoting section numbers.)
• PHUSE ADRG template — 7 required sections: (1) Introduction, (2) Protocol Description, (3) Analysis Considerations Related to Multiple Analysis Datasets, (4) Analysis Data Creation and Processing Issues, (5) Analysis Dataset Descriptions, (6) Data Conformance Summary, (7) Submission of Programs. Template at https://phuse.global/Deliverables/1 ; PHUSE working group page: https://advance.hub.phuse.global/wiki/spaces/WEL/pages/26804405/
• A worked R-based ADRG is public from Pilot 4: https://rpodcast.quarto.pub/pilot4-webassembly-adrg/ — this is the single best model for what an R-generated ADRG looks like (computing environment, package inventory, expected warnings, execution steps).
• Analysis Results Metadata (ARM) is the define.xml v2.x extension describing the purpose of each analysis, the ADaM datasets and methods used — the bridge between define.xml (structure) and the TFL (result). Background: https://www.lexjansen.com/phuse-us/2018/ds/DS10.pdf . CDISC ARS v1.0 is the modern, machine-actionable successor.
• PHUSE OSDocuMeta working group is actively building open-source metadata documentation tooling: https://phuse-org.github.io/OSDocuMeta/references/problem_statement.html

12. Gaps and openings for open.csr

1. Nobody owns the document. Every package above stops at the individual display. Assembling TFLs + prose into an ICH E3 CSR is manual/Word-based today. That is open.csr's component (3) and it is genuinely unoccupied.
2. Display metadata is not standardized across the two stacks. tfrmt's JSON is the only serializable display spec; chevron/tern encode display in R code. A neutral, version-controlled TFL definition format (ARS for the analysis + tfrmt-style JSON for the display) is a real contribution.
3. Test evidence is ad hoc. cards' warning/error columns and qcthat-style conventions exist, but no one publishes a per-TFL evidence page. The safety.viz requirements-matrix + evidence model transfers directly.
4. pharmaverseadam has no CDISCPILOT01 efficacy dataset — plan for it (derive an adadas from pharmaversesdtm::qs, or vendor the R Consortium pilot's adadas.xpt).
5. A static GitHub Pages demo is achievable — Pilot 4 proved FDA reviewers prefer browser-only WebAssembly delivery, which is exactly what a Pages site is.

Sources

All URLs are cited inline in the sections above. Primary hubs, for convenience: TLG Catalog https://insightsengineering.github.io/tlg-catalog/ · chevron https://insightsengineering.github.io/chevron/ · pharmaverse TLGs https://pharmaverse.org/e2eclinical/tlg/ · pharmaverseadam https://pharmaverse.github.io/pharmaverseadam/ · cardinal https://pharmaverse.github.io/cardinal/ · FDA ST&F https://www.fda.gov/drugs/development-resources/standard-safety-tables-and-figures-stfs · CDISC ARS v1.0 https://www.cdisc.org/standards/foundational/analysis-results-standard/analysis-results-standard-v1-0 · r2rtf https://merck.github.io/r2rtf/ · tfrmt https://gsk-biostatistics.github.io/tfrmt/ · siera https://clymbclinical.github.io/siera/ · r4csr https://r4csr.org/tlf-overview.html · R Submissions WG https://rconsortium.github.io/submissions-wg/ · Pilot 4 ADRG https://rpodcast.quarto.pub/pilot4-webassembly-adrg/ · PHUSE deliverables https://phuse.global/Deliverables/1 · FDA study data standards https://www.fda.gov/industry/fda-data-standards-advisory-board/study-data-standards-resources