R TLF Pseudocode Library 217 TLF SCRIPTS

ADaM-based R pseudocode for every Table, Listing, and Figure in an early-phase clinical-pharmacology study — a complete separate set per study design, on the pharmaverse stack (dplyr/tidyr/rtables/ggplot2/emmeans/lme4).

Status — illustrative pseudocode / skeleton, NOT validated production software. These scripts show derivation and layout logic for spec writing and double-programming. Every reported number comes only from validated tools (Phoenix WinNonlin, Pinnacle 21, EDC) per SOP — nothing here originates a reported value, and no number is ever produced by an LLM.

R twin of ../sas_tlf_pseudocode/. The higher-level pharmaverse path (admiral for ADaM, tern for pre-built TLF layouts) is the production route; this library shows the underlying logic explicitly. R runs laptop-only per SOP.

▶ Run it 📦 Stack 🗂 Designs 📋 TLF inventory 🆘 Troubleshooting 💻 Browse the code

Before you start, you'll need: R 4.2+ (with Rscript on your PATH) · the pharmaverse-style packages each script library()s — dplyr, tidyr, stringr, purrr, tibble, rtables, ggplot2, emmeans, lme4, broom, broom.mixed (plus haven to read .xpt/.sas7bdat ADaM) · read access to your study's ADaM datasets (ADSL, ADAE, ADLB, ADVS, ADEG, ADPC, ADPP, ADPD, ADIS, ADEX), CDISC-conformant · about 5 minutes to point one script at your data. No internet needed once the packages are installed. These are illustrative pseudocode skeletons: setup() defaults to placeholder paths (/data/adam) and assumes ADaM frames are loaded as tibbles — you wire in your real read step before a script produces output.

Run it

Each script source()s 00_setup_helpers.R (it expects to be run with the design folder as the working directory, e.g. parallel/), calls setup(), gets dv <- design_vars("<DESIGN>"), then builds one TLF. Install the packages once, then Rscript a program.

① install packages (once) · ② point setup() / your read step at your ADaM · ③ Rscript the program from inside its design folder.

① Install the packages once (CRAN; needs internet for this step only):

Rscript -e 'install.packages(c("dplyr","tidyr","stringr","purrr","tibble","rtables","ggplot2","emmeans","lme4","broom","broom.mixed","haven"))'

② Run a program — cd into the design folder first so the relative source("../00_setup_helpers.R") resolves, then:

cd parallel
Rscript t_demographics.R

Quick-start checklist — click each as you finish it:

Confirm Rscript --version works (R 4.2+ on your PATH)
Install the packages once (the install.packages(...) line above)
Edit setup() / the read step so it loads your ADaM (e.g. haven::read_xpt) instead of the placeholder paths
Pick the right design folder (parallel/, crossover/, single_sequence/, sad/, mad/) and the TLF script you need
cd into that design folder, then Rscript <script>.R
Compare the output against the validated tool — this is pseudocode, never the reported source

Parse-only sanity check. Every script in the library parses under Rscript as shipped. To confirm a script is syntactically intact without touching data, parse it without executing:

Rscript -e 'invisible(parse("parallel/t_demographics.R")); cat("parses OK\n")'

Stack

dplyr/tidyr/purrr (derivation) · rtables (regulatory table layout) · ggplot2 (figures) · emmeans + lme4/nlme (mixed models / BE) · broom/broom.mixed (tidy model output).

Designs (one folder each — full set per design)

Folder	Design	Key analysis differences
`parallel/`	Parallel-group	Between-group; treatment = `TRT01A`/`TRT01AN`
`crossover/`	Crossover (2×2, Williams)	Within-participant; by `APERIOD`/`TRTSEQP`; BE/ANOVA = `lmer` + `emmeans`, GMR + 90% CI
`single_sequence/`	Single-/fixed-sequence (e.g. DDI)	By `APERIOD`; fixed order, no randomized sequence; ratio + 90% CI vs reference period
`sad/`	Single Ascending Dose	Parallel cohorts (dose = treatment); dose-proportionality (power model, `lm`)
`mad/`	Multiple Ascending Dose	Parallel cohorts, repeated dosing; accumulation ratio (Rac), steady-state assessment

Design is wired via design_vars("...") in 00_setup_helpers.R (returns trtvar/trtnvar, seqvar/seqvarn, byperiod). Programs differ only where the statistics genuinely differ (PK comparison, by-period/by-sequence safety, dose escalation); safety Tables/Listings are structurally identical, re-pointed to the design's variables.

Per-script flow

source 00_setup_helpers.R→ setup()→ dv <- design_vars("<DESIGN>")→ read ADaM→ derive (dplyr/tidyr)→ layout (rtables / ggplot2)→ one TLF output

TLF inventory (each = one script per design)

Disposition / Demographics / Exposure (ADSL, ADEX, ADCM, ADMH)

t_disposition · t_demographics · t_baseline_characteristics · t_exposure · t_protocol_deviations · t_prior_con_meds · t_medical_history · l_disposition

Safety — AE (ADAE)

t_ae_overview · t_ae_by_soc_pt · t_ae_by_severity · t_ae_by_relationship · t_ae_sae_death_withdrawal · l_ae · l_sae_death

Safety — Labs (ADLB)

t_lab_summary · t_lab_shift · t_lab_marked_abnormal · f_lab_lft_scatter · f_lab_change · l_lab_abnormal

Safety — Vitals / ECG (ADVS, ADEG)

t_vitals_summary · f_vitals_change · l_vitals · t_ecg_summary · t_ecg_qtc_categorical · f_qtc_change · l_ecg

PK (ADPC, ADPP)

t_pk_conc_summary · l_pk_conc · f_pk_conc_mean (linear+semilog) · f_pk_conc_individual · t_pk_param_summary · l_pk_param · f_pk_param_boxplot · design-specific t_dose_proportionality (SAD) · t_accumulation (MAD) · t_be_anova (crossover / single-seq)

PD / Biomarkers (ADPD)

t_pd_summary · f_pd_change · f_pkpd_overlay · l_pd

Immunogenicity (ADIS)

t_ada_summary · t_ada_impact_pk · l_ada

Naming & conventions

t_ table · l_ listing · f_ figure. One script → one output. Each script source()s 00_setup_helpers.R, calls setup(), gets dv <- design_vars("<DESIGN>"), then the TLF body.

House rules baked in (the same correctness lessons as the SAS twin)

Geometric PK stats on the log scale: geomean = exp(mean(log(x))), geocv = 100*sqrt(exp(var(log(x)))-1) — never exp(mean(raw)).
Tmax = Median (Min, Max) only.
Safety/AE counts = distinct participants (n_distinct(USUBJID)), not event rows; % denominator = population N per column from bign().
Per-period denominators come from a period-bearing source (ADEX/ADIS/ADPD), NOT one-row-per-participant ADSL.
Single-sequence participant-level tables key on dv$seqvar/dv$seqvarn; period tables use dv$byperiod.
Lab "worst" category = furthest from NORMAL, with missing-limit guards (!is.na(A1HI) etc.).
No "eDISH" / "Hy's Law" terminology — neutral liver-safety scatter ("Max ALT vs Total Bilirubin, ×ULN").

ADaM variables relied on (no re-derivation)

Treatment TRT01A/TRT01AN (parallel) · TRTA/TRTAN+APERIOD/APERIODC+TRTSEQP/TRTSEQPN (crossover/seq); populations SAFFL/PKFL/ITTFL; analysis PARAMCD/AVAL/CHG/PCHG/BASE/AVISIT/ATPTN/ANRIND/BNRIND/A1HI/A1LO/R2ULN; AE TRTEMFL/AESOC/AEDECOD/AESEVN/AREL/AESER/AESDTH/AEACN; PK ADPC AVAL=conc, ADPP PARAMCD=Cmax/Tmax/AUClast/AUCinf/CL/Vz/Rac.

Status

Complete: 217 TLF scripts across all five design folders + 00_setup_helpers.R. Built mirroring the SAS twin with all its QC fixes pre-applied, then re-QC'd (6 family reviewers → per-finding verification: 24 findings, 14 confirmed P0/P1 fixed, 0 false-positives), then a deterministic sweep. All 218 scripts parse under Rscript, design_vars() matches every folder, geometric PK stats compute on the log scale, AE/safety counts are distinct participants, single-sequence participant-level tables key on dv$seqvar, period tables denominate from a period-bearing source, exposure uses ADaM AVAL/TRTDURD/NDOSES (no *DTC arithmetic), and the library is free of eDISH/Hy's Law terminology.

Pseudocode for spec / double-programming. Reported numbers still come from validated tools per SOP; an LLM never sources a number.

Troubleshooting

Symptom	Why	Fix
`Error in library(rtables) : there is no package called 'rtables'` (or `dplyr`, `emmeans`, `lme4`…)	The pharmaverse packages each script `library()`s aren't installed in this R library.	Run the `install.packages(...)` line in Run it once. Needs internet for that step. Confirm with `Rscript -e 'library(rtables)'`.
`cannot open file '../00_setup_helpers.R': No such file or directory`	You ran the script from the wrong directory; the `source("../00_setup_helpers.R")` path is relative to the design folder.	`cd` into the design folder (`parallel/`, `crossover/`, etc.) first, then `Rscript <script>.R` — so `../` points back at the shared helper.
`object 'adam' not found` / `object 'adsl' not found`	These are pseudocode skeletons: `setup()` uses placeholder paths and assumes the ADaM frames are already loaded as tibbles — nothing actually reads your data yet.	Wire in your real read step (e.g. `adsl <- haven::read_xpt(file.path(env$adam,"adsl.xpt"))`) and point `adam`/`out` at your study's ADaM and output paths before expecting output.
Table runs but every `%` / count looks wrong or empty	Population flag or treatment variable doesn't match your ADaM, so `bign()` denominators or `SAFFL=="Y"` filters drop everything.	Confirm your ADaM carries the variables the library relies on (`SAFFL`/`PKFL`/`ITTFL`, `TRT01A`/`TRTA`, `USUBJID`); for crossover/single-seq check `design_vars()` matches your `APERIOD`/`TRTSEQP` naming.
Wrong `design_vars()` branch (e.g. crossover script keyed on `TRT01A`)	The `design_vars("<DESIGN>")` argument in the script header doesn't match the folder you copied it from.	Use the script from the matching design folder; the argument must be `PARALLEL`/`CROSSOVER`/`SINGLESEQ`/`SAD`/`MAD` for that folder.
Geometric mean / CV% looks identical to the arithmetic mean	A modified copy is computing `exp(mean(raw))` instead of working on the log scale.	Keep `pkstats()`: `geomean = exp(mean(log(x)))`, `geocv = 100*sqrt(exp(var(log(x)))-1)`; filter `AVAL > 0` first. Tmax stays Median (Min, Max).
`boundary (singular) fit` warning from `t_be_anova` (crossover / single-seq)	`lmer` can't estimate a random effect (too few sequences/periods or a near-degenerate design) — expected on small or illustrative data.	It's a warning, not a failure. On real BE data confirm the model spec against the SAP; the reported GMR + 90% CI still comes from the validated tool, not this skeleton.
Numbers from a script land in a regulatory deliverable	This library is illustrative pseudocode for spec writing / double-programming — it is not validated and never the reported source.	Reported and regulated numbers come only from validated tools (Phoenix WinNonlin, Pinnacle 21, EDC) per SOP. An LLM never sources a number.

Browse the code

Shared setup

00_setup_helpers.Rsetup() · design_vars() · bign() helpers

parallel/ — Parallel-group 42 SCRIPTS

R TLF Pseudocode Library — 218 R scripts (217 TLF + 00_setup_helpers.R) across parallel / crossover / single_sequence / sad / mad. Pharmaverse-style pseudocode for spec writing and double-programming; not validated software. Reported numbers come only from validated tools per SOP — never from an LLM. Participant-level analysis throughout.