cdisc-pilot-replication

End-to-end CDISC SDTM → ADaM → TLF submission package, replicating the FDA-reviewed 2007 CDISC Pilot Alzheimer's study (CDISCPILOT01). Built in parallel R ({admiral} + pharmaverse) and SAS 9.4, with PROC COMPARE (R↔SAS equivalence) and Pinnacle 21 Community (ADaMIG 1.3) validation harnesses wired into the build.

What this demonstrates

• CDISC ADaM dataset generation - ADSL, ADAE, ADLB, ADTTE, ADQSADAS, all conformant to ADaMIG 1.3
• Dual implementation (R + SAS) with a PROC COMPARE harness for R↔SAS numerical equivalence (criterion=1e-9)
• Submission-quality TLFs - 4 tables, 2 listings, 2 figures ({rtables}, {tern}, {rlistings}, survminer, ggplot2)
• MMRM primary efficacy analysis with ICH E9(R1) estimand specification
• define.xml v2.0 generated from variable-level metadata Excel specs
• Reproducibility - renv lockfile, GitHub Actions CI, end-to-end in ~20 seconds
• Regulatory writing - mock Statistical Analysis Plan + Analysis Data Reviewer's Guide (PHUSE template)

Repository contents

Deliverable	Path	Format
5 ADaM XPT v5 datasets	data/xpt/*.xpt	binary, committed
4 submission tables	outputs/tlfs/t_*.pdf	PDF
2 listings	outputs/tlfs/l_*.pdf	PDF
2 figures	outputs/tlfs/f_*.pdf	PDF
Define-XML 2.0	outputs/define/define.xml	XML
Statistical Analysis Plan	docs/sap/sap.pdf	PDF
Analysis Data Reviewer's Guide	docs/adrg/adrg.pdf	PDF
Pinnacle 21 validation harness	python/pinnacle21_parser.py, outputs/validation/	wired - run pending
PROC COMPARE harness	sas/compare_r_sas.sas	wired - run pending

Pipeline architecture

pharmaversesdtm        cdisc-org/sdtm-adam-pilot-project
  (DM, EX, DS, AE, LB)         (QS.xpt, gitignored in data/raw/)
        │                             │
        └──────────────┬──────────────┘
                       ▼
              R/01_adsl.R - R/05_adqsadas.R           sas/adsl.sas
              (admiral pharmaverse build)             (SAS 9.4 parallel build)
                       │                                       │
                       ▼                                       ▼
              data/adam/*.rds                         data/xpt/sas_adsl.xpt
                       │                                       │
                       ▼                                       │
              R/06_export_xpt.R                                │
              (xportr labels/lengths/formats)                  │
                       │                                       │
                       ▼                                       │
              data/xpt/*.xpt ◄──── PROC COMPARE ◄──────────────┘
                       │           (criterion 1e-9)
        ┌──────────────┼──────────────┐
        ▼              ▼              ▼
   R/07-09        R/10_define_xml  Pinnacle 21
   TLF gen        define.xml v2.0  ADaMIG 1.3
        │              │              │
        ▼              ▼              ▼
   outputs/tlfs/  outputs/define/ outputs/validation/
                                       │
                                       ▼
                               python/pinnacle21_parser.py
                                       │
                                       ▼
                               README badge / JSON

Reproduce

Quickstart (clean checkout, ~30 minutes including first-time package install):

git clone https://github.com/kbd0011/cdisc-pilot-replication.git
cd cdisc-pilot-replication

# 1. Restore the exact R package versions pinned in renv.lock
Rscript -e 'renv::restore()'

# 2. Install Python validators
pip install -r requirements.txt

# 3. Download external SDTM (qs.xpt, ~33MB)
make fetch-raw

# 4. Run the whole pipeline
make all

make all runs three phases:

make pipeline   # R/00_setup → 10_define_xml (~20s, all tests green)
make validate   # python validators + xpt round-trip + P21 parse
make docs       # Quarto render SAP + ADRG to PDF

The SAS parallel build is run manually on SAS OnDemand for Academics - see sas/README.md.

Current pipeline metrics

Dataset	Rows	Vars	XPT size
ADSL	306	39	254 KB
ADAE	1,191 (1,122 TE)	31	1.3 MB
ADLB	58,700	28	41 MB
ADTTE	254 (92 events)	17	154 KB
ADQSADAS	12,241	26	8.2 MB

Validation	Status
xpt::xpt_validate() on all 5	OK
XPT vs spec round-trip (python/xpt_metadata_check.py)	OK - all datasets pass
define.xml well-formedness (xmllint, validate_define.py)	OK
testthat suite	745/745 expectations pass (90 tests across 5 ADaM datasets + XPT export)
Pinnacle 21 ADaMIG 1.3	Pending P21 run - see outputs/validation/README.md
SAS PROC COMPARE	Pending ODA run - see sas/README.md

Documentation

• docs/sap/sap.pdf - Statistical Analysis Plan (MMRM, ICH E9(R1) estimand, KM/Cox, missing-data sensitivity)
• docs/adrg/adrg.pdf - Analysis Data Reviewer's Guide (PHUSE 2019-07-18 template)
• sas/README.md - SAS OnDemand for Academics workflow
• outputs/validation/README.md - Pinnacle 21 + XSD-validation workflow
• data/README.md - Source data + reproducibility notice
• metadata/README.md - Variable-level spec format

Layout

cdisc-pilot-replication/
├── R/                      pipeline scripts (00_setup → 10_define_xml + run_all)
├── sas/                    parallel ADSL + PROC COMPARE
├── python/                 define.xml / XPT / P21 validators
├── metadata/               variable-level Excel specs (141 vars across 5 datasets)
├── data/
│   ├── raw/                external SDTM (qs.xpt) - gitignored
│   ├── adam/               R-built ADaM .rds - gitignored
│   └── xpt/                XPT v5 submission deliverables
├── outputs/
│   ├── tlfs/               4 tables, 2 listings, 2 figures
│   ├── validation/         P21 + PROC COMPARE + JSON summaries
│   └── define/             define.xml + (user-supplied) define2-0-0.{xsd,xsl}
├── docs/
│   ├── sap/sap.pdf
│   └── adrg/adrg.pdf
└── tests/                  testthat (90 tests, 745 expectations)

License

MIT - see LICENSE.