Stage 3: data standards

SDTM-like inputs, ADaM with admiral, specification checks and XPT

Executed hereNot covered: define.xml, ADRG

Synthetic data only: simulated for this worked example; not real trial data and not a regulatory deliverable.

SDTM-like source domains are simulated by 03_adam/01_sdtm.R, which reuses the sim_lifecycle/ priors and one-compartment model (N = 40, seed 2026) and takes each participant's sequence from the stage 1 randomization schedule. ADaM datasets are derived with admiral 1.5.0 following its ad_adsl.R, ad_adpc.R, ad_adpp.R and ad_adae.R templates, checked against the specification with metacore 0.3.0 and metatools 0.3.0, and written as SAS XPORT v5 with xportr 0.6.0.

SDTM-like inputs

Domain	Records	Content	File
DM	40	Demographics and reference dates	dm.csv
EX	79	One record per dose (Test or Reference)	ex.csv
PC	1185	Plasma concentrations at actual sampling times; BLQ as text	pc.csv
AE	32	Adverse events from the sim_lifecycle priors	ae.csv
VS	80	Screening height and weight	vs.csv
DS	120	Consent, randomization and disposition	ds.csv
PP	1027	PK parameters from the NCA in stage 4 (derived, then fed back into ADPP)	pp.csv

Files: dm.csv, ex.csv, pc.csv, ae.csv, vs.csv, ds.csv, pp.csv, scenarios.json. No LB domain: no output in this example needs laboratory data.

Documented scenarios that exercise the SAP rules

Participant	Scenario
SIMBE01-101-017	withdraws consent after Period 1 (no Period 2 dose)
SIMBE01-101-029	vomits 45 minutes after the Period 2 dose
SIMBE01-101-008	measurable Period 2 pre-dose concentration at 2.5% of that period's Cmax (below the 5% rule)
SIMBE01-101-035	12 h sample in Period 1 not collected
SIMBE01-101-012	4 h sample in Period 1 drawn 25 minutes late
SIMBE01-101-024	6 h sample in Period 2 reported below the LLOQ between quantifiable samples (a simulated sample-handling problem; no reassay)

ADaM datasets

Dataset	Records	Variables	Template
ADSL	40	54	ad_adsl.R pattern; derive_vars_period() for the crossover
ADPC	1264	52	ad_adpc.R, step by step (adaptations listed below)
ADPP	1027	34	ad_adpp.R; SAP exclusion rules
ADAE	32	31	ad_adae.R pattern; period from create_period_dataset()

adsl.xptSAS XPORT v5 (xportr_write) csv Dataset-JSON
adpc.xptSAS XPORT v5 (xportr_write) csv Dataset-JSON
adpp.xptSAS XPORT v5 (xportr_write) csv Dataset-JSON
adae.xptSAS XPORT v5 (xportr_write) csv Dataset-JSON

Template adaptations in ADPC (each one commented in `03_adam/03_adpc.R`)

The analyte name joins PC and EX, because EXTRT carries the formulation (SIMDRUG TEST or SIMDRUG REFERENCE).
A pre-dose sample belongs to the dose that follows it (ARRLT below 0, NRRLT = 0), as the template's DTYPE COPY records do for repeated dosing.
BLQ values are 0 in AVAL (SAP, ICH M13A); the template's LLOQ/2 records are not created.
The sample that was not collected stays in ADPC with ANL01FL not Y, so a listing can show it.
derive_var_nfrlt() and convert_xxtpt_to_hours() are labeled EXPERIMENTAL in admiral 1.5.0; NFRLT is re-derived by plain arithmetic in stage 5 (Q4).

Specification

adam_spec.xlsx uses the Pinnacle 21 specification layout (Study, Datasets, Variables, ValueLevel, WhereClauses, Codelists, Dictionaries, Methods, Comments, Documents), which metacore::spec_to_metacore() reads. adam_spec.json holds the same tables for scripts and diffs. Both come from one definition in 03_adam/spec/build_spec.R. Variable labels follow CDISC naming (for example USUBJID, "Unique Subject Identifier").

Metadata checks (executed)

metatools checks, length and mandatory checks, and an XPT read-back for each dataset.
dataset	check	pass	detail
ADSL	metatools::check_variables (strict)	pass	54 variables in data, 54 in spec
ADSL	metatools::check_ct_data	pass	all values in their codelists
ADSL	metatools::check_unique_keys	pass	STUDYID, USUBJID
ADSL	text values fit the spec lengths	pass	34 text variables
ADSL	mandatory variables populated	pass	48 mandatory variables
ADSL	XPT v5 read back: names, order and labels match the spec	pass	40 rows x 54 columns; dataset label 'Subject-Level Analysis Dataset'
ADPC	metatools::check_variables (strict)	pass	52 variables in data, 52 in spec
ADPC	metatools::check_ct_data	pass	all values in their codelists
ADPC	metatools::check_unique_keys	pass	STUDYID, USUBJID, ASEQ
ADPC	text values fit the spec lengths	pass	30 text variables
ADPC	mandatory variables populated	pass	35 mandatory variables
ADPC	XPT v5 read back: names, order and labels match the spec	pass	1264 rows x 52 columns; dataset label 'PK Concentrations Analysis Dataset'
ADPP	metatools::check_variables (strict)	pass	34 variables in data, 34 in spec
ADPP	metatools::check_ct_data	pass	all values in their codelists
ADPP	metatools::check_unique_keys	pass	STUDYID, USUBJID, APERIOD, PARAMCD
ADPP	text values fit the spec lengths	pass	24 text variables
ADPP	mandatory variables populated	pass	26 mandatory variables
ADPP	XPT v5 read back: names, order and labels match the spec	pass	1027 rows x 34 columns; dataset label 'PK Parameters Analysis Dataset'
ADAE	metatools::check_variables (strict)	pass	31 variables in data, 31 in spec
ADAE	metatools::check_ct_data	pass	all values in their codelists
ADAE	metatools::check_unique_keys	pass	STUDYID, USUBJID, ASEQ
ADAE	text values fit the spec lengths	pass	20 text variables
ADAE	mandatory variables populated	pass	30 mandatory variables
ADAE	XPT v5 read back: names, order and labels match the spec	pass	32 rows x 31 columns; dataset label 'Adverse Events Analysis Dataset'

Gap: not authored here. No define.xml and no Analysis Data Reviewer's Guide (ADRG). No Pinnacle 21 or CDISC CORE conformance run: the checks above test the datasets against this study's own specification. The source domains are SDTM-like (no trial design domains, no SUPP datasets). The Dataset-JSON v1.1 copies are a demonstration: FDA has not accepted Dataset-JSON as of Oct 2026, and SAS XPORT v5 remains the submission format.