sdtm-oak

name	sdtm-oak
description	Derives CDISC SDTM domains from raw clinical (EDC/eCRF) data using the {sdtm.oak} R package. Use when a user needs to map raw study data to SDTM Events (AE, CM, MH), Findings (VS, LB, EG), or Interventions (EX) domains following the sdtm.oak algorithm framework. Produces executable, submission- ready R code with controlled terminology recoding, ISO 8601 date derivation, sequence numbering, and study day calculation.
license	MIT
metadata	{"author":"pharma-skills contributors","version":"0.1","pharmaverse":"true"}
compatibility	Requires R with sdtm.oak (>= 0.2.0), dplyr, and tibble installed. Requires raw EDC/eCRF data and a controlled terminology (CT) specification CSV. Designed for use in a GxP-compliant environment.

sdtm-oak

Derives CDISC SDTM domains from raw clinical data using the {sdtm.oak} algorithm framework. Outputs executable R code with full derivation traceability.

See references/oak-functions.md for the full function reference.

---

Inputs

Before generating code, confirm:

Input	Required	Notes
Raw EDC dataset	Yes	e.g. ae_raw, vs_raw — raw CRF form data
CT specification	Yes	CSV in CDISC codelist format; load via read_ct_spec()
Domain specification	Yes	Variable list, CT codelists per variable, date formats
DM domain	For study day / BLFL	Provides RFSTDTC for derive_study_day() and derive_blfl()

Always inspect the raw dataset first — raw column names vary by EDC system and study. Print names(raw_dat) and head(raw_dat) before writing any derivations.

---

Core algorithms

sdtm.oak provides six mapping algorithms. Choose based on whether the target variable has controlled terminology (CT) and whether the value is derived from raw data or hardcoded.

Algorithm	CT?	Source	Use for
assign_no_ct()	No	Raw column	Free-text variables: AETERM, CMTRT, VSORRES
assign_ct()	Yes	Raw column	CT-mapped from raw: AESEV, AESER, SEX, RACE
hardcode_no_ct()	No	Fixed value	Study-constant free-text: STUDYID, custom flags
hardcode_ct()	Yes	Fixed value	Domain constants validated against CT: DOMAIN
assign_datetime()	—	Raw date col(s)	Any --DTC variable: AESTDTC, VSDTC, EXSTDTC
condition_add()	—	Condition expr	Gate any of the above to a row subset

All six functions share the same id_vars join key (default: oak_id_vars()) and the same tgt_dat pipe pattern — pass the growing SDTM dataset as tgt_dat to accumulate variables.

---

Workflow

Follow these steps in order. Write code section by section, not as a single block.

Step 1 — Setup and data inspection

library(sdtm.oak)
library(dplyr)

# Load raw data — replace with actual source
ae_raw <- <load raw AE data>   # e.g. sdtm.oak::ae_raw for the package example

# ALWAYS inspect before writing derivations
cat("Columns:\n"); print(names(ae_raw))
cat("Rows:", nrow(ae_raw), "\n")
print(head(ae_raw, 3))

Step 2 — Generate oak ID variables

generate_oak_id_vars() adds three key columns used as join keys throughout all subsequent derivations. Call this once on the raw dataset.

# REVIEW: Set pat_var to the column holding the subject/patient identifier
#   in this raw dataset. Set raw_src to a stable label for the CRF form.
ae_oak <- generate_oak_id_vars(
  raw_dat = ae_raw,
  pat_var = "patient_number",   # confirm column name from Step 1 inspection
  raw_src = "AE_FORM"           # stable label for this raw source
)
# Adds: oak_id (row key), raw_source (form label), patient_number (subj ID)

Step 3 — Load controlled terminology

# REVIEW: Replace path with the study CT spec CSV.
#   For the sdtm.oak package example data, use read_ct_spec_example().
ct_spec <- read_ct_spec_example()   # or: read_ct_spec("path/to/ct_spec.csv")

# Validate before use
assert_ct_spec(ct_spec)

Step 4 — Hardcode domain constants

Fixed values that apply to every record in the domain. Use hardcode_ct() for values validated against CT (DOMAIN); use hardcode_no_ct() for free-text constants.

ae_domain <- ae_oak |>
  # DOMAIN is a CT-controlled variable — use hardcode_ct
  hardcode_ct(
    tgt_var  = "DOMAIN",
    tgt_val  = "AE",
    raw_dat  = ae_raw,
    raw_var  = "AETERM",   # presence filter: only rows with a non-NA AE term
    ct_spec  = ct_spec,
    ct_clst  = "DOMAIN"
  )

Step 5 — Assign free-text variables (assign_no_ct)

Use for variables with no CT restriction — raw text carried directly.

ae_domain <- ae_domain |>
  assign_no_ct(
    tgt_var = "AETERM",
    raw_dat = ae_raw,
    raw_var = "ae_term"   # REVIEW: confirm raw column name
  ) |>
  assign_no_ct(
    tgt_var = "AELOC",
    raw_dat = ae_raw,
    raw_var = "ae_location"
  )

Step 6 — Assign CT-mapped variables (assign_ct)

Use for variables whose values must be recoded to CDISC controlled terminology. Supply ct_clst matching the codelist name in your CT spec.

# REVIEW: Confirm ct_clst names match the codelist_code column in ct_spec.
#   Wrong ct_clst silently returns the uppercased raw value — verify outputs.
ae_domain <- ae_domain |>
  assign_ct(
    tgt_var = "AESEV",
    raw_dat = ae_raw,
    raw_var = "severity",
    ct_spec = ct_spec,
    ct_clst = "AESEV"
  ) |>
  assign_ct(
    tgt_var  = "AESER",
    raw_dat  = ae_raw,
    raw_var  = "serious_ae",
    ct_spec  = ct_spec,
    ct_clst  = "NY"
  ) |>
  assign_ct(
    tgt_var  = "AEREL",
    raw_dat  = ae_raw,
    raw_var  = "causality",
    ct_spec  = ct_spec,
    ct_clst  = "AEREL"
  ) |>
  assign_ct(
    tgt_var  = "AEOUT",
    raw_dat  = ae_raw,
    raw_var  = "outcome",
    ct_spec  = ct_spec,
    ct_clst  = "AEOUT"
  )

Step 7 — Assign datetime variables (assign_datetime)

Use for all --DTC variables. Never use as.Date(), as.POSIXct(), or string manipulation for SDTM dates — always use assign_datetime().

# REVIEW: raw_fmt must exactly match the date format in the raw data.
#   Use "y-m-d" for ISO (2024-03-15), "d/m/y" for European (15/03/2024),
#   "m/d/y" for US (03/15/2024). Check format from Step 1 inspection.
#   Supply a list of alternatives if the format is inconsistent across records.
ae_domain <- ae_domain |>
  assign_datetime(
    tgt_var = "AESTDTC",
    raw_dat = ae_raw,
    raw_var = "onset_date",
    raw_fmt = "d-m-y"      # REVIEW: confirm raw date format
  ) |>
  assign_datetime(
    tgt_var = "AEENDTC",
    raw_dat = ae_raw,
    raw_var = "resolution_date",
    raw_fmt = "d-m-y"      # REVIEW: confirm raw date format
  )

For combined date-time (e.g. separate date and time columns):

ae_domain <- ae_domain |>
  assign_datetime(
    tgt_var = "AESTDTC",
    raw_dat = ae_raw,
    raw_var = c("onset_date", "onset_time"),   # two columns
    raw_fmt = c("d-m-y", "H:M")               # one format per column
  )

Step 8 — Conditional derivations (condition_add)

Use condition_add() to restrict a derivation to a subset of records. Wrap the target dataset in condition_add(), then pass it as tgt_dat.

# REVIEW: condition_add() criteria must reflect the study protocol.
#   Document the business rule the condition implements.
ae_domain <- ae_domain |>
  # Example: derive AEDTHFL only for fatal outcome records
  (\(dat) assign_ct(
    tgt_dat = condition_add(dat, AEOUT == "FATAL"),
    tgt_var = "AEDTHFL",
    raw_dat = ae_raw,
    raw_var = "death_flag",
    ct_spec = ct_spec,
    ct_clst = "NY"
  ))()

Step 9 — Add STUDYID and USUBJID

Derive subject-level identifiers after domain variables are built.

ae_domain <- ae_domain |>
  hardcode_no_ct(
    tgt_var = "STUDYID",
    tgt_val = "CDISCPILOT01",   # REVIEW: replace with actual study ID
    raw_dat = ae_raw,
    raw_var = "patient_number"
  ) |>
  assign_no_ct(
    tgt_var = "USUBJID",
    raw_dat = ae_raw,
    raw_var = "patient_number"   # REVIEW: confirm USUBJID construction rule
  )

Step 10 — Study day derivation

Requires DM domain (provides RFSTDTC).

# REVIEW: Confirm which DTC variable is the reference date for this domain
#   (RFSTDTC for most event domains; RFXSTDTC for findings relative to dosing).
ae_domain <- derive_study_day(
  sdtm_in      = ae_domain,
  dm_domain    = dm,
  tgdt         = "AESTDTC",
  refdt        = "RFSTDTC",
  study_day_var = "AESTDY"
) |>
  derive_study_day(
    sdtm_in      = _,
    dm_domain    = dm,
    tgdt         = "AEENDTC",
    refdt        = "RFSTDTC",
    study_day_var = "AEENDY"
  )

Step 11 — Sequence number

ae_domain <- derive_seq(
  sdtm_in = ae_domain,
  tgt_var = "AESEQ"
)

Step 12 — Supplemental domain (SUPP--)

If the study collects non-standard variables, split them to SUPPAE.

# REVIEW: Confirm which variables belong in SUPPAE vs the main domain.
#   Non-standard variables must not appear in the parent domain.
result <- generate_sdtm_supp(
  sdtm_dataset  = ae_domain,
  idvar         = "AESEQ",
  supp_qual_info = supp_spec,    # dataframe: QNAM, QLABEL, QORIG per variable
  qnam_var      = "QNAM",
  label_var     = "QLABEL",
  orig_var      = "QORIG"
)
ae_final   <- result$sdtm
suppae     <- result$supp

Step 13 — Final checks

# Required SDTM variables for AE domain
required_vars <- c("STUDYID", "DOMAIN", "USUBJID", "AESEQ",
                   "AETERM", "AESTDTC")
missing_vars <- setdiff(required_vars, names(ae_final))
if (length(missing_vars) > 0) {
  stop("Missing required AE variables: ", paste(missing_vars, collapse = ", "))
}

# No duplicate sequence numbers
stopifnot(
  ae_final |>
    count(STUDYID, USUBJID, AESEQ) |>
    filter(n > 1) |>
    nrow() == 0
)

cat("AE domain: ", nrow(ae_final), "records,",
    n_distinct(ae_final$USUBJID), "subjects\n")

---

Findings domains (VS, LB, EG)

Findings domains (one record per subject per test per visit) follow a different stacking pattern — derive each TESTCD separately, then bind_rows().

# REVIEW: Each parameter block must align with the CT codelist for VSTESTCD.
#   Stack only parameters in scope for this study per the CRF and SAP.

# Parameter 1: Systolic Blood Pressure
sysbp <- generate_oak_id_vars(vs_raw, pat_var = "patient_number",
                               raw_src = "VS_FORM") |>
  hardcode_ct(tgt_var = "VSTESTCD", tgt_val = "SYSBP",
              raw_dat  = vs_raw, raw_var = "SYSBP_result",
              ct_spec  = ct_spec, ct_clst = "VSTESTCD") |>
  hardcode_no_ct(tgt_var = "VSTEST", tgt_val = "Systolic Blood Pressure",
                 raw_dat = vs_raw, raw_var = "SYSBP_result") |>
  assign_no_ct(tgt_var = "VSORRES", raw_dat = vs_raw, raw_var = "SYSBP_result") |>
  assign_no_ct(tgt_var = "VSORRESU", raw_dat = vs_raw, raw_var = "SYSBP_unit") |>
  assign_datetime(tgt_var = "VSDTC", raw_dat = vs_raw,
                  raw_var = "visit_date", raw_fmt = "d-m-y")

# Parameter 2: Diastolic Blood Pressure — same pattern, different raw_var
diabp <- generate_oak_id_vars(vs_raw, pat_var = "patient_number",
                               raw_src = "VS_FORM") |>
  hardcode_ct(tgt_var = "VSTESTCD", tgt_val = "DIABP", ...) |>
  ...

# Stack all parameters
vs_domain <- bind_rows(sysbp, diabp, pulse, weight, height, temp) |>
  hardcode_ct(tgt_var = "DOMAIN", tgt_val = "VS",
              raw_dat = vs_raw, raw_var = "patient_number",
              ct_spec = ct_spec, ct_clst = "DOMAIN") |>
  derive_seq(tgt_var = "VSSEQ")

For findings, also derive VSBLFL (baseline flag) when applicable:

# REVIEW: Confirm baseline visit name(s) from the protocol.
vs_domain <- derive_blfl(
  sdtm_in          = vs_domain,
  dm_domain        = dm,
  tgt_var          = "VSBLFL",
  ref_var          = "VSDTC",
  baseline_visits  = c("BASELINE", "DAY 1")   # REVIEW: protocol-specific
)

---

# REVIEW: annotation rules

Place a # REVIEW: comment whenever a derivation contains a protocol-specific decision that a QC reviewer must verify. Required locations:

Location	What to annotate
generate_oak_id_vars()	pat_var column name and raw_src label
Every assign_datetime()	raw_fmt format string — must match actual raw data
Every assign_ct()	ct_clst codelist name — wrong codelist silently miscodes
Every condition_add()	The business rule the condition implements
derive_blfl() baseline_visits	Protocol baseline visit definition
derive_study_day() refdt	Reference date choice (RFSTDTC vs RFXSTDTC)
Any hardcoded tgt_val	Confirm value is correct for this study

---

Common errors to avoid

• Using as.Date(), substr(), or format() on raw date columns instead of assign_datetime() — ISO 8601 partial date handling and unknown-date placeholders ("UN", "UNK") are only correctly handled by assign_datetime()
• Skipping generate_oak_id_vars() — all assign_* and hardcode_* functions require oak_id, raw_source, and patient_number columns to be present as join keys; the dataset will silently return wrong results without them
• Passing the wrong raw_dat to an assign_* call — raw_dat must always be the original raw dataset, not the growing tgt_dat; mixing them causes incorrect joins
• Using mutate() or rename() for variable mapping instead of assign_no_ct() — mutate bypasses the oak traceability framework and does not respect id_vars join semantics
• Using assign_no_ct() for CT-mapped variables — values will not be recoded to CDISC terminology; use assign_ct() whenever a codelist applies
• Using assign_ct() with the wrong ct_clst — no error is raised; unmatched values are silently uppercased; always verify output distribution after derivation
• Skipping assert_ct_spec() before first use — malformed CT specs produce silent miscoding
• Not calling derive_seq() — --SEQ is required for all SDTM domains; do not derive it manually with row_number()
• Mixing up hardcode_ct() vs hardcode_no_ct() for DOMAIN — DOMAIN must use hardcode_ct() (validated against the DOMAIN codelist); using hardcode_no_ct() will accept any string
• For findings domains: building the full stacked dataset with bind_rows() before adding common variables (DOMAIN, USUBJID, study day) — add per-test variables in each parameter block, then add common variables after stacking

---

Output checklist

Before returning code, verify:

• Raw dataset inspected (names() + head()) before first derivation
• generate_oak_id_vars() called first with correct pat_var and raw_src
• CT spec loaded and validated with assert_ct_spec()
• DOMAIN hardcoded with hardcode_ct() — not hardcode_no_ct()
• All --DTC variables derived with assign_datetime() — not as.Date()
• Every assign_datetime() has a # REVIEW: on raw_fmt
• Every assign_ct() has a # REVIEW: on ct_clst
• Every condition_add() has a # REVIEW: on the business rule
• raw_dat in each assign_* call is the original raw dataset
• derive_seq() called to generate --SEQ
• derive_study_day() called for event/finding date study days (requires DM)
• Required domain variables present (stopifnot() or stop() guard)
• No duplicate --SEQ values (stopifnot() check)
• Findings domains: per-test blocks stacked with bind_rows() before adding common variables