Module 6 Theory — SDTM Programming with sdtm.oak

�️ Module 6 — SDTM Programming with sdtm.oak

🎯 Learning Objectives

By the end of this module, you will:

• Understand CDISC SDTM standards and domain structures
• Learn to use the sdtm.oak package for SDTM data creation
• Master metadata reading and domain specification
• Create standardized SDTM domains (DM, AE, VS, LB)
• Export datasets to XPT format for regulatory submission
• Use GitHub Copilot in RStudio to assist with SDTM programming

---

📚 1. Understanding CDISC SDTM

What is SDTM?

Study Data Tabulation Model (SDTM) is a standard for organizing and formatting clinical trial data for regulatory submission. It defines:

• Domain Structure: How data should be organized (DM, AE, VS, LB, etc.)
• Variable Names and Labels: Standardized naming conventions
• Controlled Terminology: Consistent coding of values
• Data Relationships: How domains connect via keys like USUBJID

Core SDTM Domains

Domain	Description	Key Variables
DM	Demographics	USUBJID, AGE, SEX, RACE, ARMCD
AE	Adverse Events	USUBJID, AETERM, AEDECOD, AESTDTC, AESEV
VS	Vital Signs	USUBJID, VSTESTCD, VSORRES, VSSTRESC, VSDTC
LB	Laboratory	USUBJID, LBTESTCD, LBORRES, LBSTRESC, LBDTC
EX	Exposure	USUBJID, EXTRT, EXDOSE, EXSTDTC, EXENDTC

---

🛠️ 2. Introduction to sdtm.oak

The sdtm.oak package provides tools for creating SDTM-compliant datasets in R, offering:

• Metadata Management: Read and process SDTM metadata specifications
• Domain Creation: Functions to build standard SDTM domains
• Data Validation: Check compliance with SDTM standards
• XPT Export: Export to SAS transport format for submission

Installation and Setup

# Install sdtm.oak (example - check actual installation method)
install.packages("sdtm.oak")  # if available on CRAN
# or
# remotes::install_github("pharmaverse/sdtm.oak")

library(sdtm.oak)
library(dplyr)
library(haven)
library(tibble)

---

🏗️ 3. Creating Demographics (DM) Domain

Basic DM Domain Structure

# Sample raw demographics data
raw_demo <- tibble(
  subject_id = c("001-001", "001-002", "001-003", "001-004"),
  age = c(25, 45, 67, 52),
  sex = c("F", "M", "F", "M"),
  race = c("WHITE", "BLACK", "ASIAN", "WHITE"),
  treatment_arm = c("Placebo", "Active", "Placebo", "Active"),
  ref_start_date = c("2024-01-15", "2024-01-16", "2024-01-17", "2024-01-18")
)

# Create SDTM DM domain
dm_sdtm <- raw_demo %>%
  transmute(
    STUDYID = "STUDY001",
    DOMAIN = "DM",
    USUBJID = subject_id,
    SUBJID = str_extract(subject_id, "\\d+$"),
    RFSTDTC = ref_start_date,
    RFXSTDTC = ref_start_date,
    RFXENDTC = "",  # To be populated when known
    RFPENDTC = "",  # To be populated when known
    DTHDTC = "",    # Date of death if applicable
    DTHFL = "",     # Death flag
    SITEID = str_extract(subject_id, "^\\d+-"),
    AGE = age,
    AGEU = "YEARS",
    SEX = sex,
    RACE = race,
    ARMCD = case_when(
      treatment_arm == "Placebo" ~ "PBO",
      treatment_arm == "Active" ~ "TRT",
      TRUE ~ ""
    ),
    ARM = treatment_arm,
    ACTARMCD = ARMCD,  # Actual arm (same as planned for now)
    ACTARM = ARM
  ) %>%
  # Add required SDTM variables
  mutate(
    DMDTC = "",      # Date of demographics collection
    DMDY = NA_real_  # Study day of demographics collection
  )

print("SDTM DM Domain:")
print(dm_sdtm)

---

🚨 4. Creating Adverse Events (AE) Domain

AE Domain with Study Days

# Sample raw AE data
raw_ae <- tibble(
  subject_id = c("001-001", "001-001", "001-002", "001-003"),
  ae_term = c("Headache", "Nausea", "Fatigue", "Dizziness"),
  ae_start_date = c("2024-01-20", "2024-01-25", "2024-01-18", "2024-01-22"),
  ae_end_date = c("2024-01-22", "2024-01-26", "2024-01-20", "2024-01-23"),
  severity = c("MILD", "MODERATE", "MILD", "SEVERE"),
  serious = c("N", "N", "N", "Y"),
  outcome = c("RECOVERED", "RECOVERED", "RECOVERED", "ONGOING")
)

# Create SDTM AE domain
ae_sdtm <- raw_ae %>%
  left_join(dm_sdtm %>% select(USUBJID, RFSTDTC), 
            by = c("subject_id" = "USUBJID")) %>%
  transmute(
    STUDYID = "STUDY001",
    DOMAIN = "AE",
    USUBJID = subject_id,
    AESEQ = row_number(),
    AETERM = str_to_upper(ae_term),
    AEDECOD = AETERM,  # In real scenario, map to MedDRA
    AEBODSYS = case_when(  # Body system (simplified mapping)
      str_detect(AETERM, "HEADACHE|DIZZINESS") ~ "NERVOUS SYSTEM DISORDERS",
      str_detect(AETERM, "NAUSEA") ~ "GASTROINTESTINAL DISORDERS",
      str_detect(AETERM, "FATIGUE") ~ "GENERAL DISORDERS",
      TRUE ~ "OTHER"
    ),
    AESEV = severity,
    AESER = serious,
    AEOUT = outcome,
    AESTDTC = ae_start_date,
    AEENDTC = ae_end_date,
    # Calculate study days
    AESTDY = case_when(
      !is.na(ymd(ae_start_date)) & !is.na(ymd(RFSTDTC)) ~ 
        as.numeric(ymd(ae_start_date) - ymd(RFSTDTC)) + 1,
      TRUE ~ NA_real_
    ),
    AEENDY = case_when(
      !is.na(ymd(ae_end_date)) & !is.na(ymd(RFSTDTC)) ~ 
        as.numeric(ymd(ae_end_date) - ymd(RFSTDTC)) + 1,
      TRUE ~ NA_real_
    )
  )

print("SDTM AE Domain:")
print(ae_sdtm)

---

🔬 5. Creating Vital Signs (VS) Domain

VS Domain Example

# Sample raw vital signs data
raw_vs <- tibble(
  subject_id = rep(c("001-001", "001-002"), each = 6),
  visit = rep(c("Baseline", "Week 2", "Week 4"), each = 2, times = 2),
  test = rep(c("Systolic BP", "Diastolic BP"), times = 6),
  result = c(120, 80, 118, 78, 122, 82, 135, 85, 130, 82, 128, 80),
  test_date = rep(c("2024-01-15", "2024-01-29", "2024-02-12"), each = 2, times = 2)
)

# Create SDTM VS domain
vs_sdtm <- raw_vs %>%
  left_join(dm_sdtm %>% select(USUBJID, RFSTDTC), 
            by = c("subject_id" = "USUBJID")) %>%
  transmute(
    STUDYID = "STUDY001",
    DOMAIN = "VS",
    USUBJID = subject_id,
    VSSEQ = row_number(),
    VSTESTCD = case_when(
      test == "Systolic BP" ~ "SYSBP",
      test == "Diastolic BP" ~ "DIABP",
      TRUE ~ ""
    ),
    VSTEST = test,
    VSCAT = "VITAL SIGNS",
    VSORRES = as.character(result),
    VSORRESU = "mmHg",
    VSSTRESC = as.character(result),
    VSSTRESN = result,
    VSSTRESU = "mmHg",
    VSSTAT = "",
    VSREASND = "",
    VSDTC = test_date,
    VSDY = case_when(
      !is.na(ymd(test_date)) & !is.na(ymd(RFSTDTC)) ~ 
        as.numeric(ymd(test_date) - ymd(RFSTDTC)) + 1,
      TRUE ~ NA_real_
    ),
    VISIT = visit,
    VISITNUM = case_when(
      visit == "Baseline" ~ 1,
      visit == "Week 2" ~ 2,
      visit == "Week 4" ~ 3,
      TRUE ~ NA_real_
    )
  )

print("SDTM VS Domain:")
print(vs_sdtm)

---

📊 6. Data Validation and Quality Checks

Basic SDTM Validation

# Function to validate SDTM domain structure
validate_sdtm_domain <- function(data, domain_type) {
  
  # Common required variables for all domains
  common_vars <- c("STUDYID", "DOMAIN", "USUBJID")
  
  # Domain-specific required variables
  domain_vars <- switch(domain_type,
    "DM" = c(common_vars, "SUBJID", "RFSTDTC", "AGE", "SEX"),
    "AE" = c(common_vars, "AESEQ", "AETERM", "AESTDTC"),
    "VS" = c(common_vars, "VSSEQ", "VSTESTCD", "VSORRES"),
    common_vars
  )
  
  # Check for required variables
  missing_vars <- setdiff(domain_vars, names(data))
  
  # Check for duplicate keys
  key_vars <- switch(domain_type,
    "DM" = "USUBJID",
    "AE" = c("USUBJID", "AESEQ"),
    "VS" = c("USUBJID", "VSTESTCD", "VSDTC"),
    "USUBJID"
  )
  
  duplicates <- data %>%
    group_by(across(all_of(key_vars))) %>%
    filter(n() > 1) %>%
    nrow()
  
  # Return validation report
  list(
    domain = domain_type,
    missing_required_vars = missing_vars,
    duplicate_records = duplicates,
    total_records = nrow(data),
    validation_passed = length(missing_vars) == 0 && duplicates == 0
  )
}

# Validate our domains
dm_validation <- validate_sdtm_domain(dm_sdtm, "DM")
ae_validation <- validate_sdtm_domain(ae_sdtm, "AE")
vs_validation <- validate_sdtm_domain(vs_sdtm, "VS")

print("DM Validation:")
print(dm_validation)

---

📦 7. Export to XPT Format

Creating Submission-Ready Files

# Export SDTM domains to XPT format for regulatory submission
export_sdtm_to_xpt <- function(data, domain_name, output_dir = "sdtm") {
  
  # Create output directory if it doesn't exist
  if (!dir.exists(output_dir)) {
    dir.create(output_dir, recursive = TRUE)
  }
  
  # Create file path
  file_path <- file.path(output_dir, paste0(tolower(domain_name), ".xpt"))
  
  # Add SDTM dataset label
  attr(data, "label") <- paste("SDTM", toupper(domain_name), "Domain")
  
  # Export to XPT
  haven::write_xpt(data, file_path)
  
  cat("Exported", nrow(data), "records to", file_path, "\n")
  
  return(file_path)
}

# Export all domains
dm_file <- export_sdtm_to_xpt(dm_sdtm, "dm")
ae_file <- export_sdtm_to_xpt(ae_sdtm, "ae")
vs_file <- export_sdtm_to_xpt(vs_sdtm, "vs")

---

🤖 8. GitHub Copilot in RStudio for SDTM Programming

Effective Copilot Prompts for SDTM

Comment Prompt	Expected Copilot Suggestion
# Create SDTM DM domain from raw demographics	Domain structure with required variables
# Calculate study days for adverse events	Date arithmetic with RFSTDTC
# Map raw lab values to SDTM LB structure	Standard LB domain variables
# Validate SDTM domain for missing variables	Validation function with checks
# Export multiple domains to XPT format	Loop with haven::write_xpt

SDTM Programming Best Practices with Copilot

# Good: Descriptive comment for domain creation
# Create SDTM AE domain with proper sequencing and study day calculation

# Good: Specify CDISC compliance
# Map adverse event terms to MedDRA preferred terms following CDISC conventions

# Good: Include validation requirements
# Validate AE domain for required variables and duplicate USUBJID/AESEQ combinations

---

✅ Summary: SDTM Programming Workflow

Step	Task	R Tools
1. Planning	Define domain structure and metadata	CDISC specifications
2. Data Mapping	Map raw data to SDTM variables	dplyr, case_when
3. Domain Creation	Build standardized domains	sdtm.oak, transmute
4. Validation	Check compliance and quality	Custom validation functions
5. Export	Create XPT files for submission	haven::write_xpt
6. Documentation	Document mappings and assumptions	Comments, metadata

---

🎯 Next Steps

In the demo and exercise, you’ll practice: - Creating multiple SDTM domains from raw clinical data - Implementing proper variable mappings and derivations - Using validation functions to ensure data quality
- Exporting domains to XPT format for regulatory submission - Leveraging GitHub Copilot in RStudio for efficient SDTM programming