cdiscbuilder 2.0.2

A package to convert ODM XML to SDTM/ADaM Datasets

CDISC Builder

cdiscbuilder is a Python package designed to simplify the transformation of clinical trial data from ODM (Operational Data Model) XML format into CDISC SDTM (Study Data Tabulation Model) and ADaM (Analysis Data Model) datasets.

It provides a flexible, configuration-driven approach to data mapping, allowing users to define rules using simple YAML files or Python dictionaries without harcoding complex logic.

Key Features

• ODM XML Parsing: Efficiently parses CDISC ODM strings and files into workable dataframes.
• Configurable Mappings: Define your mapping rules (source columns, hardcoded values, custom logic) in YAML.
• Schema Validation: Ensures your configuration files adhere to strict standards before processing.
• Metadata-Driven Findings: Powerful processor for Findings domains (VS, LB, FA, etc.) using granular metadata.
• Excel/Parquet Output: Generates regulatory-compliant datasets in modern formats.

Installation

pip install cdiscbuilder

Quick Start

1. Command Line Interface

You can generate datasets directly from your terminal:

# Generate SDTM datasets from an ODM XML file
cdisc-sdtm --xml study_data.xml --output ./sdtm_data

2. Python API

from cdiscbuilder.sdtm import create_sdtm_datasets

# Define paths
xml_file = "study_data.xml"
config_dir = "path/to/my/specs" 
output_dir = "./sdtm_outputs"

# Generate Datasets
create_sdtm_datasets(config_dir, xml_file, output_dir)

Configuration

For detailed and complete references on how to structure mapping specifications, see:

• SDTM Mapping Specification: Details on the intermediate odm_long.csv schema, wide vs. findings domains, regex extraction, and validation rules.
• ADaM Mapping Specification: Details on ADaM yaml configuration schema, inheritance, SQL-like derivations, conditions, and aggregations.

The package comes with standard configurations for common domains (DM, AE, VS, etc.) in src/cdisc_builder/specs. You can override these or add new ones by creating your own configuration directory.

Example YAML (DM.yaml)

DM:
    - formoid: "FORM.DEMOG"
      keys: ["StudyOID", "StudySubjectID"]
      columns:
          STUDYID:
              source: StudyOID
              type: str
          USUBJID:
              source: StudySubjectID
              prefix: "PPT-"
              type: str
          AGE:
              source: IT.AGE
              type: int
              type: str
          SEX:
              source: I_DEMOG_SEX
              type: str
              value_mapping:
                  "M": "Male"
                  "F": "Female"

Finding Domains (Dynamic Mapping)

For domains like IE, LB, FA where many input items map to a single Test Code / Test Name pair, use type: finding.

IE:
  - type: finding
    formoid: "F_ELIGIBILITY"
    # Filter rows using Regex
    item_group_regex: "IG_ELIGI_.*"
    item_oid_regex: "I_ELIGI_.*"
    
    columns:
      # Extract part of the OID for the Short Code
      IETESTCD:
        source: ItemOID
        regex_extract: "I_ELIGI_(.*)"
      
      # Use Metadata from parsed XML for the Description
      IETEST:
        source: Metadata.Question
      
      IEORRES:
        source: Value

Advanced Mapping Features

Prefixing:

USUBJID:
  source: StudySubjectID
  prefix: "PPT-"

Substring Extraction (extracts chars 3-5 before mapping):

SITEID:
  source: FULL_ID
  substring_start: 3
  substring_length: 3

**Fallback** (use secondary source if primary is missing):
```yaml
SUBJID:
  source: RFSTDTC
  fallback: USUBJID

**Default Values**:
```yaml
CUSTOM_COL:
  source: ORG_COL
  value_mapping:
    "A": "Alpha"
  mapping_default: "Other" # used if not A
  # mapping_default_source: "AnotherCol" # Fallback to column value

Case Sensitive Mapping: By default, mapping is case-sensitive. You can disable this to map values regardless of case (e.g. "Yes", "yes", "YES" -> "Y"). Unmapped values preserve their original casing.

RESP:
  source: INPUT_VAL
  value_mapping:
    "Yes": "Y"
    "No": "N"
  case_sensitive: false

Development

This project uses modern tools for testing and maintaining code quality:

1. Running Tests

Run the automated test suite using pytest:

# Using standard pip/virtual environment
pytest

# Using uv
uv run pytest

2. Code Quality

We use black for code formatting, ruff for linting, and mypy for type checking:

# Code Formatting (in-place rewrite)
uv run --with black black src/

# Linting and style checks
uv run --with ruff ruff check src/

# Type Checking
uv run --with mypy mypy src/

License

MIT License