Alterlab Redcap Cdisc

REDCap & CDISC — Research Data Capture and Standards Alignment

Designs validated electronic data capture (EDC) instruments in REDCap and aligns a study's variables to CDISC submission standards. It owns two jobs that overlap with sibling skills only at the edges:

1. REDCap project & instrument design — turn a study's requirements into a REDCap data dictionary (the importable 18-column CSV), with correct field types, validation, branching logic, longitudinal events, and survey settings.
2. CDISC alignment — map collected variables to CDASH (collection) and SDTM (tabulation) domains, and tie coded values to CDISC Controlled Terminology (NCI-EVS), so the data is submission- and reuse-ready.

This is design + standards work, not a live-API connector. Producing the CSV and the mapping is in scope; importing it into a running REDCap server is the instance owner's job (REDCap's own API/UI).

Quick Start

Build me a REDCap data dictionary for a 3-arm RCT screening + follow-up.
Add branching logic so the pregnancy question only shows when sex = female.
Lint this data_dictionary.csv before I import it.
Map my CRF variables to CDISC SDTM domains and CDASH fields.
Which SDTM domain and controlled-terminology codelist does "adverse event" go in?

→ Draft or read the dictionary, run scripts/lint_data_dictionary.py to catch structural errors, then (for CDISC) produce a variable → CDASH → SDTM mapping table. Always state which facts are design conventions vs. instance-specific.

When to Use This Skill

Use it when the request is about building the instrument or making it standards-compliant:

• "Build / design a REDCap project, form, or data dictionary."
• "Write the importable REDCap CSV for these variables."
• "Set up branching logic / show-field logic / field validation / required fields."
• "Configure longitudinal events / arms / repeating instruments / a survey."
• "Lint / validate / debug my data dictionary before import."
• "Map my study to CDISC — which SDTM domain / CDASH field / codelist?"
• "Make my CRF submission-ready (SDTM/CDASH/CDISC CT)."

Does NOT Trigger — route to the right sibling

Routing rule of thumb: survey-design owns the questions and their measurement properties; this skill owns the database that captures them and its standards mapping. If the user asks for both (e.g. "design a validated depression scale and the REDCap form for it"), do the scale work under alterlab-survey-design and the dictionary/mapping here.

REDCap Data Dictionary (the importable CSV)

REDCap projects are defined by a data dictionary — a CSV with 18 columns in a fixed order. Authoring or repairing this CSV is the core deliverable.

Field Type values: text, notes, dropdown, radio, checkbox, calc, sql, descriptive, slider, yesno, truefalse, file.

Text Validation Type values: date_ymd, date_mdy, datetime_ymd, time, integer, number, email, phone, zipcode.

Deeper field-by-field rules, the choice-string grammar, branching-logic operators, longitudinal/repeating-instrument design, and a worked dictionary are in references/redcap_design.md.

Lint before import

uv run python skills/faculty-life/alterlab-redcap-cdisc/scripts/lint_data_dictionary.py \
    path/to/data_dictionary.csv

The linter is stdlib-only (no network, no REDCap account). It checks the 18-column header, field-name syntax, field-type/validation legality, that choice fields carry a choices string, that calc/branching fields reference variables that exist, duplicate variable names, and PII fields left unflagged as Identifier?. It exits non-zero and emits a JSON report when issues are found. It is a structural linter — it does not validate against a live REDCap server, so always tell the user a clean lint still needs a test import.

CDISC Alignment (CDASH → SDTM → Controlled Terminology)

CDISC defines complementary standards across the data lifecycle. Map a study so its collected data is regulator- and reuse-ready.

SDTM observation classes: Interventions, Events, Findings, and Findings About. Common domains and their two-letter codes: DM Demographics, AE Adverse Events, VS Vital Signs, LB Laboratory, EX Exposure, CM Concomitant Medications, MH Medical History.

Controlled Terminology is maintained with NCI's Enterprise Vocabulary Services (NCI-EVS) and updated quarterly; do not invent codelist values — cite the codelist and tell the user to pull the current quarterly release.

Versions are real but move — the skill records the latest it verified (SDTM v2.1, 2024-06-10; SDTMIG v3.4, 2021-11-29; CDASHIG v2.3, 2023-09-28, referencing CDASH Model v1.3). Confirm the current version with the user / cdisc.org before asserting one in a deliverable, because these are re-released. The mapping recipe, the CDASH↔SDTM traceability idea, and the worked AE/VS examples are in references/cdisc_mapping.md.

Workflow

1. Capture requirements. Study design (cross-sectional / longitudinal / arms), the variables, value sets, PII fields, and whether it is a survey. If the measurement properties of the questions matter, hand the questionnaire design to alterlab-survey-design first.
2. Draft the data dictionary. Build the 18-column CSV: machine names, field types, choices, validation, Identifier? for every PII field, branching logic, and (longitudinal) the event/arm plan separately.
3. Lint. Run scripts/lint_data_dictionary.py; fix every error before the user imports.
4. (If submission/standards needed) Map to CDISC. Produce the variable → CDASH → SDTM table, flag value sets needing Controlled Terminology, and name the codelists — never fabricate codelist members.
5. Report honestly. State which choices are conventions vs. instance config, that the lint is structural (a test import is still required), and that CDISC versions must be confirmed against the current release.

Self-Check Before Reporting

• Does the dictionary have exactly the 18 columns in order, and is every field type one of the legal values?
• Is every PII field marked Identifier? = y?
• Do all branching-logic and calc expressions reference variables that actually exist in the dictionary?
• Did the linter run and pass — and did you tell the user a clean lint still needs a real test import?
• For CDISC: did you name codelists/versions rather than invent codelist values, and flag that versions must be confirmed against the current release?
• Did you route psychometrics to alterlab-survey-design and ELN bridging to alterlab-labarchive instead of doing them here?

References

• references/redcap_design.md — full 18-column field reference, choice-string grammar, branching-logic operators, validation types, longitudinal/repeating-instrument design, and a worked dictionary.
• references/cdisc_mapping.md — CDASH→SDTM→CT pipeline, SDTM observation classes and domains, controlled-terminology sourcing, and worked AE/VS/DM mappings, with verified versions and sources.

Part of the AlterLab Academic Skills suite.