Designs validated research data-capture instruments and aligns them to CDISC submission standards. Builds REDCap projects from a requirements spec: instrument and field design, the 18-column data dictionary (Variable/Field Name, Form Name, Field Type, Choices, Branching Logic, Text Validation, Identifier?), field validation (date_ymd, integer, number, email, phone), branching/show-field logic, longitudinal events and survey settings, and lints a data dictionary for common errors. Maps a study to CDISC: CDASH collection fields, SDTM domain mapping (DM, AE, VS, LB, EX, CM, MH across Interventions/Events/Findings classes), and NCI-EVS controlled terminology. Use when the user wants to build a REDCap project, write or lint a data dictionary, set up branching logic or validation, or map a study to CDISC SDTM/CDASH/CDISC CT. For LabArchives ELN bridging use alterlab-labarchive; for Likert/sampling/reliability use alterlab-survey-design. Part of the AlterLab Academic Skills suite.
Designs validated research data-capture instruments and aligns them to CDISC submission standards. Builds REDCap projects from a requirements spec: instrument and field design, the 18-column data dictionary (Variable/Field Name, Form Name, Field Type, Choices, Branching Logic, Text Validation, Identifier?), field validation (date_ymd, integer, number, email, phone), branching/show-field logic, longitudinal events and survey settings, and lints a data dictionary for common errors. Maps a study to CDISC: CDASH collection fields, SDTM domain mapping (DM, AE, VS, LB, EX, CM, MH across Interventions/Events/Findings classes), and NCI-EVS controlled terminology. Use when the user wants to build a REDCap project, write or lint a data dictionary, set up branching logic or validation, or map a study to CDISC SDTM/CDASH/CDISC CT. For LabArchives ELN bridging use alterlab-labarchive; for Likert/sampling/reliability use alterlab-survey-design. Part of the AlterLab Academic Skills suite.
No API key required — designs REDCap data dictionaries as offline CSV and maps to CDISC standards from local rules; the bundled linter is stdlib-only `uv run python`. Pushing a dictionary into a live REDCap instance (the REDCap API) is out of scope and delegated to the instance owner.
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:
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.
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."
alterlab-research-ethics (or alterlab-tr-research-ethics for Turkish etik kurul)
Finding registered trials by condition/NCT ID on ClinicalTrials.gov
alterlab-clinicaltrials
ISO 13485 medical-device QMS documentation
alterlab-iso13485
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.
#
Column header
Purpose
1
Variable / Field Name
Lowercase machine name (export/analysis).
2
Form Name
Instrument the field belongs to.
3
Section Header
Visual section break above the field.
4
Field Type
One of the values below.
5
Field Label
Human-readable prompt shown on the form.
6
Choices, Calculations, OR Slider Labels
1, Yes | 0, No for choice fields; the equation for calc.
7
Field Note
Helper text under the field.
8
Text Validation Type OR Show Slider Number
Validation for text; slider display.
9
Text Validation Min
Lower bound (integer/number/date).
10
Text Validation Max
Upper bound.
11
Identifier?
y marks PII/PHI (drives de-identified exports).
12
Branching Logic (Show field only if...)
Show-field condition.
13
Required Field?
y / blank.
14
Custom Alignment
Layout.
15
Question Number (surveys only)
Survey numbering.
16
Matrix Group Name
Groups matrix fields.
17
Matrix Ranking?
Matrix ranking flag.
18
Field Annotation
Action tags / annotations (e.g. @HIDDEN).
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 defines complementary standards across the data lifecycle. Map a study so
its collected data is regulator- and reuse-ready.
Standard
Role
What you produce
CDASH
Collection — standard fields to collect on CRFs; traces into SDTM.
The CRF/REDCap fields named to CDASH conventions.
SDTM
Tabulation — organizes submitted data into domains.
A variable → domain mapping.
Controlled Terminology
Codelists of valid values (NCI-EVS).
Coded values tied to the right codelist.
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
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.
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.
Lint. Run scripts/lint_data_dictionary.py; fix every error before the
user imports.
(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.
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.
