# Ethics & privacy extract — CHORUS D4D datasheet
# Source record: data/d4d_concatenated/claudecode_agent/2026-08-06_claude-opus-5-1m-generic-v3-schema2_rep1/CHORUS_d4d.yaml
# Source sha256: be2646dbe460cae84334c334bdc41ac787f485902e44e422240e026b2f8c9283
# Generated record: AI-extracted from project documentation (schema 2.0.0, run 2026-08-06_claude-opus-5-1m-generic-v3-schema2_rep1)
# Extract created: 2026-08-07
# Slots included (10): at_risk_populations, confidential_elements, ethical_reviews, human_subject_research, ip_restrictions, is_deidentified, license_and_use_terms, regulatory_restrictions, sensitive_elements, subpopulations
# Values are verbatim from the record; 'source_caveats' and 'notes'
# fields carry the record's own evidence commentary.

at_risk_populations:
  at_risk_groups_included: true
  notes: 'Minors and neonates: the current released dataset covers patient admissions from pediatric intensive
    care units (PICU) and neonatal intensive care units (NICU) alongside adult ICU admissions. The declared
    bundle records the inclusion of PICU and NICU admissions but does not describe specific assent procedures,
    guardian consent arrangements, or additional safeguards applied to these groups.'
confidential_elements:
- confidential_elements_present: true
  confidentiality_details: Every data type in the dataset (demographics, medication administration, procedures,
    nursing flowsheets, diagnoses, clinical notes, imaging, waveform telemetry, and EEG) is designated
    controlled access. Clinical notes are stored locally at contributing sites, and only tokens are made
    available in the enclave.
  notes: Community-facing ethics focus groups are conducted to determine what data is appropriate for
    public sharing.
ethical_reviews:
- notes: Patient-focused efforts determine the ethical and legal approaches to manage privacy and bias
    while accounting for Social Determinants of Health; expertise is drawn from team science, law, ethics,
    health services, biomedical science, engineering, and scientific journal publications. The declared
    bundle does not name an IRB or ethics committee of record for the dataset.
  review_details: Ethics is one of the three pillars of the CHoRUS data generation project (Ethical and
    Trustworthy AI). The project establishes a legal framework for collecting data at scale, performs
    community-facing ethics focus groups to determine what data is appropriate for public sharing, and
    analyses the existing legal and regulatory landscape to increase trustworthiness of provenance and
    privacy in AI.
human_subject_research:
  involves_human_subjects: true
  notes: The dataset consists of retrospective clinical records from critically ill patients, with an
    anticipated cohort of more than 100,000 critically ill patients.
  special_populations:
  - The current released dataset includes admissions from pediatric intensive care units (PICU) and neonatal
    intensive care units (NICU), and therefore covers minors and neonates.
ip_restrictions:
  notes: These license terms are stated for the CHoRUS software repositories; access to the dataset itself
    is governed by a separate signed licensing agreement.
  restrictions:
  - Software of the CHoRUS project published in the chorus-ai GitHub organization is licensed under the
    MIT License, with individual repositories carrying MIT or Apache-2.0 licenses (for example Chorus_SOP
    under Apache-2.0; UF-Geocoding, chorus-extract-upload, and chorus_waveform under MIT).
is_deidentified:
  deidentification_details: Data are transformed using approaches that limit re-identification. Imaging
    de-identification was in process for the larger cohort as of September 2025, with 1,000 de-identified
    images available. All modalities remain under controlled access within the cloud enclave.
  method: Tokenization of unstructured electronic health record text using the OHNLP toolkit, with full
    notes retained locally at contributing sites; de-identification of DICOM imaging supported by tooling
    including the CTP-deid repository; transformation of data using approaches that limit re-identification.
  notes: The consortium maintains de-identification and privacy tooling in the chorus-ai GitHub organization,
    including the CTP-deid repository and a privacy_scan_tool for medical records.
license_and_use_terms:
  contact_person: 'Access requests: dbold@emory.edu or jared.houghtaling@tuftsmedicine.org'
  data_use_permission: institution_specific
  license_terms: The CHoRUS dataset is controlled access. Users complete a registration form providing
    name, email, and institution, and all participants must sign a licensing agreement included in the
    registration form before gaining access to the dataset. Access to the dataset requires a ".edu" email
    address; once access is granted and compute is provisioned, an email is sent to the user.
  notes: In the AIM-AHEAD Bridge2AI for Clinical Care Training Program, the licensing agreement and .edu
    email requirements are stated not to be barriers to acceptance into the program; program administrators
    assist with access if needed.
regulatory_restrictions:
  confidentiality_level: restricted
  regulatory_restrictions:
  - All data modalities are designated controlled access; access is granted only after registration, signature
    of a licensing agreement, and verification of an institutional (".edu") email address, with data accessed
    inside a provisioned cloud enclave.
sensitive_elements:
- sensitive_elements_present: true
  sensitivity_details: The dataset comprises hospital clinical records of critically ill patients, including
    demographics, diagnoses, procedures, medication administration, nursing flowsheets, clinical notes,
    PACS imaging, waveform telemetry, EEG recordings, and social determinants of health. All listed data
    types are released under controlled access.
subpopulations:
- identification: Admissions in the current released dataset are drawn from adult intensive care units
    (ICU), pediatric intensive care units (PICU), and neonatal intensive care units (NICU).
  notes: Federated access is intended to enable sampling methods that ensure a balanced and diverse cohort,
    and data elements are intended to feature contextual factors such as geographic distance to the nearest
    hospital, alongside social determinants of health.
