AI READI all combined d4d

Datasheet for Dataset - Human Readable Format

🎯

Motivation

Why was the dataset created?

  • Response
    Better understand salutogenesis (the pathway from disease to health) in Type 2 Diabetes Mellitus (T2DM) and enable downstream AI/ML analyses across multimodal health data.
📊

Composition

What do the instances represent?

  1. Representation
    Human participants with and without Type 2 Diabetes Mellitus (T2DM).
    Instance Type
    Multiple modalities per participant, including survey responses, physical/clinical measurements, imaging (retinal), physiological signals (ECG), wearable device time series (e.g., CGM), lab results, and environmental variables.
    Data Type
    Mix of raw and processed data depending on domain; harmonized across three collection sites under a standardized acquisition protocol.
    Sampling Strategies
    • Strategies
      • Stratified recruitment to achieve approximately equal distribution across diabetes severity.
      Is Representative
      • Not fully representative in early releases due to ongoing enrollment.
      Why Not Representative
      • Pilot and periodic releases may not achieve balanced distribution across groups while enrollment is ongoing.
  • Identification
    • Subpopulations are identifiable by diabetes severity, sex, race, and ethnicity (among others).
    Distribution
    • Recruitment targeted approximately equal distribution across diabetes severity; actual distributions in pilot/periodic releases may vary during ongoing enrollment.
  • Description
    • Access via FAIRhub data portal; public subset (non-sensitive) downloadable under Health Data License; controlled-access components available via data use agreement.
    • Data modalities include tabular (surveys, labs, clinical measures), images (retinal), physiological signals (ECG), and time series (CGM); specific file formats documented per domain.
  • Description
    • 2024 11 08 (version 2.0.0; DOI
      10.60775/fairhub.2)
    • 2024 05 03 (version 1.0.0; DOI
      10.60775/fairhub.1)
🔍

Collection Process

How was the data acquired?

Flagship Dataset of Type 2 Diabetes from the AI-READI Project
Flagship Dataset of Type 2 Diabetes from the AI-READI Project
The AI-READI flagship dataset consists of multimodal data collected from individuals with and without Type 2 Diabetes Mellitus (T2DM), harmonized across three collection sites. The study was designed for AI/ML readiness, with recruitment aimed at approximately equal distribution across diabetes severity and a standardized acquisition protocol spanning multiple domains (survey data, physical measurements, clinical data, imaging, wearable device data, etc.). Public data (non-sensitive) are available under a Health Data License; full data (including sensitive elements) are available under a data use agreement. The dataset supports research into salutogenesis (pathways from disease to health) in T2DM and is versioned with periodic updates as enrollment continues.
English
  • Diabetes mellitus
  • Machine Learning
  • Artificial Intelligence
  • Electrocardiography
  • Continuous Glucose Monitoring
  • Retinal imaging
  • Eye exam
  • AI-READI Consortium
2024-11-08
  • Response
    Provides an ethically sourced, harmonized, multimodal dataset for T2DM that is not feasible to obtain from claims/EHR alone; recruitment targeted balanced distribution across diabetes severity.
RoleNameORCIDAffiliation
ContributorAI-READI Consortium-AI-READI Consortium
  • Strategies
    • Stratified sampling by diabetes severity across three data collection sites under a harmonized protocol.
    Is Representative
    • Not fully representative in early releases due to ongoing enrollment.
    Why Not Representative
    • Interim releases may have imbalances across groups while recruitment continues.
  • Description
    • Early releases (pilot/periodic) may not fully achieve intended balanced distributions across severity groups.
  • External Resources
    • Documentation portal for dataset details and domain-specific processing guidance (https://docs.aireadi.org/docs/2)
    • Some environmental exposure variables (e.g., traffic and accident reports) linked via participant geography may rely on external data sources.
    Restrictions
    • External data sources used for environmental variables may carry their own terms of use or licensing; consult source providers as applicable.
  • Description
    • Elements under controlled access include 5-digit ZIP code, sex, race, ethnicity, genetic sequencing data, past health records, and medications.
  • Description
    • Health-related data including genetic data, medications, past health records.
    • Demographic attributes such as race and ethnicity.
    • Geolocation at 5-digit ZIP code granularity.
  • Description
    • Publicly released subset excludes sensitive personal health data and is de-identified per project policies; controlled-access components include potentially re-identifiable elements and require a data use agreement.
  1. Description
    • Data comprise directly observed clinical measurements and device outputs (e.g., ECG, retinal imaging, CGM), participant-reported survey responses, and variables derived/harmonized for analysis.
    Was Directly Observed
    Yes (e.g., clinical measurements, imaging, device signals).
    Was Reported By Subjects
    Yes (e.g., surveys).
    Was Inferred Derived
    Yes (harmonized/derived variables for analysis across sites and domains).
    Was Validated Verified
    Collected under standardized, harmonized protocols across sites; domain-specific processing and quality steps are documented.
  • Description
    • Standardized multi-site protocol using clinical procedures and hardware (e.g., ECG systems, retinal imaging cameras), wearable sensors (e.g., CGM), surveys, and harmonized data processing pipelines.
  • Description
    • Data collected and harmonized across three participating collection sites by AI-READI study teams.
  • Description
    • Enrollment and data collection are ongoing; Version 1.0.0 released 2024-05-03 (pilot), Version 2.0.0 released 2024-11-08, with periodic updates anticipated.
  • Description
    • Harmonization across sites; domain-specific preprocessing described in the documentation portal (file formats, data standards, metadata, and example outputs).
  • Description
    • Domain-specific cleaning and quality control steps as described in the documentation; details vary by data modality.
  • Description
    • Domain-specific annotation/labeling (e.g., clinical assessments, test results) documented per modality.
  • Description
    • AI-READI Consortium (dataset stewardship and documentation).
    • FAIRhub data portal (hosting and access workflows).
🚀

Uses

What (other) tasks could the dataset be used for?

  • Response
    AI/ML method development and analysis for T2DM using multimodal clinical, imaging, wearable, and survey data.
  • Description
    • Multimodal representation learning; longitudinal modeling for glycemic control; risk stratification; phenotype discovery; causal inference and prognostic modeling in T2DM.
  • Description
    • Interim imbalance across groups in pilot/periodic releases and the inclusion/exclusion of sensitive attributes across access tiers may affect fairness and generalizability. Users should assess representativeness and potential bias before deployment and implement mitigations as needed.
  • Description
    • Any attempt to re-identify participants or to use the dataset outside the scope of the Health Data License or data use agreement terms.
Description
  • Publicly available, non-sensitive data are distributed under a Health Data License.
  • Full dataset access (including sensitive data such as ZIP code, genetic data, health records, and medications) requires entering into a data use agreement.
📤

Distribution

How will the dataset be distributed?

Health Data License
Description
  • Prior versions remain available on the FAIRhub portal (e.g., v1.0.0 and v2.0.0), with documentation aligned to each dataset version.
🔄

Maintenance

How will the dataset be maintained?

2.0.0
Description
  • Periodic updates planned as enrollment continues and additional data become available; major releases are versioned on the FAIRhub portal with corresponding documentation.
Generated on 2025-11-09 10:17:35 using Bridge2AI Data Sheets Schema