docs aireadi org docs-2 d4d

Datasheet for Dataset - Human Readable Format

🎯

Motivation

Why was the dataset created?

  • Response
    To enable AI/Machine Learning research to better understand salutogenesis—the pathway from disease to health—in Type 2 Diabetes Mellitus (T2DM).
📊

Composition

What do the instances represent?

  1. Representation
    Individual study participants across three collection sites
    Instance Type
    Multimodal per-participant record composed of multiple clinical and real-world data domains
    Data Type
    Survey responses, physical/clinical measurements, blood and urine lab results, retinal images, ECG, blood sugar levels, wearable device activity data, and environmental measures (e.g., home air quality)
    Sampling Strategies
    • Strategies
      • Stratified recruitment targeting approximately equal distribution by diabetes severity
  • Identification
    • With and without T2DM
    • Diabetes severity categories (targeted approximate balance)
    Distribution
    • Targeted ~equal distribution by diabetes severity; pilot data may not be balanced due to ongoing enrollment
  • Description
    • Public data distributed via a portal under license agreement; full data accessible via data use agreement
  • Description
    • Pilot release available; periodic updates to data releases as enrollment continues
🔍

Collection Process

How was the data acquired?

ai-readi-flagship-t2dm
AI-READI Dataset
AI-READI Flagship Dataset of Type 2 Diabetes
The AI-READI dataset consists of data collected from individuals with and without Type 2 Diabetes Mellitus (T2DM), harmonized across three data collection sites. The study design and sampling aimed to achieve approximately equal distribution of participants across diabetes severity categories and to acquire data across multiple domains (e.g., surveys, physical measurements, clinical data, imaging, wearable device data) to support downstream AI/ML analyses that are difficult with claims or EHR-only sources. Some data that are not considered sensitive personal health data are publicly available under a license agreement; access to the full dataset requires a data use agreement. Enrollment is ongoing; pilot and periodic releases may not be fully balanced across groups.
  • Type 2 Diabetes Mellitus
  • T2DM
  • AI
  • Machine Learning
  • Multimodal
  • Surveys
  • Clinical measurements
  • Imaging
  • Wearable devices
  • Retinal images
  • ECG
  • Blood sugar
  • Environmental variables
  • Harmonized
  • Multi-site
  • AI-READI Project
Aydan
  • Response
    Provide harmonized, multi-domain, prospective data suitable for AI/ML that are not feasible to obtain from existing sources such as claims or EHR data alone.
  • Strategies
    • Stratified sampling targeting approximate balance across diabetes severity categories
    Is Sample
    • True
    Is Representative
    • Not guaranteed; pilot release may not be balanced due to ongoing enrollment
  • Description
    • Data include directly observed clinical measurements and sensor data (e.g., imaging, ECG, wearable devices) and participant-reported survey responses
    Was Directly Observed
    includes directly observed measurements and sensor data
    Was Reported By Subjects
    includes participant-reported survey responses
  • Description
    • Multi-site clinical data acquisition
    • Clinical equipment and imaging systems (e.g., retinal imaging, ECG)
    • Wearable device data capture
    • Surveys and standardized clinical assessments
  • Description
    • Three data collection sites participated in recruitment and data acquisition
  • Description
    • Pilot study phase; ongoing enrollment with periodic data releases
  • Description
    • Data harmonized across three collection sites
  • External Resources
    • FAIRhub data portal (dataset landing page referenced)
    • AI-READI dataset documentation site
    • GitHub repository (project resources)
  • Description
    • Controlled-access elements include 5-digit ZIP code, sex, race, ethnicity, genetic sequencing data, past health records, medications, and traffic and accident reports
  • Description
    • Health and clinical data, genetic data, and other potentially identifying variables are held under controlled access
  • Description
    • Public dataset excludes sensitive personal health data; additional potentially identifying or sensitive data are available only under controlled access via a data use agreement
DescriptionName
Includes non-sensitive data such as survey data, blood and urine lab results, fitness activity level...Public dataset
Includes controlled-access elements such as 5-digit ZIP code, sex, race, ethnicity, genetic sequenci...Controlled-access dataset
🚀

Uses

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

  • Response
    Downstream AI/ML analyses across multiple clinical and real-world data domains (surveys, physical measurements, labs, imaging, wearable device data, ECG, glucose, environmental variables).
  • Description
    • Potential imbalance in pilot and early releases (due to ongoing enrollment) may affect model training and evaluation; users should check group distributions and consider stratification or rebalancing
Description
  • Public (non-sensitive) data available upon agreement with a license that defines permitted uses
  • Full dataset access requires entering into a data use agreement (DUA)
🔄

Maintenance

How will the dataset be maintained?

2.0.0
2025-09-08
Description
  • Periodic updates to data releases as enrollment continues and new data are acquired
Generated on 2025-11-09 10:17:35 using Bridge2AI Data Sheets Schema