AI READI Dataset Documentation

Description	ID	Name
Better understand salutogenesis (the pathway from disease to health) in Type 2 Diabetes Mellitus using a hypothesis-agnostic, harmonized, multi-domain dataset designed specifically for AI/ML research. The dataset aims to provide critical insights into how individuals can transition from diabetes toward health resilience through pseudotime manifold analysis.	aireadi:purpose:1	Understanding T2DM salutogenesis
Establish standards, best practices, and guidelines for collection, preparation, and sharing of medical/health data sets targeted for AI/ML applications. This includes guidance from bioethicists on ethical and equitable data collection and management practices, with adherence to FAIR principles.	aireadi:purpose:2	Establishing AI/ML data standards
Address the lack of racial and ethnic diversity in T2DM research by creating a dataset that is triple-balanced across race/ethnicity (Asian, Black, Hispanic, White), biological sex (male, female), and diabetes severity (no diabetes, pre-diabetes/lifestyle-controlled, medication-controlled, insulin-controlled).	aireadi:purpose:3	Addressing demographic inequities in T2DM research

Description	ID	Name
Self-reported Asian race/ethnicity, target approximately 1,000 participants (25% of sample)	aireadi:subpop:1	Asian participants
Self-reported Black race/ethnicity, target approximately 1,000 participants (25% of sample)	aireadi:subpop:2	Black participants
Self-reported Hispanic ethnicity, target approximately 1,000 participants (25% of sample)	aireadi:subpop:3	Hispanic participants
Self-reported White race/ethnicity, target approximately 1,000 participants (25% of sample)	aireadi:subpop:4	White participants
Participants without diabetes diagnosis, target approximately 1,000 participants (25% of sample)	aireadi:subpop:5	No diabetes
Participants with pre-diabetes or lifestyle-controlled diabetes, target approximately 1,000 participants (25% of sample)	aireadi:subpop:6	Pre-diabetes and lifestyle-controlled diabetes
Participants with diabetes treated with oral medications or non-insulin injections, target approximately 1,000 participants (25% of sample)	aireadi:subpop:7	Medication-controlled diabetes
Participants with insulin-controlled diabetes, target approximately 1,000 participants (25% of sample)	aireadi:subpop:8	Insulin-controlled diabetes

Description	ID	Name
Retinal imaging data distributed in DICOM format (converted from proprietary .fda and .sdt formats for standardization).	aireadi:format:1	DICOM for imaging
Survey data, clinical lab results, continuous glucose monitoring, environmental sensor data, and other tabular/time-series data provided in CSV format.	aireadi:format:2	CSV for tabular and time-series data
Physical activity monitoring data (from Garmin VivoSmart 5) converted from .FIT format to mHealth standard for interoperability.	aireadi:format:3	mHealth standard for wearable data
Electrocardiogram data from Philips Pagewriter TC30 exported in .xml format.	aireadi:format:4	XML for ECG data

🔍

Collection Process

How was the data acquired?

https://fairhub.io/datasets/2

Name

AI-READI

Title

Artificial Intelligence Ready and Equitable Atlas for Diabetes Insights (AI-READI)

Description

The AI-READI is a flagship dataset consisting of multimodal data collected from 4,000 individuals with and without Type 2 Diabetes Mellitus (T2DM), harmonized across 3 data collection sites (Birmingham, Alabama; San Diego, California; Seattle, Washington). The dataset was designed with future AI/Machine Learning studies in mind, including recruitment sampling procedures aimed at achieving approximately equal distribution of participants across diabetes severity (triple-balanced by race/ethnicity, biological sex, and T2DM severity), as well as a multi-domain data acquisition protocol (survey data, physical measurements, clinical data, imaging data, wearable device data, environmental sensors, biospecimens) to enable downstream AI/ML analyses that may not be feasible with existing data sources such as claims or electronic health records data. The goal is to better understand salutogenesis (the pathway from disease to health) in T2DM. The study follows FAIR principles and incorporates ethical and equitable data collection and management practices.

Page

https://fairhub.io/datasets/2

Language

Keywords

Type 2 Diabetes Mellitus
T2DM
AI-READI
Machine Learning
Artificial Intelligence
multimodal dataset
harmonized data
multi-site study
salutogenesis
FAIR principles
retinal imaging
continuous glucose monitoring
wearable devices
biorepository
biospecimens
triple-balanced sampling
health equity
Bridge2AI
cross-sectional study

Research Gaps Addressed

Description	ID	Name
Provide a large-scale, harmonized, multi-site, multi-domain dataset enabling AI/ML analyses not feasible with existing sources (e.g., claims or EHR alone). With 4,000 participants and over 10 variable domains, this is the largest publicly accessible dataset of its kind for T2DM research.	aireadi:gap:1	Lack of multimodal T2DM datasets
Address demographic inequities in T2DM research by recruiting equal proportions across four race/ethnic groups (Asian, Black, Hispanic, White) and both biological sexes, improving upon many previous epidemiological studies and clinical trials that lacked diversity.	aireadi:gap:2	Demographic underrepresentation
Create a model for future AI-ready medical datasets through comprehensive metadata, standardized data formats, FAIR compliance, and ethical data governance practices that can be replicated for other health conditions.	aireadi:gap:3	AI-readiness of medical datasets

Creators

Role	Name	ORCID	Affiliation
Contributor	Aaron Lee	aireadi:creator:1	-
Contributor	Cynthia Owsley	aireadi:creator:2	-
Contributor	Sally L. Baxter	aireadi:creator:3	-
Contributor	Christopher G. Chute	aireadi:creator:4	-
Contributor	Megan E. Collins	aireadi:creator:5	-
Contributor	Jeffrey C. Edberg	aireadi:creator:6	-
Contributor	Kadija Ferryman	aireadi:creator:7	-
Contributor	Michelle Hribar	aireadi:creator:8	-
Contributor	Samantha Hurst	aireadi:creator:9	-
Contributor	Hiroshi Ishikawa	aireadi:creator:10	-
Contributor	Cecilia S. Lee	aireadi:creator:11	-
Contributor	Alvin Y. Liu	aireadi:creator:12	-
Contributor	Gerald McGwin	aireadi:creator:13	-
Contributor	Shannon K. McWeeney	aireadi:creator:14	-
Contributor	Camille Nebeker	aireadi:creator:15	-
Contributor	Bhavesh Patel	aireadi:creator:16	-
Contributor	Sara Jean Singer	aireadi:creator:17	-
Contributor	Michael P. Snyder	aireadi:creator:18	-
Contributor	Joseph Manuel Yracheta	aireadi:creator:19	-
Contributor	Linda M. Zangwill	aireadi:creator:20	-

Subsets

Description	ID	Name
Includes data not considered sensitive personal health information, available to the public for download upon agreement with a license. Contains survey data, blood and urine lab results, fitness activity levels, clinical measurements (e.g., monofilament and cognitive function testing), retinal images, ECG, blood glucose levels, and environmental variables such as home air quality. Available at https://fairhub.io/datasets/2.	aireadi:subset:1	Public Access Dataset
Includes sensitive data accessible by entering into a data use agreement. Contains 5-digit zip code, sex, race, ethnicity, genetic sequencing data (from buffy coats), past health records, medications, and traffic and accident reports. Access requirements are being developed by the Data Access Committee.	aireadi:subset:2	Controlled Access Dataset
Biobanked samples stored at UAB Center for Clinical and Translational Science (CCTS), including plasma, serum, buffy coats, peripheral blood mononuclear cells (PBMCs), PAXgene RNA, and urine. Available to researchers for future ancillary studies according to procedures and policies in development. Finite number of samples available.	aireadi:subset:3	Biorepository

Sampling Strategies

ID
aireadi:sampling:1
Name
Triple-balanced recruitment
Description
Recruitment sampling procedures aimed at achieving approximately equal distribution of participants across three dimensions: (1) race/ethnicity (Asian, Black, Hispanic, White), (2) T2DM severity (no diabetes, pre-diabetes/lifestyle-controlled, medication-controlled, insulin-controlled), and (3) biological sex (male, female). This balanced design is critical for developing unbiased machine learning models.
Sample
True
Random Sampling
False
Representative Sample
False
Why Not Representative
- The triple-balanced design intentionally over-samples racial/ethnic minority groups and specific diabetes severity groups relative to their population prevalence, to enable unbiased AI/ML model development. Participants are volunteers, introducing volunteer selection bias that may limit generalizability to non-volunteer populations.
Strategies
- Targeted recruitment to balance demographics across race/ethnicity, sex, and diabetes severity
- Wave-based recruitment with monitoring and adjustment through under- and oversampling
- Recruitment from electronic health records screening using ICD-10 codes (E11.X for T2DM, R73.09 for pre-diabetes)
- Personalized invitation letters and emails with REDCap recruitment interface

Collection Mechanisms

Description	ID	Name
Single study encounter per participant at one of three data collection sites (Birmingham, San Diego, Seattle). Multi-domain protocol containing over 10 data collection domains performed during the visit lasting between 2.5 and 4 hours.	aireadi:collection:1	In-person data collection visits
Source population identified by screening electronic health records for patients aged 40+ who had medical encounters between 2020-2025. ICD-10 codes used to identify T2DM (E11.X) and pre-diabetes (R73.09) cases.	aireadi:collection:2	Electronic health record screening
Participants recruited in waves to facilitate efficient sampling. Composition and size of each wave influenced by observed participation characteristics to maintain balance across race/ethnicity, sex, and diabetes severity. Enrollment began July 18, 2023 and continues until November 30, 2026.	aireadi:collection:3	Wave-based recruitment
Continuous glucose monitoring (Dexcom G6, 5-minute intervals), physical activity monitoring (Garmin VivoSmart 5), and environmental sensor monitoring (temperature, humidity, air quality) conducted at participants' homes over 10-day monitoring periods.	aireadi:collection:4	Home-based wearable monitoring
Blood (53 mL) and urine collected during study visit. Local processing for plasma, serum, buffy coats at all sites. Centralized biobanking at UAB CCTS. Standardized operating procedures ensure consistent handling.	aireadi:collection:5	Biospecimen collection and biobanking

Acquisition Methods

Description	ID	Name
Self-reported data collected via REDCap interfaces including demographics, medical history, social determinants of health, depression screening (CES-D-10), diabetes-related emotional distress (PAID scale), diabetes self-care score, dietary assessment, ophthalmic survey, substance use (smoking, alcohol, vaping, marijuana), general health conditions, and current medications with RxNorm codes.	aireadi:acquisition:1	Survey and questionnaire data
Height, weight, waist and hip circumference (waist-hip ratio and BMI calculated), blood pressure (systolic and diastolic, measured twice separated by 2 minutes), and heart rate collected during in-person visit using standardized protocols.	aireadi:acquisition:2	Physical measurements and vital signs
Multi-device retinal imaging protocol capturing data from Aurora IQ (Optomed), EIDON widefield truecolor confocal fundus system (iCare), Spectralis HRA OCT and OCTA (Heidelberg Engineering), Maestro2 3D OCT-1 (Topcon), Triton DRI OCT (Topcon), Cirrus 5000 (Carl Zeiss), and Fluorescence Lifetime Imaging Ophthalmoscopy (FLIO, Heidelberg Engineering). Both eyes imaged under dilated conditions (except Aurora IQ). Output formats include DICOM, with proprietary formats (.fda, .sdt) converted to DICOM.	aireadi:acquisition:3	Retinal imaging
Visual acuity and contrast sensitivity under photopic (daylight) and mesopic (dim light) conditions using Electronic Visual Acuity tester (M&S Technology) and Mars chart (Mars Perceptrix). Autorefraction data collected using Topcon KR 800.	aireadi:acquisition:4	Visual function testing
Complete blood count (CBC) from fresh whole blood at local CLIA-certified labs. Central lab testing at UW Nutrition and Obesity Research Center (NORC) for EDTA plasma tests (NT-proBNP, Troponin-T, C-peptide, insulin), serum tests (CRP-HS, lipid panel, glucose, kidney and liver function markers), whole blood tests (HbA1c), and urine tests (creatinine, albumin, other markers).	aireadi:acquisition:5	Clinical laboratory testing
12-lead ECG data collected using Philips Pagewriter TC30 Cardiograph during study visit with participant sitting in reclining chair or lying supine at recorded position (0 degrees, 30 degrees, 60 degrees, or 90 degrees relative to supine).	aireadi:acquisition:6	Electrocardiogram (ECG)
Montreal Cognitive Assessment (MoCA) administered electronically on iPad using MoCA Duo Application (total score, section subscores, Memory Index Score, task completion times). Total possible score is 30, with higher numbers representing better performance.	aireadi:acquisition:7	Cognitive function testing
Monofilament testing performed to assess peripheral neuropathy using 10g filament at three locations on each foot (10 times per location) with participant's eyes closed, responding yes/no whether they feel the filament.	aireadi:acquisition:8	Peripheral neuropathy assessment
Dexcom G6 Continuous Glucose Monitor capturing blood glucose measurements (mg/dL) every 5 minutes for 10 days. Data exported in CSV format.	aireadi:acquisition:9	Continuous glucose monitoring
Garmin VivoSmart 5 wearable device capturing number of steps, heart rate, sleep duration (circadian and diurnal rhythm), and oxygen saturation for 10 days. Data exported in .FIT format and converted to mHealth standard.	aireadi:acquisition:10	Physical activity monitoring
Custom-designed environmental sensor (Karalis Johnson Retina Center, UW) capturing ambient temperature, relative humidity, nitrogen oxides (NO and NO2), volatile organic compounds, particulate matter (PM1.0, PM2.5, PM4, PM10), and multi-spectral light intensity (11 measurements) for 10 days. Data in CSV format.	aireadi:acquisition:11	Environmental monitoring
Non-fasting blood (53 mL) and urine collection. Processing includes whole blood for CBC, EDTA plasma, serum, buffy coats for genomic DNA extraction, peripheral blood mononuclear cells (PBMCs) from CPT tubes, and PAXgene RNA vacutainers. Samples stored at UAB CCTS biorepository at appropriate temperatures.	aireadi:acquisition:12	Biospecimen collection

Preprocessing Strategies

Description	ID	Name	Preprocessing Details
Data harmonized across three collection sites (Birmingham, San Diego, Seattle) using standardized operating procedures, common protocols, and centralized data management through REDCap. Ensures consistency and comparability across sites.	aireadi:preproc:1	Data standardization and harmonization	Standardized operating procedures across all three sites, Common protocols and equipment across sites, Centralized data management through REDCap, Standardized training of study coordinators with certification process, Manual of Procedures (MOP) for reference
Retinal imaging data converted from proprietary formats (.fda, .sdt) to DICOM standard for the dataset. Wearable device data converted from .FIT format to mHealth standard.	aireadi:preproc:2	Image format conversion	Proprietary retinal imaging formats (.fda, .sdt) converted to DICOM, Wearable device data (.FIT) converted to mHealth standard, ECG data exported in .xml format, MoCA data exported in .csv format
Standardized local processing for plasma, serum, and buffy coats using consistent protocols. Centralized processing of CPT tubes for PBMC isolation at UAB CCTS. Batch shipping of biospecimens for central clinical lab analyses.	aireadi:preproc:3	Biospecimen processing	Standardized local processing for plasma, serum, buffy coats, Centralized PBMC processing at UAB CCTS, Batch shipping for central lab analyses at UW NORC, Genomic DNA extractions from stored buffy coats performed at UAB CCTS
Multiple quality control measures including standardized training of study coordinators, equipment calibration, duplicate measurements, and data validation checks in REDCap system. Data managers at each site oversee quality control before uploading to FAIRhub.	aireadi:preproc:4	Quality control and validation	Standardized training of study coordinators with certification, Equipment calibration protocols, Duplicate measurements for quality assurance (e.g., blood pressure measured twice), Data validation checks in REDCap, Practice subjects required before enrolling participants
All data mapped to applicable data standard formats such as Observational Medical Outcomes Partnership Common Data Model for clinical data and DICOM format for retinal imaging. Data stored and shared as AI-ready enabling immediate AI/ML research without reformatting.	aireadi:preproc:5	Data mapping to standards	OMOP Common Data Model for clinical data, DICOM for retinal imaging, mHealth standard for wearable device data, RxNorm codes for medications, ICD-10 codes for diabetes classification

Cleaning Strategies

ID
aireadi:cleaning:1
Name
Multi-site harmonization
Description
Standardized protocols and procedures across all three data collection sites ensure data consistency and quality. Common equipment, training, and REDCap data management system used to maintain FAIR principles compliance.
Cleaning Details
- Cross-site harmonization procedures
- Standardized equipment and training
- REDCap data management for quality
- FAIR principles implementation
- Pilot enrollment period (July 18 - November 30, 2023) to ensure coordinator familiarity

Maintainers

ID
aireadi:maintainer:1
Name
AI-READI Consortium
Description
Multidisciplinary consortium managing dataset maintenance including data collection sites, coordinating centers, and data governance committees. Contact through the University of Washington as lead institution and data coordination center. Documentation maintained at https://docs.aireadi.org/ with version-specific guides for each data release.

Retention Limit

ID: aireadi:retention:1
Name: Data and biospecimen retention
Description: Digital data maintained according to NIH data sharing policies. Biospecimen retention subject to institutional policies and consent agreements at UAB CCTS. Finite number of biospecimen samples available for distribution to researchers. Procedures for reviewing and prioritizing biospecimen requests are under development.

Ethical Reviews

ID
aireadi:ethics:1
Name
University of Washington IRB Approval
Description
Study approved by Institutional Review Board (IRB) of University of Washington (approval number STUDY00016228), with reliance agreements from IRBs of University of Alabama at Birmingham and University of California, San Diego. Written informed consent provided by all participants.

Sensitive Elements

Description	ID	Name
Genomic DNA extracted from buffy coats, blood derivatives, and urine samples stored with potential for future genetic analyses. Available in controlled access dataset only. Includes genetic sequencing data from buffy coats, PBMCs, and PAXgene RNA.	aireadi:sensitive:1	Genetic and biospecimen data
5-digit zip code, detailed race, ethnicity, and sex information available in controlled access dataset only. Public dataset contains de-identified data.	aireadi:sensitive:2	Geographic and demographic identifiers
Past health records, medications with RxNorm codes, traffic and accident reports available in controlled access dataset only.	aireadi:sensitive:3	Medical history and records

External Resources

Description	External Resources	ID	Name
Official project website with overview and resources	https://aireadi.org/	aireadi:resource:1	AI-READI Project Website
Comprehensive dataset documentation with version-specific guides	https://docs.aireadi.org/	aireadi:resource:2	AI-READI Dataset Documentation
Dataset repository and download portal for AI-READI data	https://fairhub.io/datasets/2	aireadi:resource:3	FAIRhub Dataset Landing Page
Parent NIH Common Fund program supporting AI-ready biomedical datasets	https://bridge2ai.org/	aireadi:resource:4	Bridge2AI Program
Federal grant information and project details for grant 1OT2OD032644-01	https://reporter.nih.gov/project-details/10471118	aireadi:resource:5	NIH RePORTER Project Details
Policies and procedures for data access and sharing	https://aireadi.org/goals/data-sharing	aireadi:resource:6	Data Sharing Information
Additional dataset documentation and resources	https://doi.org/10.5281/zenodo.10642459	aireadi:resource:7	Zenodo Archive
BMJ Open publication describing study design and protocol (Owsley et al. 2025)	https://doi.org/10.1136/bmjopen-2024-097449	aireadi:resource:8	Protocol Publication (BMJ Open)
Overview of AI-READI approach and significance published in Nature Metabolism	https://doi.org/10.1038/s42255-024-01165-x	aireadi:resource:9	Nature Metabolism Commentary

🚀

Uses

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

Tasks

Description	ID	Name
Enable downstream AI/ML analyses across survey, clinical, imaging, wearable device, environmental, and biospecimen domains related to T2DM that may not be feasible with existing data sources such as claims or electronic health records data alone. The multimodal nature of the data supports complex machine learning model development.	aireadi:task:1	Enable multi-domain AI/ML analyses for T2DM
Support the development of unbiased machine learning models through balanced data collection across demographic groups and diabetes severity levels. The triple-balanced design is critical for preventing algorithmic bias in AI/ML applications.	aireadi:task:2	Develop unbiased AI/ML models
Study disease trajectories and salutogenesis pathways in T2DM through cross-sectional analysis of participants at different disease stages, enabling pseudotime manifold analysis to predict disease progression and paths to health resilience.	aireadi:task:3	Study T2DM disease trajectories

Intended Uses

Description	ID	Name
Primary intended use is development and training of artificial intelligence and machine learning models to study Type 2 Diabetes Mellitus, disease trajectories, and salutogenesis (pathways to health resilience). Designed for pseudotime manifold analysis to predict disease progression.	aireadi:use:1	AI/ML model development for T2DM
Research leveraging multiple data domains (imaging, clinical, genomic, wearable, environmental, survey) to understand complex interactions and relationships in T2DM progression and management.	aireadi:use:2	Multi-modal T2DM research
Studies examining racial and ethnic disparities in T2DM outcomes, social determinants of health effects, and development of equitable AI/ML applications for diverse populations.	aireadi:use:3	Health equity research
Discovery of novel biomarkers for T2DM progression, complications, and salutogenesis using biospecimens from the biorepository.	aireadi:use:4	Biomarker discovery
Use as an exemplar for future AI-ready medical dataset development, demonstrating best practices in data collection, preparation, sharing, and ethical governance.	aireadi:use:5	Model dataset for AI-ready data standards

Discouraged Uses

Description	ID	Name
As enrollment is ongoing until November 2026, pilot data releases and periodic updates may not have achieved balanced distribution across all groups. Early versions should be used with awareness of this limitation.	aireadi:discouraged:1	Uses during ongoing enrollment without awareness of limitations
Dataset is for research purposes. Any AI/ML models developed should undergo appropriate clinical validation before use in patient care or clinical decision-making.	aireadi:discouraged:2	Clinical decision-making without validation
Attempts to re-identify participants from de-identified data violate ethical principles and data use agreements. The license explicitly prohibits attempts to identify or contact individual data subjects or groups.	aireadi:discouraged:3	Re-identification attempts

License and Use Terms

ID: aireadi:license:1
Name: Creative Commons Attribution Non-Commercial and AI-READI Data License
Description: Public access data distributed under Creative Commons Attribution Non-Commercial (CC BY-NC 4.0) license. Permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. Controlled access data requires a separate data use agreement with the University of Washington as Licensor. The AI-READI Data License Agreement prohibits clinical treatment decisions based on the data, re-identification attempts, and sharing data with non-licensed parties. See http://creativecommons.org/licenses/by-nc/4.0/ for full CC BY-NC 4.0 license terms and https://docs.aireadi.org/ for the AI-READI specific license terms.

Motivation

Composition

Collection Process

Uses

Distribution

Maintenance

Human Subjects