Implementing the National Alzheimer's Coordinating Center Uniform Data Set (v3) within the Diabetes Prevention Program Outcomes Study

Problem
The Diabetes Prevention Program Outcomes Study (DPPOS) is a long-standing, 30-year observational follow-up of the Diabetes Prevention Program (DPP), originally designed to prevent type 2 diabetes (T2D) in adults with prediabetes. While DPPOS began limited cognitive assessments in 2009, the study lacked a standardized framework for Alzheimer's disease (AD) and AD-related dementias (ADRD) research. As the cohort aged and the focus expanded in 2022 to include cognitive impairment, the study faced the challenge of integrating rigorous, standardized AD/ADRD data collection into an existing, complex longitudinal infrastructure that was not originally designed for this purpose. The specific problem was how to implement the National Alzheimer's Coordinating Center Uniform Data Set version 3 (NACC-UDSv3)—the standard used by Alzheimer's Disease Research Centers (ADRCs)—into the DPPOS electronic data capture (EDC) system while maintaining fidelity to both the NACC standards and the study's 25-year longitudinal data history.

Methodology
The authors implemented the NACC-UDSv3 forms into MIDAS (Multimodal Integrated Data Acquisition System), the proprietary EDC system used by DPPOS. The methodology involved several key technical steps:

• Data Harmonization: The team compared the 16 NACC-UDSv3 data forms against existing DPPOS forms. Overlapping items were integrated, and answer options for existing DPPOS questions were expanded to ensure they could be mapped to corresponding UDSv3 items. New UDSv3 items not previously collected were added to the existing forms.
• Assessment Protocol: The study utilized the UDSv3 neuropsychological battery (forms C2 for in-person and C2T for virtual visits), supplemented by a neurologic exam, the Quick Dementia Rating System (QDRS), and the NIH Toolbox Oral Reading Recognition Test.
• Adjudication Workflow: A centralized team of five adjudicators, supported by managers and a coordinator at Columbia University, performed cognitive syndrome ascertainment. The process included:
  • Auto-adjudication: An automated protocol classified participants as "normal" if they met predefined normative criteria, streamlining the workflow.
  • Dual Review: Cases not auto-adjudicated were reviewed by two independent adjudicators who completed electronic NR1 forms.
  • Consensus and Committee Review: A consensus summary report compared the two reviews. Discrepancies were resolved by the adjudicators or, if disagreement persisted on diagnosis, by a five-member adjudication committee via remote consensus meetings.
• Quality Control: All assessments were double-scored by Program Coordinators (PCs). A 10% random sample underwent QC checks by the adjudication core. The MIDAS system enforced skip patterns and inter-field checks to ensure data integrity.

Key Contributions

• Successful Integration: The paper demonstrates the successful integration of NACC-UDSv3 forms with full fidelity into a non-AD/ADRD cohort (DPPOS) that had been operating for over 25 years with a focus on diabetes.
• Technical Framework: The authors provide a practical framework for harmonizing legacy data with new standardized forms, adapting existing EDC systems (MIDAS) for complex adjudication workflows, and implementing automated adjudication protocols.
• Data Compatibility: The resulting dataset is compatible with data received by the NACC from all ADRCs in the United States, allowing DPPOS data to be shared and analyzed alongside national AD/ADRD research data.
• Operational Scale: In the first wave of the DPPOS-AD/ADRD study (2022–2024), the team successfully completed 1,561 cognitive adjudications, with 291 (19%) processed via auto-adjudication and 1,270 reviewed by human adjudicators.

Results

• Completion Rates: Of 1,683 consenting participants, 1,561 completed cognitive assessments. 83% completed in-person visits, 16% remote, and 2% hybrid.
• Adjudication Outcomes: Of the 1,270 participants reviewed by adjudicators, 1,233 (97%) received a diagnosis.
  • First-pass agreement between two adjudicators occurred in 53% of cases.
  • Second-pass agreement after reviewing minor disagreements occurred in 21% of cases.
  • 26% of cases required review by the full adjudication committee.
• Diagnosis Distribution: The study successfully classified participants into categories including normal cognition, various forms of dementia (e.g., amnestic multidomain, behavioral variant frontotemporal), mild cognitive impairment (MCI), and "cognitive impairment – not MCI."

Significance
The paper claims that this project represents the first time UDSv3 forms have been integrated into a non-AD/ADRD cohort with full fidelity. The significance lies in the ability to add valuable information on AD/ADRD to a high-risk population (persons with T2D and prediabetes) within the national NACC database. The authors posit that the harmonization, electronic capture, and automated adjudication processes described provide a practical framework for other long-standing longitudinal cohorts seeking to incorporate NACC-UDSv3 to align with national research standards. The study acknowledges limitations, such as the use of non-clinician data collectors and the lack of clinical biomarkers at the time of the report (though research biomarkers are being collected), but maintains that the methodology offers a robust model for expanding cognitive research in diverse cohorts.