NACC data: Who is represented over time and across centers, and implications for generalizability

Kwun Chuen Gary Chan, Fan Xia, Walter A. KukullAlzheimer 2025 · DOI

TL;DR

This study examines the demographic and clinical characteristics of participants in the National Alzheimer's Coordinating Center (NACC) Uniform Data Set (UDS) over two decades and across multiple research centers. It reveals significant temporal shifts and substantial between-center variability, highlighting challenges and opportunities for generalizing findings from this valuable, albeit non-nationally representative, dataset.

What Problem Does This Paper Address?

The Alzheimer's Disease Research Centers (ADRCs) have been collecting data through the NACC UDS since 2005, providing a rich resource for Alzheimer's research. However, concerns persist about the generalizability of findings derived from this data. This is because the NACC UDS is a volunteer sample, not a statistically representative sample of the broader U.S. older population. Furthermore, each ADRC employs its own recruitment strategies, leading to potential heterogeneity in the participant pool across different centers. This study aims to address this gap by systematically investigating how participant characteristics have evolved over time and how they vary across the NACC's participating centers, ultimately informing how we interpret and apply research findings from this dataset.

Methods

The researchers utilized data from the NACC UDS, focusing on participants recruited between 2005 and 2024. To ensure data stability, they included only centers with at least three years of recruitment. Their analysis involved assessing temporal trends and between-center variation in several baseline characteristics: age, sex, race, education level, clinical diagnosis, referral source, family history of cognitive impairment, and co-participant relationship. For temporal trends, they employed the multinomial Cochran-Armitage test. To assess inter-center variation, they used the Chi-square test of homogeneity. Statistical analyses were performed using R.

Datasets

The primary dataset used in this study is the National Alzheimer's Coordinating Center (NACC) Uniform Data Set (UDS). This dataset contains longitudinal clinical measurements and diagnostic data from participants enrolled in Alzheimer's Disease Research Centers (ADRCs) across the United States. The data used for this analysis was extracted from a September 2024 data freeze, covering participants recruited from 2005 to 2024.

Key Insights and Ideas

  • Dynamic Cohort: The NACC UDS is not a static snapshot but an evolving cohort. Participant characteristics have shifted over time, reflecting changes in research priorities and outreach efforts.
  • Heterogeneity is the Norm: There's significant variation in participant profiles not only over time but also between different ADRCs. This means no single center perfectly mirrors the overall NACC population.
  • Generalizability is Nuanced: While the NACC UDS isn't nationally representative, its findings can still be generalizable. The key lies in employing appropriate analytical strategies to account for temporal and site-level variations.
  • Opportunities in Diversity: The very heterogeneity that complicates direct generalization also offers opportunities. It allows for the study of diverse subpopulations and potentially more robust insights into the mechanisms of Alzheimer's disease.

Key Results

  • Age: Early recruitment focused on older adults, but more recent enrollment shows a broader age distribution, with an increasing proportion of participants under 70. This trend is statistically significant (p < 0.001).
  • Sex: While 43% of participants were male overall, there was no significant time trend. However, the proportion of male participants varied significantly across centers (p < 0.001), ranging from 23% to 53%.
  • Race: There was a modest increase in the proportion of African American and other racial/ethnic minorities over time (p < 0.001). Most centers maintained a high proportion of White participants (65-95%), but two centers had substantially lower proportions. Significant inter-center variation was observed (p < 0.001).
  • Education: The proportion of participants with a high school education or less has declined over time (p < 0.001). Centers showed significant variation in educational representation, with proportions of high school or less education ranging from 6% to 42%.
  • Clinical Diagnosis: Participants were more likely to be cognitively normal at enrollment over time (p < 0.001). However, baseline clinical diagnosis distribution varied substantially across centers (p < 0.001), with normal cognition ranging from 18% to 76%.
  • Referral Source: Healthcare provider referrals increased over time, peaking around 2021 (p < 0.001). Referral type also showed substantial heterogeneity across centers (p < 0.001).
  • Family History: 58% of participants had at least one first-degree relative with cognitive impairment, with no significant time trend (p = 0.21). This proportion varied significantly across centers (p < 0.001), from 48% to 80%.
  • Co-participant Relationship: There was a shift towards more child co-participants and fluctuating proportions of spouse and other co-participants over time (p < 0.001). Substantial heterogeneity was also noted across centers, with spouse co-participants ranging from 29% to 79%.

Limitations and Future Work

The authors acknowledge that the NACC UDS samples are not representative of the local regions or the U.S. population in a traditional statistical sense. This limits direct generalization. However, they argue that the diversity within these samples broadens the scope of generalizable insights.

Future work should focus on:

  • Documenting Center-Specific Practices: More detailed documentation of recruitment practices and their evolution over time at each center is needed.
  • Transportability of Estimates: Further research is required to understand how to transport findings from non-representative samples to target populations, potentially using advanced analytical methods.
  • Mechanisms of Selection Bias: Investigating the specific mechanisms of selection bias for various exposures and outcomes is crucial.
  • Advanced Analytical Techniques: Developing and applying advanced statistical methods, such as sensitivity analyses and meta-analytic techniques, can help assess and enhance the generalizability of NACC findings. Treating centers as separate studies in meta-analyses is suggested as a feasible approach given the large number of ADRCs.

Summary updated

Abstract

The National Alzheimer's Coordinating Center (NACC) Uniform Data Set has enrolled participants for 20 years across more than 40 centers. We identified temporal trends and site-level variation in participant characteristics in the initial visit. Despite being a volunteer sample, modern epidemiologic and biostatistical approaches can help assess and enhance the generalizability of scientific findings derived from NACC data.
BibTeX
@inproceedings{chan2025nacc,
  author = {Kwun Chuen Gary Chan and Fan Xia and Walter A. Kukull},
  title = {{NACC data: Who is represented over time and across centers, and implications for generalizability}},
  booktitle = {Alzheimer},
  year = {2025},
  doi = {10.1002/alz.70657},
}
Figure