Substituting Blood-Based Biomarkers for Imaging Measures in Alzheimer’s Disease Studies

Implications for Sample Size and Bias

Sarah F. Ackley, Renaud La Joie, Michelle Caunca, Shubhabrata Mukherjee, Seo-Eun Choi, Emily H. Trittschuh, Paul K. Crane, and Eleanor Hayes-Larson, for ADNI

The Journals of Gerontology: Series A · 2026;81(5):glag068 · doi:10.1093/gerona/glag068

---

Key Takeaway

Blood biomarkers can improve study scale and access, but substituting plasma p-tau for amyloid-PET may require 1.5–6.5× larger samples and can substantially attenuate exposure associations.

---

Background and Objective

Blood-based biomarkers (BBMs) for Alzheimer’s disease pathology are attractive because they are lower cost, minimally invasive, and easier to collect in community and non-clinical settings.

However, BBMs have lower test–retest reliability than neuroimaging measures such as amyloid-PET Centiloids. The resulting trade-offs in statistical power and potential bias have not been well characterized.

Research Question

What happens to statistical power, the sample size required for 80% power, and bias when plasma p-tau181 or p-tau217 is used in place of amyloid-PET Centiloids?

---

Data and Simulation Design

• Empirical data came from ADNI and A4, which included both amyloid-PET and blood-based biomarker measures.

• Key measures were Centiloids, plasma p-tau181 or p-tau217, age, and harmonized memory z-scores.

• Simulations varied sample size and random measurement error across 165 scenarios.

• Each scenario included 5,000 simulated datasets to estimate power and bias.

---

Analytic Roles Compared

Biomarker as outcome:
Age → biomarker

Biomarker as exposure:
Biomarker → memory

• Measurement error in an outcome primarily reduces precision.

• Measurement error in an exposure can reduce precision and bias estimates toward the null.

---

Biomarker as Outcome: Sample Size

When the biomarker was the outcome, replacing Centiloids with plasma p-tau reduced power—especially for p-tau181.

1.5–6.5× Larger Sample Required

Plasma p-tau required approximately 1.5–6.5 times more participants than true Centiloids to achieve 80% power.

Interpretation

• Blood collection may enable larger studies, but that advantage must offset lower biomarker precision.

• Study planning should estimate power for the specific assay, cohort, and analytic question.

• A cheaper biomarker is not necessarily a cheaper study when substantially larger samples are required.

---

Biomarker as Exposure: Bias

When the biomarker was the exposure, measurement error produced regression-dilution bias and reduced power.

Model	Bias
p-tau181 in ADNI exposure models	−88%
p-tau217 in ADNI exposure models	−34%
p-tau217 in A4 exposure models	−62%

Interpretation

• Larger samples improve power but do not eliminate regression-dilution bias.

• Sample-size requirements and bias depend on the p-tau isoform and the biomarker’s analytic role.

• p-tau217 generally outperformed p-tau181.

---

Overall Interpretation

Blood-based biomarkers are valuable tools for scalable and potentially more representative Alzheimer’s disease research, but they are not one-to-one substitutes for imaging measures.

Researchers should:

• Match the biomarker to the scientific question and analytic role.

• Consider assay-specific power calculations during study planning.

• Use quantitative bias analysis when biomarkers are exposures.

• Include PET imaging in a validation subset when feasible.

---

Funding and Data Sources

National Institute on Aging Grants:
R00AG073454, R01AG082730, R00AG075317, R13AG064971, and R01AG072681.

Data Sources:
Alzheimer’s Clinical Trial Consortium A4/LEARN, funded by NIA/NIH, Eli Lilly and Co., the Alzheimer’s Association, GHR Foundation, and philanthropic donors.

Alzheimer’s Disease Neuroimaging Initiative (ADNI), funded by the National Institute on Aging/NIH (U19AG024904), with past additional support from public and private partners.