Simulation Methods to Inform Study Design and Understand Potential Bias: The Example of Random Error in AD Biomarkers

TL;DR

The paper uses simulations to show how random errors in blood-based Alzheimer's biomarkers can affect study design and results compared to traditional measures.

Contribution

The novel contribution is using simulation methods to quantify the impact of random error in blood-based AD biomarkers on study design and bias.

Findings

Blood-based biomarkers like p-tau 181 have higher random measurement error than neuroimaging measures.

Random error in blood biomarkers can reduce statistical power and increase required sample sizes.

Using error-prone biomarkers as independent variables can introduce bias in estimating relationships with cognitive outcomes.

Abstract

Novel blood-based biomarkers for Alzheimer’s disease pathology are increasingly available and are appealing options for inclusion in population-based studies due to numerous advantages related to cost and low invasiveness. However, they are subject to more random measurement error (i.e., noise) relative to more traditional biological measures such as neuroimaging or CSF. We leverage data from the Alzheimer’s Disease Neuroimaging Initiative and use a simulation study to highlight potential implications of random measurement error for power and sample size requirements when using blood-based biomarkers (e.g., p-tau 181) vs. neuroimaging (amyloid PET) to understand the effects of risk factors like age on biomarker levels, and additional implications for bias when using the biomarker as an independent variable to evaluate the relationship between biomarkers and cognitive outcomes (e.g. Clinical Dementia Rating Sum of Boxes). More broadly, we highlight the benefits of simulation studies to inform study design, understand sources of bias, and investigate other methodological issues in population aging research.