Data-Driven Thresholding Statistically Biases ATN Profiling across Cohort Datasets.

Area of Science:

• Neuroscience
• Biomarker Research
• Alzheimer's Disease Diagnostics

Background:

• The amyloid/tau/neurodegeneration (ATN) framework is widely used in Alzheimer's disease (AD) research.
• Uncertainty exists regarding the concordance and interchangeability of thresholds derived from different methods across datasets and cohorts.

Purpose of the Study:

• To investigate the robustness of data-driven thresholding methods within the ATN framework.
• To assess the generalizability of ATN profiling across diverse cohort datasets.

Main Methods:

• Applied five common thresholding methodologies to ATN profiling across multiple AD cohort datasets.
• Included participants with CSF amyloid-β 1-42, phosphorylated tau, and total tau measurements from eleven AD cohorts.
• Clustered individuals from different cohorts assigned to the same ATN profile to evaluate disease pattern comparability.

Main Results:

• Observed significant variability in ATN thresholds across methods and datasets, impacting participant profile assignments.
• Identified comparable disease patterns across most cohorts when clustering individuals within the same ATN category.
• Found that biases from distinct thresholding and data representations were largely insignificant across most ATN profiles.

Conclusions:

• Thresholding method selection significantly impacts ATN profiling, affecting sensitivity and specificity.
• Thresholds are not directly interchangeable between independent Alzheimer's disease cohorts.
• A thorough understanding of thresholding methods, their statistical implications, and result validation is crucial for robust ATN framework application.