Toward a dynamic biomarker model in Alzheimer's disease

Area of Science:

• Neuroscience
• Biomarker Research
• Medical Imaging

Background:

• Alzheimer's disease (AD) progression is monitored using biological and imaging biomarkers.
• Understanding the precise relationship between these biomarkers and disease severity is crucial for accurate staging and prognosis.
• Previous hypotheses suggested a sigmoidal relationship between biomarkers and AD severity.

Purpose of the Study:

• To determine the precise shape of the association between various Alzheimer's disease biomarkers and disease severity.
• To compare the fit of six different statistical models (linear, quadratic, robust quadratic, local quadratic regression, penalized B-spline, sigmoid) to population data.
• To utilize large-scale data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) for this analysis.

Main Methods:

• Analysis of cross-sectional data from 576 subjects in the ADNI study (229 controls, 193 AD, 154 MCI converters).
• Inclusion of baseline data on cerebrospinal fluid (CSF) amyloid-β (Aβ)42, phosphorylated tau (p-tau), total-tau (t-tau), hippocampal volumes, and FDG-PET.
• Model comparison using the Akaike Information Criterion (AIC) to assess goodness-of-fit.

Main Results:

• Local quadratic regression was 42% more likely than sigmoid to best model Aβ42.
• Penalized B-spline showed superior fit for p-tau (22% more likely) and t-tau (73% more likely) compared to sigmoid.
• Linear models for FDG-PET (3500% more likely) and Penalized B-spline for hippocampal volumes (6700% more likely) demonstrated significantly better fits than sigmoid models.

Conclusions:

• The association between Alzheimer's disease biomarkers and disease severity is predominantly non-linear.
• Different biomarkers exhibit distinct non-linear relationships with disease severity.
• These findings challenge the generalized sigmoidal hypothesis and highlight the complexity of AD biomarker dynamics.