Characterizing Alzheimer's Disease Biomarker Cascade Through Non-linear Mixed Effect Models

Alzheimer's Disease (AD) research has shifted to focus on biomarker trajectories and their potential use in understanding the underlying AD-related pathological process. A conceptual framework was proposed in such modern AD research that hypothesized biomarker cascades as a result of underlying AD pathology. In this paper, we leverage the idea of biomarker cascades and develop methods that use a non-linear mixed effect model to depict AD biomarker trajectories as a function of the latent AD disease progression. We tailored our methods to address a number of real-data challenges present in BIOCARD and ADNI studies. We illustrate the proposed methods with simulation studies as well as analysis results on the BIOCARD and ADNI data showing the ordering of various biomarkers from the CSF, MRI, and cognitive domains. We investigated cascading patterns of AD biomarkers in these datasets and presented prediction results for individual-level profiles over time. These findings highlight the potential of the conceptual biomarker cascade framework to be leveraged for diagnoses and monitoring.