An unsupervised learning approach for clustering joint trajectories of Alzheimer's disease biomarkers: An application to ADNI Data.

Introduction: Current models of Alzheimer's disease (AD) progression assume a common pattern and pathology, oversimplifying the heterogeneity of clinical AD.

Methods: We define a syndrome as a unique biomarker progression pattern and develop a lag measure to cluster pre-dementia individuals, reflecting their pathology's multi-dimensionality. The technique uses the time-ordering of events to group individuals based on their position along the disease process and the relative positions of their markers.

Results: An application using Alzheimer's Disease Neuroimaging Initiative (ADNI) data highlights the need for our novel approach to clustering individuals into syndrome groups.

Discussion: Accurately characterizing biomarker curves associated with brain damage requires an initial step that groups individuals on a syndrome basis, accounting for the heterogeneity of underlying pathologies in clinical AD.

Highlights: Developed a novel distance measure and clustering approach for AD biomarker trajectories. Identified distinct subgroups with different biomarker progression patterns in ADNI data. Findings challenge the traditional amyloid cascade hypothesis and suggest AD heterogeneity. Clustering approach accounts for shifts in time and emphasizes progression patterns. Results have implications for AD diagnosis, targeted interventions, and clinical trials.