Probabilistic clustering using shared latent variable model for assessing Alzheimer's disease biomarkers.

Area of Science:

• Neuroscience
• Biostatistics
• Biomarker Research

Background:

• Preclinical neurodegenerative diseases, like Alzheimer's disease (AD), have long preclinical stages with subtle biomarker changes.
• Early detection is challenging due to unobservable disease states and individual variability in biomarkers and disease progression.
• Hypotheses suggest patient subgroups with distinct biomarker patterns exist, influenced by comorbidities and brain resilience.

Purpose of the Study:

• To identify systematic patterns within the heterogeneous biomarker-disease cascade specific to Alzheimer's disease.
• To quantify disease progression using a dynamic latent variable model that accounts for patient subgroups.
• To enhance early diagnosis and intervention strategies for neurodegenerative diseases.

Main Methods:

• Developed a dynamic latent variable mixture model to represent patient subgroups.
• Employed Hamiltonian Monte Carlo for model estimation and the Bayesian Information Criterion for determining the number of clusters.
• Applied the model to the longitudinal Biomarkers of Cognitive Decline Among Normal Individuals (BCDN) dataset.

Main Results:

• The proposed model successfully identified distinct patterns in Alzheimer's disease biomarker progression.
• Analysis revealed two subgroups with significantly different disease-onset trajectories.
• Findings support the hypothesis of heterogeneity in the biomarker-disease relationship within AD.

Conclusions:

• The study successfully identified distinct preclinical Alzheimer's disease subgroups, offering new insights into disease heterogeneity.
• The developed dynamic prediction approach can improve the precision of prognoses for individuals in the preclinical stage.
• Understanding subgroup dynamics is crucial for developing targeted early diagnostic and therapeutic strategies.