A Latent-Class Model for Time-To-Event Outcomes and High-Dimensional Imaging Data

Area of Science:

• Neuroimaging
• Biostatistics
• Computational Biology

Background:

• Structural magnetic resonance imaging (MRI) is crucial for predicting Alzheimer's disease (AD) risk and enabling precision medicine.
• Existing models often lack flexibility in capturing population heterogeneity and dynamic disease progression in AD research.

Purpose of the Study:

• To propose a novel latent-class model for analyzing structural MRI data in Alzheimer's disease research.
• To address population heterogeneity and model varying covariate-survival outcome relationships in AD progression.
• To develop a robust computational framework for implementing the proposed statistical model.

Main Methods:

• A latent-class model incorporating bivariate splines on triangulated brain image domains was developed.
• A generalized expectation-maximization (EM) algorithm was designed, integrating logistic regression and penalized proportional hazards models.
• The method was validated through extensive simulation studies and applied to the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.

Main Results:

• The proposed latent-class model effectively captures heterogeneity in Alzheimer's disease patient populations.
• Bivariate splines accommodate the complex, irregular domains of brain imaging data.
• The generalized EM algorithm provides an efficient computational approach for model implementation.

Conclusions:

• The developed latent-class model offers a flexible and powerful tool for Alzheimer's disease research using structural MRI.
• This approach facilitates the identification of distinct AD subtypes or disease stages, advancing precision medicine.
• Application to the ADNI study demonstrates the model's utility in revealing disease heterogeneity.