Grouped multi-trajectory modeling using finite mixtures of multivariate contaminated normal linear mixed model

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Machine Learning

Background:

• Multivariate longitudinal data analysis is crucial for understanding complex biological processes.
• The Alzheimer's Disease Neuroimaging Initiative (ADNI) presents challenges due to diverse progression patterns and atypical observations.
• Existing methods struggle with modeling and clustering such heterogeneous grouped data.

Purpose of the Study:

• To propose novel statistical models for modeling and clustering multivariate longitudinal trajectories.
• To address the complexities of grouped longitudinal data, including multimodality and atypical observations.
• To extend existing models to incorporate concomitant covariates, enhancing flexibility.

Main Methods:

• Developed a finite mixture of multivariate contaminated normal linear mixed model (FM-MCNLMM).
• Introduced an extended version (EFM-MCNLMM) allowing mixing weights to depend on covariates.
• Employed alternating expectation conditional maximization algorithms for maximum likelihood estimation.

Main Results:

• The proposed FM-MCNLMM and EFM-MCNLMM models effectively handle multivariate longitudinal data.
• Demonstrated the models' utility and effectiveness through comprehensive simulations.
• Successfully applied the methodology to analyze the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort data.

Conclusions:

• The proposed mixture models offer a robust framework for analyzing complex grouped longitudinal data.
• The methodology provides valuable tools for identifying subgroups and understanding disease progression patterns.
• The models are effective in handling data with diverse features, including atypical observations, as shown in ADNI data analysis.