Generalized Multilevel Multisource Functional Regression With an Application to Alzheimer's Disease

Area of Science:

• Biostatistics
• Neuroscience
• Data Science

Background:

• Population aging increases Alzheimer's disease (AD) prevalence.
• Longitudinal studies like ADNI generate complex, multilevel, multisource functional data.
• Understanding disease progression requires advanced analytical methods for this data.

Purpose of the Study:

• To propose a novel estimation method for multilevel multisource functional principal components.
• To develop a generalized multilevel multisource functional regression model.
• To compare two-stage estimation with Bayesian estimation for improved AD classification.

Main Methods:

• Applied multilevel functional principal component analysis to univariate data.
• Estimated multisource principal components using relationships between univariate and multisource models.
• Developed a two-stage estimation method for functional regression.
• Incorporated Bayesian estimation based on the two-stage approach.

Main Results:

• Both proposed two-stage and Bayesian estimation methods demonstrated effectiveness in simulations and empirical analyses.
• Bayesian estimation significantly outperformed two-stage methods in classification performance.
• The methods effectively handle multilevel and multisource functional data for disease progression studies.

Conclusions:

• The novel two-stage estimation method provides a robust approach for analyzing complex functional data in AD research.
• Bayesian estimation offers superior classification accuracy, highlighting its potential for clinical applications in Alzheimer's disease.
• These advancements aid in understanding and potentially predicting Alzheimer's disease progression using diverse datasets.