Dynamic Factor Analysis for Multivariate Time Series: An Application to Cognitive Trajectories

Area of Science:

• Epidemiology
• Biostatistics
• Neuroscience

Background:

• Large-scale epidemiological studies generate complex datasets.
• Analyzing longitudinal cognitive data presents statistical challenges.
• Existing models may not efficiently handle large numbers of subjects with limited temporal data.

Purpose of the Study:

• To propose a dynamic factor model suitable for large epidemiological datasets.
• To develop an efficient estimation algorithm for such models.
• To evaluate the model's performance using real-world Alzheimer's data.

Main Methods:

• A two-cycle iterative Expectation-Maximization (EM) algorithm for parameter estimation.
• Application of the dynamic factor model to National Alzheimer Coordinating Center (NACC) data.
• Comparison of the dynamic model against a non-dynamic counterpart.

Main Results:

• The dynamic factor model demonstrated superior fit statistics compared to the non-dynamic version in simulations.
• The proposed estimation algorithm effectively handles large datasets with sparse temporal information.
• The dynamic model showed increased statistical power to detect differences in cognitive decline rates.

Conclusions:

• Dynamic factor models are advantageous for analyzing longitudinal cognitive data in large epidemiological studies.
• The developed algorithm provides an efficient method for parameter estimation in these complex datasets.
• This approach enhances the ability to track and understand cognitive trajectories, particularly in neurodegenerative diseases.