Modeling and Prediction of Multiple Correlated Functional Outcomes

Area of Science:

• Statistics
• Functional Data Analysis
• Biostatistics

Background:

• Analyzing functional data with multiple correlated outcomes presents statistical challenges.
• Heterogeneous shape characteristics in functional outcomes require flexible modeling approaches.
• High dimensionality in multi-outcome functional data necessitates efficient parameter estimation.

Purpose of the Study:

• To propose a novel copula-based framework for analyzing functional data with multiple, correlated outcomes.
• To develop a computationally efficient method for parameter estimation in high-dimensional functional data.
• To demonstrate the methodology's utility in identifying group differences using diffusion tensor imaging data.

Main Methods:

• A two-step estimation procedure using the skew t family for marginal distributions.
• Gaussian copula modeling to capture dependence structures within and across outcomes.
• Karhunen-Loève expansion and EM algorithm for dimension reduction and efficient parameter estimation.

Main Results:

• The proposed method effectively estimates marginal and dependence parameters for multi-outcome functional data.
• Demonstrated accurate prediction of unknown functional outcomes based on known ones.
• Identified significant differences in marginal distributions and dependence structures between multiple sclerosis patients and controls using DTI data.

Conclusions:

• The developed copula-based approach offers a robust and efficient tool for analyzing complex functional data.
• The methodology successfully differentiates patient groups based on functional imaging data.
• The approach is generalizable to diverse functional data applications in biology and beyond.