Estimation in multivariate linear mixed models for longitudinal data with multiple outputs: Application to PBCseq...

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Biomedical studies often involve multiple response variables measured repeatedly over time on the same subjects.
• Analyzing such complex longitudinal data requires robust statistical methods that can handle potential outliers and select relevant variables.

Purpose of the Study:

• To develop a simultaneous variable selection and estimation method for multivariate t linear mixed-effects models (MtLMM).
• To enhance the analysis of longitudinally measured multi-outcome data by accommodating fat tails and identifying insignificant variables.

Main Methods:

• Utilized a multivariate t linear mixed-effects model (MtLMM) for longitudinal multi-outcome data.
• Employed the smoothly clipped and absolute deviation (SCAD) penalty function for simultaneous variable selection and estimation.
• Implemented an expectation conditional maximization (ECM) algorithm for parameter estimation and an information-based method for standard error calculation.

Main Results:

• The proposed penalized MtLMM demonstrated robustness and flexibility in handling data with fat tails.
• Analysis of the Primary Biliary Cirrhosis (PBCseq) data revealed that 'drugs' and 'sex' were not significant predictors and could be eliminated.
• The study confirmed disease progression over time in the PBCseq cohort.

Conclusions:

• The developed methodology provides an effective approach for variable selection and estimation in complex longitudinal studies.
• The penalized MtLMM offers a valuable tool for identifying key factors influencing disease progression and managing multi-outcome data.
• The findings highlight the importance of robust statistical modeling in biomedical research, as exemplified by the PBCseq data analysis.