Bayesian multivariate linear mixed-effects models with varied association structures

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Medical Statistics

Background:

• Analyzing multiple, repeatedly measured continuous outcomes over time is crucial for assessing disease progression.
• Unbalanced longitudinal data presents challenges in understanding associations between various health outcomes.
• Multivariate linear mixed-effects models (MLMMs) are commonly used but understanding outcome associations remains difficult.

Purpose of the Study:

• To enhance MLMMs by incorporating interpretable association structures for analyzing longitudinal outcomes.
• To investigate the relationship between primary outcomes and other longitudinal outcomes based on current or cumulative effects.
• To apply these enhanced models to understand associations in Pompe disease, linking patient-reported outcomes with physical measures.

Main Methods:

• Proposed enhanced MLMMs with novel association structures.
• Incorporated current value, cumulative effect (total or partial), or combined effects.
• Utilized a Bayesian framework with Hamiltonian Monte Carlo for model fitting.

Main Results:

• Demonstrated a positive association between patient-reported outcome measures and physical outcomes in Pompe disease.
• The enhanced MLMMs provide a more nuanced understanding of the connections between longitudinal health indicators.
• The proposed association structures offer interpretable insights into complex relationships within medical data.

Conclusions:

• Enhanced MLMMs with interpretable association structures improve the analysis of longitudinal outcomes.
• The study confirms a positive link between physical improvements and patient-reported quality of life in Pompe disease.
• This approach offers valuable tools for medical researchers studying complex disease progression and treatment effects.