Bayesian mixed-effects location and scale models for multivariate longitudinal outcomes: an application to ecological...

Area of Science:

• Statistics
• Biostatistics
• Health Data Science

Background:

• Traditional statistical methods focus on single outcomes, but health studies increasingly require joint analysis of multiple variables.
• Technological advancements generate complex datasets, necessitating sophisticated approaches to model multivariate outcomes and their correlations.

Purpose of the Study:

• To propose a generalized Bayesian framework for the joint analysis of multivariate continuous outcomes.
• To simultaneously model the impact of explanatory variables on outcome variations and capture inter-outcome correlations.

Main Methods:

• A Bayesian approach incorporating random effects at both location and scale levels to model outcome variation and correlations.
• Utilizing a spherical transformation for random location and scale effects to enable prior elicitation and ensure covariance matrix properties.

Main Results:

• The proposed method effectively models changes in variation based on explanatory variables.
• It successfully captures correlations between multivariate continuous outcomes using random effects.
• Demonstrated applicability through an ecological momentary assessment study in adolescents.

Conclusions:

• The developed Bayesian framework provides a comprehensive method for joint analysis of multivariate health outcomes.
• This approach is valuable for health-related studies dealing with complex, correlated data, enhancing statistical modeling capabilities.