Serial correlation structures in latent linear mixed models for analysis of multivariate longitudinal ordinal...

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Mixed-Effects Models

Background:

• Multivariate longitudinal data with ordinal variables are common in clinical research.
• Traditional methods may not adequately capture the underlying continuous latent structures and serial dependencies.
• Misspecification of variance components can lead to inaccurate conclusions in longitudinal studies.

Purpose of the Study:

• To propose a latent linear mixed model for analyzing multivariate longitudinal ordinal data.
• To investigate the effects of covariates on latent variables while accounting for serial correlation.
• To develop a graphical tool for detecting serial correlation in latent stationary linear mixed models.

Main Methods:

• Developed a latent linear mixed model incorporating serial correlation in the variance component.
• Focused analysis on the latent variable level to assess covariate effects.
• Introduced latent empirical semi-variograms for serial correlation detection.

Main Results:

• Demonstrated that misspecifying the variance component can yield misleading inferences.
• The proposed graphical tool aids in identifying serial correlation in latent models.
• Application to amyotrophic lateral sclerosis (ALS) data showed a treatment effect.

Conclusions:

• The proposed latent linear mixed model effectively analyzes complex longitudinal ordinal data.
• Accounting for serial correlation is crucial for valid statistical inference.
• The treatment investigated demonstrated a capacity to decelerate the progression of latent cervical and lumbar functions in ALS patients.