Factor analytic models of clustered multivariate data with informative censoring

Area of Science:

• Statistics
• Biostatistics
• Multivariate Data Analysis

Background:

• Clustered multivariate data often present challenges due to missing observations.
• Informative missingness, where missingness depends on unobserved data, complicates standard analysis.
• Existing methods may not adequately address the complex dependencies in such data.

Purpose of the Study:

• To propose a general class of factor analytic models for analyzing clustered multivariate data with informative missingness.
• To develop a flexible framework that accounts for dependencies between primary outcomes and the censoring process.
• To provide a method that can be implemented using computational techniques like Markov chain Monte Carlo.

Main Methods:

• Utilizing factor analytic models to capture latent structures within clustered data.
• Employing a hierarchical model to link cluster-level latent variables to outcomes and censoring.
• Integrating linear and generalized linear models to relate covariates and latent variables to outcomes and censoring probabilities.
• Leveraging Markov chain Monte Carlo (MCMC) for posterior estimation.

Main Results:

• The proposed factor analytic model effectively handles clustered multivariate data with informative missingness.
• The hierarchical structure successfully models dependencies between latent variables influencing outcomes and censoring.
• The methodology allows for multiple latent variables and covariate effects, offering flexibility.
• Demonstrated successful application using data from a spermatotoxicity study.

Conclusions:

• The developed factor analytic models provide a robust approach for analyzing complex clustered data with informative missingness.
• The method offers a flexible framework for understanding relationships between covariates, latent variables, outcomes, and censoring.
• The approach facilitates robust statistical inference in the presence of missing data patterns.
• The study highlights the utility of advanced statistical modeling in biological and toxicological research.