Hierarchical mixture models for longitudinal immunologic data with heterogeneity, non-normality, and missingness.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Bayesian Modeling

Background:

• Longitudinal data in medical studies often exhibit complex trajectories with multiple phases (changepoints).
• Standard mixed-effects models struggle with analyzing data featuring changepoints, population heterogeneity, non-normality, and non-ignorable missingness simultaneously.
• Existing literature has limited work addressing all these complex data characteristics concurrently.

Purpose of the Study:

• To develop a novel Bayesian approach to simultaneously address multiple challenges in longitudinal data analysis.
• To introduce Finite Mixture of Changepoint (piecewise) Mixed-Effects (FMCME) models incorporating skew distributions.
• To estimate model parameters and class membership probabilities at both population and individual levels.

Main Methods:

• Developed Finite Mixture of Changepoint (piecewise) Mixed-Effects (FMCME) models using a Bayesian mixture modeling approach.
• Incorporated skew distributions to handle non-normality in the data.
• Utilized simulation studies to evaluate the performance of the proposed FMCME models.
• Applied the methodology to an AIDS clinical data example.

Main Results:

• The proposed FMCME models effectively handle longitudinal data with changepoints, heterogeneity, non-normality, and missingness.
• Simulation studies demonstrated the robustness and reliability of the Bayesian approach.
• Analysis of AIDS clinical data showcased the practical application and comparative performance of the mixture models.

Conclusions:

• The developed Bayesian FMCME models provide a flexible and powerful framework for analyzing complex longitudinal data in medical research.
• This approach offers improved quantification of treatment effects and patient care management by accounting for multiple data complexities.
• The methodology addresses a critical gap in the literature for simultaneous analysis of these challenging data features.