A stochastic model for the analysis of bivariate longitudinal AIDS data

Area of Science:

• Biostatistics
• Epidemiology
• Immunology

Background:

• Multivariate longitudinal data analysis is crucial for understanding complex health trajectories.
• Existing models may not fully capture the dynamics of correlated biological markers over time.
• Accurate modeling is needed to investigate relationships between immune markers in diseases like HIV/AIDS.

Purpose of the Study:

• To introduce a flexible statistical model for multivariate repeated measures incorporating random effects and stochastic processes.
• To generalize existing univariate longitudinal data models to a multivariate setting.
• To analyze the dynamic relationship between CD4 T-cell counts and beta-2-microglobulin levels in HIV seroconverters.

Main Methods:

• Developed a multivariate generalization of longitudinal data models using the multivariate integrated Ornstein-Uhlenbeck process.
• The model accommodates random effects, correlated stochastic processes, and measurement errors.
• Applied the model to analyze CD4 and beta-2-microglobulin data from the Multicenter AIDS Cohort Study, allowing for unequally spaced and missing observations.

Main Results:

• The analysis of CD4 and beta-2-microglobulin measurements suggested a bivariate Brownian motion process.
• The model demonstrated significant serial correlations and correlations between random effects for CD4 and beta-2-microglobulin.
• Model fitting indicated that an increase in beta-2-microglobulin is associated with a subsequent decrease in CD4 cell counts.

Conclusions:

• The proposed multivariate model effectively captures the complex dynamics of repeated measures, including immune markers.
• The findings support the immunologic postulate that elevated beta-2-microglobulin predicts declining CD4 counts in HIV infection.
• This statistical framework provides a robust tool for investigating biomarker relationships in longitudinal health studies.