Quantile regression for censored mixed-effects models with applications to HIV studies

Area of Science:

• Biostatistics
• Epidemiology
• Infectious Diseases

Background:

• Viral load measurements in HIV studies often have detection limits, leading to censored data.
• Standard mixed-effects models rely on distributional assumptions that may not hold, impacting inference robustness.
• Censored longitudinal data analysis is crucial for understanding HIV disease progression and treatment efficacy.

Purpose of the Study:

• To propose a fully Bayesian quantile regression approach for analyzing censored longitudinal HIV viral load data.
• To offer a more robust statistical inference method compared to conventional mean regression models.
• To characterize the entire conditional distribution of viral load, not just the mean.

Main Methods:

• Utilized Markov Chain Monte Carlo (MCMC) methods for Bayesian inference.
• Developed a hierarchical Bayesian model assuming an asymmetric Laplace distribution for error terms.
• Applied quantile regression to handle left and right censored responses in longitudinal models with random effects.

Main Results:

• The Bayesian quantile regression demonstrated robustness to outliers and distributional misspecification.
• The method effectively analyzed censored viral load data from HIV/AIDS studies.
• Provided insights into the full distribution of viral load, beyond the median.

Conclusions:

• Bayesian quantile regression offers a powerful and robust alternative for analyzing censored longitudinal HIV viral load data.
• This approach enhances the reliability of statistical inference when standard assumptions are violated.
• The methodology is valuable for HIV research, offering a more comprehensive understanding of viral dynamics.