Bayesian Two-Part Tobit Models with Left-Censoring, Skewness, and Nonignorable Missingness

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Clinical Trials

Background:

• Longitudinal HIV/AIDS studies often face challenges with nonignorable missing data due to patient dropouts.
• Response variables can exhibit skewness and left-censoring, particularly when values fall below the limit of detection (LOD).
• Existing models may not adequately address the simultaneous presence of these complex data features.

Purpose of the Study:

• To develop and illustrate an advanced statistical model for analyzing longitudinal data in HIV/AIDS studies.
• To account for left-censoring, skewness, nonignorable missingness, and measurement error in covariates.
• To provide a robust inferential framework for complex clinical trial data.

Main Methods:

• Extension of the random effects Tobit model.
• Incorporation of a mixture model for undetectable observations and skew-normal distributions.
• Application of Bayesian inference to handle left-censoring, skewness, nonignorable missingness, and covariate measurement error.

Main Results:

• The proposed unified modeling approach effectively handles multiple data complexities in longitudinal HIV/AIDS studies.
• The methods were successfully illustrated using real-world data from an AIDS clinical study.
• The model provides a more accurate assessment of outcomes compared to standard methods when dealing with censored and missing data.

Conclusions:

• The developed statistical model offers a powerful tool for analyzing complex longitudinal HIV/AIDS data.
• This approach enhances the reliability of findings in clinical trials with challenging data characteristics.
• The study highlights the importance of using appropriate statistical methods for accurate interpretation of HIV/AIDS research data.