Extending t linear mixed models for longitudinal data with non-ignorable dropout applied to AIDS studies

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Longitudinal data often exhibits dropout, where participants withdraw prematurely.
• Dropout can be non-ignorable, depending on unobserved data, complicating analysis.
• Existing models may not adequately handle non-ignorable dropout alongside outliers or heavy-tailed distributions.

Purpose of the Study:

• To extend t linear mixed models for continuous longitudinal data with non-ignorable dropout and outliers.
• To develop a robust statistical framework for analyzing complex longitudinal datasets.
• To improve the accuracy of parameter estimation and prediction in the presence of missing data mechanisms.

Main Methods:

• Utilized a selection modeling strategy with a logistic link function to model the probability of dropout.
• Developed a Monte Carlo Expectation Conditional Maximization (MCECM) algorithm for maximum likelihood estimation.
• Employed the Monte Carlo empirical information matrix for standard error calculation.

Main Results:

• The proposed t linear mixed-effects (tLME) model demonstrated capability in handling non-ignorable dropout and outliers.
• Simulation studies showed improved performance compared to normal counterparts in estimation precision and predictive accuracy.
• The methodology provided practical insights for analyzing longitudinal data with missingness.

Conclusions:

• The tLME model offers a robust approach for longitudinal data analysis when non-ignorable dropout and outliers are present.
• The MCECM algorithm effectively estimates model parameters and missingness indexing parameters.
• The findings have significant implications for real-world applications, such as clinical trials.