Joint modeling of survival time and longitudinal outcomes with flexible random effects

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Joint models with shared Gaussian random effects are standard for longitudinal and survival data.
• Normality assumption violations in random effects can bias parameter estimation and prediction.

Purpose of the Study:

• To develop robust joint models for longitudinal outcomes and survival endpoints.
• To address the limitations of the normality assumption for shared random effects.

Main Methods:

• Proposed a joint model accommodating unknown distributions for shared random effects.
• Utilized a mixture of Gaussian distributions as an approximation for inference.
• Employed the Expectation-Maximization (EM) algorithm for computation.
• Applied Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) for model selection.

Main Results:

• Simulation studies demonstrated the proposed method's effectiveness.
• The approach showed improved performance compared to methods assuming normality.

Conclusions:

• The proposed mixture of Gaussian distributions offers a flexible and robust alternative for joint modeling.
• This method enhances the reliability of parameter estimation and prediction in complex biomedical data analysis.