Computationally efficient Bayesian inference for semi-parametric joint models of competing risks survival and skewed...

Area of Science:

• Biostatistics
• Medical Informatics
• Epidemiology

Background:

• Joint modeling analyzes longitudinal biomarkers and survival outcomes simultaneously, crucial for public health interventions.
• Existing models are computationally intensive, especially for competing risks and skewed longitudinal data.
• There's limited research on joint modeling of competing risks, survival, and skewed longitudinal data using Integrated Nested Laplace Approximations (INLA).

Purpose of the Study:

• To present a computationally efficient inference approach for joint modeling of competing risks survival and skewed longitudinal data.
• To enable prompt decision-making in clinical and epidemiological settings through efficient statistical methods.

Main Methods:

• Developed cause-specific competing risks joint models with semi-parametric mixed-effects longitudinal submodels and random walk hazards.
• Reformulated models as latent Gaussian models for efficient Bayesian inference using INLA.
• Compared INLA with Markov-Chain Monte-Carlo (MCMC) using INLAjoint and R2WinBUGS R packages, evaluating various smoothing splines, distributions, and association structures.

Main Results:

• Evaluated computational efficiency and estimation performance using chronic kidney disease (CKD) data and simulation studies.
• Compared models with different specifications for smoothing splines, skewed distributions, and association structures.
• Both INLA and MCMC provided accurate statistical estimation and inference.

Conclusions:

• Integrated Nested Laplace Approximations (INLA) significantly reduces the computational burden for joint models of competing risks survival and skewed longitudinal data.
• The proposed INLA approach ensures robust statistical inference and accurate estimation, particularly beneficial for complex medical research.