Bayesian blockwise inference for joint models of longitudinal and multistate data with application to longitudinal...

Area of Science:

• Biostatistics
• Health Informatics
• Epidemiology

Background:

• Multistate models are crucial for analyzing complex event history data in clinical research.
• Joint modeling extends these to incorporate informative longitudinal covariates like biomarkers.
• Computational challenges hinder the practical application of these advanced models.

Purpose of the Study:

• To introduce novel Bayesian inference approaches for joint multistate models.
• To address computational challenges in analyzing large-scale longitudinal health records.
• To enable accurate modeling of multimorbidity and disease progression.

Main Methods:

• Developed Bayesian inference approaches decomposing estimation into smaller, parallelizable blocks.
• Utilized simulation studies to assess estimation accuracy and computational efficiency.
• Applied methods to a large UK health records dataset (Clinical Practice Research Datalink Aurum).

Main Results:

• Proposed methods demonstrate satisfactory estimation accuracy.
• Achieved significant computational efficiency gains compared to standard Bayesian strategies.
• Identified distinct, previously unrecognized association structures between systolic blood pressure and chronic disease transitions.

Conclusions:

• The novel Bayesian approaches offer a computationally efficient and accurate solution for joint multistate modeling.
• These methods facilitate the analysis of complex, large-scale longitudinal health data.
• The findings provide new insights into the interplay between blood pressure and chronic disease progression.