Bayesian Joint Modeling of Multivariate Longitudinal and Survival Data With an Application to Diabetes Study

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Research

Background:

• Traditional joint models for longitudinal and time-to-event data often assume normality for longitudinal outcomes.
• Deviations from normality (skewness, heavy tails) and correlated multiple longitudinal outcomes can bias traditional analyses.
• Ignoring correlations among multiple longitudinal outcomes can lead to inaccurate estimations in epidemiological and clinical studies.

Purpose of the Study:

• To introduce robust multivariate joint (MVJ) models using Bayesian inference for correlated multiple longitudinal outcomes.
• To address departures from normality in longitudinal data using a skew-normal (SN) distribution.
• To improve the analysis of time-to-event data linked to multiple, correlated longitudinal exposures.

Main Methods:

• Developed a Bayesian joint modeling approach for MVJ models.
• Coupled a multivariate linear mixed-effects (MLME) model with a skew-normal (SN) distribution.
• Integrated the MLME-SN model with a Cox proportional hazards model for time-to-event analysis.

Main Results:

• Simulation studies demonstrated the effectiveness of the proposed Bayesian MVJ models.
• The models successfully handled correlated multiple longitudinal outcomes with skewness.
• Application to a diabetes study showcased the practical utility of the approach.

Conclusions:

• The proposed Bayesian MVJ models offer a robust framework for analyzing complex longitudinal and time-to-event data.
• These models effectively account for non-normality and correlations in multiple longitudinal measurements.
• The methodology provides a valuable tool for epidemiological and clinical research, enhancing the reliability of inference.