Multivariate piecewise joint models with random change-points for skewed-longitudinal and survival data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Traditional joint models for longitudinal and time-to-event data often use simplified linear mixed-effects and Cox models.
• These standard approaches may not adequately represent complex biological processes, such as nonlinear longitudinal trajectories or non-normal data distributions.
• Ignoring correlations among multiple longitudinal outcomes can lead to biased results in survival analysis.

Purpose of the Study:

• To develop and apply Bayesian multivariate piecewise joint models for correlated longitudinal data.
• To address departures from normality in longitudinal measures using skewed distributions.
• To incorporate random change-points to better model nonlinear longitudinal trajectories and their association with event times.

Main Methods:

• Proposed Bayesian multivariate piecewise joint models incorporating skewed distributions and random change-points.
• Utilized a piecewise (broken-stick) nonlinear approach to model longitudinal trajectories.
• Linked multivariate longitudinal outcomes to time-to-event processes within a Bayesian framework.

Main Results:

• Demonstrated the methodology using a real-world motivating example.
• Simulation studies confirmed the performance and accuracy of the proposed joint models.
• The models effectively handle correlated multivariate longitudinal data and non-normal distributions.

Conclusions:

• The proposed Bayesian multivariate piecewise joint models offer a flexible and robust approach for analyzing complex longitudinal and survival data.
• These models provide improved accuracy in quantifying treatment effects and disease progression by accounting for nonlinear trajectories and data complexities.
• The methodology enhances clinical decision-making by providing more reliable estimates of the association between longitudinal health measures and event times.