Backward joint model and dynamic prediction of survival with multivariate longitudinal data

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Dynamic prediction of time-to-event outcomes relies on joint modeling of longitudinal and time-to-event data.
• Shared random effect models are common but computationally challenging with many longitudinal predictors.

Purpose of the Study:

• To develop an alternative joint modeling approach for longitudinal and time-to-event data.
• To address the computational limitations of existing models when incorporating numerous longitudinal predictors.
• To improve the accuracy of dynamic prediction in complex scenarios.

Main Methods:

• Proposed a novel joint model formulation for longitudinal and time-to-event data.
• The new model requires only one-dimensional integration, irrespective of the number of longitudinal variables.
• Employed pseudo maximum likelihood estimation, Expectation-Maximization algorithm, and convex optimization for model fitting.

Main Results:

• The proposed model demonstrates computational tractability and stability.
• The methodology effectively handles a large number of longitudinal predictors.
• Evaluated predictive accuracy through simulations and a primary biliary cirrhosis dataset.

Conclusions:

• The developed joint model offers a computationally efficient alternative for dynamic prediction with extensive longitudinal data.
• This approach is particularly advantageous for complex prediction tasks involving multiple longitudinal variables.
• The findings support the utility of this new methodology in biostatistical applications.