Dynamic prediction using joint models of longitudinal and recurrent event data: A Bayesian perspective.

Area of Science:

• Biostatistics
• Cardiovascular Disease Research
• Clinical Epidemiology

Background:

• Cardiovascular disease (CVD) studies often involve recurrent events, where individuals experience multiple occurrences.
• Longitudinal measurements collected during follow-up are crucial predictors of these recurrent events.
• Personalized prediction of recurrent events is vital for informed clinical decision-making.

Purpose of the Study:

• To propose a novel joint model for analyzing longitudinal and recurrent event data in CVD studies.
• To develop a Bayesian inference approach and a dynamic prediction framework for personalized risk assessment.
• To enable accurate prediction of future outcome trajectories and the risk of subsequent recurrent events.

Main Methods:

• Development of a joint model integrating longitudinal measurements and recurrent event data.
• Application of a Bayesian approach for robust model inference.
• Utilization of an embarrassingly parallel Markov Chain Monte Carlo (EP-MCMC) method for computational efficiency, involving data partitioning and random partition trees.

Main Results:

• The proposed joint model effectively integrates longitudinal data with recurrent event processes.
• The dynamic prediction framework allows for real-time risk assessment based on individual patient data.
• The EP-MCMC method significantly improves computational efficiency for large datasets.

Conclusions:

• The developed joint modeling framework provides a powerful tool for personalized risk prediction in cardiovascular disease.
• This approach enhances clinical decision-making by offering dynamic and individualized predictions of recurrent events.
• The study demonstrates the successful application of the method to the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT).