Bayesian modeling of multiple episode occurrence and severity with a terminating event

Area of Science:

• Biostatistics
• Epidemiology
• Perinatal Health

Background:

• Individual health events, such as bleeding during pregnancy, can signal risks for adverse outcomes like preterm delivery.
• Understanding the dynamic interplay between recurrent health episodes and event timing is crucial for predictive health modeling.

Purpose of the Study:

• To develop a joint statistical model for analyzing multiple recurrent health episodes and a related event time.
• To dynamically model individual episode intensity and its predictive power for terminating events, such as delivery.

Main Methods:

• A latent variable model characterizes the frequency and severity of recurrent episodes, allowing for time-varying intensity.
• A discrete-time model incorporates this latent episode intensity as a predictor for the terminating event.
• A Bayesian framework with conjugate priors and Gibbs sampling facilitates posterior computation.

Main Results:

• The proposed joint model effectively integrates dynamic episode intensity with event time prediction.
• Time-varying coefficients allow for the examination of effects at different stages, such as during gestation.
• The methodology is validated using real-world data on bleeding episodes and gestational length.

Conclusions:

• The joint modeling approach provides a robust framework for analyzing recurrent events and their association with critical time-to-event outcomes.
• This method enhances the prediction of adverse events, like preterm delivery, by accounting for dynamic individual health trajectories.