A Bayesian joint longitudinal-survival model with a latent stochastic process for intensive longitudinal data.

Area of Science:

• Biostatistics
• Digital Health
• Behavioral Science

Background:

• Mobile health (mHealth) technology facilitates intensive longitudinal data (ILD) collection, offering insights into dynamic health outcomes.
• Existing joint models for longitudinal and event-time data struggle with ILD's complexity and computational demands.

Purpose of the Study:

• To propose a novel joint longitudinal and time-to-event model optimized for analyzing intensive longitudinal data (ILD).
• To assess the model's performance through simulations and application to a smoking cessation mHealth study.

Main Methods:

• Summarizing multivariate longitudinal outcomes into time-varying latent factors using an Ornstein-Uhlenbeck stochastic process.
• Parametrically modeling the risk of a time-to-event outcome within a hazard model framework.
• Employing a Bayesian approach for model fitting and performance evaluation.

Main Results:

• The proposed model effectively analyzes ILD, summarizing complex emotional states (positive and negative affect) from 9 emotions.
• These latent states were found to capture the risk of smoking lapse following a quit attempt.
• The model demonstrated suitability for analyzing data from mHealth smoking cessation interventions.

Conclusions:

• The developed joint model offers an efficient and effective method for analyzing intensive longitudinal data in mHealth research.
• Understanding dynamic psychological states is crucial for predicting and intervening in smoking relapse.
• This approach advances the analysis of complex behavioral data in digital health studies.