τ $$ \tau $$ -Inflated beta regression model for censored recurrent events.

Area of Science:

• Biostatistics
• Survival Analysis
• Longitudinal Data Analysis

Background:

• Recurrent event data analysis presents challenges, especially with censored observations.
• Heterogeneity in event susceptibility and event-free periods complicates modeling.
• Existing methods may not adequately capture these complexities.

Purpose of the Study:

• To introduce a novel multivariate zero-inflated beta regression (zero-IBR) model.
• To analyze censored recurrent event data with a mixture of susceptible and non-susceptible individuals.
• To provide interpretable outputs for understanding event patterns and durations.

Main Methods:

• Development of a multivariate zero-IBR model for censored recurrent event data.
• Application to restructured longitudinal data with overlapping follow-up windows.
• Integration of multiple imputation (MI) and expectation-solution (ES) for model fitting.
• Generation of parameter estimates, mean event-free duration estimates, and heat maps.

Main Results:

• The zero-IBR model effectively handles censored recurrent event data.
• Provides insights into factors influencing event susceptibility and duration.
• Demonstrates good statistical performance through simulations.
• Offers practical application via an example from the Azithromycin for Prevention of COPD Exacerbations Trial.

Conclusions:

• The proposed zero-IBR modeling approach is a valuable tool for analyzing complex censored recurrent event data.
• It enhances the understanding of patient heterogeneity in event occurrence and timing.
• The method provides clinically relevant outputs for patient risk stratification and treatment evaluation.