Dependence modeling for multi-type recurrent events via copulas.

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Analyzing recurrent events with multiple types presents challenges in understanding dependence structures.
• Existing models may not adequately capture heterogeneity in risk across different event types.

Purpose of the Study:

• To propose a novel multivariate mixed-Poisson model for analyzing multiple recurrent events.
• To model the dependence structure between different types of recurrent events using type-specific random effects and a Gaussian copula.
• To apply the model to assess the impact of nutritional supplements on infection rates in malnourished children.

Main Methods:

• Development of a multivariate mixed-Poisson model incorporating type-specific random effects.
• Utilizing a Gaussian copula to associate random effects and model dependence.
• Employing semiparametric inference based on composite likelihood for computational efficiency.

Main Results:

• The proposed model effectively captures marginal features and provides interpretable results.
• It accounts for heterogeneity in risk for each event type.
• The model offers insights into the dependence structure between different types of recurrent events.

Conclusions:

• The multivariate mixed-Poisson model with Gaussian copula is a flexible and interpretable tool for analyzing multiple recurrent events.
• This approach is valuable for understanding complex event patterns and risk factors in health studies.
• The application demonstrates the model's utility in evaluating interventions like nutritional supplements.