Determination of correlations in multivariate count data with informative observation times

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Recurrent event data often involves complex dependencies between event occurrences and observation times.
• Existing methods struggle to directly quantify these intricate correlation structures.
• Nonparametric models are available but lack direct correlation assessment.

Purpose of the Study:

• To propose a novel multivariate frailty model for analyzing correlated recurrent events and random observation times.
• To explicitly model and quantify the dependence structure between event processes and observation times.
• To apply the model to a real-world skin cancer prevention study.

Main Methods:

• Development of a multivariate frailty model linking event and observation processes via shared frailty variables.
• Utilizing a multivariate normal distribution to implicitly specify the joint distribution of frailties.
• Employing Bayesian inference for estimating regression coefficients and correlation parameters.
• Using trigonometric functions for efficient positive-definite covariance matrix representation.

Main Results:

• Simulation studies confirmed the model's utility and effectiveness.
• In a skin cancer study, treatment significantly affected examination time.
• Prior tumor counts, age, and gender were significant predictors of tumor occurrence rates.
• Analysis revealed positive correlations among event types and a notable association between basal cell counts and examination times.

Conclusions:

• The proposed multivariate frailty model effectively captures complex dependencies in recurrent event data.
• The model provides a robust framework for quantifying covariate and association effects in correlated event processes.
• Findings from the skin cancer study highlight key factors influencing disease progression and monitoring.