Identifiable Copula-Double-Cox Models: A Fully Parametric Framework for Dependent Right-Censored Survival Data.

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Dependent censoring, common in medical studies with informative dropout, violates the independent censoring assumption required by standard Cox regression.
• Existing copula-based methods for modeling dependence can suffer from identifiability issues in parametric extensions.

Purpose of the Study:

• To introduce a novel, fully identifiable parametric model, the copula-double-Cox model, to address dependent censoring in medical studies.
• To synergize double-Cox marginal structures with copula dependence for flexible and identifiable modeling.

Main Methods:

• The copula-double-Cox model utilizes Weibull or generalized exponential (GenExp) distributions within a double-Cox framework, linking scale and shape parameters to covariates.
• The model establishes identifiability under dependent censoring and derives consistent estimators for baseline parameters, regression coefficients, and copula association.
• Leverages Cox-type regressions for covariate effects on both failure and censoring times, accommodating non-proportional hazards.

Main Results:

• Identifiability is established under dependent censoring, with consistent estimators derived for key model parameters.
• Simulation studies demonstrate robustness to association structure misspecification and over-parameterization.
• The model accurately estimates covariate effects on both failure and censoring times, as confirmed by asymptotic theory and application to a real-world dataset.

Conclusions:

• The copula-double-Cox model offers a practical and interpretable solution for analyzing data with dependent censoring, improving upon existing methods.
• The proposed method provides a comprehensive characterization of covariate influences on both survival and dropout processes.
• An open-source R implementation is available, facilitating broader application in biostatistical research.