A D-vine copula-based model for repeated measurements extending linear mixed models with homogeneous correlation...

Area of Science:

• Statistics
• Biostatistics
• Data Science

Background:

• Longitudinal data analysis presents challenges, especially with unbalanced observations and complex dependence structures.
• Existing methods like linear mixed models have limitations in flexibility and handling missing data.
• D-vine copulas offer a powerful tool for modeling intricate multivariate dependencies.

Purpose of the Study:

• To propose a novel statistical model for unbalanced longitudinal data using D-vine copulas.
• To extend the flexibility of linear mixed models for longitudinal data analysis.
• To provide a robust method for handling missing data and enabling accurate predictions.

Main Methods:

• Development of a D-vine copula-based model for longitudinal data.
• Implementation of a sequential estimation approach as an alternative to joint maximum-likelihood.
• Adaptation of the Bayesian information criterion for model selection.
• Validation through simulation studies and application to real-world data.

Main Results:

• The proposed D-vine copula model offers a highly flexible extension to linear mixed models for homogeneous correlation structures.
• The sequential estimation approach is validated and performs well.
• The model effectively handles missing data without data loss.
• Predictions for future events are analytically tractable.
• The model demonstrated superior performance compared to linear mixed models in a heart surgery data application.

Conclusions:

• The D-vine copula model provides a flexible and powerful framework for analyzing unbalanced longitudinal data.
• This approach enhances predictive capabilities and data handling, outperforming traditional methods.
• The model is suitable for various applications requiring sophisticated analysis of time-dependent data.