Estimating regression parameters and degree of dependence for multivariate failure time data

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Multivariate failure time data arise in longitudinal studies with multiple events or clustered individuals.
• Existing methods like marginal hazard and frailty models address specific aspects of dependence and regression parameters.
• A need exists for a unified approach to estimate both mean regression parameters and dependence structures.

Purpose of the Study:

• To propose a new, combined approach for estimating mean regression parameters and dependence in multivariate failure time data.
• To develop a more efficient and simpler estimation procedure compared to existing methods like the EM algorithm.
• To validate the proposed method's robustness, consistency, and large-sample properties through simulations.

Main Methods:

• A novel two-step estimation procedure combining marginal hazard and frailty models.
• Implementation of the combined model for simultaneous estimation of regression coefficients and dependence measures.
• Simulation studies to assess the performance and properties of the proposed estimators.

Main Results:

• The proposed two-step method provides a quicker and simpler alternative to the EM algorithm for frailty model estimation.
• Simulation results demonstrate the robustness, consistency, and desirable large-sample properties of the new estimators.
• The method effectively estimates both mean regression parameters and the degree of dependence within units.

Conclusions:

• The combined marginal hazard and frailty model offers an efficient approach for analyzing multivariate failure time data.
• This method enhances the ability to assess treatment effects and intra-unit dependence simultaneously.
• Application to a diabetic retinopathy study highlights its utility in understanding disease progression and bilateral effects.