An approach to model clustered survival data with dependent censoring

Area of Science:

• Statistics
• Biostatistics
• Survival Analysis

Background:

• Survival data analysis often assumes independence between failure and censoring times.
• Dependent censoring can bias results in clustered survival data.
• Existing methods may involve complex computations.

Purpose of the Study:

• To develop a likelihood-based method for analyzing clustered survival data with dependent failure and censoring times.
• To account for the correlation between failure and censoring through a latent frailty model.
• To simplify inference procedures and reduce computational burden.

Main Methods:

• Utilized Weibull and piecewise exponential distributions for modeling.
• Introduced a latent frailty model to handle dependent censoring.
• Developed a Monte Carlo Expectation-Maximization (EM) algorithm for inference.
• Ensured continuous survival functions through baseline distribution assumptions.

Main Results:

• The proposed method effectively accommodates the dependence between failure and censoring times.
• The likelihood specification simplifies inference compared to previous methods.
• Simulation studies demonstrated the performance of the developed models.
• A real-world application using the Dialysis Outcomes and Practice Patterns Study was explored.

Conclusions:

• The new statistical methodology provides a robust approach for survival data with dependent censoring.
• The method offers computational advantages and ensures continuous survival functions.
• This work contributes to more accurate analysis of clustered survival data in various fields.