Frailty models with missing covariates

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Missing covariate data is common in survival analysis.
• Complete-case methods can introduce bias in parameter estimation.
• Existing frailty models have limitations in accommodating diverse random effects distributions.

Purpose of the Study:

• To develop a robust method for parameter estimation in random effects survival models with missing covariates.
• To extend the flexibility of frailty models by incorporating a wider range of random effects distributions.
• To reduce bias associated with missing data in survival data analysis.

Main Methods:

• Developed a generalized random effects model for survival data.
• Incorporated random effects as an offset in the linear predictor, similar to generalized linear mixed models.
• Utilized a Monte Carlo Expectation-Maximization (EM) algorithm combined with the Gibbs sampler for parameter estimation.

Main Results:

• The proposed method provides parameter estimates that are less biased than complete-case analyses.
• The methodology accommodates a broad spectrum of random effects distributions.
• Demonstrated the method's utility in a real-world clinical trial setting.

Conclusions:

• The novel method effectively handles missing covariate data in random effects survival models.
• This approach offers a more general and less biased alternative to traditional frailty models.
• Applicable to clustered survival data with unobserved characteristics, enhancing statistical reliability.