A calibrated Bayesian method for the stratified proportional hazards model with missing covariates

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Missing covariate data is a common challenge in survival outcome analyses.
• Excluding missing data can introduce bias and lead to erroneous conclusions.
• Existing methods like inverse probability weighting have complexities in variance estimation.

Purpose of the Study:

• To propose a new approximate Bayesian method for handling missing covariates in survival data.
• To address limitations of frequentist approaches, particularly in variance estimation.
• To develop a method applicable to both single and multiple missing data patterns.

Main Methods:

• An approximate Bayesian approach is utilized, avoiding Taylor expansion.
• A stratified proportional hazards model is employed to accommodate non-proportional hazards.
• The method is evaluated for single and multiple missing covariate patterns.

Main Results:

• The proposed estimators demonstrate consistency and asymptotic normality, aligning with frequentist properties.
• Simulation studies confirm asymptotic unbiasedness and accurate credible regions.
• The algorithm is computationally efficient and straightforward to implement.

Conclusions:

• The novel approximate Bayesian method effectively handles missing covariates in survival data.
• This approach provides a robust and efficient alternative to traditional methods.
• The method shows promise for real-world applications, as demonstrated with stem cell transplantation data.