Dependent censoring in piecewise exponential survival models

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Data Analysis

Background:

• Dependent censoring is common in medical survival data due to prognosis-based treatment decisions.
• Identifiability issues necessitate sensitivity analyses to evaluate the impact of independent censoring assumptions.
• Existing methods for sensitivity analysis in survival data have limitations in model flexibility.

Purpose of the Study:

• To introduce a novel sensitivity analysis method for piecewise exponential survival models.
• To assess the sensitivity of survival model results to minor dependence between failure and censoring times.
• To provide a computationally simple method that allows for more flexible marginal distributions.

Main Methods:

• Developed a sensitivity analysis technique for piecewise exponential survival models.
• Employed the same dependence assumptions as prior sensitivity analyses for parametric and Cox models.
• Utilized a simulation study to evaluate the method's accuracy and applicability.

Main Results:

• The proposed sensitivity analysis method demonstrates good performance across various scenarios.
• Accuracy is compromised when dealing with survival datasets exhibiting a high proportion of censored observations.
• The method offers greater flexibility in modeling marginal distributions compared to previous approaches.

Conclusions:

• The new sensitivity analysis is a valuable tool for assessing dependent censoring in survival data.
• Researchers should exercise caution when applying this method to datasets with substantial censoring.
• The approach enhances the robustness of survival analyses by accommodating flexible marginal models.