Covariate adjustment using propensity scores for dependent censoring problems in the accelerated failure time model

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Research Methodology

Background:

• Standard survival analysis methods assume independent censoring, which is often violated in medical studies.
• Dependent censoring in semicompeting risks data complicates the analysis of treatment effects and covariate impacts.
• Existing artificial censoring techniques can lead to unstable estimations due to confounders with high variability.

Purpose of the Study:

• To propose a novel weighted estimation strategy for analyzing semicompeting risks data.
• To address the limitations of existing methods, particularly excessive artificial censoring caused by confounders.
• To provide a computationally simpler and more numerically stable approach for estimating treatment-outcome associations.

Main Methods:

• Developed a weighted estimation strategy within the accelerated failure time model framework.
• Utilized propensity score modeling of treatment, conditioned on confounder variables, to derive weights.
• Employed Monte Carlo simulation studies to evaluate the proposed methodology.

Main Results:

• The proposed weighted estimation strategy effectively handles dependent censoring in semicompeting risks data.
• Propensity score-based weighting avoids excessive artificial censoring, leading to more stable and reliable estimations.
• The method demonstrated improved performance in simulation studies and practical applications.

Conclusions:

• The novel weighted estimation approach offers a robust solution for analyzing treatment effects in the presence of dependent censoring.
• This method enhances the reliability of survival analysis in medical research, particularly in fields like AIDS and cancer.
• The use of propensity scores simplifies computation and improves numerical stability compared to traditional artificial censoring techniques.