Linear regression analysis of survival data with missing censoring indicators

Area of Science:

• Biostatistics
• Survival Analysis
• Clinical Trial Data Analysis

Background:

• Linear regression in random censorship settings typically requires observed censoring indicators.
• Missing censoring indicators present a challenge for accurate parameter estimation in survival models.
• Existing methods often fail when censoring information is incomplete.

Purpose of the Study:

• To develop and evaluate synthetic data methods for linear regression with missing censoring indicators.
• To provide robust estimators for regression parameters in the presence of incomplete censoring data.
• To assess the performance of these methods in a real-world clinical trial setting.

Main Methods:

• Development of estimators using regression calibration, imputation, and inverse probability weighting.
• Asymptotic normality proofs for the proposed estimators.
• Simulation studies to evaluate finite-sample performance.
• Application to a brain cancer clinical trial dataset.

Main Results:

• All three proposed estimators (regression calibration, imputation, inverse probability weighting) were proven to be asymptotically normal.
• Simulation studies demonstrated the finite-sample performance of each estimator.
• The methods were successfully applied to analyze time to non-ambulatory progression in brain cancer patients.

Conclusions:

• Synthetic data methods offer a viable solution for linear regression analysis with missing censoring indicators.
• The developed estimators provide reliable parameter estimates even with incomplete censoring information.
• These methods enhance the analysis of survival data in clinical research, as demonstrated in the brain cancer trial.