Estimating differences in restricted mean lifetime using observational data subject to dependent censoring

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Estimating mean lifetime is crucial in epidemiologic studies.
• Non-randomized treatments and complex censoring present challenges in causal effect estimation.
• Existing methods may not adequately address both confounding and dependent censoring.

Purpose of the Study:

• To propose novel methods for estimating group-specific differences in restricted mean lifetime.
• To address confounding from baseline covariates and time-dependent predictors of censoring.
• To provide a robust approach for causal inference in observational time-to-event studies.

Main Methods:

• Hybrid approach combining treatment-specific proportional hazards models and inverse probability of censoring weighting.
• Proportional hazards models used to adjust for baseline confounders.
• Inverse probability of censoring weighting applied to handle time-dependent censoring predictors.

Main Results:

• The proposed methods provide consistent estimators for average causal effects.
• Large-sample properties of the estimators are theoretically derived.
• Simulation studies demonstrate the finite-sample applicability and performance of the methods.

Conclusions:

• The developed methods offer a powerful tool for estimating causal effects on restricted mean lifetime in observational studies.
• Applicable to real-world health data, such as liver transplant wait list mortality.
• Enhances the ability to draw valid conclusions from non-randomized comparative studies.