Doubly-robust methods for differences in restricted mean lifetimes using pseudo-observations.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Restricted Mean Lifetime (RML) is a key metric for comparing survival between treatment groups in clinical studies.
• Estimating RML differences in observational studies requires adjusting for confounding factors due to non-randomized treatments.
• Existing methods can be complex and may require strong modeling assumptions.

Purpose of the Study:

• To propose a novel, doubly-robust pseudo-value approach for estimating the difference in Restricted Mean Lifetime (RML) between two treatment groups.
• To account for confounding factors in observational studies when comparing survival times.
• To provide a robust and interpretable method for causal inference in survival analysis.

Main Methods:

• Utilizes a doubly-robust pseudo-value estimation strategy.
• Combines group-specific regression models for time-to-event data and covariate information.
• Incorporates inverse probability of treatment assignment weights and pseudo-observations to handle censoring.

Main Results:

• The proposed estimator is double-robust, ensuring consistency if at least one working model is correct.
• Explores the use of machine learning algorithms to mitigate bias in complex survival-covariate associations.
• Simulation studies demonstrate the finite-sample performance of the pseudo-value causal effect estimators.

Conclusions:

• The doubly-robust pseudo-value method offers an effective and robust way to estimate RML differences in the presence of confounding.
• The approach is particularly valuable for observational studies where treatment assignment is not randomized.
• The methodology is accessible through the R package drRML, facilitating its application in breast cancer research and beyond.