Augmented and doubly robust G-estimation of causal effects under a Structural nested failure time model.

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Time-dependent confounding is a challenge in survival analysis.
• Structural nested failure time models (SNFTMs) and G-estimation adjust for this.
• Inverse probability of censoring weighting (IPCW) handles informative censoring but can lose efficiency with high dropout.

Purpose of the Study:

• To enhance the efficiency of G-estimators in SNFTMs.
• To develop an augmented estimator robust to censoring model misspecification.
• To improve precision of survival outcome estimates in the presence of significant dropout.

Main Methods:

• Derivation of an augmented estimator for SNFTMs.
• Utilizing both censored and uncensored observations.
• G-estimation with inverse probability of censoring weighting (IPCW).

Main Results:

• The proposed augmented estimator increases efficiency compared to standard IPCW G-estimators.
• The method demonstrates robustness against misspecification of the censoring process model.
• Simulation studies confirm the empirical properties of the new estimators.

Conclusions:

• The augmented estimator provides a more efficient and robust approach for SNFTMs with informative censoring and dropout.
• This method improves the precision of effect estimates in survival analysis.
• The approach is applicable to real-world epidemiological surveillance data.