One-step targeted maximum likelihood estimation for time-to-event outcomes

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Observational survival analysis requires accurate estimation of survival curves and statistical inference.
• Existing methods may produce non-monotone survival curve estimators, which is statistically undesirable and inefficient.
• Covariate information can influence treatment and censoring, necessitating methods that account for these effects.

Purpose of the Study:

• To develop a novel estimator for the counterfactual average survival curve in the presence of right-censored data.
• To address the limitation of non-monotone estimators in existing observational survival analysis methods.
• To improve the efficiency and reliability of survival curve estimation by incorporating prior knowledge of the estimand's shape.

Main Methods:

• Development of a one-step Targeted Maximum Likelihood Estimator (TMLE) for counterfactual average survival curves.
• The proposed TMLE is designed to be executed recursively through small, local updates.
• Utilizes observational data with right-censored failure times and considers the influence of baseline covariates on treatment and censoring.

Main Results:

• The new one-step TMLE yields monotone survival curve estimators, overcoming a key limitation of previous methods.
• Demonstrated finite sample performance through simulation studies.
• Successfully applied the TMLE to a real-world dataset concerning monoclonal gammopathy.

Conclusions:

• The one-step TMLE provides a statistically sound and efficient method for estimating counterfactual average survival curves.
• The monotonicity constraint enhances the reliability and interpretability of survival estimates in observational studies.
• This approach offers a valuable tool for researchers in biostatistics and epidemiology dealing with complex survival data.