Weighted Kaplan-Meier estimators for two-stage treatment regimes

Area of Science:

• Biostatistics
• Clinical Trials
• Survival Analysis

Background:

• Two-stage randomization designs involve sequential treatment allocation based on patient response.
• Accurate statistical inference for survival data is crucial in these complex trial designs.
• Existing methods like marginal mean models and weighted risk set estimates have limitations.

Purpose of the Study:

• To propose novel weighted Kaplan-Meier (WKM) estimators for survival data in two-stage randomization trials.
• To evaluate the performance of these WKM estimators against standard methods.
• To address bias issues associated with standard estimators in specific scenarios.

Main Methods:

• Development of two inverse-probability-weighted Kaplan-Meier estimators: one with fixed weights and another with time-dependent weights.
• Comparison with standard Kaplan-Meier (SKM), marginal mean model-based (MM), and weighted risk set (WRS) estimators.
• Conducting simulation studies to assess estimator properties and performance.

Main Results:

• Both proposed weighted Kaplan-Meier estimators demonstrate asymptotic unbiasedness.
• WKM estimators provide coverage rates comparable to marginal mean models and weighted risk set estimators.
• The standard Kaplan-Meier estimator exhibits bias when second-stage randomization rates differ between initial treatment responders and non-responders.

Conclusions:

• Weighted Kaplan-Meier estimators provide a robust and unbiased approach for survival data analysis in two-stage randomization trials.
• These novel methods offer an improvement over the standard Kaplan-Meier estimator, particularly in complex trial designs.
• The proposed methods are validated using a leukemia clinical data set.