A note on rank-preserving structural failure time models to account for crossover

Area of Science:

• Biostatistics
• Survival Analysis
• Clinical Trials

Background:

• Estimating treatment effects in randomized trials with treatment crossovers is crucial.
• Rank-preserving structural failure time models offer a method without modeling crossover behavior.

Purpose of the Study:

• To investigate unusual behavior observed in the rank-preserving structural failure time model's acceleration parameter estimation.
• To understand the factors influencing the stability of differential treatment effect estimates.

Main Methods:

• Application of rank-preserving structural failure time models in practice.
• Examination through simple examples and limited simulation studies.
• Analysis of estimator behavior with varying survival distributions (exponential, Weibull) and censoring.

Main Results:

• The estimator for the acceleration parameter can exhibit extreme values, especially when intent-to-treat analysis favors the standard arm.
• Censoring paradoxically reduced estimator variability for exponential distributions.
• A Weibull distribution with short tails resolved the unusual behavior.

Conclusions:

• Rank-preserving structural failure time model estimators are sensitive to data characteristics, not relying on joint ranks.
• Instability is linked to long-tailed survival distributions.
• Model choice and data properties significantly impact the reliability of estimated treatment effects.