Laboratoire de Mathématiques Raphaël Salem

UMR 6085 CNRS - Université de Rouen Normandie

In many clinical studies, defining a clear time origin is essential: eligibility, treatment assignment, and the beginning of follow-up are ideally synchronized. In analyses based on observational data, however, these time points are often misaligned. This misalignment can create periods during which events cannot occur by design, effectively introducing a form of left-truncation or guaranteed survival that biases effect estimates.

A popular way to correct this problem is the “clone-censor-weight’’ (CCW) strategy introduced by Hernán et al. (2016). CCW realigns treatment assignment with the start of follow-up by allowing a short window for initiating treatment, then censoring individuals whose observed data deviate from pre-specified treatment strategies. While useful, this approach does not solve misalignment involving the eligibility time, which can lead to biased estimation when the variables driving eligibility and treatment decisions evolve over time.

In this talk, we explore these issues through an observational study of the effect of lung transplantation on survival among patients with cystic fibrosis. We propose a framework that constructs a nested sequence of trials, separating the tasks of (i) rebalancing the study population and (ii) estimating the treatment effect. This separation permits analyses that are more flexible than standard parametric approaches such as marginal structural models. Using the case study, we ponder a question around the formal representation of the CCW procedure to investigate conditions under which the resulting estimator is consistent under realistic data-generating mechanisms.