Immortal bias from different lengths of grace period in a clone-censor-weight design: a simulation study.

Background: The clone-censor-weight design is intended to address immortal time bias in studies using a target trial emulation framework. The determination of the length of the grace period for this design may affect the reduction of immortal time bias. However, there is no consensus on the optimal length of grace period.

Objectives: To examine the magnitude of immortal time bias resulting from different lengths of grace periods in a clone-censor-weight design.

Methods: We conducted a simulation study across 100 datasets with potential immortal time bias using an extensive Monte Carlo method, each dataset including 1000 individuals. We employed the clone-censor-weight design and used a data-driven approach to determine the lengths of grace periods that resulted in 10%, 30%, 50%, 70%, and 90% of the study sample being selected as the treatment group, respectively. We set five true effect sizes as hazard ratios: 0.5, 0.7, 1, 1.3, and 1.5. The difference in restricted mean survival time (RMST) between the treatment and control groups was calculated, and a 100-bootstraps method was used to determine the 95% confidence intervals. The results of the difference of RMST from different lengths of grace periods and the true effect sizes were compared to determine the magnitude of immortal time bias.

Results: For true effects of 1.3 and 1.5, longer grace periods were associated with unbiased results. We identified that a grace period length resulting in 70% or more patients in the treatment group was a cut-off point for unbiased results (greater than 1.0). With true effects of 0.5 and 0.7, we found nearly 100% of estimated results were less than 1.0 regardless of grace period length. However, longer grace periods were not associated with more precise estimations of true values.

Conclusions: The results indicated that the reduction of immortal time bias varied with different lengths of grace periods. This finding suggests that considering the appropriate length of grace period, especially when the true effect is higher than 1.0, is important when conducting a clone-censor-weight design.