Methods to address selection bias in post-trial studies of legacy effects were evaluated

Post-trial follow-up studies have become increasingly important to investigate the long-term effectiveness of interventions after randomized controlled trials (RCTs). Legacy effects refer to intervention effects that are only observed after the trial has ended and are not the direct effects observed during the trial period. However, limited attention has been given to the potential selection bias in post-trial studies.

Using directed acyclic graphs, we illustrated potential sources of selection bias in post-trial studies of cardiovascular disease preventative interventions. We constructed scenarios where selection bias was present and undertook simulations to assess the ability of different modeling approaches to correct for this bias: no adjustment, adjustment for trial baseline covariates, adjustment for post-trial covariates and inverse probability weighting (IPW) methods. Using empirical data from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial and post-trial study, we demonstrated application of different modeling approaches.

In the presence of selection bias, modeling without adjustment always resulted in biased estimates. Modeling with adjustment and IPW methods were able to correct the selection bias. The ACCORD study also demonstrated that while the direct effects were potentially beneficial, all models attempting to address selection bias revealed larger potential legacy effects when compared to unadjusted estimates.

Post-trial follow-up studies have the potential to provide valuable information for clinical practice by detecting legacy effects. However, it is important to consider and address selection bias that may arise from the post-trial study. This study highlights the importance of using an appropriate analysis method and identifying the potential bias sources to ensure that the findings are reliable and generalizable to the target population.