Estimated legacy effects from simulated post-trial data were less biased than from combined trial/post-trial data

"Legacy effects" describe the phenomena where treatment effects are apparent during the post-trial period that are not attributable to the direct effects observed within the trial. We investigate different approaches to analysis of trial and extended follow-up data for the evaluation of legacy effects.

We conducted a simulation to compare three approaches, which differed in terms of the time period and selection of trial participants included in the analysis.

The most common approach used for estimating legacy effects in the literature, which combines initial trial and post-trial follow-up data, gave the most biased estimates. Approaches using post-randomized controlled trial data had better performance in most scenarios. When the size of the legacy effect was set to differ according to whether or not drugs were taken after trial, the stratified approach using post-trial data but only from participants taking the drug after trial was less biased but often had lower power to detect a legacy effect.

When estimating legacy effects, approaches to analysis that are restricted to post-trial follow-up data are preferred. If data are available on participant drug use after trial, then both stratified and unstratified approaches to analysis of the post-trial data should be investigated.