|
Abstract:
|
Since Addendum R1 of the ICH E9 Guidance, for many trials the regulator has requested the objective of assessing the effect of the randomized treatment policy, irrespective of adverse events, protocol deviations or treatment actually administered. For such trials, the occurrence of events that change the interpretation of the estimate of treatment effect is eliminated (R1 calls such events intercurrent events (ICEs)); missing data is similarly reduced. Data are collected until end of scheduled follow-up time; and outcomes analyzed as randomized. However, post-study-withdrawal measurements will remain unavailable. For such uncollectible data, R1 suggests that a kind of reference-based imputation could be employed, with the reference group being those subjects who discontinue study treatment but remain on the study. If visit of discontinuation is taken into account, sparsity of outcomes for this reference group may make such an approach difficult in practice. We compare the model that uses time of discontinuation as a continuous covariate with the standard by-visit imputation model. Results assessing bias and variance will be presented, based on a simulation study.
|
