|
Abstract:
|
An addendum to the International Council for Harmonisation E9 guideline proposed that an estimand, or treatment effect of scientific interest, should be defined for clinical trial objectives, and that estimands for the trial should determine all aspects of the trial. Estimand definitions should address intercurrent events (i.e., events after starting treatment that prevent measurement of or affect interpretation of the outcome). A common intercurrent event is discontinuation of the randomized treatment. In this situation, the estimand may be the following: a comparison of the outcome among the treatments in the target trial population, if no one had discontinued their randomized treatment. We propose to estimate this estimand using an inverse probability weighting (IPW) approach: (1) fit a Cox model for the time-to-discontinuation conditional on covariates, (2) calculate IP weights based on that model, and (3) fit a weighted Cox model for the outcome of interest using the IP weights and only including participants in the risk set prior to discontinuation. We present simulations comparing this proposed IPW approach to alternative approaches for estimating the desired estimand.
|
