Abstract
The objective of this article is to bring together the key current information on practical considerations when conducting statistical analyses adjusting long-term outcomes for treatment switching, combining it with learnings from our own experience, thus providing a useful reference tool for analysts. When patients switch from their randomised treatment to another therapy that affects a subsequently observed outcome such as overall survival, there may be interest in estimating the treatment effect under a hypothetical scenario without the intercurrent event of switching. We describe the theory and provide guidance on how and when to conduct analyses using three commonly used complex approaches: rank preserving structural failure time models (RPSFTM), two-stage estimation (TSE), and inverse probability of censoring weighting (IPCW). Extensions and alternatives to the standard approaches are summarised. Important and sometimes misunderstood concepts such as recensoring and sources of variability are explained. An overview of available software and programming guidance is provided, along with an R code repository for a worked example, reporting recommendations, and a review of the current acceptability of these methods to regulatory and health technology assessment agencies. Since the current guidance on this topic is scattered across multiple sources, it is difficult for an analyst to obtain a good overview of all options and potential pitfalls. This paper is intended to save statisticians time and effort by summarizing important information in a single source. By also including recommendations for best practice, it aims to improve the quality of the analyses and reporting when adjusting time-to-event outcomes for treatment switching.