All Times EDT
Keywords: Estimand, Inverse probability of censoring weight, two-stage estimation, statistical inference, overall surval
For ethical reasons, patients in control arm can switch to the investigational drug in some randomized clinical trials. This treatment switch complicates interpretation of analyses results based on intent-to-treatment population. Some analyses methods, e.g., inverse probability of censoring weight (IPCW), have been used for addressing complications of this type. Another possible method for addressing this complication is the two-stage estimation. These methods reside at the intersection of real word studies and randomized trials, and the practical aspects of their implementation sometimes herald challenges, often involving subjective decisions. Importantly, some of these methods are not without untestable assumptions. This talk hopes to address the following questions: 1) What are the implication of using these methods on benefit of randomization? 2) For IPCW with extreme weights, what are the implications of trimming and not trimming extreme weights on interpretation of the estimad? What are the implications of delayed (for oncology trials) in confirming event of interest on successful implementation of these methods? How are unmeasured confounder assumption in these methods addressed? What aspects of study design will foster successful implementation of these methods? Best analysis practice will be discussed.