INSP-04. Confirmatory adaptive designs for survival trials with several time-to-event endpoints

Abstract

Confirmatory adaptive designs comprise a range of statistical methods that allow to modify the sample size of an ongoing trial in a data-dependent way without compromising control of the type I error rate. For short-term endpoints (e.g., 3-month response rate), comprehensive methodology of adaptive designs exists. However, clinical trials in oncology often have a special focus on long-term outcome and therefore often choose a time-to-event endpoint as the primary endpoint. Typical examples are progression-free survival (PFS) or overall survival (OS). But subtle statistical problems arise when adaptively analysing survival trials. Classical designs for survival trials are therefore commonly limited to a single primary endpoint, which combines the occurrence of progression, toxicities, deaths, and other events of potential interest into a single statistical measure (composite endpoint). However, the complexity of oncological diseases can be mapped more accurately using multi-stage models, where the occurrence of progressions, toxicities and deaths is modelled jointly instead of combining them into a single composite endpoint. We present and discuss adaptive design methodology for single-arm phase II survival trials for testing hypotheses on the joint distribution of several time-to-event endpoints in the context of multi-state models. We illustrate the methodology using the example of adaptive hypothesis tests for the joint distribution of progression-free survival (PFS) and overall survival (OS) in the context of an illness-death model. The methodology is motivated from application in pediatric oncology.