|
Abstract:
|
The draft E9(R1) estimands addendum framework divides “censoring” into three concepts, confounding intercurrent events, data excluded from the effect of interest, and non-confounding missing data. Only the latter can be assumed at random. The 2007 FDA Cancer Endpoint Guidance censored possible confounding events and assumed further risks for patients with such events are the same as patients without, mapping to a hypothetical strategy. The 2012 and later EMA PFS (Appendix 1) guidelines focus on the comparison between arms and ignore intercurrent events, mapping to a treatment policy strategy. The 2018 FDA guidance allows both. The estimands framework includes additional strategies for TTE endpoints not previously accepted by regulatory authorities. This talk discusses how the framework maps existing and possible new strategies. Assumptions about intercurrent event (e.g. withdrawal) mechanisms, and sensitivity analyses evaluating those assumptions, become critical under the framework. Clinical and statistical expertise must be combined, together with clear understanding of research goals, to define estimands appropriate for the context. Alternatives and examples are discussed.
|
