Keywords: multiple hypotheses, multiple tests, IFU modification, AI
Room: Harding
With the advance of artificial intelligence technology, a sponsor (e.g. medical diagnostic device maker) may generate a large number of (correlated) hypotheses and, correspondingly, multiple alternative IFU (Indication For Use) proposals. To minimize the study failure risk and maximize the efficacy claim underpinned by potential technical capacity, we may structure (and gate) the test hypotheses in a cascading fashion, from easy to hard, in an effort to reach the lowest (hardest) level that can be validated, a trial conclusion that also indicates the appropriate IFU choice among pre-specified IFU candidates. We will discuss a (hypothetical) example in medical imaging where the extent of validation for an automated diagnostic assistance technology is unknown (e.g. reporting the category, severity, and location of various disease forms) but could be systematically investigated through such strategy.