A1420
Title: Futility bounds for drop-the-losers designs
Authors: Dominique-Laurent Couturier - University of Cambridge (United Kingdom) [presenting]
Boyun Jiang - University of Cambridge (United Kingdom)
Thomas Jaki - University of Cambridge and University of Regensburg (Germany)
Abstract: Multi-arm trials comparing experimental treatments against a common control are increasingly used to identify effective therapies. The drop-the-losers (DTL) design sequentially eliminates inferior arms at each stage, but lacks a mechanism to stop early when no arm shows promise. Prior research showed DTL performs poorly relative to multi-arm multi-stage (MAMS) designs when all treatments are ineffective. DTL has been extended by introducing a futility boundary permitting early termination when interim statistics fall below a threshold. Closed-form expressions for the familywise error rate and power are derived, characterised by multivariate normal probabilities over regions corresponding to futility and efficacy boundaries, and sample size distributions across designs are compared via simulation. For the same power and FWER, DTL with futility requires a median sample size slightly higher than MAMS and classic DTL under the alternative, but substantially lower under the global null and when arms perform worse than control. It also produces a narrower spread of sample sizes compared to MAMS, offering greater predictability. Whilst sacrificing fixed sample sizes, the futility boundary improves resource allocation by preventing continuation of futile trials. Future work will extend this to non-constant boundaries such as O'Brien-Fleming-type rules.
