Adaptive Designs for Group Sequential Clinical Survival Experiments

Randomized two-group clinical survival experiments now commonly allow at least one interim look, enabling possible early stopping to meet ethical concerns. Various authors have also studied the possibility of interim design modifications to adapt to unexpected accrual or control-group mortality rates. This paper formulates trial design as a decision theoretic problem with a finite number of interim looks and a large class of loss functions, in the setting of a statistic with the asymptotic behavior of Brownian motion with drift, as in Leifer and Slud (2002). A more general action space can specify adaptive designs allowing the option of continued followup without new accrual past an interim look, as was introduced in Koutsoukos, Rubinstein and Slud (1999). An optimal two-look design is displayed in the first formulation, and a sevenlook design in the second, and both types of adaptation are given a unified decision-theoretic motivation.