Abstract:
|
Two sample size (SS) recalculation methods applied to 3-arm (T for experimental treatment, A for active comparator and P for placebo arm) non-inferiority (NI) trials with normally distributed endpoint were investigated with operational characteristics compared via simulations. Method I (modified from a previously published) recalculates SS at the interim stage based on the observed ?AP. Method II (newly proposed) is a group sequential design that recalculates SS only when the conditional power for NI test falls into a promising zone. We show that Method I provide substantial power gain when the ?AP is over-estimated at the design stage. However, it cannot handle uncertainties in ?TP. Also, type I error inflation is observed when the SS allocation ratio does not meet certain criteria. In contrast, Method II controls type I error at all investigated SS allocations. It provides moderate power gain if the preserved effect ?TP/ ?AP is over-estimated. If the interim result is promising the recalculated SS increases power considerably. When the ?TP is under-estimated, the average SS reduces compared to fixed sample design due to a high probability of rejection at the interim stage.
|