|
Abstract:
|
This article proposes a method to overcome limitations in current methods that address multiple comparisons of adaptive interventions (AIs) embedded in sequential multiple assignment randomized trial (SMART), a clinical trial design for comparing multiple AIs. In mental health research, specifically, SMART is particularly useful for evaluating multi-stage depression interventions. However, since a SMART typically consists of numerous AIs, inferences based on pairwise comparisons may suffer substantial loss in power after multiplicity adjustment. Also, most traditional methods in comparing non-adaptive treatments require that the correlation structure is known a priori. Since it is not the case for analyzing SMART data, they cannot be directly applied in SMART settings. This article addresses these problems by proposing a gate-keeping test and derives the sample size formula to formally justify SMART sample sizes. Simulations show that the asymptotic approximation is accurate with a moderate sample size. The proposed method outperforms the multiple testing procedures based on Bonferroni's correction in terms of power for testing and probability of selecting better interventions.
|
