Randomization Inference for Stepped Wedge Cluster Randomized Trials

• Activity Number: 610
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Xinyao Ji* and Dylan Small
• Companies: The Wharton School and University of Pennsylvania
• Keywords: stepped wedge ; randomization inference ; mixed model

Abstract

A stepped wedge cluster randomized trial is a one-way crossover trial in which an intervention is never removed once it has been implemented. This feature makes the stepped wedge design useful for settings in which it would not be ethical, healthy or practical to withdraw the treatment or to implement the intervention in many clusters at once. A classical approach in this design is to use a mixed effect model (Hussey and Hughes 2007) which assumes no cluster by time interactions. However, cluster by time interactions are common in certain settings such as pandemics that spread from place to place or when the clusters are communities that are affected by different local economic and political developments. In this paper, we develop another strategy using randomization inference to account for such interactions. As significance level is always guaranteed by randomization, our goal is to find a test statistic with good power in the cases of constant and distributional interventions. We conduct simulation studies comparing our approach to other analytical approaches and apply it to a study of a community-based insurance program in rural Burkina Faso.