Abstract:
|
Cluster randomized trials test interventions delivered at a cluster level (e.g., schools, hospitals, clinics). The ideal for these trials is first to recruit and consent participating clusters, then obtain consent from individuals participating from each cluster (if needed), and finally randomize each cluster to treatment. This ideal preserves the internal validity of the trial. However, this ideal requires that clusters agree to be randomized, which in practice can result in clusters choosing not to participate in the study. Such cluster nonparticipation may limit the external validity of study findings. Motivated by a school-based trial of physical activity promotion programs, we discuss strengths and weakness of an alternative approach where clusters are randomized first and then approached for consent to study participation. This approach has the potential to improve cluster consent rates, but risks bias due to differential participation. In this work, we provide a framework for when this bias may occur and its potential magnitude, thus providing guidance for future studies about the statistical tradeoffs between the traditional and alternative randomization timing.
|