|
Abstract:
|
Interventions that have multiple components, such as both a provider-level intervention and a patient-level one, are particularly common in health behavior research. For example, a cluster randomized trial was recently completed in Tanzania (Cundill et al, 2015) where health care facilities were randomly assigned to three arms: control, health worker intervention (interactive workshops), and health worker intervention plus patient materials (leaflets, clinic posters). We refer to these arms as control (C), treatment (T), and treatment plus (T+). Commonly, the initial primary hypotheses test is treatment plus against control with a secondary comparison (either superiority or non-inferiority) of T+ compared to T. In these circumstances, we argue that a traditional three arm study is inelegant, as it does not leverage the structure of the interventions nor is it designed to answer the most common hypotheses. We consider an alternative design that leverages sequential analysis methods, and compare the efficiency of the alternative design to the traditional three arm approach. We also discuss strengths and weaknesses of both designs from scientific, ethical, and financial perspectives.
|
