Identifying Less Burdensome and More Cost-Efficient Incomplete Stepped Wedge Designs for Continuous Outcomes Collected via Repeated Cross-Sections.

Stepped wedge trials can be costly and burdensome. Recent work has investigated the iterative removal of cluster-period cells from stepped wedge designs, producing a series of candidate incomplete designs that are less burdensome. We propose a novel way to explore the space of incomplete stepped wedge designs, by considering their cost efficiency, seeking to identify designs that retain high power while limiting the total trial cost. We define the cost efficiency of a design as the ratio of the precision of the treatment effect estimator to the total trial cost. Total trial cost incorporates the costs per cluster, costs per participant in intervention and control conditions, and the costs of restarting data collection in a cluster under intervention and control conditions following a pause. We consider linear mixed models for continuous outcomes with a repeated cross-sectional sampling scheme and use an iterative procedure to remove individual cells with the lowest contribution to the cost efficiency metric, producing a series of progressively reduced designs. We define the optimal design within this design space as that which maximizes the cost efficiency relative to the complete design, subject to a minimum acceptable power constraint. We illustrate our methods with an example motivated by a real-world trial. Our methods enable trialists to identify incomplete stepped wedge designs that are less burdensome and more cost-efficient than complete designs. We find that "staircase"-type designs, where clusters only contribute measurements immediately before and after the treatment switch, are often particularly cost-efficient variants of the stepped wedge design.