Alan Girling University of Birmingham, UK A.J.Girling@bham.ac.uk Optimality in stepped wedge cluster trials: When and Where to take measurements Introduction Alan Girling University of Birmingham, UK A.J.Girling@bham.ac.uk Funding support (AG) from the NIHR through: The NIHR Collaborations for Leadership in Applied Health Research and Care for West Midlands (CLAHRC WM). The HiSLAC study (NIHR ref 12/128/07) SCT/ICTMC Liverpool, May 2017
Standard Stepped Wedge Layout Clusters Time 1 0: Control (untreated) Cell 1: Treated Cell Aim: maximise the precision of the treatment effect estimate by changing the design in some way
Essential Constraints Allow for Time Effects Otherwise get a “Before & After” Design Treatment implementation cannot be reversed If it can a Cross-Over Design is best
1. Optimising the layout in a ‘complete’ trial Assume uniform sampling over time in each cluster Choose when to implement the treatment in each cluster. For example, Is better than ? And what is the best layout overall? Lawrie J, Carlin JB, Forbes AB. (2015) Optimal stepped wedge designs. Statistics & Probability Letters , 99: 210-214. Girling, A. J., and Hemming, K. (2016) Statistical efficiency and optimal design for stepped cluster studies under linear mixed effects models. Statist. Med., 35: 2149–2166.
For ‘large studies’ the best design looks like: Parallel Clusters 100R% Stepped-Wedge Clusters A ‘Hybrid’ combination of Parallel and Stepped Wedge clusters
2. Optimising the sampling rate in each cluster over time In a standard SW design, some of the cells contribute almost nothing to the treatment effect estimate Better estimates can be got by re-allocating observations from less to more influential cells Optimal allocation depends on constraints E.g. Fixed total number of observations; Fixed number in each cluster…. Solution can be a sparse design
E.g. A Sparse Design with 6 groups of clusters 5 8 9 This is Optimal for fixed total number of observations unless this number, or the ICC, is ‘too large’. (unpubl)
The Way Ahead Exact Results Algorithmic Approaches Useful designs (Optimal, or near-Optimal)
Speakers Jennifer Thompson (LSHTM) Jessica Kasza (Monash): The optimal design of the stepped-wedge trial and a comparison to other trial designs Jessica Kasza (Monash): Information content of cluster-periods in stepped-wedge trials Richard Hooper (Queen Mary) Trimming the fat from stepped wedge trials: an algorithmic search for the trial design requiring the fewest participants