Practical Implementation of Randomization in a Complex Site-Level Stepped-Wedge Cluster Pragmatic Randomized Trial

• Activity Number: 385 - SPEED: Statistical Methods and Applications in Medical Research, Risk Analysis, and Marketing Part 1
• Type: Contributed
• Date/Time: Wednesday, August 10, 2022 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Wen Wan* and Linda Rosul and Theodore Karrison and Neda Laiteerapong
• Companies: Section of General Internal Medicine, University of Chicago and Access community health network and University of Chicago and Section of General Internal Medicine, University of Chicago
• Keywords: Stepped-wedge cluster randomized design ; Randomization implementation; Federally qualified health centers (FQHCs)

Abstract

The use of the stepped-wedge design is growing rapidly. Often, difficult issues arise with regard to randomization implementation, such as uneven number of sites for stratification, risk for intervention contamination by personnel shared between sites, and potential confounding effects of site-level characteristics across waves. We developed a new randomization method accounting for these. A total of 36 sites (32 FQHC and 4 non-FQHC) were randomized into 5 wedges. To avoid clinicians’ contamination, we grouped sites into 19 bundled FQHC and 3 non-FQHC. Due to the uneven number of sites per FQHC type, we randomized them together. We avoided assigning those bundled sites with relatively larger numbers of clinicians into one wave. We also calculated an imbalance score across multiple site-level characteristics. We performed 20,000 randomization and selected randomization (n=3870) that have the 3 bundled non-FQHC spread over waves 1, 3 and 5 for stratification. Among the 3870, we chose the randomization with the smallest sum of imbalance scores. This randomization method helps consider uneven-size stratification and clinician size per site, and balances site characteristics.