How to Make Scientific Inferences and Conduct Power Analyses for Randomized Implementation Rollout Trials

C. Hendricks Brown, PhD
Northwestern University

Ian Cero, PhD
University of Rochester Medical Center

ABSTRACT:
This two-part presentation continues the virtual presentations on Randomized Implementation Rollout Designs and Trials, which include Stepped Wedge Implementation Designs.  These designs are commonly used to examine how well an evidence-based intervention or package is being implemented in community or healthcare settings.  The multitude of implementation research questions and specific hypotheses suggest the need for diverse randomized rollout implementation trial designs, assignment principles and procedures, and statistical modeling.  In the first part we discuss key research questions and identify mixed effect models for randomized implementation rollout trials involving 1) a single implementation strategy that tests how this strategy varies over time and/or resources that are allocated, 2) comparison of two distinct implementation strategies, and 3) three distinct strategies or components tested in a single trial. 

In the second part of the presentation we present the use of Rollout, a general statistical package written in R that can be used to conduct detailed statistical power and sample size analyses for diverse rollout designs.  Users specify both the underlying generative data model as well as the analytic model and output includes power and bias in the parameters.  We discuss how the package can account for misspecified modeling and robustness.  Only limited knowledge of R is necessary to use this package, and we provide examples for planning new implementation trials and examining the effects on power when modifications of a design during the conduct of a trial are necessary.