Lead/Presenter: Thomas Glorioso, COIN - Seattle/Denver
All Authors: Glorioso TJ (VA Eastern Colorado Healthcare System and Department of Biostatistics and Informatics, Colorado School of Public Health, University of Colorado Denver)
Grunwald GK (VA Eastern Colorado HCS and Department of Biostatistics and Informatics, Colorado School of Public Health, University of Colorado Denver)
Baron AE (VA Eastern Colorado HCS and Department of Biostatistics and Informatics, Colorado School of Public Health, University of Colorado Denver)
Caplan L (VA Eastern Colorado HCS and University of Colorado School of Medicine)
Objectives:
Stepped wedge trial designs are often used to avoid simultaneous roll out of an intervention to all available sites, yet time and monetary constraints may limit the number of study sites. In a multi-faceted trial using proactive real-time monitoring of patients in the VA to improve medication adherence, data on all sites are passively collected through the electronic health record. Thus, study power may be augmented in a cost-efficient manner by modifying the traditional stepped wedge design to include sites acting strictly as controls throughout the study.
Methods:
Based on previously collected data, we simulated data from models using a continuous adherence outcome, a traditional stepped wedge design rolling out an intervention to 4 sites over 4 waves (plus 1 baseline period), and varying amounts of site level variation. Power was estimated by applying linear mixed models to the simulated data, accounting for site level variation and time trend. Control sites were then added and the additional power generated was compared to a strategy in which the number of intervention sites was increased.
Results:
Initially, the addition of a strictly control site provided more power compared to the addition of an intervention site. As more sites were added, the effect of additional control sites plateaued and was eventually surpassed by the addition of intervention sites. Increasing site level variation decreased both power and the asymptote for the additional control sites; however, power with additional intervention sites increased towards 100%.
Implications:
The use of sites acting strictly as controls in a stepped wedge designed trial targeting medication adherence may provide more power at lower cost. Simulation provides results needed to balance required power, hypothesized site level variation, and resources required for additional control versus intervention sites to determine which type of sites to include.
Impacts:
Within VA, stepped wedge designed trials are practical for evaluating implementation of innovative programs across multiple sites, but are limited by temporal and monetary constraints. This work suggests that when data are passively collected at sites, power can be augmented at low cost by including strictly control sites, promoting further exploration of implementations in VA.