Power Considerations for Multiple-Group (Controlled) Interrupted Time Series Analysis: A Comprehensive Simulation Study

Currently there is little guidance on the power considerations for the multiple-group (controlled) interrupted time series design (MG-ITSA). In this study, simulations estimated power based on the number of time periods, the number of control units, when the treatment is introduced, and the degree of autocorrelation. The measures of effect were the difference in differences (DID) in level and the DID in trend . Power was evaluated at three different effect sizes. Higher power was generally associated with longer studies, more control units, larger effect sizes, and decreasing autocorrelation. Introducing the treatment at the halfway point in the study typically produced higher power than elsewhere for DID in trend. DID in level required fewer time periods to achieve the desired power than the DID in trend. The results show that to increase power, a researcher can increase the number of control units, increase the number of time periods, utilize the DID in level as the measure of effect, and maximize the effect size. Autocorrelation cannot be readily manipulated, and therefore must be accounted for in the time series regression model. Health researchers must consider the many factors highlighted here that uniquely affect power when determining the most efficient way to conduct an MG-ITSA study.