A Comprehensive Simulation Study to Evaluate the Effect Size and Study Length Relationship in Single-Group Interrupted Time Series Analysis.

Abstract

Single-group interrupted time-series analysis (ITSA) is a popular non-experimental study design in healthcare research. However, little guidance is available to inform the power requirements of ITSA studies under most common usages. We performed simulations to estimate the number of time periods (ranging from 10 to 100) required for percentage increases in level and trend (from baseline), to achieve statistical significance (p < 0.05, p < 0.01) with >80% and >90% power, when autocorrelation ranges from -0.90 to 0.90, and the intervention is introduced at 33%, 50% and 67% of the time series. Larger effect sizes were required for shorter studies, as well as with increasing autocorrelation, and when the intervention was introduced earlier or later than halfway in the time series. The required effect sizes were generally lower for estimating a change in the level of the time series as compared with the change in the trend, but the opposite was true when the number of time periods was larger. Simulations of studies with 10 time periods consistently produced unreliable estimates. The tables created from these analyses as well as a new community-contributed Stata package called POWER_ITSA will guide healthcare researchers in determining the most efficient way to achieve anticipated treatment effects in single-group ITSA studies.