Comparison of estimation methods and sample size calculation for parameter-driven interrupted time series models with count outcomes

Interrupted time series (ITS)—a quasi-experimental design—is often used to evaluate the effectiveness of a health policy intervention. When the outcome of interest is rare, for example, for certain hospital-acquired infections, a common practice is to focus on aggregated count outcomes. However, analyzing ITS with count outcomes is challenging due to the needs to consider possible overdispersion and to account for serial correlation. In this paper, we compare the performance of three estimation methods, the generalized estimating equation (GEE) method, the generalized linear model (GLM) method, and the composite likelihood (CL) method, to fit parameter-driven time series models with count outcomes, and develop a simulation-based approach to calculate the sample size and power for designing such studies.