Simulation-based power and sample size calculation for designing interrupted time series analyses of count outcomes in evaluation of health policy interventions

Summary

This methodological paper presents simulation-based tools for calculating statistical power and sample size in interrupted time series studies evaluating health policy or environmental interventions using count outcomes. The authors developed ready-to-use computer programmes for two model types (Poisson and negative binomial) and demonstrated that power to detect policy effects varies substantially depending on whether the intervention affects baseline level, trend, or both, and on the degree of autocorrelation in the data. The work provides practical guidance for investigators designing ITS studies with count data.

UK applicability

This statistical methodology is applicable to UK health services research and policy evaluation, particularly for studies examining the effects of NHS interventions or public health policies on countable health outcomes such as clinical events or healthcare utilisation. The tools would be valuable for designing rigorous evaluations of UK health system changes.

Key measures

Statistical power, sample size requirements, autocorrelation coefficients (−0.9 to 0.9), segmented autoregressive error models, level change and trend change detection

Outcomes reported

The study developed and validated simulation-based approaches to calculate statistical power and required sample sizes for interrupted time series (ITS) analyses using count outcome data. Power calculations were performed across varying autocorrelation structures and effect sizes for both Poisson and negative binomial models.

Topic tags