Interrupted time series analysis using autoregressive integrated moving average (ARIMA) models: a guide for evaluating large-scale health interventions

Abstract

Background Interrupted time series analysis is increasingly used to evaluate the impact of large-scale health interventions. While segmented regression is a common approach, it is not always adequate, especially in the presence of seasonality and autocorrelation. An Autoregressive Integrated Moving Average (ARIMA) model is an alternative method that can accommodate these issues. Methods We describe the underlying theory behind ARIMA models and how they can be used to evaluate population-level interventions, such as the introduction of health policies. We discuss how to select the shape of the impact, the model selection process, transfer functions, checking model fit, and interpretation of findings. We also provide R and SAS code to replicate our results. Results We illustrate ARIMA modelling using the example of a policy intervention to reduce inappropriate prescribing. In January 2014, the Australian government eliminated prescription refills for the 25 mg tablet strength of quetiapine, an antipsychotic, to deter its prescribing for non-approved indications. We examine the impact of this policy intervention on dispensing of quetiapine using dispensing claims data. Conclusions ARIMA modelling is a useful tool to evaluate the impact of large-scale interventions when other approaches are not suitable, as it can account for underlying trends, autocorrelation and seasonality and allows for flexible modelling of different types of impacts.

Changes in systemic cancer therapy in Australia during the COVID-19 pandemic: a population-based study

“…For more information see the statistical supplement to this paper and our previous paper describing the use of ARIMA methodology for evaluating public health interventions . [22] We included variables representing temporary changes (only occurring during the month) for March 2020, April 2020, and July 2020; and a level shift (permanent change) from April 2020 through December 2020 in our ARIMA models. We modelled temporary changes in March and April as Australian COVID restrictions did not come into full effect until late March and the impacts may have carried over into April; and we modelled a temporary change in July 2020 to capture the impacts of the second outbreak in Victoria and subsequent lockdowns there.…”

National Trends and Policy Impacts on Provision of Home Medicines Reviews and Residential Medication Management Reviews in Older Australians, 2009–2019

“…Trends were stratified by age group (65–74, 75–84, and ≥85 years) and sex. Interrupted time series analysis was conducted using seasonal autoregressive integrated moving average (ARIMA) modelling [ 13 ]. Program changes for HMRs (October 2011, March 2013, March 2014) and RMMRs (October 2011, March 2014) ( Table 1 ) were modelled as change points.…”

Changes in Purchases for Intensive Care Medicines During the COVID-19 Pandemic

“…This method is preferred in drug use research because purchasing trends often are autocorrelated and violate linear regression assumptions. 31 , 32 , 33 We performed a seasonal difference by lagging each series by 12 months ( d = 12). To maximize stability, prepandemic trends were estimated using all 6 years of available data (August 2014-February 2020).…”

“…We tested the stationarity assumption using augmented Dickey Fuller tests (α = 0.1). 33 We did not report estimated coefficients from the ARIMA models (eg, AR or MA coefficients), because these are not interpretable directly in relationship to changes in purchasing rates.…”