Assessing the interactions between COVID-19 and influenza in the United States

Casey M Zipfel; Shweta Bansal

To assess whether ILI incidence during January-February 2020 deviated from expectations, we conducted an intervention analysis using the CausalImpact R package [12] . Briefly, this method uses data from control time series to construct structural models for a counterfactual time series, or what is projected to occur if a disruption ("intervention") had not occurred. In particular, we use past influenza seasons as controls to characterize counterfactual 2019-2020 influenza season dynamics if COVID-19 risk perception had not occurred; the comparison between this counterfactual influenza time series and the true influenza time series for 2019-2020 allow us to identify any deviations from expectations. We used three control time series of previous influenza seasons (the 2015-2016, 2008-2009, and 2002-2003 seasons) individually, and together. The disruption initiated by COVID-19 was modeled as an "intervention" starting February 3, 2020. Thus, we considered October 7, 2019 to February 3, 2020 as the pre-intervention period, and February 10, 2020 to March 2, 2020 as the post-intervention period.