Study population and setting
This study assessed the spread of SARS-CoV-2 and social behaviors related to a large annual motorcycle rally event in Sturgis, South Dakota in early to mid-August 2020. The population consisted of both local residents of Sturgis as well as attendees of this rally, who travelled to Sturgis from all over the United States. The study examined a number of outcomes, mainly COVID-19 cases and social distancing behaviors, using a combination of county-level aggregated cases from the CDC and personal movement information based on cell phone location data from SafeGraph, from July 6 to August 31, 2020. The main methods used to estimate the impact of the rally on COVID-19 cases were based on comparing changes in regional COVID-19 cases relative to a synthetic control (a method for constructing a “control” from weighted average of other locations without exposure to the rally) to estimate what would have happened had the rally not taken place. Comparator regions were selected across the US (excluding adjacent regions), based mainly by matching on case rates, urbanicity, and population-related characteristics. Additional analyses were performed to see if impacts were larger for places that had more of its residents attending the rally. Finally, the authors estimated the costs of infections attributed to this rally by multiplying a cost per-case estimate (from another study) with their estimate of the number of additional cases resulting from the rally.
Summary of Main Findings
The study makes a large number of claims regarding the impact of the Sturgis event on behaviors, the spread of SARS-CoV-2, and costs to the healthcare system. First, the study finds substantial increases in COVID-19 cases by an additional 6 to 7 cases per 1,000 population locally within Meade County (the county in which the event took place), 3.6 and 3.9 cases per 1,000 population in the state of South Dakota, and ultimately over 266,796 additional cases nationwide attributable to the rally. The study found that social movement, both from residents and non-residents, increased substantially due to the rally. Costs attributable to these additional infections were estimated to be $12.2 billion.
The key strength of this study is in the aggregation and documentation of cases over time and place, in combination with movement data. At least for the local results in Sturgis, the data demonstrates some impact. The case data show relatively stable trends prior to the event and clear changes around the event, with little reason to believe that the changes in cases could have been caused by anything but the event. The overall conclusions that the Sturgis event caused a large increase in COVID-19 cases and infections are likely to be relatively robust to the specific statistical methodologies used.
While large and substantial increases in the local cases are clear from the observed data, the construction of the synthetic counterfactuals and the associated data analyses used to obtain nationwide estimates were relatively weak. In the synthetic control construction, there appear to be very few actual regions matched, and they are regionally clustered (e.g., in Texas or the North East), exacerbating the risk of confounding by concurrent changes and unobserved characteristics. While the assumptions underlying the synthetic control are plausible for the analyses considering the local area around Sturgis, they become decreasingly plausible as the analyses move to larger geographic regions. In addition, the analyses do not take changes in testing behaviors into account. People at and surrounding the rally may have been more likely to seek testing for SARS-CoV-2 due to the rally and public health messaging. This increase in testing could lead to an overestimate of the impact of the rally on infections. Ranges of uncertainty on estimates of rally impact nationwide are not reported. These are likely to be large, given the relatively small numbers of regions that contributed to the overall analysis. The methods behind estimating the cost to the healthcare system due to SARS-CoV-2 infections are largely undiscussed, and rely entirely on a cost estimate from a separate paper. We note that cost estimates are provided only at the national level, which was the least reliable level of aggregation, and as such should not be considered informative.
This analysis provides evidence of how large gatherings such as the Sturgis rally can greatly increase the local spread of SARS-CoV-2. It achieves this through combining data from routinely collected aggregated case data over time, in combination with evidence gathered from personal location data over time from cell phones.
This review was posted on: 11 September 2020