How can we improve measures of human interaction?
The reopening of businesses in states across the United States is resulting in frequent human interactions, which in turn could lead to an increase in COVID-19 cases in a community. Mobility in a region has been an important feature of many COVID-19 forecasting models. Often, models use mobility measures that are very broad and only measure movement within and between communities as a proxy for human interaction, without directly measuring human interaction. However, mobility to an outdoor park where everyone is socially distant should be measured differently than mobility to a crowded, indoor business. Not all types of mobility within a community contribute equally to increased risk of transmission. Thus it is important to directly measure human interaction when weighing the costs and benefits of reopening and lifting restrictions on businesses.
What type of businesses are people going to as communities reopen? How much are people practicing social distancing as they’re frequenting businesses and restrictions are lifted? When restrictions are lifted, do people actually change their behaviors, or do they continue practicing social distancing even though they are no longer as restricted in their movements? The answers to these questions vary widely by state and even by communities within the same states. Measuring how individuals are behaving as states reopen can help policymakers weigh the costs and benefits of lifting different restrictions and allow states to reopen in the safest way possible.
The tool: how we monitor risk of transmission from business traffic
As states reopen and restrictions on business capacity are lifted, individuals may visit businesses that are more crowded and they may linger longer in businesses than they did during the restrictions. In order to quantify how individuals are actually behaving as states reopen, we use anonymized GPS data to measure the number of visits and the duration of visitors to businesses in the northeastern United States and California. To measure visits per square foot, we calculate how many visits there are to a business each week and adjust for the size of the business. This gives us a measurement of how crowded a business is at a given time. To measure the amount of time individuals linger in the business, we calculate the median length of visits to each business. Using these two metrics – visits per square foot and median dwell time – we built a business risk index tool to monitor the weekly potential risk of transmission from business traffic in each community. We also examine the risk of transmission over time in four potentially high-risk industries: restaurants, bars, universities, and personal care (nail salons, hair salons, and barbershops). These outputs are updated with a one-week lag, so policymakers can have a near-real-time understanding of the activity in their community.
Can this tool be used to forecast future COVID-19 cases?
We found that the average Business Risk Index in a county was useful for forecasting positive COVID-19 cases in that county with a 1-week lag. This suggests that there is a relationship between the risk of transmission from business traffic in a county and their future COVID-19 cases. Thus, our tool could be useful for policymakers trying to weigh the costs and benefits of lifting different restrictions in their communities. Additionally, hospital decision-makers can monitor the business traffic in their area and understand the potential for changes in COVID-19 cases and hospitalizations in their service area.
The color density of the plots is based on the total COVID-19 case rates per capita for all counties in the study, with the darker blue counties representing the highest COVID-19 cases. Potential high-risk businesses are also displayed on the map as red dots.
Ongoing Forecasting Work
Our index has successfully been used as a feature in a COVID-19 forecasting model for a large health system in Massachusetts that is monitoring potential future hospitalizations in their service area. Thus, our index can strengthen forecasting models to better quantify not just the mobility, but also the level of human interaction in an area, which is an important predictor of COVID-19 transmission and can help to identify a potential second wave. We continue to update the Massachusetts index every week and it continues to be used in forecasting models for hospital decision-makers.