Super-Spreader Businesses and Risk of COVID-19 Transmission

Importance: The United States has the highest number of confirmed COVID-19 cases in the world, with over 150,000 COVID-19-related deaths as of July 31, 20201. The true risk of a COVID-19 resurgence as states prepare to reopen businesses is unknown. Objective: To quantify the potential risk of COVID-19 transmission in business establishments by building a risk index for each business that measures transmission risk over time. Design: This retrospective case series study uses anonymized cell phone GPS data to analyze trends in traffic patterns to businesses that may be potentially high-risk from January 2020 to June 2020. Setting: Massachusetts, Rhode Island, Connecticut, New Hampshire, Vermont, Maine, New York, and California. Participants: 1,272,260 businesses within 8 states from January 2020 - June 2020. Exposure(s): We monitored business traffic before the pandemic, during the pandemic and after early phases of reopening in 8 states. Main Outcome: Our primary outcome is our business risk index. The index was built using two metrics: visitors per square foot and the average duration of visits. Visitors per square foot account for how densely visitors are packed into businesses. The average duration of visits accounts for the length of time visitors are spending in a business. Results: Potentially risky traffic behaviors at businesses decreased by 30% by April. Since the end of April, the risk index has been increasing as states reopen. On average, it has increased between 10 to 20 percentage points since April and is moving towards pre-pandemic levels of traffic. There are some notable differences in trends across states and industries. Conclusion: Our risk index provides a way for policymakers and hospital decision makers to monitor the potential risk of COVID-19 transmission from businesses based on the frequency and density of visits to businesses. Traffic is slowly moving towards pre-pandemic levels. This can serve as an important metric as states monitor and evaluate their reopening strategies.


. Introduction
The United States has the highest number of confirmed COVID-cases in the world to date, with over , COVID--related deaths . A primary reason has been the emergence of various clusters of COVID-transmission from super-spreader events and establishments . Identifying potential super-spreader businesses has important implications for policy-makers as they decide when and how to safely reopen non-essential businesses . Baicker et al aimed to determine which industries or business establishments had a higher risk of transmission. The study raised important questions that individuals may face as businesses reopen, including the comparative risk of visiting di erent business establishments .
There is a pattern to the events and places that have a high risk of transmission that can be deemed as super-spreaders. They are o en indoor events with people in extremely close proximity to each other for a long duration of time. The risk of transmission in a closed establishment is . times higher than in an event in an openair establishment . Even though the public and states are ready to to reopen the economy, experts cautioned on resurgence of the virus and death tolls if we open our economy prematurely . Given the empirical evidence on the potential impact of super-spreaders in the spread of COVID-, it is crucial to evaluate which businesses, events, establishments, and industries should reopen first and which ones may have a higher risk of spreading the virus. In this study, we sought to identify the businesses that have the potential to be super-spreaders before reopening the business establishments in di erent counties to aid the decision-making process for the policymakers. We tested the hypothesis that US counties with higher densities of super-spreader businesses, as defined by our index, were at a higher risk of COVID-transmission and thus may require a careful reopening of businesses to minimize a resurgence of COVID-cases.

. Data
We use data from SafeGraph Monthly Patterns 0 from January , -May , and SafeGraph Core Points of Interest data to measure business characteristics and SafeGraph is a data company that aggregates anonymized location data from numerous applications in order to provide insights about physical places. To enhance privacy, SafeGraph excludes census block group information if fewer than five devices visited an establishment in a month from a given census block tra ic. Data on county-level COVID-cases and tests are from Johns Hopkins University and the New York Times. Socio-economic and demographic characteristics are collected from the American Community Survey from the United States Census Bureau. Businesses are classified by their -digit North American Industry Classification System (NAICS) code, developed by the United States Census Bureau.

. Setting
This study focused on counties in states (Massachusetts, Rhode Island, Connecticut, New Hampshire, Vermont, Maine, New York, and California). There are counties, with a total population of , , . We examine tra ic to , businesses from di erent -digit NAICS codes from January , -May , . We analyze COVID-cases in these counties from January , -May , .

. Index Construction and Super-Spreader Classification
We constructed a COVID-Business Transmission Risk Index using data on business characteristics and tra ic by NAICS code from January , -May , . The index was built using data on visitors per square foot, frequency of visits, and the average duration of visits. Visitors per square foot account for how densely visitors are packed into businesses. Businesses that are more densely packed may have a higher risk of COVID-transmission. The average duration of visits accounts for the length of time visitors are spending in a business. Businesses where visitors linger for longer periods of time could be riskier for COVID-transmission than businesses where visitors are quickly in and out of the business. The COVID-Business Transmission Risk Index is calculated for each -digit NAICS code in our sample by weighting the total visit time across all visitors from January , -May , by the square footage of the business establishment. NAICS codes which fall in the top % of the Index are classified as super-spreader industries. We classify businesses in these industries as super-spreader businesses. This classifies , individual businesses as super-spreaders out of a total of , businesses.

. Study Variables
The outcome measure is the cumulative number of COVID-cases each week per county. The independent variable is the density of super-spreader businesses in a county, which is measured as the number of super-spreader businesses out of the total number of businesses. Covariates included are counties' racial composition (Black and White), population above years, population below the poverty line, and population density per square mile. group.

. Statistical Analysis
Univariate analyses were conducted to produce overall baseline characteristics and most common super-spreader businesses. Data were analyzed using a negative binomial regression at the county-level. The natural log of the total county population was included as an o set term. The model was adjusted for counties' racial composition, percent of population above years, percent of population below the poverty line, and population density. Additionally, an indicator variable for each state was included in an e ort to adjust for di erences in testing practices across states, though this will not account for di erences in testing across states that varies over time. Standard errors were clustered at the state level.
Coe icients were transformed into incidence rate ratios (IRRs) and are reported with 95% confidence intervals (CIs). Statistical significance was determined by a pvalue ≤ 0.05. All tests were two-tailed. Statistical analysis was performed using Stata SE version . (StataCorp). .

. Summary Statistics
Summary statistics are reported in Table . In our sample, there were an average . cumulative cases of COVID-per , by May , . The average density of super-spreader businesses in a county was . per businesses. On average, 18.55% of a county was above the age of , 84.86% was White, 2.95% was Black, and 8.96% was below the federal poverty line. The average population density of a county was . people per square mile. Our study covered states, counties, with a total population of , , , and , businesses.  The color density of the plots are based on a percentile rank of the total COVID-case rates for all counties in the study, ranging from to , . cases per , people. Superspreader businesses are also displayed on the map as red dots with their relative size reflecting the estimated total dwell time for people in each location.

. Main Results
Table reports the most common super-spreader business types by NAICS code. The most common type of super-spreader business in our sample is full-service restaurants. These are restaurants where you're seated, typically have a server, and pay after your meal is completed. There are , full-service restaurants in our sample. The second most common type of super-spreader business is limited-service restaurants with , in our sample. These are restaurants where you may pay at a counter prior to your meal. This would include fast food, delicatessens, sandwich shops, takeout restaurants, and pizza delivery. The third most common type of super-spreader business in our sample is hotels (except casino hotels) and motels with , of these businesses in our sample. Table reports the main results of our negative binomial regression measuring the association between super-spreader density and COVID-cases. In table , p <  Hotels) and Motels , 0.01. The estimates are incidence rate ratios, weighted by total county population, adjusted for population over age , racial distribution, population below the poverty line, and population density. Includes an indicator variable for each state. Standard errors clustered at the state-level. We find a positive association between the density of super-spreader businesses and COVID-cases (adjusted IRR=1.05; 95% CI: 1.02 − 1.07). Our results suggest that an increase in super-spreader businesses by 1 percentage point results in a 5% increase in COVID-cases, all else equal. .

Super-spreader Businesses and COVID-
Our index attempts to quantify the risk of COVID-transmission at businesses based upon the frequency and duration of visits as well as the density of visitors in the businesses. Businesses with more visitors that stay for longer and are more densely packed are likely to have higher risks of transmission.
Knowing the density of super-spreader businesses will be very useful for policymakers. This can allow policy-makers to help plan to reopen these super-spreader businesses in the safest way possible. Our index classifies restaurants as the most common type of super-spreader business. When planning to reopen, policymakers can consider more options to help restaurants reopen while mitigating the risk to the public. This could include more outside seating, limitations on the number of visitors at a time, and monitoring tra ic to potential super-spreader businesses.
This study can also be useful for hospital decision-makers. Knowledge of the density of super-spreader businesses and monitoring tra ic to these businesses may help hospitals prepare for a potential second-wave if tra ic increases to these businesses very quickly.

. Limitations
There are several limitations to this study. First, COVID-cases are based upon a positive COVID-test. Thus, this will not account for individuals who may be COVIDpositive but did not receive a test, either because of scarcity of tests or because they were asymptomatic. To help mitigate some of this bias, we also plan to explore other measures of COVID-incidence at the county-level, such as the percent of total tests that are positive.
Second, while we control for population density at the county-level, there is variation in population density within counties that is likely correlated with both the variation in super-spreader business density and COVID-cases within counties. Thus, we are currently seeking out more granular data on COVID-cases in order to more accurately adjust for potential confounding by population density.

. Future Work
We are continuing to work on this study. First, we plan to incorporate all states into the next iteration of our analysis. Second, we also plan to incorporate business airflow into the Index, such as outside seating options for restaurants.
Third, we are currently building an online decision-support tool that will allow policy-makers and hospital decision-makers to visualize potential super-spreader businesses in their area and monitor weekly tra ic to these businesses. This can help policy-makers and hospital decision-makers plan for a potential second wave.
Finally, as states begin to reopen non-essential in phases, we plan to evaluate the e ects of these reopenings on COVID-transmission. We plan to implement a di erence-in-di erences event study framework to measure the dynamic e ects of reopening on COVID-cases. Knowing the e ects of reopening can help future policy-makers and hospital decision-makers plan for the potential impact of reopening.

. Conclusion
In conclusion, we built a COVID-Business Transmission Risk Index based on the frequency, density, and duration of visitors to businesses in states. We find a positive association between the density of super-spreader businesses and COVID-cases in a county. We control for several socio-demographic and economic characteristics of counties, population density, and attempt to account for di erences in testing across states.
This study can have important implications for policymakers as they consider how to most safely reopen these potential super-spreader businesses. We continue to work on acquiring more granular data to better account for confounding from population density. We also are in the process of building a tool for policymakers and journals/jama/articlepdf/ /jama\_angulo\_ \_vp\_ .pdf. Available from: https: //doi.org/ . /jama. . . doi: . /jama. . .