Should I stay or should I go
Massive dataset reveals 4 superspreader sites to avoid this winter
"Our work highlights that it doesn't have to be all or nothing."
In recent history, no other global event has radically transformed our daily movements quite like Covid-19. To limit the coronavirus's insidious spread, people have been advised to stay home, social distance, and at times, lockdown.
To show how these mobility shifts influence disease transmission, scientists have just released a far-reaching, yet fine-grained, dynamic model. The data maps how 98 million Americans in ten of the nation's largest metro areas moved through half a million different establishments — from bodegas to wine bars to shopping malls.
The data paints a sobering picture of what might happen if people abandon social distancing and resume normal life amid fluctuating case counts. But it also illuminates a future that doesn't require total economic shutdown — if we choose a path associated with minimizing infections.
These findings, which offer critical insight to policymakers designing public health guidance, were published Tuesday in the journal Nature.
Instead of advising lockdown for months on end, the data pinpoints four "superspreader" sites that pose the highest risk of infection:
Avoiding these establishments — if they are operating at full capacity — could be lifesaving as people navigate a cold, virus-laden winter, the study suggests.
"Without any reduction in mobility — any staying at home in a single month — 30 percent of the entire population of these cities would be infected," study co-author Jure Leskovec, a computer scientist at Stanford University, explained in a related press call.
These infections are happening very unevenly, Leskovec added: About 10 percent of the locations examined accounted for over 80 percent of all infections. These places, on average, are smaller, more crowded, and are occupied for longer periods of time. The researchers predict tamping down on foot traffic in these high-risk locations could have outsized positive effects on curbing infections.
"There is an important trade-off between wanting to restart the economy but also wanting to minimize the number of Covid cases," Leskovec said.
"Our work highlights that it doesn't have to be all or nothing."
The study design — To determine how mobility patterns influence the coronavirus' spread, the team used a massive set of cell phone data collected by SafeGraph, a company that aggregates anonymous location data from mobile applications.
The mobility data ranged from March 1 to May 2, 2020. It captures the hourly movements of people from local neighborhoods, technically called census block groups, each containing around 600 to 3,000 people.
Leskovec and his colleagues looked at three factors: where people go in the course of a day, how long they linger at each location, and how many other people are visiting the same place at the same time.
They subsequently built a computer model that accounts for people's demographic background, the neighborhood where they live, and how crowded each establishment is on a daily basis. After running their simulation with multiple parameters, the team contrasted their predictions with real-time confirmed coronavirus case counts gathered by The New York Times from the 10 metro areas. These infections occurred between March 8 to May 9, 2020.
"Our model's ability to capture the true case curve, despite only capturing changes in mobility, suggests that mobility played a really large role in determining the infection curve," study co-author Serina Chang, a researcher at Stanford University, said on the call.
Using this detailed model, the researchers say they can predict the likelihood of new infections occurring at any given place or time, down to the hour.
Superspreader sites and targeted occupancy caps — While much attention has focused on singular "superspreaders" — individuals who inadvertently pass the virus along to large groups of people — these researchers turned that idea on its head. Instead of looking at individual human superspreaders, the team explored "superspreader sites," places where the risk of transmission is dangerously high due to crowds or confined spaces.
The team confirmed that most Covid-19 transmissions occur at four superspreader sites: full-service restaurants, gyms, hotels, and cafes. These places are especially risky because groups of people tend to remain in close quarters for extended periods of time.
In Chicago, for example, only 10 percent of locations accounted for 85 percent of the predicted infections.
Based on these findings, the team doesn't say it's necessary to shut these places down for months on end. Instead, policymakers should institute targeted occupancy caps to limit the foot traffic at each place.
The exact threshold would vary city by city and community by community. Still, the team says lowering the occupancy to about 20 up to 50 percent capacity could dramatically lower the risk of catching or spreading the virus.
"This is an especially effective strategy because it targets points of interest during high-risk time periods," Leskovec said. "With 20 percent occupancy, the restaurant still gets 60 percent of their visitors, but we are preventing 80 percent of infections."
In Chicago, capping at 20 percent maximum occupancy cut down predicted new infections by more than 80 percent but only lost 42 percent of overall visits. The team saw similar trends across other cities.
The team also saw that reopening certain places was riskier than others. Across all cities in the study, full-service restaurants, gyms, hotels, cafes, religious organizations, and limited-service restaurants produced the largest predicted increases in infections when reopened.
Restaurants are, by far, the riskiest — about four times riskier than the next categories, which are gyms and coffee shops, followed by hotels, Leskovec said.
"Fully reopening restaurants, after a month, would lead to about six percent of the entire population being infected," he explained.
Who is most at risk — Just like various locations aren't hit equally across the board, minority communities and low-income groups also suffer disproportionate risks of transmission. Past research links these disparities to unequal access to health care and varying rates of pre-existing conditions.
This model suggests mobility is a stronger driving factor behind this disparity than previously hypothesized.
In the model, people living in lower-income and in less white neighborhoods were not able to reduce their mobility as much as higher-income neighborhoods, likely because they were essential workers or did not have the luxury of working from home. In turn, these groups were substantially likelier to have been infected by the end of the simulation.
This predicted disparity was also driven by visiting full-service restaurants and establishments that tended to be smaller and more crowded. Scientists saw the same disparity play out in neighborhoods with fewer white people.
Reopenings also didn't affect communities equally. In Chicago, researchers predicted that full reopening would result in an additional 39 percent of low-income neighborhood populations becoming infected within a month, compared to 32 percent of the overall population.
What comes next? — Based on this comprehensive analysis, the team also laid out five targeted policy interventions that could make a meaningful difference in reigning in Covid-19's spread:
- More stringent caps on establishments' capacity.
- Emergency food distribution centers that can reduce foot traffic in high-risk stores.
- Free and widely available testing in neighborhoods predicted to be high risk (especially given known disparities in access to tests).
- Improved paid leave policy or income supports that allow essential workers to curtail mobility and stay home when sick.
- Improved workplace infection prevention for essential workers, including high-quality PPE, good ventilation, and distancing when possible.
These precautions can limit further catastrophic transmission, and keep cities from locking down.
Abstract: The COVID-19 pandemic dramatically changed human mobility patterns, necessitating epidemiological models which capture the effects of changes in mobility on virus spread . We introduce a metapopulation SEIR model that integrates fine-grained, dynamic mobility networks to simulate the spread of SARS-CoV-2 in 10 of the largest US metropolitan statistical areas. Derived from cell phone data, our mobility networks map the hourly movements of 98 million people from neighborhoods (census block groups, or CBGs) to points of interest (POIs) such as restaurants and religious establishments, connecting 57k CBGs to 553k POIs with 5.4 billion hourly edges. We show that by integrating these networks, a relatively simple SEIR model can accurately fit the real case trajectory, despite substantial changes in population behavior over time. Our model predicts that a small minority of “superspreader” POIs account for a large majority of infections and that restricting maximum occupancy at each POI is more effective than uniformly reducing mobility. Our model also correctly predicts higher infection rates among disadvantaged racial and socioeconomic groups2–8 solely from differences in mobility: we find that disadvantaged groups have not been able to reduce mobility as sharply, and that the POIs they visit are more crowded and therefore higher-risk. By capturing who is infected at which locations, our model supports detailed analyses that can inform more effective and equitable policy responses to COVID-19.