Rather than a shut-down, what if we knew which places to close during a pandemic to curtail the spread of a virus and lessen the economic impact to a locality? A model showing how this can be done, created by researchers at the University of Virginia Biocomplexity Institute and Stanford and Northwestern universities, has won a Best Paper Award in the Applied Research track at the recently concluded ACM KDD conference. ACM is the world's largest educational and scientific computing society.
The researchers won the award for their paper on “Supporting COVID-19 Policy Response with Large-scale Mobility-based Modeling.” Stanford’s Serina Chang is the paper’s first author, and she, along with her adviser, Jure Leskovec, created the dynamic model and worked closely with the UVA team.
The UVA Biocomplexity Institute authors who contributed to the award-winning paper are: Mandy Wilson, Zakaria Mehrab, Bryan Lewis and Madhav Marathe. Stanford University researchers who contributed to the paper include: Serina Chang, Jure Leskovec, David Grusky, Emma Pierson and Pang Wei Koh; and from Northwestern University, Jaline Gerardin and Beth Redbird. Komal Dudakiya, a UVA contractor with Persistent Systems, also contributed to the paper, which is published on medRxiv.
Wilson, the Institute’s principal author for the paper and who helped create the dashboard, was thrilled to win such a prestigious award. “With an acceptance rate of 17.7 percent, we were excited that SIGKDD [ACM’s Special Interest Group on Knowledge Discovery and Data Mining] accepted our paper submission, let alone recognized it as a best paper,” she said. Winning the award validates the “real-world applicability of our approach.” Hopefully, it will encourage other scientists to consider how mobility data can contribute to their work, as well as to give them ideas on how their theoretical research can be made accessible for solving real-world problems, she said.
The impetus for UVA’s contribution to the paper came from the Institute’s Bryan Lewis and Madhav Marathe after reading a November 2020 Nature journal article on Chang and Leskovec’s research. As co-leads for the Institute’s COVID-19 Response Team, Lewis, a research associate professor, and Marathe, distinguished professor in biocomplexity and the Institute’s Network Systems Science and Advanced Computing division director, wondered if they could build on the Stanford team’s work. They brought the paper to the attention of the Virginia Department of Health, who they’ve been working with since January 2020 to provide projections for COVID-19.
“We and the VDH were intrigued by the paper’s findings,” Lewis said. It showed how mobility data was used to refine an epidemiological model to improve projections of COVID-19 across 10 metropolitan areas. It found that not only did the mobility-guided model outperform the base model, it showed that minorities and lower-income people were more likely to be impacted by the pandemic, and that selective closures of some places, like restaurants or essential retail, rather than blanket closures might be effective in reigning in the virus without the economic impact of a blanket closure.
Lewis and Marathe proposed that a dashboard could be made that would use a modified version of Stanford's model to do projections of COVID-19 prevalence under different levels of mobility to categories of places of interest. “Virginia is not as urban as the metropolitan areas Stanford examined,” Wilson noted. “We limited our model to restaurants, retail, essential retail, fitness centers and places of worship.”
The dashboard the team built was reviewed by the VDH and revised based on guidance from Lewis so that it would be easier to assess the impact of increasing or decreasing mobility to points of interests of different types, Wilson said. “Once we were satisfied with the revisions to the model and the dashboard, we decided to submit a paper about the experience to KDD.”
“While Bryan and I came up with the idea of applying the model within the context of VDH and were able to convince VDH of its value, Serina and Jure’s model was critical, and we could not have done this otherwise, at least not in the short period,” Marathe said.
Of the collaboration, Chang said, “It was a pleasure to work with UVA and others on this research. Team science is deeply important when you and your teammates have complementary skills and backgrounds. This was certainly the case for our team: on the Stanford/Northwestern side, we had recently published a paper in Nature on our novel computational model that leveraged large-scale mobility networks to capture the fine-grained spread of COVID-19; meanwhile, UVA had been advising the Virginia Department of Health and other public health agencies for over a year, and uniquely understood the needs of policymakers and how to communicate with them. So, we naturally came together to extend our original model into a tool that policymakers could directly use, to support their decision-making around COVID-19 response. In doing so, we developed our tool's dashboard with feedback from VDH, greatly extended the technical capabilities of our model, and worked together to build a computational infrastructure that allowed us to efficiently communicate results from thousands of model experiments.”
“Being able to identify which places of interest offer greater opportunity to spread the disease, and being able to quantify the impact that partial closures and re-openings could have on the active number of cases, will allow policymakers to make informed decisions about which places to impose restrictions upon, while allowing relatively low-impact places of interest to remain open, reducing the economic impact,” Wilson said.
Of the award, Marathe said, “Using data to advance scientific knowledge and address real-world problems to save lives, lessen economic loss and help policymakers make informed public health decisions is incredibly rewarding, and to be recognized for this work by your peers with such a prestigious award is both humbling and validating. We are honored to have been part of this important research, and our team and other researchers will continue to build on these findings to improve outcomes for all. The work serves as an excellent example of transdisciplinary team science.”