Researchers and students from the University of Michigan gathered via Zoom Thursday to discuss the importance models and predictions played in the fight against the spread of COVID-19. The meeting included presentations on how models were used in the early stages of the pandemic; how models can be used to predict the efficacy and benefits of COVID-19 vaccinations; and how models can apply to a campus or university environment.
F. DuBois Bowman, Dean of the School of Public Health, introduced the morning session of the symposium by speaking to the importance of research and the progress the University has made in response to the COVID-19 pandemic. Bowman, a biostatistician, said data research and statistics were critical to the progress made in controlling the pandemic.
“Thinking back to March of 2020, I’ve seen the general public come to understand how important data are in a public health crisis and I’ve seen a newfound appreciation for modeling statistics, and for those who do this critical work,” Bowman said. “The modeling community has been indispensable to our COVID response efforts and I want to recognize the tremendous work that they’ve done to guide us.”
The symposium then moved to presentations focusing on how data visualization and modeling have been used in the past year to combat COVID-19. Bhramar Mukherjee, public health professor and chair of the Biostatistics Department, spoke to her research on COVID-19 transmission in India. Mukherjee said she and her team of Public Health students and faculty developed an app titled covind19.org in March 2020 that recorded the daily transmissions, hospitalizations, and deaths due to COVID-19 and made it available to Indian residents and policymakers.
“I think in terms of public engagement and everyday updating this app has been the truly remarkable contribution, and the University of Michigan School of Public Health students have made a great difference with data in this case,” Mukherjee said.
Mukherjee said her team’s app contributed to lockdown and other preventative measures in India. She said that through this work, she was able to have a role larger than a statistician and contribute to the improvement of her home country during its current unprecedented COVID-19 outbreak.
“I jumped into this process and it’s too hard to back out right now, and I became an activist, …writing columns, talking to media, emphasizing the need for global collaboration somewhere — I was not just a statistician,” Mukherjee said. “These numbers were not just numbers. These were people, these were faces of loved ones, people that we have lost in the last two weeks. So India is having a grim future but … we are striving towards normalcy every day.”
LSA senior Sabrina Corsetti then presented on her group’s use of data machinery to better predict weekly statewide and national COVID-19 cases, hospitalizations and deaths. The machinery used past COVID-19 data and matrices to adjust for variables such as social distancing measures and testing capacities. Corsetti said her group used their knowledge of physics and machinery to break free of the traditional model and look at data science from a different perspective. The model sends this data to the Office of the Governor of Michigan, the national COVID-19 Forecast Hub and the Metro Health University of Michigan hospital system.
“We’ve seen that our model has behaved very well historically with national case errors under 50% and death areas under 20%,” Corsetti said. “In addition, we found that our model principles and its execution are applicable not only to other areas in epidemiology, but also other fields, such as psychology or quantitative genetics.”
The morning session ended with presentations on models pertaining to recent vaccine developments in the fight against COVID-19. Professor Lisa Prosser, who directs the Susan B. Meister Child Health Evaluation and Research Center, and public health professor David Hutton presented their model. The professors showed the cost-benefit analysis of giving COVID-19 vaccinations to children under the age of 16, given children typically have a much lower risk of developing severe complications from COVID-19. The model analyzed probabilities to compare possible costs of the vaccine such as adverse side effects, transportation to get a vaccination and physical costs of administering the vaccine with benefits like less transmission of the virus and fewer hospitalizations and deaths.
“The preliminary results show that we do anticipate that it’s very likely to be cost-effective within general use of thresholds of cost-effectiveness,” Prosser said. “But I think a major question that we don’t know the answer to yet is what we will see in terms of vaccine effectiveness for pediatric populations….I don’t think there are any indications that it’s not likely to be just as effective as it is in healthy adults.”
U-M President Mark Schlissel introduced the afternoon session of the symposium, which focused primarily on how modeling could impact the University and other colleges’ campuses, by thanking the speakers at the event and other researchers across the University for their collaboration in finding the best course of action to limit the spread of COVID-19.
“We can see why the modeling work done here at Michigan and elsewhere is so remarkable,” Schlissel said, “Our researchers use knowledge and expertise from multiple disciplines, and they brought a higher degree of certainty to the many decisions we faced.”
Marisa Eisenberg, associate professor of epidemiology, complex systems and mathematics, spoke next about her model, which splits the Ann Arbor population into four distinct groups: residential students, off-campus students, faculty and staff, and the broader community. The focus of this model is to examine ways to de-densify the residence halls, determine the most effective approach to testing and plan for vaccinations in fall 2021. Eisenberg emphasized the importance of testing in order to limit the spread of COVID-19.
“The real key to effective testing is to get infectious individuals from onset to isolation as fast as possible,” Eisenberg said, “Reducing that time from onset to isolation depends not just on how many tests you have, but it depends on student engagement, those testing resources themselves, contact tracing, case investigation, and quarantine and isolation resources.”
The presentations then shifted to tracking risk for students and Ann Arbor residents with preexisting health conditions. Jenna Wiens, associate professor of electrical engineering and computer science and co-director of Precision Health, has been using machine learning techniques to develop patterns to determine the appropriate treatments and determine which patients are at greatest risk of complications or adverse outcomes. These outcomes include in-hospital mortality, invasive mechanical ventilation, heated high-flow nasal cannula and receipt of intravenous vasopressors due to COVID-19. Wiens helped to create the Michigan COVID-19 Utilization and Risk Evaluation System, or M-CURES, which repeatedly assesses the risk of a patient over time.
“Such a predictive model could help clinicians and the hospital manage a surge in COVID-19 patients and anticipate needs well in advance,” Wiens said.
Seth Guikema, professor of industrial and operations engineering, also spoke about his model that has used risk analysis to determine the risk of students contracting COVID-19 in the fall with in-person classes. Guikema’s model simulates a full semester of classes in the College of Engineering to track individuals to look at health, educational, mental health and financial outcomes with different measures, including testing, cleaning surfaces, masks, air filtration and vaccines put in place.
“We have decisions that have to be made for (the) fall semester soon,” Guikema said, “When we think about risk, it’s about more than transmissions and infections, we have to think about some measure of the severity of the health outcomes for those individuals in the institution and the public.”