When COVID-19 arrived in Michigan, decisions about shutdowns, reopening and resource allocation couldn’t wait for perfect data. MIDAS moved quickly, funding rapid-response projects, convening modelers and public health experts, and building tools that showed where the virus was spreading and who was being left behind. The Zoom call stretched late into the night in early April 2020. Scattered across the screen, University of Michigan administrators were weighing whether, and if so, how to bring students back to campus. Amongst them, epidemiologists and data scientists were sharing maps and models that made the decision anything but straightforward.
Zelner and his colleagues showed COVID-19 spreading across Michigan counties in alarming waves. Another visualization, created using anonymized WiFi data, traced how students moved through campus building (where they clustered and how long they stayed), trying to predict which spaces posed the highest risk for transmission.
The data were imperfect. Testing was scarce and uneven across regions. As case counts rose, No one knew how much of the rising case counts reflected real spread or increased testing. Hospital capacity projections also swung wildly depending on what assumptions were being used.
“We were making high-stakes decisions with incomplete information,” said H. V. Jagadish, MIDAS Director from 2019 to 2025. “The question wasn’t whether to use data; it was how to use messy, biased, rapidly changing data responsibly, and be honest about what we did and didn’t know.”
The tension between urgency and uncertainty defined Michigan’s (and the world’s) early pandemic response. What helped set the university apart was how quickly MIDAS mobilized its network of data scientists, public health researchers, clinicians, and modelers to tackle problems that could not wait for traditional grant cycles.
Between April and May 2020, MIDAS reviewed 49 proposals and funded seven interdisciplinary COVID-19 projects, launching all of them within 30 days.
Why case counts weren’t enough
In the pandemic’s early months, public attention focused on confirmed cases, hospitalizations, and deaths. But those numbers told an incomplete story.
Biostatistician Bhramar Mukherjee spelled out a “testing paradox.” When testing is limited and selective, raw case counts severely underestimate true infection rates and make trends difficult to interpret. Everyone was flying blind. Some counties reported hundreds of cases and others almost none, but that often reflected access to testing more than actual spread.
With MIDAS funding, Mukherjee developed methods to estimate underlying infection prevalence and optimize how limited tests should be allocated. Her team combined infectious disease models with survey sampling techniques to infer who was being missed. The team was able to also predict where testing would be most informative, helping public health officials interpret positivity rates and decide where to deploy mobile testing units. It wasn’t just about the math, it was about communicating uncertainty in ways that supported real decisions.
Following students through WiFi signals
As state-level researchers tracked the virus, another team focused on the campus itself. If students returned, where would transmission risk be highest?
Quan Nguyen, Christopher Brooks, and Daniel Romero used anonymized WiFi connection data to map how students moved through campus spaces. Aggregated patterns revealed which buildings became crowded and for how long. It allowed the team to see empirically where risks were highest. not just in terms of occupancy, but also in duration and overlap between groups.
Across universities, classroom schedules were adjusted, study spaces reconfigured to reduce proximity, and high-risk building usage managed to reduce simultaneous occupancy. The insights from this and similar studies fed directly into such reopening plans.
Seeing the virus across the state
Zelner’s lab zoomed out to analyze COVID-19 spread across Michigan using high-resolution spatial modeling. Beyond mapping cases, the team integrated demographic data and structural factors such as housing density, occupational exposure, and healthcare access.
Their interactive Michigan COVID-19 Tracker became a widely used public resource for journalists, health officials, and residents.
“Place matters because of what place represents,” Zelner said. “Social and economic structures that shape who gets exposed, who gets sick, and who survives.”
The work showed that predominantly Black neighborhoods in Detroit and Flint faced compounded risk— patterns tied to decades of policy decisions rather than chance. These insights guided targeted testing, vaccination outreach, and resource deployment.
Zelner now contributes to the CDC-funded research collaborative, the Michigan Public Health Integrated Center for Outbreak Analytics and Modeling, which extends this work to future outbreaks.
Mapping inequality
Early mortality data showed that Black Michiganders were dying from COVID-19 at disproportionately high rates, but the numbers alone did not explain why. Epidemiologist Nancy Fleischer, whose work focuses on health disparities, used MIDAS funding to build a probability sample of people who tested positive for COVID-19 and survey them about illness, care, and recovery.
The results revealed inequities at every stage of the disease experience. Black respondents reported more severe symptoms, longer hospital stays, worse treatment experiences, and less follow-up care—even after accounting for underlying health conditions and socioeconomic factors. Disparities also extended to mental health impacts and what would later be recognized as long COVID.
“We knew from mortality data that Black communities were bearing a disproportionate burden,” Fleischer said. “But our new research showed how deeply those inequities were embedded.”
The findings informed resource allocation decisions and broadened conversations about structural racism in healthcare.
Inside the hospital
At Michigan Medicine, clinicians faced a different challenge: predicting which hospitalized patients would deteriorate.
A MIDAS-funded team led by Andrew Admon and Christopher Gillies developed real-time models using electronic health record data to track patient trajectories. The approach combined transfer learning with ordinal regression to monitor risk hour by hour.
“COVID overwhelmed hospitals in ways we weren’t prepared for,” Admon said. “Even imperfect early warnings helped us be proactive rather than constantly reacting.”
The tools supported staffing and ICU planning and informed later efforts to design clinical decision systems that account for equity and population differences.
The 30-day turnaround
What made this response possible was not just expertise, but institutional agility.
Traditional grant cycles take months. MIDAS compressed the entire process, from proposal submission to review and to funding, into 30 days. “We accelerated everything,” Jing Liu, MIDAS Executive Director, said. “Researchers and reviewers understood the urgency and the potential impact. While everyone was trying to also pivot how to teach and do research and adjust to a new way of life, they helped MIDAS make this rapid funding happen.”
MIDAS also convened cross-campus working groups that met throughout 2020, connecting public health
The Legacy
The pandemic emergency has passed, but its infrastructure remains. Data pipelines linking clinical, mobility, and demographic data persist, alongside stronger cross-campus relationships and a shared understanding of what crisis-ready analytics require: speed, transparency, equity, and connection to decisions.
Methods developed during COVID-19 now inform broader work: Mukherjee’s approaches support ongoing surveillance, Zelner’s modeling extends to future outbreak analytics, and Fleischer’s research continues through long-COVID studies.
