No one knows for sure how bad COVID-19 will get.
Data modeling by Minnesota experts predicted as many as 50,000 deaths in the state, while a University of Washington model estimated fewer than 2,000.
Nationally, the Imperial College of London predicted 2.2 million deaths if the U.S. did nothing to slow the coronavirus causing the pandemic, while the same Washington analysts recently lowered their estimates from 200,000 deaths to 93,000 because mitigation strategies such as Minnesota’s two week “stay-at-home” order appear to be working.
While the variations have caused confusion in an increasingly homebound and anxious public, data modelers said they are to be expected as the nation’s top health care analysts plug differing assumptions into their formulas about this unfolding COVID-19 pandemic.
“It’s not really that one model is that much better or that much worse,” said Dr. Henry Ting at Mayo Clinic, which has examined 13 COVID-19 models in the development of its own. “The key is the assumptions you’re putting into these models. Those are driving the differences.”
Modeling by the Minnesota Department of Health and the University of Minnesota persuaded Gov. Tim Walz to announce the current stay-at-home order. The modeling is conservative, perhaps pessimistic, about the course of the outbreak, said Stefan Gildemeister, state health economist, but the Washington model may be optimistic and overlooking risk factors in the United States that could make the outbreak worse.
“Saying that the Washington approach is optimistic is not saying that we think we’re right,” he said. “Some of our assumptions might have turned out to be unusually conservative. In fact, we’ve been saying this from the beginning. We will continue to test our assumptions and change them.”
The Minnesota modeling wasn’t designed to forecast deaths but rather how much they would decline under social distancing restrictions. Gildemeister and colleagues concluded that deaths could decline by one-third under the current stay-at-home strategy — primarily by delaying the surge of cases so that hospitals have more time to prepare.
When it comes to the death numbers, any change in assumptions makes a dramatic difference. The Minnesota researchers used hospitalization and death rates that increase with age and were first used by colleagues at Imperial College.
The spreading factor
The local researchers also used a so-called R-naught estimate that one Minnesotan carrying the coronavirus would spread it on average to 2.5 people, whereas initial global estimates were only 2.2.
In the world of R-naughts, tenths can make a substantial difference in estimating how rapidly a virus can spread, said Dr. Frank Rhame, a virologist for Allina Health. “Over a couple of months, 2.2 and 2.5 is a big difference.”
With no interventions at all, the Minnesota model estimated 74,000 deaths in the state. Under the current stay-at-home strategy, it estimated around 50,000. However, Rhame said the reduced face-to-face contact in Minnesota could be reducing the R-naught, which in turn would reduce infections and deaths.
The addition of ventilators could help as well, as the Minnesota model predicts local hospitals will run out before the COVID-19 outbreak peaks. Globally, only around 5% of COVID-19 cases have needed hospital intensive care, and fewer yet have needed ventilators to compensate for the severe respiratory symptoms that can occur. But death risks increase dramatically when people don’t have ventilators available when needed.
The Minnesota model also assumed that people with COVID-19 needing intensive care would remain in hospital ICU units on average for 23 days. The actual length of stay could be shorter, which would mean more availability of beds and ventilators.
The length of stay “affects available resources, and that affects the timing of when we reach overcapacity, and that affects [the rate of] mortality,” Gildemeister said.
The differing models agree that hospitals are slated to be overrun. ProPublica published data from the Harvard Global Health Institute that Minnesota would need 2.9 times as many hospital beds as it has available if only 40% of its population were infected over the next year.
“Models are models,” said Dr. Rahul Koranne of the Minnesota Hospital Association. “They’re all going to be somewhat wrong. But if they all say a surge is coming, that you need to prepare now, and that social distancing matters, then let’s do that and work hard at doing that.”
Modeling by the University of Washington’s Institute for Health Metrics and Evaluation has grown in prominence and was highlighted this week by Deborah Birx, the White House coronavirus response coordinator.
The results of that model are updated daily with new information from states on both their deaths and the dates on which they implement prevention strategies.
Deaths vs. infections
A key difference from the approach of the Minnesota model and others is that it is based on deaths, rather than infection counts, said Ali Mokdad, University of Washington chief strategy officer for population health. “We have much better data on death than we have on diagnosis, because people who need to be diagnosed are not being tested, because we didn’t have enough tests early on.”
The results suggest that interventions work, with death rates ebbing in hard-hit cities such as Seattle and San Francisco that reacted quickly, he said. “We were concerned that a shutdown in China doesn’t mean exactly the same as a shutdown in the U.S. Now we’re seeing it works here.”
The Washington model suggests an improving situation in Minnesota. It predicted a death toll of more than 2,000 in the state a week ago but only 932 as of Wednesday because of the apparent impact of Walz’s stay-at-home order.
While Minnesota’s real-life death tally is increasing, it hasn’t been at the exponential rate expected — which suggests the interventions are working, Mokdad said.
One concern about the Washington model is that it doesn’t directly account for the high level of obesity and chronic disease in the U.S., which could result in more severe illnesses and deaths than in China or other countries, Gildemeister said.
The Minnesota model was based on prior studies that demonstrated that people with chronic diseases were 7.6 times more likely to die from a COVID-19 illness than people without those conditions.
Minnesota modeling has predicted only a delayed peak in COVID-19 cases and hospitalizations because of the mitigation strategies, and not a reduction of cases, but updated analysis may show differently, Gildemeister said.
“The hope is that we were wrong,” he said, “and that the peak is farther out and that mitigation in fact reduces the degree to which people transmit the disease.”