Nonprofit, nonpartisan journalism. Supported by readers.


Here’s what we know about Minnesota’s model for predicting the toll of COVID-19

The model looks at six different policy scenarios and attempts to time the peak of cases and ICU needs. This chart depicts scenarios 2 vs. 3 vs. 4.
Minnesota Department of Health
The model looks at six different policy scenarios and attempts to time the peak of cases and ICU needs. This chart depicts scenarios 2 vs. 3 vs. 4.

On Friday, state health officials unveiled more information about the model they’re using to predict how different strategies designed to contain the spread of COVID-19 might change the course of the disease in Minnesota.

The model was built by researchers at the University of Minnesota’s School of Public Health and the Minnesota Department of Health, and officials say it’s one of many tools being used to chart policy in response to the coronavirus.

Here’s what we know about Minnesota’s model (you can watch the briefing yourself here and see slides from the presentation here).

What’s the purpose of the model?

As we all know by now, COVID-19 is a fast-spreading disease caused by a novel coronavirus that is believed to have first infected humans late last year and reached global pandemic status last month.

Unlike the viruses we regularly come into contact with, nobody has immunities to COVID-19, meaning we’re all susceptible to it. That sets it apart from strains of the flu, to which parts of the population have varying degrees of immunity.

An estimated 80 percent of COVID-19 cases are mild. But for the other 20 percent of people, the disease can be severe, even deadly, generally because of respiratory complications.

Minnesota does not have enough hospital beds or ventilators to accommodate the number of people who could need care in short order if COVID-19 hit the state’s population en masse.

The model is designed to predict, in general terms, how different policies could change the rate of infection, giving the state more time to respond. It’s limited to estimating things like the spread of the disease and mortality from it; it doesn’t model, say, how different policies would affect the economy.

What are some of the underlying assumptions about COVID-19 in the model?

Any model is only as good as the assumptions it makes: How fast is the virus spreading? How at-risk are different populations? How many resources do we have now, and how will that look in the future?

Here are some of the assumptions about COVID-19 the Minnesota model makes:

  • Latent period: 5 days. This is the average amount of time between exposure and when someone has become infected.
  • Infectious period: 8 days. Following the latent period, this is how long, on average, someone would be able to transmit the virus to others.
  • R0 : Pronounced “R naught,” this number refers to how many others the average person with COVID-19 infects. The Minnesota model predicts an R0 of about 3.87 (That’s an estimate; researchers acknowledge that the real value could likely be from 2.5 to 4.7). That means the average person with COVID-19 infects approximately 3.9 other people. That makes COVID-19 less infectious than measles, which has an R0 between 12 and 18, but more infectious than the flu which has an R0 between 0.9 and 2.1.
  • Average days in hospital: 13.3 (But ranging between 7 and 23)
  • Average days in ICU: 10.3 (Ranging between 4 and 17)
  • Mortality: Is higher for some than others. Depending on age, it’s between 1.5x and 16.5x higher for those who need ICU care and can’t get it. It also increases by 7.6x for people with underlying health factors that heighten the risk of death from COVID-19, such as obesity or COPD.

Here are assumptions about risk of hospitalization and mortality by age:

Age groupCases needing hospitalizationHospitalized cases requiring ICUICU mortality rate (per 10-person days)
0-9 years0.1%5.0%0.000
10-19 years0.3%5.0%0.002
20-29 years1.2%5.0%0.001
30-39 years3.2%5.0%0.002
40-49 years4.9%6.3%0.003
50-59 years10.2%12.2%0.009
60-69 years16.6%27.4%0.024
70-79 years24.3%43.2%0.056
80+ years27.3%70.9%0.111

Based on the number of people who have died of COVID-19, health officials believe the number of lab-confirmed COVID-19 cases represents about 1 percent of the cases in the state right now. With 1,336 positive cases as of Friday, that means the true number of cases the state has seen is probably more like 134,000.

Given those assumptions, how could different mitigation strategies change how COVID-19 hits in Minnesota? 

The model looks at six different policy scenarios and estimates the time of peak of cases and peak need for ICU beds  (unless noted, the model assumes the state has 2,200 ICU beds including ventilators available for COVID-19 patients) and the number of Minnesotans expected to die throughout the course of the epidemic. (The ranges for these estimates are noted parenthetically.)

Scenario 1: No mitigation (no longer applicable since mitigation started mid-March). Assumes only 235 ICU beds, which was the number available in early March:

  • Date of peak cases: May 11, or 7 weeks (5-10 weeks)
  • Date of peak ICU need: April 20, or 4 weeks (2-5 weeks)
  • Number of beds needed at peak ICU: 3,300 (2,000-4,800)
  • Mortality: 50,000 (34,000-68,000)

Scenario 2: Two weeks of stay-at-home (i.e. what was called for in Walz’s original stay-at-home order, which has since been extended) followed by physical distancing similar to what was asked of Minnesotans before the stay-at-home order, lasting until May 1:

  • Date of peak cases: June 8, or 11 weeks (9-15 weeks)
  • Date of peak ICU need: or June 8, or 11 weeks (8-14 weeks)
  • Number of beds needed at peak ICU: 4,500 (3,200-6,000)
  • Mortality: 41,000 (22,000-59,000)

Scenario 3: Two weeks of stay-at-home followed by physical distancing until May 1, followed by a requirement that the vulnerable (such as people who are older), stay home until about July 10:

  • Date of peak cases:or June 8, or 11 weeks (9-15 weeks)
  • Date of peak ICU need: June 8, or 11 weeks (9-15 weeks)
  • Number of beds needed at peak ICU: 3,700 (2,700-4,900)
  • Mortality: 22,000 (9,000-36,000)

Scenario 3.1: Two weeks of stay-at-home followed by physical distancing until June 12, followed by requirement that the vulnerable stay home until about Aug. 14:

  • Date of peak cases: July 6, or 15 weeks (10-20 weeks)
  • Date of peak ICU need: June 29, or 14 weeks (9-19 weeks)
  • Number of beds needed at peak ICU: 3,300 (2,600-4,000)
  • Mortality: 20,000 (9,000-33,000)

Scenario 3.2: Two weeks of stay-at-home followed by physical distancing until May 1, followed by a low form of social distancing, such as requiring businesses to create conditions where people can be spaced apart, until the end of the epidemic, plus a requirement that the vulnerable stay home until about July 31.

  • Date of peak cases: June 15, or 12 weeks (9-18 weeks)
  • Date of peak ICU need: June 15, or 12 weeks (9-17 weeks)
  • Number of beds needed at peak ICU: 3,400 (2,400-4,600)
  • Mortality: 22,000 (6,000-35,000)

Scenario 4: Six weeks of stay-at-home (until May 8), physical distancing until June 6, plus stay-at-home for vulnerable populations until approximately Aug. 28. This is the scenario closest to what Gov. Tim Walz ordered, with the extension of the stay-at-home, although it assumes stay-at-home goes on for a few days longer than Walz has ordered it.

  • Date of peak cases: July 13, or 16 weeks (13-21 weeks)
  • Date of peak ICU need: July 13, or 16 weeks (12-21 weeks)
  • Number of beds needed at peak ICU: 3,700 (2,700-4,800)
  • Mortality: 22,000 (9,000-36,000)

How is this different from other models?

Minnesota health officials have stressed that the Minnesota model is one of many inputs being used to make decisions about the state’s response to COVID-19. It’s also looking at other states’ responses and other models, such as modeling by health systems in Minnesota, which tend to be more pessimistic than this one.

MDH officials have said they believe the oft-cited University of Washington model, which predicts fewer than 500 deaths for Minnesota, makes unrealistically optimistic assumptions about how well Americans’ success at social distancing and about the accuracy of death data from China.

The University of Washington model also only goes out four months, which means it doesn’t account for peaks MDH officials expect to see past that time period. The Minnesota model follows the course of the epidemic, which could be a year.

“We have to help people understand that this is not a one-time event,” Health Commissioner Jan Malcolm said Friday. “It isn’t you hit the peak and then everything goes back to normal. This is going to be with us in a really challenging way and multiple waves until there are treatments and a vaccine and that is just the hard reality that we also need to factor into this into this planning.”

What else do I need to know?

Pay attention to the ranges of the estimates. Health officials have said over and over that this model is designed to help predict a range of possible outcomes — not exact numbers. They are not designed to tell you that on x day, y people are expected to die.

And the model will change, becoming more refined as more data become available.

“I fully expect that our understanding of the disease will change every week,” said State Health Economist Stefan Gildemeister. “It will change every week, and as we incorporate that new understanding in the data, our data will change.”

Comments (17)

  1. Submitted by Annette Caruthers on 04/10/2020 - 05:35 pm.

    This is the most clearly written explanation I have found anywhere. Thank you!!

  2. Submitted by Monique Venne on 04/10/2020 - 06:13 pm.

    Thank you so much for this information. This article is far more detailed than the similar Strib report. The actual model numbers and scenarios help in making long-term plans, as well as an estimate of how long one has to stay at home. For this 60+ person, it looks like “normal” life will not resume until the fall (unless the virus mutates, there is a second surge of COVID in the fall, etc.)

  3. Submitted by Leonard Foonimin on 04/11/2020 - 07:09 am.

    agree, very good explanation

  4. Submitted by Leon Webster on 04/11/2020 - 08:44 am.

    Thanks so much. Excellent job of explaining the model and it’s assumptions

  5. Submitted by Joel Stegner on 04/11/2020 - 09:17 am.

    For the Republicans for complained about not seeing the model, now you have it. If you want to be critical of Walz’s leadership be prepared to intelligently talk about the models and deaths involved. All models involve big death totals. Let’s work to keep the numbers as low as possible.

  6. Submitted by Joel Fisher on 04/11/2020 - 09:19 am.

    Thanks for the solid information. Perhaps sometime in the future a story which includes the potential impact of soon to be available antibody tests.

  7. Submitted by Kathryn Hahne on 04/11/2020 - 09:33 am.

    Greta, thank you for the work you have put into this concise, clear explanation. As a 70+ person, it is depressing, but better to have a realistic understanding of what we are/may be facing. In addition, political battle lines are now being drawn around these issues, and we all need to really understand them.

  8. Submitted by Ray J Wallin on 04/11/2020 - 11:25 am.

    The curves presented do not match Governor Walz’ statements about flattening the curve. Walz said increasing stay at home times/distancing will lower and widen the curve. The green (strategy 3) curve is lower, not wider, than the red (strategy 2) curve.

    In addition, the green curve (ICU demand) is about 17% lower than the red, yet the predicted deaths are almost 50% lower. This cannot all be due to ICU availability/demand. This discrepancy does not make sense.

    I know data is sparse and often contradictory, but a model need not be. A model should explain any discrepancies. Unless there are cumulative effects (and they would have to be substantial), the UofM model needs tweaking even at this most basic level.

  9. Submitted by John Ferman on 04/11/2020 - 02:05 pm.

    One of the assumptions is that the virus dies out. The fact is we do not know. What is thought to be true is that virus leapt from bats to humans in China then was transported by unknowing humans. The Chinese bats are still there. Might the US virus infected transmit the virus to animals and birds and bats. This is not known. What we do know is that the corovid19 virus is aggressive and nimble and it might have properties yet to be realized.

  10. Submitted by Kate Kocher on 04/11/2020 - 02:41 pm.

    None of these projections is reflecting what is actually happening, though. The Governor feels that the University of Washington model is too optimistic…and yet we are outperforming it. Look at the data – hospitlizations per day and deaths per day. Reality trumps models.

  11. Submitted by Steve Gilbert on 04/11/2020 - 05:52 pm.

    In describing the different scenarios Ms. Kaul frequently refers to “the vulnerable” who will be asked (required?) to shelter at home for extended periods of time, the duration of which changes with each scenario (and, I assume, could change again with more extensive testing). Which Minnesotans fall into the “vulnerable” category and, more importantly, who will determine that? Will the”vulnerable” be defined by Executive Order? Or by personal choice? If by the state, is it likely to be everyone over 60? Over 70? With specified health conditions? Will the state issue requirements or just guidelines?

  12. Submitted by Kevin Powers on 04/13/2020 - 06:53 am.

    Has there been any discussion of sensitivity analysis of the model for the various input parameters?
    Do we have data yet to support the latent and infectious period numbers?
    I know that historically the MDH has promulgated more conservative numbers than other states and federal agencies (see PFAS standards for example), is that the case with this model?

  13. Submitted by Ray Schoch on 04/13/2020 - 07:34 am.

    I agree with Annette Caruthers – nice work, Greta. Ms. Kocher should read Mr. Ferman, and Mr. Gilbert asks good questions to which I’ve not yet seen credible answers, at least not yet.

  14. Submitted by joe smith on 04/13/2020 - 07:51 am.

    All models are wrong, the only true numbers are what has happened, not what may happen. You would think folks, who can actually think for themselves, might get a bit skeptical when the death toll for USA goes from 2.2M down to a new estimate of 60,000. When your estimate (taken from Neil Ferguson of UK model) is 97% high, you would think someone may question that? There was plenty of evidence that this COVID 19 virus was not “the big one” if you bothered to look. The folks dying from it were old with underlying health issues. The same as every flu season, 2017-18 flu season caused 60k+ deaths. The cruise ship Diamond Princess, with 3,700 passengers and crew, floated for a week before they were diagnosed with Coronavirus on board. With an average age of 68 and being contained on a ship, you would think the infection rate would be sky high (this is the most highly contagious COVID virus yet the models said), 83% of passengers DID NOT get the disease.
    Bottom line is models are not accurate and we shut down an entire world for a flu. While 60,000 folks dying of this virus is bad, just like it was bad in 17-18 when 60,000 died from flu, shutting down an economy is worse. 17,000,000 have had to go to the Government for unemployment checks, countless small businesses will fold and our economy will be rocked for months while we recover…… So much for believing models!

  15. Submitted by K. Knutson on 04/13/2020 - 11:44 am.

    So the model is assuming 235 ICU beds are available for the no mitigation scenario and 2,200 ICU beds (i.e., 10x as many) are available in every other scenario? This feels like cooking the books.

    The reason stated for the assumption is that number of beds was the amount available in early March, but according to the model, peak need for ICU beds under scenario one does not come until mid-April.

    Curiously, according to the model, the no mitigation approach (Scenario 1) has the lowest peak need for ICU beds — tied with the most stringent approach, which is Scenario 4. If the same number of ICU beds was used for Scenario 1, would it have the second lowest number of fatalities?

  16. Submitted by Andrew Gaspard on 04/13/2020 - 03:09 pm.

    Thank you for writing this piece. I am having a hard time understanding the estimated death tolls in all scenarios of 20,000+. It is confusing in the context of overall U.S. estimates, currently forecast as 60K (See

    Also is out of proportion to death tolls in other countries to date (for example, Italy 20k, Spain 17K, etc.). Is the expectation MN would have a number of cases similar to Italy despite having a only a fraction of the population?

  17. Submitted by John Sandberg on 05/04/2020 - 09:46 am.

    Any word on when they’ll release updated modeling information, incorporating extension of the Stay At Home order and the last several weeks of testing/hospitalization/mortality data?

Leave a Reply