The Star Tribune’s Jeremy Olson is a Pulitzer Prize-winning reporter covering health care. Trained in investigative and computer-assisted reporting, Olson has covered politics, social services, and family issues. I took a critical look at Olson’s first look at the model underlying the current shutdown ordered by Minnesota Governor Tim Walz in “Coronavirus in one state (3)” and in “No one here gets out alive.”
Olson returns today with the 1,300-word page-one story “Minnesota’s COVID-19 response shows promise in helping to slow virus.” Subhead: “Walz said he will be looking this week for updated modeling and any signs of the infection ebbing before deciding on extend his stay-at-home order.”
I have read Olson’s story closely several times. I may have read it more closely than he himself has. In his mess of verbiage he fails to note that that the governor predicated his current shutdown order on the assertion that, according to the model relied on by the governor, Minnesota would experience 74,000 deaths by the virus absent his order. Last week, in his earlier page-one story, Olson reported that, with the order, Minnesota would experience 50,000 deaths. I do believe the governor left that out in the announcement of his March 25 order.
Olson’s story today posits a tentative optimism. He forgets about the projected 50,000 deaths based on the model reported in his last story. As of today, the Minnesota Department of Health attributes 29 deaths to the coronavirus. Olson also omits the arithmetic that would suggest we have 49,971 deaths to go if everything goes as projected! (Projected by the governor, that is.)
Olson instead works with the IHME model:
Modeling by the University of Washington’s Institute for Health Metrics and Evaluation suggests this is working. Deaths so far haven’t been increasing in Minnesota at the expected exponential rate, prompting the institute to lower its forecasted COVID-19 deaths in the state from around 2,000 two weeks ago to 932.
“We’re seeing the impact of these measures and how early they are” put in place, said Ali Mokdad, chief strategy officer for population health at the University of Washington.
I asked my friend Kevin Roche for his reaction to Olson’s story today. Kevin is the former general counsel of UnitedHealthGroup and chief executive officer of its Ingenix division. Kevin responded:
Olson does not make it clear, and this is the critical point, that when the Governor says his extreme lockdown is working, he means it is working according to the model he based it on. As Olson’s story on Friday in the Star Tribune makes clear, that model was predicated on only delaying deaths, except for those allegedly caused by overrunning the health system with cases. Under the model, 50,000 deaths will still occur, just spread out over a longer time. So a more accurate statement would be the Governor’s extreme shutdown is working to delay deaths.
Now I don’t believe at all that we are preventing 24,000 deaths. The model appears to be using an erroneous infection rate, an erroneous mortality rate, an erroneous ICU stay length and other factors that led to that worst case projection. And if the model is flawed in its predictions of overwhelming the health system, every one of those deaths that doesn’t occur comes off the top of the 74,000 number; it isn’t a proportional reduction. So if there is no overwhelming of health resources, 50,000 deaths will occur, it is just a matter of when. This can be clearly seen in the University of Washington IMHE chart…The virus isn’t disappearing. [Please see clarifications per the update below.]
So one of two things is true, and they are mutually exclusive. Either the numbers in the model are wildly off and were never going to happen — so the justification for the extreme lockdown doesn’t exist — or the model is right, which is what the governor said he believed and is consistent with his actions, in which case we are going to have 50,000 deaths no matter what with no lives were saved.
I think if Minnesotans understood that at the start, they would have been outraged at both the deception and at the enormous economic, non-economic and social damage being done to Minnesotans.
I urge interested readers to look over my shoulder at Jeremy Olson’s long page-one story today to see just how little light a Pulitzer Prize-winning reporter covering health care can shed when he is disinclined, for whatever reason, to “question authority.”
UPDATE: Kevin adds this clarification regarding the IMHE:
A couple of very good comments made me realize that maybe my assumptions about the IMHE model and projections weren’t right and I should go back to poke around on the site. Having been distracted much of the day with other analysis and grandkids, I didn’t explain myself well in my original reaction to the Star Tribune article or the comments. Here is what I hope is a better explanation.
The main purpose of the IMHE site was supposedly to help states plan for whether or not they might have cases requiring care in excess of the resources –hospital beds, ICU beds and ventilators –available to provide that treatment. It was basically a near-term forecast and I assumed it was what I would call a runout model. In other words, they assumed social distancing (I would actually call the set of measures they were using an economic lockdown) for a set period of time, in this case through May, and they made assumptions about the number of cases you would see through May under those social distancing restrictions. Then you just run out the outcomes of those cases. So in their projection, there are no new cases after the end of May and the cases have had their outcomes by August. This is what the cumulative chart reflects. The day charts I don’t regard as particularly useful other than as a check against the reality of what actually happened versus their projections.
My poking around verifies that I was correct in my assumptions, but also indicates a far more dire situation that I realized. If you look in the frequently asked questions section here, you will see that they were in fact only forecasting first wave, and that for that first wave they only assumed 3 percent of the population became infected, or as they put it, 97 percent of the population remained susceptible. They note that this likely means there will be a resurgence if restrictions are lifted. So we will face the ongoing choice of keeping the lockdown in place or seeing a bunch of new cases in June and beyond.
Commenters and others assume that the approximately 93,500 total US deaths they are forecasting was it. It isn’t by a long shot. These are just the deaths from the first wave, that 3 percent of the population. And if you do the math, they are assuming about a 1 percent fatality rate in the infected population. That is a big number if you apply to whatever remaining part of the population you think will be infected. Now personally I think that fatality rate is much too high. But the key point is that forecast is only the first wave. And as they say, that certainly isn’t all there will be to the epidemic.
And this episode has reminded me again of the need for people to be clear in what they are modeling and the assumptions they are making. You shouldn’t have to go to FAQs to figure that out. The graphs should be labeled “First Wave Only.”
I have slightly edited Kevin’s original message to confine the discussion based on the IMHE projections to this update.