100,000–200,000 COVID-19 Deaths. Where Does That Estimate Come From?

President Trump said at yesterday’s coronavirus briefing that the country would be doing well if the numbers from the COVID-19 were held down to 100,000.

And Anthony Fauci, MD, said 100,000 to 200,000 deaths is entirely conceivable, a range he had mentioned earlier in the day during an interview on CNN with Jake Tapper. In his interview with Tapper, Fauci misspoke initially and said cases instead of deaths: “Looking at what we are seeing now, I would say between 100,000 and 200,000 cases-but I don’t want to be held to that because-excuse me, deaths. We are going to be have millions of cases.”

Deborah Birx, MD, mentioned a range of 80,000 to 160,000 deaths at the press conference yesterday afternoon. On the Today show this morning, Birx said, "If we do things together well, almost perfectly, we could get in the range of 100,000 to 200,000 fatalities."

These projections of the numbers of deaths the might result from the COVID-19 outbreak are coming from models constructed by epidemiologists and other disease experts. The models are complex and virtually impossible for the layperson to understand in any detail, but, in essence, they take information about the number of cases, the rate of spread, and the fraction of those infected who will die from the disease to arrive at estimates of the COVID-19 death toll. Like all models, they are based on some assumptions from past experience-which, in this case, involves use of data from how COVID-19 behaved in other countries-and some others based on the researcher’s understanding of how the virus is behaving in this country. 

“A model is only as good as the assumption you put into the model,” Fauci said at the press conference yesterday, a maxim he has used several times.

“We are hoping that the models aren’t completely right-that we can do better than what the predictions are,” said Birx, who last week drew some sharp criticism from one of the modelers, Harvard’s Marc Lipsitch, when she suggested that the models weren't matching what is occurring in the U.S. and unduly frightening people.

Yesterday Birx mentioned looking at 12 different models and also referred to specifically to a model published by Christopher J.L. Murray of the Institute for Health Metrics and Evaluation at the University of Washington. Murray is a highly regarded disease researcher who is well known for his burden of disease studies.

Murray's model projects that there will be 82,141 deaths from COVID-19 by August 4 with the deaths per day peaking at 2,271 on April 15. Like all such models, Murray’s has a range-and a wide one, which is reflective of uncertainty. The lower bound of his estimate of the number of deaths is 38,242, the upper bound, 162,106

Murray’s model and its projections are on the lefthand of the distribution of what other modelers are projecting. Thomas McAndrew at the University of Massachusetts, Amherst, has been surveying infectious disease modelers and aggregating their results during the COVID-19 outbreak, and Nate Silver’s FiveThirtyEight data journalism website has been highlighting and explaining his work with some helpful charts.

McAndrew surveyed 20 modelers (one of whom is Lipsitch) on March 23rd and 24th (last Monday and Tuesday). According to his calculation, the midpoint of the modelers’ models project that there will be 245,500 deaths in the U.S. from COVID-19 by the end of the year, although there is a wide range around that number with a lower bound of 61,900 and an upper one of 784,100.

When McAndrew surveyed 18 modelers on March 16th and 17th, the midpoint of their death estimates was considerably lower at 195,000 with an even wider range (lower bound of 19,000 and an upper bound of 1.2 million).

One of the main factors in calculating the number of deaths are estimates of the number of cases McAndrew reports that in the March 23-24 survey, the modelers estimate that there were 361,900 total cases in the U.S. on March 23, a figure that includes their estimates of the subclinical and asymptomatic cases as well as symptomatic ones. That case number is nearly three times the 122,653 cases than the CDC was reporting as of March 29.