On Tuesday, networks were once again ready to proclaim the day on which Donald Trump became president by noting that he stopped using third-grade nicknames long enough to deliver the message that as many as a quarter of a million Americans may die. Then he went right back to talking about the important things—like how great the ratings are when he steps out to tell Americans how they’re going to die. It all gives the distinct impression that, had Trump known that reading a daily death sentence out for hundreds of thousands of viewers, Trump would have scheduled a pandemic long before this.
But as Trump runs his fingers across a chart that turns killing 240,000 Americans into a “goal,” it’s worth examining … where are these numbers coming from? And is the doom Trump is pronouncing really as certain as it seems?
The answer to the “where” question seems to be here, the Institute for Health Metrics and Evaluation at the University of Washington. Working with public data and the most commonly used statistics about COVID-19, the IHME has prepared a study looking at the need for hospital beds, ICU suites, and respirators over the next four months. That model currently predicts that COVID-19 in the United States will peak just two weeks from now, on April 16, when the disease will demand 84,000 more hospital beds than are available, and deaths will reach 2,600 a day. Their model calls for a steep decline in new cases during May.
By August, they project total number of deaths across the country between a low of 40,000 and a high of 175,000. If those numbers don’t completely align with what Trump was pointing at in his last session, it’s because the model is regularly updated with new information and in response to new actions.
The numbers are horrific—though less horrific than some that have been put forward. To get a better sense of what’s driving them, this open source epidemic calculator allows users to fiddle with many of the basic numbers. Push up the basic reproductive number R0, also known as transmissibility, and the speed of contagion rises rapidly. Other factors include how long a patient is contagious, how long those who are hospitalized are likely to stay, and at what stage in the epidemic government acts to impose restrictions that cut down on transmission.
In the Washington IHME analysis, they looked at each state, using these factors and others to model the expected course of the disease while comparing those results to the number of available hospital beds, ICU rooms, and respirators. This was done not as a single national model, but by looking at state by state resources. The Washington model also takes into account the dates on which each state put in place a statewide lockdown, closures of schools, limits on services, and restrictions on travel.
Unlike most of the values that you’re seeing, the numbers aren’t “confirmed cases,” but really resource utilization and the deaths that result both from the disease, and from the failure to protect the healthcare system from being overrun by cases. The factors that are used in the model are laid out in detail, and the full data set behind the model is available for download.
Not every state hits its peak resource needs on the same day, and in general those who have taken action to push their peaks later, or simply been lucky enough to not be an early hot spot, benefit from that delay by spreading cases out over a longer period of time. Those that have a peak in the next couple of weeks get the biggest crunch.
Taking a few examples, at the very top of the alphabet, Alabama shows what happens with a state that has done very little to enforce social distancing.
The result in Alabama is that the state crashes through its available hospital beds in just the next few days, greatly exceeds the available resources on every front, and sees a death rate of over 300 a day in the next three weeks, leading to a total of around 7,300 deaths. See that purple line all the way down near the bottom of the chart? That’s the total number of hospital rooms in the state. It barely registers against the projected number of cases.
It’s a hideous outcome, but one that still has quite a bit of potential change—those totals could run from 1,100 dead to a staggering 17,000, depending on actions still to be taken. The broad swath of light purple on the chart still shows how much variability remains in this outcome.
On the other end of the spectrum, Kentucky, where Governor Andy Beshear has gotten considerable recognition for taking early, forceful action to protect the citizens of his state, has a very, very different outcome.
Kentucky doesn’t reach peak demand until mid-May, never exceeds the available hospital or ICU beds (on this chart, available beds are way up there at the top), and ends up with a projected total of 800 deaths. And in case you’re wondering, the populations of Kentucky and Alabama are quite similar.
In larger states, the differences are similarly drastic. California, where local lockdowns in the Bay Area played a huge role in pushing back against an early hot spot, reaches peak demand on May 1 with a caseload that pushes, but does not exceed, the available resources. The estimated deaths come to over 5,000—daunting, but a fraction of what could have happened had California’s government reacted the way Alabama’s has.
The most amazing result may be Washington state, where Governor Jay Inslee never let up on testing, isolation, suppression … everything that could be done. The result is that, despite starting out as the state with the first large concentration of community-spread cases, and the highest death rate in the country, Washington looks set to go out on the other end of the scale, as one of the states with the lowest number of deaths … and not just per-capita. Total projected deaths for the state are 1,233 … and that’s including the 243 that have already occurred. Again, the death of 1,000 Americans—the avoidable deaths, had there been a competent federal response—is nothing to celebrate. But the difference brought on by firm action is certainly worth praise.
Unfortunately for some states that are taking it hard already, there’s not much good news other than … it should start to get better soon. On these projections, New York hits its peak in just 8 days, with a need for 11,000 more ICU beds than currently exist and deaths exceeding 800 a day. I am so sorry.
These are just projections. This is just a model. The numbers have already changed in the last day since Trump was pointing at the chart. They will change again. And, like weather models, there are many factors not included that may alter the outcomes severely in either direction.
Thursday, Apr 2, 2020 · 11:47:39 AM +00:00
·
Mark Sumner
Yesterday, projecting just one day ahead from the last update, the IHME model undershot the number of actual deaths in the United States by 17% … which doesn’t exactly make it’s rather simplistic assumptions seem all that capable of projecting the next four months.