Ricochet is the best place on the internet to discuss the issues of the day, either through commenting on posts or writing your own for our active and dynamic community in a fully moderated environment. In addition, the Ricochet Audio Network offers over 50 original podcasts with new episodes released every day.
Coronavirus Update: A New Projection for the US
There’s a new COVID-19 projection for the US, released by the Institute for Health Metrics and Evaluation (IHME) at the University of Washington. Our friend Rodin has an excellent post on the subject, earlier today (here). Special thanks to Rodin for all of his work on this, keeping us updated with daily posts for over a month now.
The IHME has an interactive page about its projections here. It has excellent graphs and tabular information, with a drop-down menu allowing you to focus on a particular state. The purpose of this post is to provide you with a state-by-state summary.
Before getting to this, I’m going to comment on one of the IHME graphs posted by Rodin, showing their projection of total deaths in the US:
As you can see, the projection follows the expected S-curve shape, and reaches a total of about 80,000. This is significant, but is pretty close to the total for a bad seasonal flu. The shaded areas show their high and low projections, while the dotted line is their main projection. The low projection is 32,766, and the high projection is 179,726.
For the rest of this post, I’ll use their main projections, but you should realize that these are estimates within a range. You should also realize that this is just a projection, and may be correct or incorrect. The IMHE data can be downloaded, in Excel-compatible format, via a link at the bottom of their page.
Here is a chart of IHME projected deaths by state, in alphabetical order.
This chart shows total counts by state, without adjusting for population, so naturally the big states like California, Texas, and New York have the highest figures. To put it in perspective, here is the same data expressed in projected deaths per million. This chart also shows the nationwide projection at the top.
This chart allows us to see the states projected to suffer the most, with tiny Vermont expected to have the most deaths per capita, followed by New York, Missouri, New Jersey, Louisiana, and Michigan.
In addition to deaths, the IMHE projects the shortfall of available hospital beds and ICU beds. I’ve focused on ICU beds. The IMHE page lists a peak shortfall of 14,601 ICU beds, but this is because they report figures on overall peak utilization day based on total hospital beds needed (which is April 14 in their projection). The ICU bed shortfall actually peaks a few days earlier, on April 9 at a shortfall of 17,380.
Many states are projected to have little or no shortfall in ICU bed needs. Here is a chart of the shortfall, by state.
Most of the states projected to have an unusually high number of deaths, per capita, stand out on this chart — New York, New Jersey, Michigan, and Louisiana. Vermont does not stand out, because of its tiny population. But Vermont is worst in this category, having a peak ICU bed shortfall of 466 per million, followed by New York (373), Michigan (273), New Jersey (267), and Louisiana (217).
Note that the state subtotals for the ICU bed shortfall do not add up to the national total. This is because the IHME report include different projections for each state, by day, and the chart above reports the maximum figure for each state. The maximum ICU bed shortfall can occur on different days for different states, while the national total is the sum of all of the state shortfalls for each date.
Let me know if you have any questions about the IHME projections. I do not have answers about methodology, but may have the information to respond on other issues.
ChiCom delenda est.
Published in Healthcare
I think it’s worth mentioning the second line on their website:
So they say coronavirus will be equivalent to a bad flu season if draconian measures are kept intact.
I’m not making any judgment on the validity of these projections – I really have no way of judging whether they’re plausible or not – but I do think this is a very important aspect to understanding their forecasts and should not be overlooked.
Mendel, we don’t even know this. We don’t know what protective measures they used in their projections.
Man, I messed up the acronym — I used IHME the first time, then IMHE thereafter. I fixed it in the post, but I made the same mistake in the charts, and it’s too much trouble to fix it. Sorry.
Sometimes I hate acronyms.
Yes we do. The quote in boldface in my comment is from their website. Two more clicks leads to the description of their methods:
http://www.healthdata.org/sites/default/files/files/research_articles/2020/COVID-forecasting-03252020_4.pdf
Under “Data identification and processing”
Thanks for putting this together in this helpful form.
Two things strike me about the projection compared to a “normal” flu season. According to the CDC, flu deaths since the 2010-11 season have varied from about 12,000 to 61,000. Most of this is in our mental model of “normal” risk. We are used to it, for better or worse.
The IHME projection shows daily death counts hitting 1,000 on April 2, staying above that level until May 5 and about 60,000 deaths in that period. The sheer number, pace, and visibility of those deaths is going to have a much bigger public impact than even the worst flu season. What that impact is, and how it plays out, I don’t know, but if this happens as IHME projects it will be significant.
The other is something I raised on Rodin’s post. If the early peak and quick decline is attributable to our interventions, including social distancing, but the result is we haven’t built sufficient group immunity against the virus, what happens when we stop those interventions?
I don’t like that the graphs show that isolation lowers the curve, but does not flatten it. Reducing contact should lower the effective R0 and that should make the curve wider and flatter. Instead, it is the same duration and shorter. Something is wrong there.
I’m not sure that “draconian” measures must be maintained in order to achieve “strong social distancing.” Given that the at-risk population skews heavily toward the elderly and retired/unemployed, I don’t know that the universal nature of the current stay-at-home shutdown procedures are really required. Right now, no one knows.
We could invest a lot of money — a lot of money — in protecting the most at-risk population, and still come out ahead if many of the rest of us were able to resume working.
Yup. I caught that, too. EDIT: Those are what you might call aspirational death tolls…
The charts make it really clear what officials are worrying about.
By the way, though I know that “anecdote” does not equal “data,” I now know two people who have been hospitalized with this thing. One is old-ish (70s) but the other is maybe 55? I’ve never known anyone to be hospitalized with the ordinary flu.
Okay, but… you know that they are, right? Every year, many people die that way. I don’t assume they all die on the street or at home or in an Applebees or whatever.
I remain generally wary of models, and more so when there is so little data available to inform them.
I found this comment from the site interesting:
Given the degree to which the most densely populated (and, I suspect, most dependent on mass transit) metropolitan area in America dominates all of the US statistics, this seems like a potentially important limitation of the model. The heavy reliance of the model on mortality data from China and Italy is also not reassuring.
The error range is sufficiently large (40K to 160K) that I won’t be surprised if the model hits it, but I’ll be very surprised if its specific projection is anywhere near correct.
24,000 so far this flu season.
Typically, the error bands will cover 95% of of the outcomes, so I expect that you’ll be right.
It’s not a model with a mechanism simulation, it’s curve fitting, based on previous outcomes from US, Italy, and early Chinese numbers. They should be able to keep it updated as new data comes in, but their mapping of distancing measures to curve changes is more than a bit of a SWAG.
Yes, I know. But still. Oh, and as of this evening it’s four.
It’s interesting that they regard some states as wild unknowns, while for others their estimates are within a tight margin. Eg, Florida will peak at about 10 deaths/day on Apr 4 or so, or possibly it’ll peak at 150/day on May 15, with its totals between 390 and 9,100 (!!!) GA, LA, and CA have similarly wide variance in projections.
OH on the other hand is expected to peak between 85 and 100 deaths/day in mid April, with totals at 2,100-3,400. That’s a pretty precise guess considering there’s still pretty low numbers in OH at this stage to discern patterns.
The southern states seem to have the wider range of possibilities, on balance. Possibly suggests that the effect of weather is a great unknown. (???)
That’s relatively easy: the NYC
doublingdeaths doubled in two days and the total is exceeding their maximum projections.As for the US as a whole,
So I’d say stick a fork in that model. It’s done.
I share your skepticism about the accuracy of the IHME projections. I’d also caution against using any data that purports to reflect national fatalities on a timescale shorter than a full day, given the inconsistency of the various reporting jurisdictions.
New York continues to skew the total. We don’t know how much those results will be replicated across the country; we don’t know how big a factor the density, mass transit, and demographic characteristics of NYC influence the progression of the disease there. And we are just entering the time when effects of the social distancing and shutdown procedure would begin to be counted in the mortality statistics.
I think it’s too early to tell.
Vermont is interesting.
Who is the governor?
Who are the senators?
New York City is an outlier — thank goodness for the rest of the other 49 states
retired is not the same as unemployed
Agreed.
BTW, Rhode Island is trying to keep New Yorkers out:
This report which claims to be, and it sounds plausible, from an ER doc in New Orleans is alarming:
Another datum:
you know what would be interesting?
number of chinese students in each state
number of chinese tourists in each state
number of chinese <fill in the blank> in each state
As an American of Korean descent, I’m calling it a Chinese virus since it started in Wuhan.
I’m offended by Asian virus because it’s simply incorrect.
I know. That’s why I mentioned both.
California has routinely blocked the construction of new hospitals.
When was the last hospital built in California?
I have no idea and have lived her all my life.
One hospital in Los Angeles recently closed in January.
Don’t block the construction of new hospitals/beds and then complain about the lack of beds, Governor hair gel
WE know what happens on a cruise ship called the diamond princess that docked in Japan
We are learning, certainly.
But the Diamond Princess is not representative of America either. It’s a great test case, because everyone on it was tested, whether or not they had symptoms. We can learn a lot from that. But it isn’t America. It’s a few thousand mostly older adults packed into close quarters together for a month, sharing facilities and trying to avoid infection.
So about 20% of them caught the virus under those conditions. We don’t know how that translates to a socially-distancing America. Half of those who tested positive were asymptomatic. That might transfer to America, but we don’t know because they were an older than average crowd; maybe young people are more likely to be asymptomatic.
We don’t yet know what “normal” America looks like. I’m sure we’ll learn soon.
it means that America will fare better than the cruise.
the cruise is the worst case scenario.
its the closest thing we have to a controlled experiment.
I live near Penn State University, where my wife is a professor. She teaches a massive online course, so essentially nothing has changed for her. There are as of this morning 15 confirmed in Centre County. PSU has lots of Chinese students. I’m not sure how much traveling back and forth they do (certainly none now). There were many in Wegmans (regional grocery chain) when I went yesterday. So just based on that the correlation can’t be that big.
My wife got essentially the same email from many of her students that went back to China, which said something to the effect of “China has conquered the virus, medical care is so much better here, so I’m staying here, can I finish the course from here?” Reminded me of my grad school days at Cornell (’76-’82), where someone would distribute Maoist propaganda to our mailboxes. There was a major earthquake in China that killed 10’s of thousands, but not according to the Maoist rag. “The Chinese people, under the revolutionary guidance of Chairman Mao, leapt to the ground and wrestled the earthquake into submission, and no one died”.