Monday, March 23, 2020

Can you say "inexorable"?

I now have two blogs: https://tomalrichblog.blogspot.com/ (devoted to cybersecurity and compliance in the electric power industry) and https://thepandemicblog.blogspot.com/ (devoted to...well, you know what!). While I will continue to put pandemic posts on both blogs in the near future, I will soon also start putting cyber posts up on this blog, and they will be the one that gets picked up by the email feed. So if you're only concerned with pandemic issues - that's also a limited audience: only human beings, absolutely no animals - you might want to go to that blog and subscribe to the email feed, so you'll always be sure to have pandemic posts emailed to you.

Yesterday’s numbers (as of noon EDT):
Total US confirmed cases (from Worldometer): 33,546
Increase in cases since previous day: 9,339
Percent increase in cases since yesterday: 39%
Percent increase in cases since 7 days previous: 812%
Expected cases 7 days from yesterday (March 29): 272,251
Expected cases 14 days from yesterday (April 5): 2,209,527

Total US deaths: 388
Increase in deaths since yesterday: 86
Percent increase in deaths since yesterday: 28%
Expected US deaths over course of pandemic: 1,006 (assumes no new cases starting today, and case mortality rate of 3%. Based on yesterday’s figures)
Expected US deaths 7 days from yesterday (March 29): 8,168 (same number as above, but calculated 7 days ahead)
Expected US deaths 14 days from yesterday (April 5): 66,286 (ditto, but 14 days ahead)
(This means that, on April 5, there will be enough confirmed cases, that we can safely say – based on a 3% case mortality rate – that more people will die in the US in the pandemic than military personnel who died of any cause during the Vietnam War. Sound bad? I would that were the worst of it! Read on, if you dare)

As you can see from the above, I’ve expanded the numbers I’m tracking. Here’s what I’ve done:

  1. In the spreadsheet I first created yesterday (which is expanding exponentially, faster than the virus itself) I’ve expanded the data on total US cases (using the Worldometers site, which seems to share data with Johns Hopkins and vice versa. The CDC site has almost exactly the same number, except it’s reported as today’s number, while Worldometers has jumped another 5,000. But I’m going to solve these problems by not even considering any numbers for the day I write the post) back to March 2. I chose that date for two reasons: a) It’s three weeks before today, and b) It was my grandson’s first birthday. I believe it was mainly b) that led me to this decision.
  2. I got burned with yesterday’s post, when I quoted the site’s number of 38,000 cases for the day, which appeared in the afternoon, whereas today it was down to 35,000. So I’m not going to even attempt to say what one day’s number is until the next day. I also found that, after first entering all the daily case numbers, while I was writing this post, every one of those numbers changed twice (first they all went down by a small amount and then back up to something close to the old amount, but not exactly what it had been). The site apologized that they were adjusting data, and I don’t blame them – there’s confusion everywhere in the US.
  3. I’m now tracking 7-day percent changes in cases, not just day-to-day ones.
  4. I’m using those 7-day changes to project out total cases 7 and 14 days ahead. I do this by taking the 7-day percent change I just calculated and applying it to the next seven days, then growing the seven-day figure by the same percentage, to arrive at a 14-day-ahead projection.
  5. How accurate are these projections? I now have enough data to track what my 7-day projections would have been, had I made them a week ago. For the first few days, the projections were very close to the actual numbers. Then they became consistently lower, so that my “projection” for yesterday’s number from 7 days prior was not much more than two thirds of the actual number. The reason for this is clear from the 7-day actual growth rates I’ve calculated. They have pretty consistently gone up. They start on March 9 (14 days ago) at 604%, and yesterday that number is the one I reported above: 812%. So my projections are probably underestimates, for this reason alone (and there are other reasons. Keep reading).
  6. Keep in mind, in all of the case numbers, historical or projected, they’re based on the reported test results. As of last week, epidemiologists were saying that the reported numbers were probably around a tenth of the actual numbers, meaning every number up above (cases and deaths) should be multiplied by say 5 at least. However, as testing very sloooowly rolls out, presumably some of these previously unidentified cases are being reported – although that statement needs to be heavily tempered by the fact that all the unidentified infected people have been walking around infecting others, so that the multiple might now be above ten, not below it.
  7. As I started to do yesterday, I’m reporting day-to-day reported deaths, and the percent change from day to day. I’m not going to report 7-day changes in deaths, since I don’t think that’s very useful.
  8. As I did yesterday, I’m projecting total expected deaths over the course of the pandemic, based on yesterday’s case number. The formula for that is simple: the case number times the estimated case mortality rate. I’m going to keep using three percent, but if you want to use a lower or higher number, just multiply my number by the appropriate ratio (e.g. if you want to see a 2% rate, multiply my number by 2/3). Given that the main determinant of the case mortality rate is the state of readiness of the hospitals and especially ventilators, and that we are blessed with a president who is still waiting to see if things become bad enough to require the government to mandate production of ventilators, 3% may prove to be an underestimate, not an overestimate (in Spain and Italy, they’re now triaging patients for treatment. It’s very possible that in New York City next week that may also be necessary).
  9. As previously when I’ve estimated expected deaths, I’ve assumed that a) massive social distancing – i.e. lockdown - will be put in place in every place in the US where it’s needed – and this will be done today (it’s only 11:28 Central time, but I somewhat doubt I’ll read that Trump ordered this later today – so obviously this isn’t a likely assumption at all); and b) the social distancing will be 100% successful, so that every case in the US will be identified and quarantined (of course, this requires freely available tests, which itself is a very questionable assumption).
  10. But the two most important numbers for me are the projections of total deaths 7 and 14 days out. Remember, these are estimated deaths over the course of the epidemic (with the same restrictive assumptions just discussed), computed by multiplying the 3% case mortality rate times the projected 7- and 14-day total case numbers listed above. The same thing I said about projected cases applies to projected deaths: These are more likely under- rather than over-estimates.
So what conclusions can I draw from all of these numbers? You’ll notice that both projected cases and projected total US pandemic deaths grow roughly 8-fold over the next seven days and another 8-fold after that (not surprising, since they’re calculated using the same growth rate of 812% per week). What happens if we project them out into the future? Keeping in mind that all sorts of things might happen as we get farther out.

To do this, I first found the mean of all of the 7-day reported case growth rates that I can calculate now (13 of them – the mean was a 650% 7-day growth rate). Then I took the projected case number for April 5 that I projected above and grew that by 650% for every seven days, and multipled all of the case numbers by 3% to get total estimated deaths (again, deaths over the course of the epidemic, based on the assumption that a total lockdown was imposed the day the estimate was made and it was 100% successful).

What I got was really, really scary. Even though the total case numbers – starting April 6 - initially are close to what I projected for April 5 above, they grow exponentially because…well, that’s what they’ve been doing so far. By April 3, we cross the line into more than a million expected deaths for the pandemic (in the US, of course). By April 24, close to the entire US population is infected, and there are 9.3 million deaths guaranteed for the pandemic (and it would really be much more than that, because the mortality rate would soar. The UK Imperial College report, used by the White House to justify the measures announced last Monday, estimated the US hospital system might be overwhelmed by eight times. This sounds like the scenario I’m talking about).

Not being an epidemiologist, I imagine there are a lot of factors that would limit this growth, including that a lot of people would be recovered at that point, and would presumably not get infected again. This is probably why the same report that estimated an 8X overwhelming of hospitals also estimated total US deaths at “only” around 2.2 million; I’m certainly not going to argue against this. I guess what I take from this so far – and of course, I’ll be revising all the numbers every day, based on the latest number of cases – is that the 2.2 million number for deaths is certainly nothing to be dismissed out of hand.

So what does this all mean? Exponential growth is unimaginably powerful. The scariest part is that, even though you might put in place massive social distancing for say two or three weeks, if that’s not almost 100% successful in stopping the spread, as soon as you relax that the growth quickly ramps right back up. You end up with the same total numbers for the infection, but just delayed by a few weeks (and if you don’t believe me, just try this yourself in a spreadsheet, using almost any number that grows by 650% a week).

In the last few days, both liberals and conservatives have been saying we need to carefully weigh the cost of going with all-out distancing vs. a much less intrusive approach that would be less burdensome on people. Great, we’ll do all of this, and we’ll successfully delay the 2.2 million deaths by at least a month! In other words, we’re giving 2.2 million people another month to live. Now, there’s a job well done…

We are in big, big, BIG trouble. At this point, for every day we delay doing the right thing, literally thousands of lives will be lost.


2 comments:

  1. Tom
    the actual data does not support these worst case scenarios. Asia is in steep decline already as is Washington State. Epidemiologists do not support forecasting tens of thousands of deaths in the US. Treatments are seeing success in trials in multiple cities as of a few days ago. I hope and believe your forecasts are based on extreme assumptions

    ReplyDelete
  2. I wish that were the case, however it's just the opposite. For example, my projections of total deaths, whether based on today's cases, or total cases 7 days ahead or 14 days ahead, rely simply on the observed 7 day growth rate in cases, updated every day, and they assume that a total lockdown will be implemented that day (i.e. today, 7 days forward or 14 days forward).

    Today that 7-day rate is 830% - that's what's driving these astronomical increases. Unfortunately, it doesn't matter what you make the starting point. If we had 10 cases today (which is about what we had 3 weeks ago), it would still reach the same levels I'm talking about, just 3 weeks later.

    My post will be on this subject today.

    ReplyDelete