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.
COVID-19 Projection Charts Oversold Pandemic
Better late than never. I’ve been tracking the pandemic in the US for a couple of weeks, and I took publically available data and started charting its progress. As early as 10 days ago the data already showed it’s likely peak and then decline. Using data provided by Worldometer and the University of Washington’s Institute of Health Metrics and Evaluation, and a spreadsheet, the “overselling” of Armageddon and needless promotion of an exponential death curve “popped” from the graphs. And now that the alarmists have started to calm, more than a few are looking foolish.
So here is the good news, in graphs:
Figure 1 – Running Total of Confirmed Cases of COVID-19
The first graph charts the total confirmed cases, by date, as provided by the Worldometer. The aforementioned IHME model does not publish or forecast the number of confirmed cases. The key takeaway is that the actual trends, and best fit projected trends, are not best represented as a fixed exponential curve (note the light purple curve). The trends are “best fit” (see the R2 score) with a polynomial trend line(s). For those lurid exponential growth diehards who fret about a “sky-rocketing” pandemic, tell them to chill … that may have been true in the epidemic in its earliest stage but that’s not how S-curved epidemics work.
And while you are talking down your hysteric buddy with some free life counseling, also point out to him (or her) that while you can’t see it, on 4/5 started to bend to the right and will continue to do so until flat. How do we know? Read on.
Figure 2
The chart below shows what the realists waited for, the daily growth rate of new infections falling to “1” and below. This is a common epidemiological measurement. When today’s count of “new COVID cases” is divided by yesterday’s count of “new COVID cases,” it provides a growth rate (or a ratio). If today there is 1,100 new cases, and yesterday were 1,000 new cases then the “growth rate” is 1,100/1,000 = 1.1. If growth is declining, however, if today there are only 1,000 new cases and yesterday was 1,100 new cases, then the rate is 1000/1,100 = .91. So on April 5, the red line (a moving average of the rate of growth/decline of new COVID cases) crossed the inflection point of “1,” meaning that the daily growth of new cases is lower and lower (declining).
Better yet, on April 8 or 9, it will cross again and likely remain below that inflection point (I don’t pay attention much to the blue line because, for mathematical reasons, the total case rate can never drop below 1). The trend lines on the above graph will start to “bend over” and eventually flatten out.
Figure 3
This chart tracks the daily changing of cumulative US death totals from COVID-19 as reported by Worldometer. The IHME does provide long-term forecasts of the anticipated total fatalities by date. The solid red and green trend lines are based on Worldometer’s actual trends of cumulative deaths.
What does the IHME and the trend-lines forecast? It gets more complicated. There have been three IHME forecasts in the last week; what I call ‘the longer standing initial forecast’ (IF) followed by the ‘ forecast Rev 2′ (4/5) (FR2) and forecast Rev 3’ (4/7) (FR3).
The IHME initial forecast (IF) projected 35,000 total deaths by 4/16. The forecast Rev 2 (FR2) of 4/5 projected 40,000 deaths by the same date. The forecast Rev 3 (FR3) projects 30,000 deaths at the peak, 4/12. Each of these IHME forecasts is represented by dashed red, blue, and green forecast lines respectively (see figure below). All fall within my own Worst and Best case trend lines. While in my view FR3 is the most likely forecast, you can see that the trend lines don’t exactly support any of them. They will eventually, and most likely it will be where the green trendline now crosses on 4/16. (The importance of 4/12 and 4/16 is explained in the section on Figure 4, below).
Hence on April 27, the fatality totals will be 50,000, after which the total will only climb to 60,000. And how do we know it will bend to the right, showing slowing the rise in fatalities? See Figure 4.
Figure 4
This chart provides the moving average of daily growth rates of new fatalities. Although choppy, as with new COVID cases the rate has been declining. At one time it was estimated to drop below 1 on 4/16, the projected peak of daily deaths. Now the IHME says it deaths will peak on 4/12 … three days from this writing. After 4/12 the number of new deaths will be in decline (below “1”). This is good news, the earlier the better.
Figure 5
You don’t need this chart to see the good news. But if you want to understand the trends on a more detailed graphic level, here it is. There are three IHME forecasts after 4/5, represented by three IHME lines. The most important is the green line with little green triangles. This is the latest forecast, with the earliest peak date (4/12) and the lowest number peak deaths.
Summary
The COVID pandemic in the US will soon be in decline. After 4/12 daily death report counts will decrease. Far fewer are projected to die than first projected, about 60,000. By May 15, the daily COVID death counts will be very small.
But then, some already knew that.
Published in Healthcare
I wish I understood better what you went to such great lengths to explain, so all I can do is say thanks, and I hope and pray you are correct.
Thanks for not calling your curve-fitting a “model” like the IHME folks do. Do your curve fits assume current social behaviors are the same? If so, that is a bad assumption. If not, the curve fit is not sufficient.
President Trump’s Easter miracle. :-)
Actually my curve fitting is not a model, its merely a method using trend lines to forecast where the data is headed. However I have found that the IHME folks forecasts are pretty accurate (if you leave out their hospitalization, ICU, and medical usage predictions) and the trend lines generally echo their take.
So yes, I do assume that the future won’t create a different set of behaviors. If, for example, the lockdown ended today, then none of this is likely.
But…that’s what IHME is mostly predicting: hospitalizations, ICU, medical use…
In New Hampshire (where I live), the IHME model was predicting 337 total death before August 4. But then over the weekend they dropped that down to 31 deaths (90%+ difference). Then yesterday they “doubled” the number to 66. They’re all over the place. They have literally now idea what is going on. The IHME model is, in my opinion, complete garbage.
Mark, good and interesting post. I have a few questions.
(1) In your confirmed cases graph (Fig. 1): how do the projections play out if you extend them well into the future? If I’m understanding you correctly, you used a second and a third degree polynomial, and got a very strong fit with both. But the coefficient on the x^2 term in the 2nd degree polynomial model would have to be positive in order give the shape that you show (I think), which would imply endless growth. Is the coefficient on x^3 term in the 3rd degree polynomial model negative?
I don’t think that we can use a polynomial model for this projection, in the long term, because any such model will eventually head toward infinity (positive or negative).
I think that the key issue in charting an S-curve is predicting the inflection point — which leads to your other graphs.
(2) Your Fig. 2 label says “moving averages” (this is one of the graphs of cases). Are the dark blue and orange trend lines moving averages of the growth ratios, or are they the daily figures? If they are moving averages, how many days.
(3) Your Fig. 4 label also says “moving average” (this is one of the graphs of deaths). How many days are averaged?
(4) I like your Fig. 5, through I don’t trust the IHME models. The real trick is figuring out the location of the “hump” in the daily deaths curve. Your Fig. 4 suggests April 16, so I’m not sure why you go with the latest IHME forecast (which predicts an earlier peak on Apr. 12) rather than one of the other two IHME forecasts (which both predict a peak on Apr. 16, per your graph). If the peak is a bit later, the death toll will be a bit higher (probably, though this depends on the sharpness of the peak).
I’m not sure if you saw it, but I posted my own projection on Monday, Apr. 6 (here), for what it’s worth. My methodology was pretty similar to yours, though I focused on rates of growth in total reported deaths, rather than the change in the ratio of reported deaths each day.
In my projection, the peak of daily deaths is April 16 — which is exactly the same day shown by your Fig. 4 (if I’m reading it correctly).
The IHME model is fluctuating pretty widely in the details, as it appears to do a separate calculation for each state, and then totals the individual state estimates. So I have less confidence in the IHME model than you appear to, though just about any well-calibrated S-curve is going to fit the data that we have so far.
The real tricky question is whether we’ll actually hit that hoped-for peak next week.
Mark, thanks for the response. My questions are answered, and I follow your calculations.
Mark, Nice post.
From Zerohedge a comparison of forecasts for use of hospital beds for COVID-19 in New York:
• Gates funded IMHE: 73,000, peaking around 4-6
• McKinsey Moderate growth scenario : 55,000 peaking around 4-15
• McKinsey Severe growth Scenario: 110.000 and still growing in the foreseeable future
• Columbia University NYC only: 136,000 peaking around 5-1
Actual number of beds used at present: 18,279 and projected to peak at 27-28K around 4-15
c