Another curve fitting exercise

[The_jackalope]

I just graphed the global COVID-19 case data from https://www.worldometers.info/coronavirus/ from Feb 19 to today in Excel.

I fitted two trendlines to the data, second-order and third-order polynomials, both projected out one month. The fit for the second-order trendline is 0.9909, and the projection indicates half a million cases worldwide in a month. The third-order trendline has a fit of 0.9992, and the projection for April 15 is 1 million cases.

My expectation is that the final number will be between the two - closer to the smaller number if social distancing is relatively successful around the world, but closer to the upper number if it is not.

The CFR for resolved cases (recovered+died) is currently around 7%, but has climbed from 6.3% over the last two or three days. The percentage of recovered cases has dropped from 53% to 49% in the same time, meaning that the spread of active cases is accelerating.

No warrantee express or implied, only one per customer, offer void where prohibited, yadda yadda.

[Steelrolled]

1. How does this compare to previous pandemics?

What does it tell us about the future and when we turn the corner?

[The_jackalope]

8. I posted some comparisons below

But they don't tell us much if anything about how it will unfold. All we know is that it will depend on the severity of the measures we take.

My gut feeling (which is probably as good as Trump's) is that it will take a couple of years to burn out world-wide - like the Spanish flu in 1918-1920.

[uponit7771]

2. Thx, that CFR seems very high no? tia

[The_jackalope]

3. It's higher than seasonal flu, but lower than H1N1

Seasonal flu seems to be on the order of 0.1%, so this is 70 times higher.
The CFR for H1N1 was 20-40%.
the CFR for Ebola is about 40%.

https://about.futurelearn.com/blog/covid-19-how-does-coronavirus-compare-to-other-outbreaks

It's worth remembering that theis way of calculating CFR produces what amounts to a worst-case number. The naive technique of using total cases instead of resolved cases as the denominator produces a number that tends to be too low, especially in an ongoing epidedmic. That number is currently 3.7%.

The final CFR when all is said and done will probably be somewhere between 3.7% and 7%.

My bet is that it will settle around 6% when all cases have resolved.

[uponit7771]

4. Got it, my understanding is the infection rate higher than h1n1? Seems because of asymptomatic

... people walking around this spreads faster.

We need testing quick fast, we're so behind !!

[The_jackalope]

5. The R0 for H1N1 was 1.6 to 1.8

We don't know the value for this coronavirus for sure, but estimates are 2 to 3. So yes, it's more more infectious than H1N1, by maybe 2x. It's also 2-3x more transmissible than seasonal flu.

https://www.ncbi.nlm.nih.gov/pubmed/19545404

[uponit7771]

6. Shit (Aretha Franklin gif)

[Ms. Toad]

7. My trendline agrees pretty closely with yours.

It's slighly faster than exponential - I tried an exponential cuve first, but the last couple of days are way short.

Mine hits a half million on between Apr 15 and Apr 16.

What throws everything off, though, is that many of the new cases aren't new - they are just newly recognized because we now have (some) testing capabilities. So some of the growth over the last few days is paper growth only.

BUT - for the same reason, the infected population (the basis for exponential growth) is also much larger (so we're already closer to half a million than we think)

[The_jackalope]

9. I agree that we're already closer to 500,000 at this point.

I think my third-order million-case projection may be closer to reality, unless some major players clamp down like Italy has. But I'm a catastrophist at heart, so I always try to damp down my enthusiasm for eschaton when talking in public fora. So somewhere in between half and one million by April 15-16 is my bet.

[Steelrolled]

10. Interesting stuff. But of course any smooth function can be approximated by a polynomial.

The higher the order the better the fit. Taylor's theorem rules!

Eventually we will turn the corner, and a more complex model will be needed to fit the data. This is why I was curious how previous pandemics behave.

[The_jackalope]

11. Yes, we don't know how, when or why the curve will inflect

We just know that it will. The Chinese example demonstrates that possibility. I do these exercises to get a feeling for the progress of the epidemic now and over the near term.

Also, this is a global aggregation that says little to nothing about local progressions or risks.