Models for Covid-19

Paul · May 7th, 2020, 9:18 pm

Logistic function better. Or an asymmetrical version of it.

Paul · May 8th, 2020, 1:00 am

I think most people who died from Covid-19 on the Diamond Princess were over 70. Over 70s made up about 30% of the passengers. Now assume all passengers had the virus (although only about 20% actually tested positive). You get a death rate of 0.35% of infected.

Now take Singapore. They have had gazillions of cases but very few deaths. Let's assume that they are doing a much better job of testing than anyone else, and getting numbers about right. Their numbers give a death rate of infected of 0.095%. But only about 8% of their population is over 70. (I think.)

Both of these calculations give a death rate of over 70s infected with Covid-19 of 1.2%.

(The number is about 1.8% for NYC.)

Many, many assumptions in the above.

Anyway, leads to this number infected:

Belgium 38%
Spain 29%
UK 24%
Italy 23%
Ireland 22%
Netherlands 22%
France 21%
Sweden 17%
USA 15%
Switzerland 11%
Luxembourg 10%

Alan · May 8th, 2020, 2:09 pm

Have you seen any good statistics on the death rate for the flu for over 70s? I was looking for US values, but apparently the CDC doesn't even track flu deaths for those over 65.

Cuchulainn · May 8th, 2020, 2:13 pm

A software critique of the Ferguson code

https://lockdownsceptics.org/code-review-of-fergusons-model/

It misses the real issues and problems. Too low-level and grungy. The author is a programmer (nothing wrong with that) but is not 1) domain expert, 2) project manager, 3) designer, 4) mathematician.
The real problems are cultural and organisational.

On a personal level, I’d go further and suggest that all academic epidemiology be defunded. This sort of work is best done by the insurance sector. Insurers employ modellers and data scientists, but also employ managers whose job is to decide whether a model is accurate enough for real world usage and professional software engineers to ensure model software is properly tested, understandable and so on. Academic efforts don’t have these people, and the results speak for themselves.

Ah yes, the SANTY CLAUSE!!

Cuchulainn · May 8th, 2020, 2:19 pm

Here's a nice modelling study (Univ. Washington) cited in my local paper (LA Times):
Forecasting COVID-19 impact on hospital bed-days, ICU-days, ventilator days and deaths by US state in the next 4 months

No differential equations, just fitting a sigmoidal-type (actually a CumNormal) to the cumulative death rate, particularized to the local details of each US state. A 4-page Appendix gives the details, and there are online visualizations of all the projections. I like the wide error bands! It all sounds rather plausible.

Here's a link to the state-by-state projections and one for the country as a whole. Total projected US deaths from COVID-19: (38,000 - 162,000), with new daily deaths peaking around Easter, and with the (current) episode essentially over by June 1.

California projections: 800 - 17,000 deaths, with new daily deaths peaking around the last week of April.

A point of comparison for the US total is the CDC estimate of 26,000 - 53,000 US deaths for seasonal flu (2018-2019 season). Preliminary estimates for the 2019-2020 season are similar. So, another way to frame the UW COVID-19 projection is that, under the current lockdown policies, we'll have a multiplier of 1.5x - 3x deaths on top of recent seasonal flu deaths.

Alan,
Are you still enthusiastic about this IHME. Some say it is completely useless.

Cuchulainn · May 8th, 2020, 2:23 pm

Logistic function better. Or an asymmetrical version of it.

Why? Better at what?
Nothing special about this function, meiner meinung nach.

Alan · May 8th, 2020, 2:59 pm

Alan,
Are you still enthusiastic about this IHME. Some say it is completely useless.

It is like predicting the stock market: it's crazy to focus on point estimates. So, their current US projection (May 4, 2020) is for 95,000 - 243,000 deaths by Aug 4. Their first projection (March 25, 2020) was: 38,000 - 162,000 by Aug 4. Those are big ranges (good for them), but there is significant overlap. I wouldn't call it useless. Does somebody else have better estimates? (Honest question).

Plus, I think they were pretty good in calling the (first?) peaks.

Their Mar 25 prediction that the US outbreak would be essentially over by June 1 was wrong. Now they predict a much longer tail.

When their mid-point US estimates doubled, that certainly surprised me. But, I looked and saw they made a major model revision. Probably after a lot of internal debate. I'm sure they knew they would take a lot of heat for that change, esp. the mid-point doubling. Good for them.

The media critics focus too much on the point estimates.

Now, with the model revision and new data, their upper bound to the uncertainty interval has risen significantly. Is that new bound -- unfortunately -- plausible?

My gut feel, just from following the situation like everybody else, is that the upper bound should rise significantly in the US. The reason is that I think the political authorities have come to realize that the economic pain is so great that these various "17 step" etc programs to reopen, as nice and scientific as they sound, are just too slow. So, my (two cents) prediction is that all the US states (including California), like it not, are going to essentially rip off the band-aids and become like Sweden by the end of May. This is a change from the IHME (lockdown-type) scenarios of their first predictions. As Keynes famously may or may not have said, "When the facts change, I change my mind ..."

So, weighing all of the above: yes,-- I am still a big fan.

Cuchulainn · May 8th, 2020, 6:05 pm

It's just what I don't understand is that these self-professed (by Trump and Johnson) world-class scientists can't get further than cubic splines in Excel.

Paul · May 8th, 2020, 7:43 pm

Logistic function better. Or an asymmetrical version of it.
Why? Better at what?
Nothing special about this function, meiner meinung nach.

The logistic function comes out of many math biol models (as well as other fields), including some epid. models. Or it's at least a decent approximation to what comes out of these models. So it's not unreasonable to fit that function. By "asymmetrical" I just mean that the basic logistic function doesn't capture the asymmetry in the growth and decay phrases of an epidemic. Hence needing a slightly more general function for fitting.

But taking an arbitrary function and fitting it seems a bit silly. It's usually nice to have some plausible excuse for a functional form!

I don't know what the "cubic" model does. But cubics do go off to plus/minus infinity, so I hope it's not modelling deaths!

Cuchulainn · May 9th, 2020, 2:11 pm

Alright then, I downloaded the revamped Imperial Covid C code, built the project and got it running in console mode (it uses GDI which I haven't used since 1992 ... it is deprecated big time and is impossible to find GDI drivers),

I will write up a precise and detailed report in the coming days (BTW I have written code on projects large and small since 1972).

It is going to be a very rough ride; fasten seatbelts. BTW the Microsoft/Github code is still C running under Visual Studio.

Even this new ported code is not fit for purpose
Even this new ported code is not fit for purpose

Watch this space.

Reminds me of the quote from Local Hero

“My parents were Hungarian immigrants. They took the name MacIntyre 'cause they thought it sounded more American!”

Cuchulainn · May 9th, 2020, 2:57 pm

I reckon kats and I could write a correct version together before lunchtime.

What is needed is a coupling to SqlServer database (you know, traceability and audit trail for various stakeholders).

From what I see, there are many headless chickens running around.

Cuchulainn · May 10th, 2020, 12:44 pm

Alan,
Are you still enthusiastic about this IHME. Some say it is completely useless.

So, weighing all of the above: yes,-- I am still a big fan.

I think a big problem is that there is no one(?) in Congress with STEM degree. So, being comfortable with numbers is non-existent.
Same with leaders who studied Greek and the Classics.

So, how do they link cause and effect?

The consequences are dire: it means software projects will fail (what's new) because of politics, lack of right people on the team (from requirements analyst and beyond) and lack of expertise on all fronts.

A counterexample: I was chief architect of the largest social services projects in NL in1990. We had a very small core team and we were quarantined from external noise. The project was a success mainly because of this fact. We fired our first project leader and replaced him by one who wanted to get the job done, and he lived up to his promise.
These days, government projects are horror stories.

Cuchulainn · May 10th, 2020, 1:12 pm

Anyone know what spatial gravity models are in the current context?
(they are in the code to model people belonging to places).

https://www.researchgate.net/publicatio ... _Structure

//Sounds plausible, I suppose (I am not a epidemiologist, I can barely spell it).

katastrofa · May 10th, 2020, 2:57 pm

The same idea as in economy, but here applied to the flows of people between places. In Ferguson's model the probability of transmitting the virus between two individuals is generated by kernel functions, whose values decrease with the distance between those individuals. I'm not sure why they call it a gravity model, but ok.

bearish · May 10th, 2020, 4:02 pm

Alan,
Are you still enthusiastic about this IHME. Some say it is completely useless.

So, weighing all of the above: yes,-- I am still a big fan.
I think a big problem is that there is no one(?) in Congress with STEM degree. So, being comfortable with numbers is non-existent.
Same with leaders who studied Greek and the Classics.

So, how do they link cause and effect?

The consequences are dire: it means software projects will fail (what's new) because of politics, lack of right people on the team (from requirements analyst and beyond) and lack of expertise on all fronts.

A counterexample: I was chief architect of the largest social services projects in NL in1990. We had a very small core team and we were quarantined from external noise. The project was a success mainly because of this fact. We fired our first project leader and replaced him by one who wanted to get the job done, and he lived up to his promise.
These days, government projects are horror stories.

Well, not exactly. The STEM people make up around 10% of Congress. Still not a lot.

https://www.npr.org/2018/11/18/66794400 ... w-congress

Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19

Re: Models for Covid-19