Our Own Covid-19 Epidemiological Model

13
877

From our epidemiology correspondent:

Building a model is not easy.

In order to produce a really good quality model you would like all these factors to be “High Confidence”.

31 Disease Specific Factors Required to Build a High Confidence Covid 19 Model

Model Criteria Confidence

How many people have it

No symptoms Low

Minor Symptoms Low

Moderate Symptoms Low

Severe Symptoms Moderate

Hospitalisation High

Dying High

How Easily Does It Transmit

Surfaces Moderate

Open Air Low

Weather Effects Moderate

Indoors Moderate

Ventilation Effects Moderate

Personal Contact Moderate

Fluid Transfer High

Protecting Factors

Previous Illnesses Low

Vaccinations Low

How Good Are Barriers

Hand Washing High

Face Touching High

Other Hygene Low

PPE Masks Moderate

Non PPE face covers Low

Spectacles Low

Other PPE Moderate

How Serious Will It Be

Co Morbidities Moderate

Age Factors Moderate

Gender Factors Low

Race factors Low

Other Genetic Factors Low

Physical Traits Low

Viral Load Moderate

Recovery & Defence

Full Recovery Time Moderate

Antibody Lifecycle Low

Here’s a demonstration of the problem:

Just starting with “How Many People Have it” the only thing we know with any confidence is the number in hopsital. In England that is currently 16k (mid April, High confidence, counted).

We know that there are a number outside hospital with it in a serious way, or developing a serious case of it, and additionally in care homes. No idea how many though, maybe 2-5x that number (low confidence, not tracked). That would give you 32-80k current “serious” cases.

We also know that the “Serious Symptoms” to “Minor Symptoms” ration is about 15-30% (low confidence, Chinese data). that gives us 100k to 500k currently with some symptoms.

Now it gets really rough – some people possibly get it with no symptoms at all. Stanford did a study (not peer reviewed, low confidence) suggesting for every symptomatic patient they identified between 50 and 80 with no symptoms at all. Which means possibly 40m in England might currently have Covid, That is maybe, 40m current cases (very low confidence). There’s a tail in front of that of those that have already been through it and recovered.

So at the high end, it’s already blown through, infection has infected. Of the 80k probable current serious cases expect another 20k fatalities, back to work.

At the low end, it’s only in 100k people. with everyone else still to go, panic stations.

Our current range is “Somewhere between 100k and 40m”. We didn’t even get onto how it spreads….

So if someone says they modelled this accurately, with any degree of confidence, they are telling Socialist grade porkies.

Much more on this over at Hector’s place.

13
Leave a Reply

avatar
8 Comment threads
5 Thread replies
15 Followers
 
Most reacted comment
Hottest comment thread
10 Comment authors
NDReaderJohn BBlokeInNormandyFromTejasDavid BennPat Recent comment authors
  Subscribe  
newest oldest most voted
Notify of
Charles
Guest
Charles

While much of the analysis involves uncertainty, it’s not to the extent of the range between the high end and the low end you mention. In countries which have not lost control (China, South Korea, New Zealand, etc), contact-tracing is used to find people potentially infected by known cases and efforts are made to check where an infected person has acquired the infection. If there were a huge number of asymptomatic cases, the symptomatic cases would seem to appear at random. Instead the path of contagion is usually clear, so this limits the proportion of asymptomatic cases. Note also that… Read more »

Phoenix44
Guest
Phoenix44

Contact tracing is a bizarre myth that some governments have used to make themselves look good and ours is about to use because it can’t think of anything else. There is good evidence a large proportion of people are asymptomatic and so never get tested. There’s good evidence the sensitivity of the test might be as low as 70%. So let’s say 50% are never tested and 30% of the 50% who are tested are false negatives. The first 100 infections then go: 50 not spotted, 15 not traced as false negative. So already we are at only 35% success.… Read more »

Charles
Guest
Charles

Well then, consider the Diamond Princess. nearly 20% of the people on board tested positive, with about 400 asymptomatic. If there are 50 asyptomatic cases for each symptomatic one, there should have been only about 14 symptomatic cases. There is plenty data which all points in the same direction – the asymptomatic cases are only sightly more than the symptomatic ones.

Michael van der Riet
Guest
Michael van der Riet

Dear Charles, if you believe the numbers coming out of China, I have a bridge to sell you.

Phoenix44
Guest
Phoenix44

Say thirty assumptions, each with ten plausible ranges. So the possible combinations are huge (30 million?). So the chance Ferguson got it right is the same chance as him winning the Lottery tomorrow night. Zero. And the chances the infection will look different are essentially 100%. Yet nobody in government or advising government seems remotely interested in that simple fact.

Chester Draws
Guest
Chester Draws

It’s worse than that. Many of these factors work exponentially, so even if you got every single factor to within 10% of the true value, your results would still be utter shit.

The difference between a re-infection rate (R) of 1.9 and 2.1 (i.e. 10%) is incredibly significant.

See also climate models.

Michael van der Riet
Guest
Michael van der Riet

The problem with these epidemic models is the suggestion that analysis using aggregate functions will yield meaningful results. Radioactive decay in even a gram of an element is precisely predictable. Human behaviour is utterly unpredictable, which is why economic predictions always fail to see the next crash coming even when it’s right around the corner. People sicken and die when they feel like it, not because an atomic clock tells them to. “Excess deaths” is just another statistic in an illusory quest for certainty.

Esteban
Guest
Esteban

A highly complex model that is extremely dependent on assumptions (as all models are) is based on a few critical ones (infection and lethality rate) we don’t understand. Yeah, let’s re-arrange the world based on that. The more I read about the Kung Flu, the more I’m struck by how confusing the information is. Some people report data that includes China, who we know lies about the virus. Our lethality info is dependent on understanding how many people have been infected, which is bollocks on stilts. And everyone is keeping score differently. In the U.S. if you die and there’s… Read more »

David Benn
Guest
David Benn

“…In the U.S. if you die and there’s a chance you were exposed, you’re scored as a virus fatality…”

I read an article at the weekend which suggested that US hospitals could claim from Medicare/Medicaid at the rate of $5000 per flu patient, $13000 per ‘standard’ Covid-19 patient, and $39000 per Covid-19 patient treated in an intensive care unit.

If that is true (or even approximately true), then the US statistics do not come as a total surprise…..

Pat
Guest
Pat

It seems obvious to me that we’re far too far down the line to use track and trace. Lets take a low end estimate of 100k infected. Someone has to find all of them, and then find all the people each of them met in the last fortnight, and then check their contacts. It’s vastly too many to deal with. And if the number of infected people is higher it gets worse. Track and trace can only work at the beginning of an outbreak when there are few cases to follow up. And even then you will have to keep… Read more »

BlokeInNormandyFromTejas
Guest
BlokeInNormandyFromTejas

I think the article is less about constructing “a model” and more about “constructing the picture of where we are”.

A model will allow you to see the varying outputs as you vary the assumptions.

If you knew the values of the key parameters you could get an output, and by wriggling the values of the key parameters (to account for measurement errors) you can see the volume of uncertainty arising.

Then you wriggle all the other parameters and you end up almost certainly discovering that the space of likely outcomes is so enormous as to be useless…

John B
Guest
John B

The number affected by the disease (as opposed to infected by the virus) serious enough to warrant medical attention ranging from little to intensive, is less than 0,5% of any population and deaths are around 0,02% of any population. You don’t need a computer model, just Darwin and look at the evidence. Then with pen, paper, pocket calculator, and the published figures of reported cases, deaths and population sizes work out the percentages. In other words the insistance that Covid-19 was/is a dangerous virus to the population at large was/is false, and was never was supported by the data. Certainly… Read more »

NDReader
Guest
NDReader

“In fact the virus is a classic example of Darwin’s Theory. Genetic mutation gave it characteristics to be best adapted to its environment to endure most successful reproduction rate.”
It’s probably nowhere near that yet. Viruses make the step from animals to humans quite often, I expect, but almost always have a reproduction rate less than 1. Maybe variants of this one have made the leap before, and failed to stick. This time, Covid-19 has a reproduction rate >1 in human societies. Lucky virus. But it isn’t fine tuned for transmission yet.