09 April 2020

In advance of the data

So there have been a couple possible revisions to R₀  for COVID-19 (it could be almost six), and various people talking about "not until a vaccine", and other people going on about how the lethality rate is less, and so on.

The short answer is that, as yet, we don't know a whole lot.  We certainly don't know any of those things with confidence.

The reason R₀ matters is that the degree of immunity you need in the population to extirpate the disease is greater the greater the value for R₀ (or, really, R-effective); the formula is 1 - 1/R₀, so if R₀ is 2 (the flu), you get a lot of benefit from 50% immunity.  If R₀ is 12 (low end of measles estimates), it's  92%.  If we're looking at a COVID-19 R₀ of 6 (ish), that means, absent other efforts to reduce the effective value, you need 83% of the population to be immune.

Only... we don't know any of that.  Everything we've got so far is a rough estimate.

To know R₀ with confidence, you have to antibody test the whole population.  (with the antibody test we have not yet got!) You have to go back and test the tissue samples that weren't kept from folks who died of pneumonia in March. You have to do a lot of work there simply hasn't been time to do.  (We do not, for example, know how much immune response amounts to functional immunity, or how long it lasts.  We can't know that; it takes time to find that out.)  Once you know R₀ with confidence, you know how effective the vaccine has to be; if R₀ is as  high as the folks thinking there's been an enormous number of asymptomatic cases implicitly think it is, any vaccine has to be nearly perfect to be effective.  ("can we create a vaccine at all?" doesn't have an empirical answer yet, either.)

Then you have to look at a few things other than "got well"; what happens when you catch it again?  Is the immune system panic reaction worse?  Has your immune function been generally compromised, as it will be with the flu or measles?   Do mild cases confer immunity? Do any cases confer immunity?

If you want to plan, you need to be able to answer all of those questions.  Nobody can answer those questions today; a lot of work needs to be done first.

So, really; isolate enough, and long enough, that the rate at which people are dying is decreasing.   We can measure that.  Make sure there's enough testing to tell what the rate of people dying really is, that the measure we're taking is accurate.  Support efforts to get enough information to plan.  Don't go trying to plan when we don't know anything.

If you have to plan -- if you just can't cope with not planning -- the only thing to plan for right now is the Uncommon Cold; in an environment with no transmission barriers, you get it every year, there's no vaccine, there's no direct treatment, and it's got at least a one percent chance of killing you every time you catch it.

The transmissions barriers are not only important, they're long term.  Minimum time to a vaccine is something like a year.  (There's no way to test a vaccine that doesn't involve waiting to see what happens over time to the immune response of the test subjects.)  Probable time is much more like "at least two years".  This just isn't a short term problem.  Getting through it absolutely involves making large changes to social structures and institutions.


Peter T said...

The true impact of this will not be known until we have total death numbers to compare against background. These don't seem to be collated or reported as they happen, so it might be a while.

Thomas Castiglione said...

It's an interesting demonstration of the fog of war. We're living through history and we know less about it than historians will, not from any incuriosity.

Daniel O'Neil said...

So let's assume a WILDLY hypothetical situation where a giant country with a completely incmpetent central government fails to supply adequate testing and creates no federal guidelines for surveillance or quarantine.

Assume the worst case of both infection rate and lethality.

Beyond the worst case scenario, which is cycling from regional outbreak to lockdown and back again for the next twelve months, is there a way that we "low tech" our way to a better solution?

I'd really like help mapping this out, because in about five weeks it's going to be much more than a thought experiment.

Graydon said...

@Peter T -- well, that's the death rate, and yeah, we can get a decent approximation of it that way.

How many people die in the first wave of a greenfield pandemic tells you some stuff, but not (alas) anything that important; what you really really want to know is degree and duration of immunity in survivors, and the old-fashioned way to get that involves waiting through the subsequent waves.

Graydon said...

+Daniel O'Neil --

Let's consider three different childhood diseases.
Measles -- if you get it, and live, you never get it again
Chickenpox -- if you get it, and live, it may come back in middle age as something different and worse
Flu -- if you get it, and live, you can get it again; there isn't a whole lot of difference between "has had it" and "has not had it" in terms of the likelihood of catching it again

COVID-19 could be in any of those categories. If a nation makes public plans on "oh, it's like Measles" (it might be horrid (greenfield measles empties cities) but then it is over) you might decide (like the UK government of Boris Johnson) to let the disease run its course. Lots of people die; the health care system may collapse. And if it's like flu, not measles, it comes back until with the passing of several generations the selection pressure of the disease produces a population in which the disease is less likely to cause death. (Which might-maybe be exactly what happened with the Common Cold, a long time since.)

Worse-case scenario involves collapse of supply chains; that'll kill a lot more people than the disease will directly.

And, again, "infection and lethality" aren't really the concern; it's "do the recovered stay immune?" and "do the recovered gain sensitivity to the disease?" If the former isn't so, it's not "twelve months on lockdown", it's "social distancing is permanent (until effective vaccines can be developed)"; if the later IS so, it means that the second wave of the disease is going to be much more lethal and instead of wobbling (as they now are) various global supply chains will get disrupted long enough to be unrecoverable. Everybody's economy loses capability, with the general consequences for everybody who doesn't have a lot of personal surplus. (That's at least 90% of the population.)

So aside from a desire to mutter about how vaccines are low tech (vaccines for coronaviruses are not low tech), there isn't anything to do but strong social distancing. If we could figure out how to do it universally for two months (the longest anyone's recorded from exposure to symptoms is about five weeks), we could plausibly extripate it, but as a practical matter that can't be done.

Daniel O'Neil said...

I always come here for the darkest timeline and you always deliver. Thanks!

Zeborah said...

Can people get *the same variant* of flu a second time? Or just other variants?

Latest I saw suggested that the variants of Covid19 out there aren't that diverse so probably-we-think-so-far-maybe would be treated by the immune system as the same thing and/or a hypothetical vaccine could cover them all-in-one. [Other researchers probably have different opinions on this.]

Of course if Certain Countries let the disease just run rampant (I speak of countries that would have the means to do otherwise if they got their A into G; other countries sadly do not even have the means) then we'll get more variants popping up, increasing the risk that this will no longer be the case, so, great, even if New Zealand does manage to eliminate it internally [plausible if we keep up current efforts!] we may have to quarantine all incoming visitors for two weeks forever, and still risk occasional asymptomatic cases slipping through from time to time.

Peter T said...

On the death rate, comment on another blog sent me poking around in the US CDC stats database. US deaths from pneumonia and flu usually run between 1300 and 1600 weekly. From the first week of February this year they run over 4000 weekly. This tells us first that estimates in the absence of testing are almost certainly severely under-count mortality and second that the disease reached the US in early January - probably at about the same time that the Chinese identified it (they sequenced the genome 7 January).

Graydon said...

+Zeborah That's a tough question about flu because the half-life of the variants is plausibly less than your immune time once you fight the thing off. My understanding is that yes, you can; flu does bad things to the immune system (which is why flu vaccination is an economic net win; it wouldn't be if all it suppressed was flu. But like measles, flu resets your immune system so statistical-you is then considerably sicker than you might otherwise have been.)

SARS-CoV-2 (the virus that causes COVID-19) doesn't mutate much, true, but the comparable disease is "the common cold" (some of the viruses causing the common cold are coronaviruses) rather than the flu. There we see no long term immunity and previous attempts to produce a vaccine (back before gene sequencing, with much less economic incentive) died in a pit.

It's possible testing could replace quarantine, but not until there's a large enough set of baseline data to tell how good the tests are. I expect New Zealand is going to be quarantining new arrivals for the foreseeable and I expect the quarantine time to be sixty days after the first "asymptomatic for a month" case (several already recorded in Wuhan) slips through.

Graydon said...

+Peter T --
One month, 30 days, is about 10 doublings; one person to a thousand. To be seeing significant general upticks in the national death rate from respiratory disease, it'd have to have started spreading in the US in November and gone general by February.

That kind of timeline has it spreading from the US to Wuhan, assuming that the medical system in Wuhan caught it about when it happened. (Which despite the initial local official response it looks like they did.) There were a few reports on the net of a first case there being pushed back to November, but it's unlikely they had widespread transmission earlier than their official timeline, and it's still very much an open question how transmission from bats to humans might have happened. Horseshoe bats are not considered a commestible and weren't present in the market, and that's the identified animal reservoir.

Which runs into, as a thing, how fast UW Virology figured out what was going on, but then again that was significantly because of the news out of China and that they were looking. Plus how the genes match up, though that doesn't give chronology and it's a political minefield so I'm not expecting to see that in the literature this year or next.

It also brings up the question of comorbidity. This year's flu is unusually bad; about twice as bad. And flu can be, and usually is, diagnosed by reasonably accurate testing. So there is an uncomfortable "is it this year's flu really that bad, or is it co-morbid?" question that (so far as I know) hasn't been answered yet.

Peter T said...

It could well have reached the US in late November. In any event, the world is now so connected that by the time it was firmly identified it was well and truly global.

There does seem to be significant co-morbidity (see UK death rate: https://www.theguardian.com/world/2020/apr/14/coronavirus-pushes-england-death-rate-to-highest-level-ever). Just how much will come out in time.

johne said...

When did COVID-19 begin?
"More on that Alleged Intel Report, and When Did COVID-19 Really Begin?"

Comorbidity (in the widest sense):
"How Many People Have Died In NYC During The COVID Pandemic?"

Graydon said...

It looks very much like we're seeing a near-perfect greenfield scenario; the mutation that made it infectious in humans happened in a human in a populated place. The timing will have some error bars -- molecular clocks always have error bars -- but in this case many of the calibration values for the molecular clock are very well known, so the error bars will be unusually small.

The interesting thing to me is that the US excess death numbers look like they tick up in February. That implies that the disease spread between the US and China extremely quickly; one of the first few dozen infected got on a plane. It's not absolutely certain which way the plane was going.

Peter T said...

More TPM on excess deaths:


Graydon said...

+Peter T
Yup. Which is unsurprising; not only do many of the elderly simply not get to hospitals, everyone who might be concerned for causes of death is very busy right now. There may be some large statistical studies in a few years.

But right now, it's looking vaguely like the effective mortality rate is much more like 2% than 1%. What it looks like in a greenfield away from major centres with large hospitals won't be better than that.