I note that this particular checklist results in an alarm bell which basically cannot go off until a pandemic is already well under way. Like, the “3 continents” item or the “medical supply shortages” or “quarantine of a city” or “front page news” are essentially hindsight indicators; by that point the pandemic has already reached significant scale. In hindsight, February 20 2020 was very late to start paying attention to covid.
My starting point was based on the efficient markets hypothesis. I figured that it would be hard to come to correctly calibrated confidence about the economic consequences of a pandemic faster than the market. So I tried to integrate the most relevant-seeming information we had just prior to the market crash of 2020, and see if it was possible to do a little bit better. Predicting the outcome with correctly calibrated confidence in early February, mid-January, or early January would have been progressively more impressive, but I wanted to set myself an easier task.
Hopefully, this checklist retains utility as a tool for earlier warning, as it can serve as a sort of dashboard for monitoring a developing outbreak. For example, as an increasing number of checklist items move from “no” to “yes,” or as they become closer to “yes,” we become increasingly concerned. There’s tons of room for improvement in this checklist!
As a side note, this checklist would have beaten the FDA to declaring Sars-CoV2 a pandemic by 3 weeks. The WHO declared Sars-CoV2 a pandemic on March 11, whereas Sars-CoV2 would have met 13 of these 14 criteria no later than Feb 20, 2020.
Note also that Wuhan was locked down on Jan. 23 2020, after the Chinese government first identified a cluster of sick people in mid-December. So there was a 1-month time lapse between first detection and major city lockdown, which occurred a month prior to Feb. 20.
This points out the dilemma. A major city lockdown seems like a hindsight indicator, but what about 18 deaths?
The New York Times had at least two articles by Jan. 8 on the “mystery flu,” though they were on page A13. Only the Jan. 23 articles made page A1. So perhaps the “front page” criteria should be relaxed.
Prospectively, you can just look at the first handful of patients, and simply infer epidemiological parameters in a rough way, and make an adequate mechanistic projection of what will happen (at least… in the absence of functional human responses).
This points out the dilemma. A major city lockdown seems like a hindsight indicator, but what about 18 deaths?
If you have 18 deaths, and every patient’s social network has been contact traced and tested and all the rest are negative, then even if the mortality is 90%, if the infection sequence looks like this...
...then the R0 is very low and also the epidemic is over! <3
On the other hand: Suppose your patient zero is the first death, then what if the other 17 deaths had direct exposures to patient zero, and especially what if some of those 17 dead people’s second order contacts are fleeing from quarantine doctors...
...then the R0 > 17 most likely, and I really really hope Madagascar closes the ports, so that at least some human society survives the biohorror that is about to unfold.
(Logically, all of the second order contacts of those 18 dead should be racing towards the quarantine doctors, because in a coherently sane system they would obviously have a PERSONAL INCENTIVE to seek out the amazing expert care and paid vacation that are properly owed to the first thousand or so patients in any potential epidemic… because the doctors reasoning about how to head-off the expansion of the new disease would understand incentives and have the budget to bribe patients into cooperating with what is best for the health of the herd… but people are not logical, in general, so… yeah… Madagascar closing its ports would be a much more realistic thing to actually hope for AFTER most of the biohorror is inevitably baked in and AFTER the madness starts to occur in a way that is legible to people who cannot reason but can “get a vibe” from other people.)
The R0 is the main exponential parameter for a disease, within the field of epidemiology.
As the saying goes “never turn your back on an exponentially growing process”.
Once you know the likely near future R_t of a disease, you know how exponential it will be, and therefore you know whether the disease will exist in the future.
If the R_t goes down to 0.5 and stays there for long enough, then the disease will go extinct and the R_t will become undefined <3
The rest of the parameters in epidemiology (like attack rate and mortality and so on) are useful for figuring out, conditional on the disease existing in the wild, how bad things will be in terms of health impacts...
....but since the health impacts of a non-existent disease are negligible, a low R0 absolves nearly every other sin that a disease might otherwise be exhibiting.
Using the lock-down of the city in China as a proxy for “when to worry”, you are OUTSOURCING “your ability to mechanistically reason” to competent reasoners in the (authoritarian and unelected) government of China.
They wouldn’t be able to use your instrument to make the decision that your instrument takes as an input. That would be circular! ;-)
But your instrument can piggy back on them simply directly reasoning about reality directly.
Suppose they get a convenience sample of medical data to estimate the R0, and made two mechanistic predictions using exponential growth: 1) “what the country looks like (with this R0) if we don’t lock down our city” and 2) “what the country looks like (with this R0) if we successfully lock down our city” and
then choosing the policy that causes fewer deaths in the wider country.
That produces a signal. That signal can go into your instrument.
But if you can just directly get the patient data yourself, you can predict what they predict BEFORE they make their announcements.
Also, suppose that those government officials just thought it would be funny for everyone else to get the disease too. Then they wouldn’t lock down their city and you wouldn’t get the signal.
Or suppose they thought that “overpopulation exists and is bad” and were pro-actively in favor of death for practically everyone? Murder monkeys like this exist! They’re not even ashamed of it, and newspapers like that are not even protested for espousing pro-genocide ideologies.
Another way to break the signal might be if hypothetical city-lockdown-deciders specifically authorize people to travel around the world on purpose if they were sufficiently morally monstrous (or sufficiently aware that almost everyone high enough on the food chain of OTHER countries is ALSO basically psychopaths, and then they consciously spread disease to other places according to a “pre-emptively defensive” logic based on realistic doctrines that insist that if they have a new weakness then everyone else should also have that weakness for their own relative safety).
All such morally monstrous calculations or behavioral changes could BREAK the “city lockdown” signal, for your instrument.
It is cleaner and less noisy and cheaper and less entangled with extraneous signals to just LOOK AT MECHANISTIC REALITY instead of having to compute who is evil and what their incentives are before you try to think about whether to copy them for reasons you don’t yet mechanistically understand.
(at least… in the absence of functional human responses).
This is the limiting factor of our ability to infer R0 from reported cases at this early stage in practice. This monkeypox outbreak in Europe provoked an immediate and intense social response, both to identify cases and to prevent further spread.
The number of cases reported is a function of both the actual extent of disease spread and the increased amount of testing and public awareness, which relates in a complicated way to disease spread and to earlier public awareness efforts.
Clearly, we can experience such a huge spike of viral spread that increased testing can’t possibly account for it, as we saw in Omicron.
In the first few days of the monkeypox outbreak, increased case reports were probably a function of disease spread. Now, though, I am very uncertain about whether to attribute increased cases to better testing and social awareness for a disease that was already there, or to actual viral spread.
If we see about 1,000 cases or more in the next couple weeks, though, or see it achieving community spread outside Europe, I’ll definitely start to think this is getting out of hand. By then, we’ll also have more information about its genetics, how it spreads, and the CFR among the European population.
Great info, thanks!
I note that this particular checklist results in an alarm bell which basically cannot go off until a pandemic is already well under way. Like, the “3 continents” item or the “medical supply shortages” or “quarantine of a city” or “front page news” are essentially hindsight indicators; by that point the pandemic has already reached significant scale. In hindsight, February 20 2020 was very late to start paying attention to covid.
For sure!
My starting point was based on the efficient markets hypothesis. I figured that it would be hard to come to correctly calibrated confidence about the economic consequences of a pandemic faster than the market. So I tried to integrate the most relevant-seeming information we had just prior to the market crash of 2020, and see if it was possible to do a little bit better. Predicting the outcome with correctly calibrated confidence in early February, mid-January, or early January would have been progressively more impressive, but I wanted to set myself an easier task.
Hopefully, this checklist retains utility as a tool for earlier warning, as it can serve as a sort of dashboard for monitoring a developing outbreak. For example, as an increasing number of checklist items move from “no” to “yes,” or as they become closer to “yes,” we become increasingly concerned. There’s tons of room for improvement in this checklist!
As a side note, this checklist would have beaten the FDA to declaring Sars-CoV2 a pandemic by 3 weeks. The WHO declared Sars-CoV2 a pandemic on March 11, whereas Sars-CoV2 would have met 13 of these 14 criteria no later than Feb 20, 2020.
Note also that Wuhan was locked down on Jan. 23 2020, after the Chinese government first identified a cluster of sick people in mid-December. So there was a 1-month time lapse between first detection and major city lockdown, which occurred a month prior to Feb. 20.
Jan. 23 is also the first day for which Our World In Data begins tracking worldwide COVID-19 deaths. There were 18 deaths by that day.
This points out the dilemma. A major city lockdown seems like a hindsight indicator, but what about 18 deaths?
The New York Times had at least two articles by Jan. 8 on the “mystery flu,” though they were on page A13. Only the Jan. 23 articles made page A1. So perhaps the “front page” criteria should be relaxed.
Prospectively, you can just look at the first handful of patients, and simply infer epidemiological parameters in a rough way, and make an adequate mechanistic projection of what will happen (at least… in the absence of functional human responses).
If you have 18 deaths, and every patient’s social network has been contact traced and tested and all the rest are negative, then even if the mortality is 90%, if the infection sequence looks like this...
Patient0--patient1--p2--p3--p4--p6--p8--pA--pB--pC--pD--pF--pG (end)\--p5--p7--p9 (end) \--pE--pH--pI (quarantined)...then the R0 is very low and also the epidemic is over! <3
On the other hand: Suppose your patient zero is the first death, then what if the other 17 deaths had direct exposures to patient zero, and especially what if some of those 17 dead people’s second order contacts are fleeing from quarantine doctors...
...then the R0 > 17 most likely, and I really really hope Madagascar closes the ports, so that at least some human society survives the biohorror that is about to unfold.
(Logically, all of the second order contacts of those 18 dead should be racing towards the quarantine doctors, because in a coherently sane system they would obviously have a PERSONAL INCENTIVE to seek out the amazing expert care and paid vacation that are properly owed to the first thousand or so patients in any potential epidemic… because the doctors reasoning about how to head-off the expansion of the new disease would understand incentives and have the budget to bribe patients into cooperating with what is best for the health of the herd… but people are not logical, in general, so… yeah… Madagascar closing its ports would be a much more realistic thing to actually hope for AFTER most of the biohorror is inevitably baked in and AFTER the madness starts to occur in a way that is legible to people who cannot reason but can “get a vibe” from other people.)
The R0 is the main exponential parameter for a disease, within the field of epidemiology.
As the saying goes “never turn your back on an exponentially growing process”.
This maxim applies to diseases no less than it applies to CPU speeds, or the endowments of perpetual trusts, or oxytocin feedback loops.
Once you know the likely near future R_t of a disease, you know how exponential it will be, and therefore you know whether the disease will exist in the future.
If the R_t goes down to 0.5 and stays there for long enough, then the disease will go extinct and the R_t will become undefined <3
The rest of the parameters in epidemiology (like attack rate and mortality and so on) are useful for figuring out, conditional on the disease existing in the wild, how bad things will be in terms of health impacts...
....but since the health impacts of a non-existent disease are negligible, a low R0 absolves nearly every other sin that a disease might otherwise be exhibiting.
Using the lock-down of the city in China as a proxy for “when to worry”, you are OUTSOURCING “your ability to mechanistically reason” to competent reasoners in the (authoritarian and unelected) government of China.
They wouldn’t be able to use your instrument to make the decision that your instrument takes as an input. That would be circular! ;-)
But your instrument can piggy back on them simply directly reasoning about reality directly.
Suppose they get a convenience sample of medical data to estimate the R0, and made two mechanistic predictions using exponential growth:
1) “what the country looks like (with this R0) if we don’t lock down our city” and
2) “what the country looks like (with this R0) if we successfully lock down our city” and
then choosing the policy that causes fewer deaths in the wider country.
That produces a signal. That signal can go into your instrument.
But if you can just directly get the patient data yourself, you can predict what they predict BEFORE they make their announcements.
Also, suppose that those government officials just thought it would be funny for everyone else to get the disease too. Then they wouldn’t lock down their city and you wouldn’t get the signal.
Or suppose they thought that “overpopulation exists and is bad” and were pro-actively in favor of death for practically everyone? Murder monkeys like this exist! They’re not even ashamed of it, and newspapers like that are not even protested for espousing pro-genocide ideologies.
Another way to break the signal might be if hypothetical city-lockdown-deciders specifically authorize people to travel around the world on purpose if they were sufficiently morally monstrous (or sufficiently aware that almost everyone high enough on the food chain of OTHER countries is ALSO basically psychopaths, and then they consciously spread disease to other places according to a “pre-emptively defensive” logic based on realistic doctrines that insist that if they have a new weakness then everyone else should also have that weakness for their own relative safety).
All such morally monstrous calculations or behavioral changes could BREAK the “city lockdown” signal, for your instrument.
It is cleaner and less noisy and cheaper and less entangled with extraneous signals to just LOOK AT MECHANISTIC REALITY instead of having to compute who is evil and what their incentives are before you try to think about whether to copy them for reasons you don’t yet mechanistically understand.
This is the limiting factor of our ability to infer R0 from reported cases at this early stage in practice. This monkeypox outbreak in Europe provoked an immediate and intense social response, both to identify cases and to prevent further spread.
The number of cases reported is a function of both the actual extent of disease spread and the increased amount of testing and public awareness, which relates in a complicated way to disease spread and to earlier public awareness efforts.
Clearly, we can experience such a huge spike of viral spread that increased testing can’t possibly account for it, as we saw in Omicron.
In the first few days of the monkeypox outbreak, increased case reports were probably a function of disease spread. Now, though, I am very uncertain about whether to attribute increased cases to better testing and social awareness for a disease that was already there, or to actual viral spread.
If we see about 1,000 cases or more in the next couple weeks, though, or see it achieving community spread outside Europe, I’ll definitely start to think this is getting out of hand. By then, we’ll also have more information about its genetics, how it spreads, and the CFR among the European population.