2 April 2020
Updating my priors
I haven’t publicly updated the distribution of my current, subjective probabilities, against the three high-level scenarios I’ve previously described (which I call “expected, better, worse”), in a while. If you haven’t already read that post, you might want pause and skim through it to have context for what comes next. I also updated my thoughts on these scenarios a few days later, here.
As a reminder, “expected” is the still-unhappy case where, in many wealthy countries, we go through a period of lockdown for (what I previously called) 4-12 weeks (but now seems like it could be longer); where the lockdown is ultimately successful in slowing/controlling the epidemic; but where we still say large-scale, if temporary, economic disruption, a final attack rate of 25-40% of the population, and a Infection Fatality Rate of 0.5-1.0%. So there are a huge number of cases, and a significant number of deaths, but we can look ahead to a time, a year or two from now, when things are mostly normal. (Emerging economies are likely to suffer a much worse path.)
“Better” and “worse” do what they say on the tin; see prior posts for more detail.
On 19 March (the first post linked above), I put the distribution of probabilities as 60% expected, 20% better, 20% worse.
On 23 March (the second post linked above), I concluded that the tails had grown fatter: that the chances of better and worse outcomes were higher than I’d thought, and moved to 50% expected, 25% better, 25% worse.
In the last 8 days I’ve see-sawed between greater optimism and greater pessimism, but not shared specific updates to my priors. Today, I want to share what has behind the see-sawing, and why I significantly updated my views today in the last 24-hours.
Until a few days ago, I was concluding that Europe was on a high-probability path for Expected with a chance of Better, given the strong flattening of the curve, discussed here. At the same time, I feared that the risk of Worse outcomes in the US was rising rapidly, as discussed here.
This week, we’ve seen the US not only demonstrate a willingness to deploy unlimited (and steadily increasing) firepower on the economic front, but also radically shift both rhetoric and behavior towards stricter, medium-term control measures. That made me upgrade the chance of the US having an Expected outcome versus Worse.
If you’d asked me even two days ago, I would probably have been around 65% Expected, 25% better, 10% Worse for rich Western countries. (Rich Asia as well as developing economies are in different situations, where I’m largely more optimistic about the former and extremely pessimistic about the latter.)
Yesterday and today, I’ve been thinking and reading about exit strategies, and ended up depressed again. Depressed not because I concluded that I should increase the probability of Worse outcomes — I am moving even more strongly towards Expected — but because I realized that my thinking about the Expected case was sloppy (to the point of hand-waving) on the question of exit strategy, and that Expected cases are likely quite a bit more painful that I had thought.
My thoughts are very much in formation on this, but I’ll share my updated thinking about why I fear the most likely exit strategies are painful.
When I talk about exit strategies, I mean paths, post-lockdown, to outcomes where life can mostly return to normal: where most economic activity resumes, and where the risk of serious adverse health consequences from COVID-19 for most people has decreased to a level where it is not much greater than the risks that we accept every day, without radical intervention, from diseases like seasonal flu.
(As an aside, it’s remarkable that so many of us would rather accept somewhere between a 1 in 10,000 and 1 in 100 chance of dying from the flu, as a rough order of magnitude and excluding 65+, than get a flu vaccine. In the US, less than half of the <65 population was vaccinated in a 2018 survey. We don’t shut down nursing homes, let alone the entire economy, in a worse-than-average flu season. There is something extraordinary about the salience, the perceived threat, the alienness, and the many unknowns around COVID-19 that makes us think about it in an entirely different way.)
What I’ve said previously, in an extremely hand-waving way, is that once the epidemic is under control, we can move to a different stage of aggressive contract tracing, quarantine of infected and suspected infected individuals, and tight border controls including quarantine for new entries. In other words, we can try to be China, South Korea, Hong Kong, or Singapore.
I’ve now realised that this is very naive for a few reasons.
Before I go there, let’s remind ourselves of how an epidemic of a highly contagious virus ends. There are really only four ways:
- The virus for some reason becomes less contagious, or so much less lethal that we accept the risk of infection as we do with more familiar infectious diseases. The lower risk (or perceived risk) could be because of mutation or because of response to changing weather, for example.
- We develop herd immunity because a high enough proportion of the population acquire immunity through being infected and recovering, assuming that immunity does indeed result and last. (For basic reproduction number R0, we need (R0-1)/(R0) to be immune; so for R0=3, that’s 2/3 of the population.)
- We develop herd immunity through developing and globally deploying an effective vaccine that (combined with acquired immunity) results in the level of immunity described in (2) above.
- We drive Rt (the effective reproduction number at time t) down to below 1, and keep it there for an extended period of time. This could be through some combination of lockdowns, behavioral changes like hand-washing and voluntary social distancing, as well as extensively testing for COVID-19 and quarantining those who are infected.
I don’t know any scientist who thinks we should count on (1) above.
Despite the vast resources being put towards developing a vaccine (3), I’m sceptical that we will have one in 2021, or even 2022, for three reasons.
- First, we’ve tried for many years, and spent a lot of money, to develop vaccines against other viruses (common cold, AIDS, SARS, MERS) without success.
- Secondly, if the true IFR (infection fatality rate) really is 1% or less, as seems likely, we will need to have confidence that severe side effects occur in well below 1 in 100 people who are vaccinated. It will take human trials of significant scale and time to conclude that.
- Third, we need to overcome the challenge of manufacturing, distributing, and administering billions of doses around the globe; and we also need to get high (probably >2/3) compliance; something that we struggle with for some vaccines.
(2) is clearly the scenario we’ve been trying to avoid, or at least to delay: overwhelmed hospitals, higher-than-necessary death rates.
And so the question is, if the least bad options are (4) or a very slow version of (2) — which I think at some level are almost equivalent — what do they look like?
Nicholas Davies, along with the influential epidemiologist / modeler Adam Kucharski and colleagues from the Centre for Mathematical Modelling of Infectious Diseases COVID-19 working group have just published a really depressing paper focused on the UK; and although it is UK-specific, the conclusions will directionally apply in many countries.
The paper appears designed to demonstrate to UK policymakers that they have made the right decision to go down (however slowly) the lockdown route versus other options. And to that end, the lockdown option in this set of simulations is shown to be less bad than other options.
What really hit me in reading the paper is how bad the lockdown option is in reality.
The problem is that after an intense lockdown of, say, two months, what happens? As we’re starting to see in Asian countries that initially controlled the virus, new outbreaks begin to appear, either because there are a significant number of infected people in the population whom have not been detected, or because new cases enter from outside the country.
This new paper suggests that the least-bad option requires us to “pulse” in-and-out of strict control measures in order to keep demand for ICU beds within supply, even if the supply of ICU beds doubles.
In the charts above, the “Lockdown 1000-bed trigger” (top row, second chart from the left) is the least-bad scenario when measured by total deaths, peak ICU beds needed, and for staying closest to within ICU bed capacity (though note that to do so the number of beds needs to double). (The name of the scenario comes from the idea that we would relax lockdowns whenever we brought active cases within then-current health-care system capacity — double todays’ — but would re-impose them to avoid breaching that capacity.)
But note that the lockdowns need to recur, albeit with decreasing frequency, through the summer of 2021! And that we are more often in lockdown than not for the rest of 2020. In this scenario, we are in lockdown for 5 of the next 8 months, and for 7 of the first 8 months of 2021; and “intensive interventions” (social distancing) are a constant through March 2021.
Needless to say, the economic impact of (taking this “Lockdown 1000-bed trigger” scenario as a specific example) would be extreme; much more than the already-dire forecasts you’ve been hearing anticipate.
Now, the authors are very careful to make clear that this is only one modeling study; that there are important assumptions made that may not obtain in real life; and that other outcomes are possible.
So the question we need to ask is: what are the alternatives, and how likely are they?
The best alternative
The best alternative, I think, goes something like this — effectively the China / Hong Kong / Singapore / South Korea strategy:
- Take a lot of pain up from to get new cases very near zero.
- Do extremely broad, mandatory testing in the population to detect new cases, and force-quarantine those suspected or confirmed of being infected, including removing them from their families.
- Close the borders except to small numbers of people, and require long quarantine periods for those who do come in.
- When in doubt about whether an individual might have been exposed, enforce quarantine.
But as I understand about how this has been implemented in some Asian countries (and about the potential limitations on the effectiveness of the strategy), I’m growing increasingly concerned about whether such a strategy is feasible in a country like the US (it might be in some European countries, especially smaller and highly cohesive ones).
Any country will find it hard to impose the economic and social pain from on-again, off-again lockdowns for an extended period.
But is it possible to imagine the compliance, the perceived invasion of privacy, the separation of families in such an individualistic culture? Would we submit to an app that rated us as red/yellow (amber)/green and imposed restrictions on our movements as a result? Would we permit a family member showing symptoms to be removed to government-run quarantine?
Another, potentially complementary path, is through serological testing.
(This free-to-read FT article gives a good and recent summary of testing.)
If we acquire immunity after recovering, and if we can detect it broadly and reliably, we will have an increasing number of “superheroes” who can (in theory) return to work, work on the front lines, etc. without significant risk to themselves or to society. In this scenario, we can permit more and more of the economy to ramp as the proportion of society known to have immunity increases.
This is clearly a better scenario than shutting down the economy for most of the next 16 months. But even in this scenario, a significant proportion of the population is subject to strict control measures for most of the rest of 2020, and parts of 2021 as well.
I shared a draft of this post with a few friends, and thought this was an interesting way of looking at it (minor formatting edits are mine):
[Could] therapeutic (non-vaccine) medical developments [change] society’s calculus on the issue? Surely there has to be some fatality/severe case rate that is tolerable, and this will not only go up over time as economic damages mount; but also the actual data on the fatalities can be pushed down if a therapeutic development manifests (which I believe is much more probable in the next 6-12 months than a vaccine).
Maybe that is one ‘better’ equilibrium here that doesn’t solely involve the ‘R’s – between some measure of social distancing and the like we bring attack rates down from a hypothetical X to a hypothetical X -20% and then we have a therapy that reduces fatality from 1-2% to .5% or something like that.
My friend is making two points that both seem highly relevant.
First, as therapeutics get better (and I am much more optimistic that we will have helpful therapeutics quite soon, than that we will have a vaccine within a few years), the Infection Fatality Rate will come down, and COVID-19 will look more like familiar infectious diseases like seasonal flu.
Secondly, in some countries, over time we might choose a different balance between health outcomes and economic outcomes, particularly as the economic costs mount and as the potential health impact reduces thanks to effective therapeutics.
Among all of the posts I’ve written so far, I feel the least confident about my thinking this one. I’d really appreciate hearing reactions, and in particular ideas about better exit strategies.