microCOVID.org: A tool to estimate COVID risk from common activities

post by catherio · 2020-08-29T22:28:05.612Z · score: 66 (27 votes) · EA · GW · 23 comments

This is a link post for https://www.microcovid.org/

This is a linkpost for a model and web tool (that I and several friends created) to quantitatively estimate the COVID risk to you from your ordinary daily activities:

This website contains three outputs of our work:

1. a web calculator that you can use to calculate your COVID risk (in units of microCOVIDs, a 1-in-a-million chance of getting COVID).

2. a white paper that explains our estimation method. EAs might be particularly interested in the footnotes throughout, and the detailed research sources section.

3. a spreadsheet to compute your COVID risk in more detail and to track your risk over time. EAs might find this more customizable and powerful than the web calculator.


We hope this will directly help the EA community by resolving some of the issues highlighted in an earlier post:

https://forum.effectivealtruism.org/posts/q3v8CpAiuNhZzJ7Gt/customized-covid-19-risk-analysis-as-a-high-value-area [EA · GW]

"[Many EAs] are doing [COVID modeling] work themselves: it's time-costly, and it is mentally draining and stressful. It's also wasteful if a lot of this analysis work ends up getting replicated privately across many people. At the same time, [if people don't do these analyses,] households don't have ways of analyzing risks and deciding on acceptable behaviors"


If you have different beliefs than us and would like to use a version of the model that reflects your beliefs rather than ours, you can make modifications to your copy of the spreadsheet, or fork the repository and make a personal copy of the web calculator. We also hope you will submit suggestions, either by emailing us or by making issues or pull requests directly on github.

23 comments

Comments sorted by top scores.

comment by PeterMcIntyre · 2020-09-01T12:43:58.550Z · score: 11 (9 votes) · EA(p) · GW(p)

This is such a useful public good. Thank you!!

comment by Milan_Griffes · 2020-08-29T23:08:35.758Z · score: 4 (3 votes) · EA(p) · GW(p)

Thank you for creating this!

I was surprised by the riskiness estimates in the table on this page: https://www.microcovid.org/paper/2-riskiness

If an event that accrues 1000 μCoV is considered "borderline reckless", wouldn't that imply a very low risk tolerance for everyday activities like driving a car? (Because driving is fairly high risk and some of that risk is externalized.)

comment by Linch · 2020-08-30T10:16:26.665Z · score: 8 (6 votes) · EA(p) · GW(p)

Hmm, so a ballpark is that if you think you getting covid results in ~0.5% deaths in expectation (adjusting for lower likelihood due to typical EA demographics but also incorporating some chance of infecting others), 1000 μCoV ~= 5 micromorts. This is equivalent to driving 2000km, or 50km on a motorbike. I know very few people who think the former is an everyday activity, though I can imagine some people (none who I personally know) will commute by motorbike 25km one way for work?

That said, skydiving is ~8 micromorts, and marathon running ~7, and neither quite seem reckless to me? But "borderline reckless" seems approximately right as a description?

(If you think the number is instead ~0.1% deaths in expectation, we get to ~1 micromort. If we think it's ~1%, we get to ~10 micromorts)

comment by Benjamin_Todd · 2020-08-31T16:25:01.699Z · score: 8 (5 votes) · EA(p) · GW(p)

If R = about 1, then each infection in expectation infects another person.

If the disease last two weeks, then each infection results in another 31 infections per year that R stays around 1.

Given that it seems like R is going to stay around 1 for at least the next ~6 months, I think the 0.5% expected deaths seems a bit low.

If someone ~30 who's healthy gets infected, they have maybe a 1 in 2000 chance of dying, but then will also in expectation perhaps infect a chain of ~15 people. The morality among the broader population is more like 1.5%, so that's 0.22 expected deaths from the chain, about 40-times your estimate.

comment by Linch · 2020-08-31T20:43:40.786Z · score: 9 (2 votes) · EA(p) · GW(p)

This depends a lot on where and when you're situated!

For example, California's total infected numbers are plausibly >14% now, so reinfections aside, a chain of ~15 people from the marginal infection as of August 31 California is implausibly high.

Numbers aren't that different for London (as of late May), see page 22 of this report.

Likewise, the empirical fatality rate used to be >1% in the US in March/early April, but is likely lower than 0.5% in the US now, partially due to better treatment and mostly due to changing demographics in who gets infected (younger people less cautious and more likely to be infected, etc).

In contrast, I can totally believe that a marginal infection outside of East Asia/Oceana in mid-March will result in >20 infections.

comment by Benjamin_Todd · 2020-08-31T21:02:44.744Z · score: 4 (2 votes) · EA(p) · GW(p)

Ah that makes sense, thank you!

comment by Linch · 2020-08-31T22:20:27.165Z · score: 2 (1 votes) · EA(p) · GW(p)

Glad to help! :)

comment by Benjamin_Todd · 2020-09-17T12:02:39.697Z · score: 2 (1 votes) · EA(p) · GW(p)

Thinking a bit more, I'm not sure this argument works, though I might have misunderstood.

In London, 5-10% have been infected. Prevalence currently is ~1 in 2000 , R = 1 and let's assume transmission time is 1 week. That means that in 6 months time, about another 1.5% of people will have been infected (30/2000)

If I get infected now, then I there will be an extra chain of infections 30 people long.

I don't see how the overall prevalence levels block the chain I cause. If in 6 months, another 1.5% of people have been infected, that's not enough to meaningfully change R.

If 5% of people were infected now, and R = 1, then we'd be saturated and reach herd immunity in a matter of weeks, which would cut off the chain. But instead, the prevalence is sufficiently low that it seems like it is possible for each individual to cause a long chain.

comment by Linch · 2020-09-18T00:08:20.547Z · score: 2 (1 votes) · EA(p) · GW(p)

Right, I think the argument as written may not hold for the UK (and other locations with very low prevalence but R ~=1). My intuitions, especially in recent months, have mostly been formed from a US context (specifically California), where R has never been that far away from 1 (and current infectious prevalence closer to 0.5%).

That said, here are a bunch of reasons to argue against "Alice, an EA reading this forum post, being infected in London means Alice is responsible for 30 expected covid-19 infections (and corresponding deaths at 2020/08 levels)."

(For simplicity, this comment assumes an Rt ~= 1, a serial interval of ~one week, and a timeframe of consideration of 6 months)

1. Notably, an average Rt~=1 means that the median/mode is very likely 0. So there's a high chance that any given chain will either terminate before Alice infects anybody else, or soon afterwards. Of course, as EAs with aggregatively ethics, we probably care more about the expectation than the medians, so the case has to be made that we're less likely on average to infect others. Which brings us to...

2. Most EAs taking some precautions are going to be less likely to be infected than average, so their expected Rt is likely <1. See Owen's comment [EA(p) · GW(p)] and responses. Concretely, if you have a 1% annualized covid budget for a year (10,000 microcovids), which I think is a bit on the high side for London, then you're roughly exposing yourself to 200 microcovids a week. At a baseline population prevalence of 500 microcovids, this means you have a ~40% chance of getting covid-19 in a week conditional upon your contacts having it, which (assuming a squared term) means P(Alice infects others | Alice is infected) is also ~40%.

Notably a lot of your risk comes from model uncertainty, as I mentioned in my comment to Owen [EA(p) · GW(p)], so the real expected Rt(Alice) > 0.4

As I write this out, under those circumstances I think a weekly budget of 200 microcovids a week is possibly too high for Alice.

However, given that I live in Berkeley, I strongly suspect that E(Number of additional people infected, other than Linch | Linch being infected) is < 1. (especially if you ignore housemates).

3. If your contacts are also cautious-ish people, many of them who are EAs and/or have read this post, they are likely to also take more precautions than average, so P(Alice's child nodes infecting others | Alice's child nodes being infected) is also lower than baseline.

4. There's also the classist aspect here, where most EAs work desk jobs and aren't obligated to expose themselves to lots of risks like being essential workers.

5. Morally, this will involve a bunch of double-counting. Eg, if you imagine a graph where Alice infects one person, her child node infects another person etc, for the next 6 months, you have to argue that Alice is responsible for 30 infections, her child node is responsible for 29, etc. Both fully counterfactual credit assignment and proposed alternatives have some problems in general [EA · GW], but in this covid-y specific case I don't think having an aggregate responsibility of 465 infections when only 30 people will be infected will make a lot of sense. (Sam made a similar point here, which I critiqued because I think there should be some time dependence, but I don't think time dependence should be total).

6. Empirical IFR rates have gone down, and are likely to continue doing so as a) medical treatment improves, b)people make mostly reasonable decisions with their lives (self-select on risk levels) plus c) reasonable probability of viral doses going down due to mask usages and the like.

7. As a related point to #3 and #6, I'd expect Alice's child nodes to be not just more cautious but also healthier than baseline (they are not randomly drawn from the broader population!).

8. There's suggestive evidence of substantial behavioral modulation (which is a large factor keeping Rt ~=1). If true, this means any marginal infection (or lack thereof) has less than expected effect as other people adjust behavior to take less or more risks.

___

Counterarguments, to argue that E(# of people infected| Alice is infected)>>30:

1. Maybe there's a nontrivial number of worlds where London infections spike again. In those worlds, assuming a stable Rt~=1 is undercounting. (and at 0.05% prevalence, a lot of E(#s infected) is dominated by the tails).

2. Maybe 6 months is too short of an expected bound for getting the pandemic under control in London (again tail heavy).

3. Reinfections might mess up these numbers.

___

A nitpick:

In London, 5-10% have been infected

Where are you getting this range? All the estimates I've seen for London are >10%, eg this home study and this convenience sample of blood donors.

comment by Benjamin_Todd · 2020-09-21T13:23:43.834Z · score: 4 (2 votes) · EA(p) · GW(p)

These seem like interesting points, but overall I'm left thinking there is still a significant chance of setting off a long chain that wouldn't have happened otherwise. (And even a lowish probability of a long chain means the bulk of the damages are on other people rather than your self.)

I think the argument applies to California too. Suppose that 20% have already been infected, and 0.5% are infected currently, and R = 1.

Then in 6 months, an extra 0.5%64 = 12% will have been infected, so 32% will have had it in total. That won't be enough to create herd immunity & prevent a long chain.

An extra infection now would in expectation cause a chain of 641 = 24 infections, and if a vaccine then came and the disease were stamped out, then those 24 people wouldn't have had the disease otherwise.

What seems to matter is that we're in a "slow burn" scenario, where we're a decently long way from ending it, but R ~ 1, but we're not sure we're going to reach herd immunity as the end game.

PS My figure for London was a rough ballpark from memory - your figures are better. (Though like I say I don't think the argument is very sensitive to whether 10% or 30% have already had it.)

comment by Linch · 2020-09-22T00:41:15.490Z · score: 2 (1 votes) · EA(p) · GW(p)
And even a lowish probability of a long chain means the bulk of the damages are on other people rather than your self

Sure, but how large? At an empirical IFR of 0.5%, and expected chain size of 5 (which I think is a bit of an overestimate for most of my friends in Berkeley), you get to 2% fatality rate in expectation (assuming personal risk negligible).

If you assume local IFRs of your child nodes are smaller than global IFR, you can easily cut this again by 2-5x.

This is all empirical questions, before double-counting concerns in moral aggregation.

comment by Owen_Cotton-Barratt · 2020-08-31T20:20:25.741Z · score: 7 (5 votes) · EA(p) · GW(p)

I like the point about chains of onwards infections. 

Actually there might be a dynamic where the number of people you expect to infect is also relatively proportional to your exposure, so the total cost goes with something like the square of your exposure? (If your exposure is small, your personal risk dominates as it doesn't have the squared term.)

comment by Linch · 2020-08-31T20:52:21.836Z · score: 4 (2 votes) · EA(p) · GW(p)

I think this is definitely partially true.

That said, I have some discount factor in my intuitions but much less than a squared term. Part of the issue is that your (Bayesian) chances of infecting others is not independent of your chances of being infected, a fair fraction of my "probability of being infected" comes from model uncertainty so there's a substantial error term for correlated reasons to think that we're in some way wrong about how we are modeling risk.

comment by catherio · 2020-08-31T23:11:13.705Z · score: 6 (4 votes) · EA(p) · GW(p)

Just want to chime in and say

1) yes, we think that thinking about chains of onwards infections is important, and

2) we haven't done this in great detail, and

3) we would ***LOVE*** if someone wrote up an analysis of this. Issue for it: https://github.com/microcovid/microcovid/issues/17

comment by Linch · 2020-09-01T01:58:52.516Z · score: 7 (4 votes) · EA(p) · GW(p)

What's the best way to give feedback? Your contact page said that tweeting is fine so I just left a small comment there.

I think doing chain analysis is hard because you basically need a full epi model, which isn't easy to do (especially in places where % infected is low) at an interesting granularity, since (from reading your white paper) your budget/target for model uncertainty seems to be <3x.

comment by Milan_Griffes · 2020-08-30T15:57:37.898Z · score: 2 (1 votes) · EA(p) · GW(p)

Thanks for running the conversion into micromorts – that's helpful.

fwiw back in 2003 the average US commute was ~30 miles/day (I couldn't find more recent data). So that's about one micromort/week from commuting.

Here's some case fatality rate data by age. 0.5% chance of death seems a bit high, though maybe reasonable depending on how you're incorporating the externality.

comment by Risto_Uuk · 2020-08-30T07:45:07.631Z · score: 4 (3 votes) · EA(p) · GW(p)

It would also be more informative to assess risks of death from COVID-19. 'Micromorts' normally stand for one-in-a-million chance of death because the word is combined from micro and mortality. If 1000 μCoV were a thousand-in-a-million chance of death, then engaging in activities with such a risk would be quite reckless indeed. That would be about similar to climbing quite high mountains and doing a couple of base-jumps.

I have calculated COVID-19 risks for myself in the context of Estonia where I am currently. My numbers right now are about: risk of getting COVID-19: 1^-4 and risk of dying of COVID-19: 4^-6 (about 4 micromorts). These are probably overestimates as I'm young, healthy, and very cautious and I'm using nasal swab data rather than antibody data, which indicate about 10 times larger infection rate than the nasal swab data (meaning 10 times smaller death rate in Estonia). These numbers are of course smaller in Estonia than in the Bay Area.

Another interesting topic here is what counts as too risky? I think that my risk threshold is about traveling 10 km by motorbike, which is about 1 micromort. I would engage in such activities once in a while, but in general 1 micromort seems too large in the context of activities that are easily substitutable. Can't ride a motorbike for entertainment? Easy, just play some less risky sport and get as much pleasure.

comment by Misha_Yagudin · 2020-08-30T21:33:33.421Z · score: 4 (3 votes) · EA(p) · GW(p)

I believe it is "borderline reckless" because 1000 μCoV per event = 0.1% Cov per event and their default risk tolerance is 1% per year [another available option is 0.1% per year]. So you can do such events about one once per month [or per year] before exhausting your tolerance.

Another question is whether 1% or .1% risk tolerance is reasonable. It might be for some age/health cohorts; or for someone really worried/confused about long-term effects [s.a. chronic fatigue from SARS or some unknown-unknowns].

On the other hand, while being cautious, one shouldn't neglect gradual negative effects on mental health and so on.

comment by Milan_Griffes · 2020-09-23T18:47:38.966Z · score: 3 (2 votes) · EA(p) · GW(p)

Here's a pragmatic return-to-work plan (a) that makes use of microCOVID.org

comment by Milan_Griffes · 2020-09-04T18:35:26.820Z · score: 3 (2 votes) · EA(p) · GW(p)

In other covid news, we seem to be learning that Vitamin D supplementation is helpful.

A small RCT was recently published: Castillo et al. 2020

From Masterjohn's commentary (a):

The trial was conducted at the Reina Sofía University Hospital in Córdoba, Spain. The trial included 76 patients with COVID-19 pneumonia. Although this is no longer the standard of care, all patients were treated with hydroxychloroquine and azithromycin and, when needed, a broad-spectrum antibiotic. Admission to the ICU was determined by a multidisciplinary committee consisting of intensive care specialists, pulmonologists, internal medicine specialists, and members of the ethics committee.
The patients were randomly allocated to receive or not receive vitamin D in a 2:1 ratio. This resulted in 50 patients in the vitamin D group and 26 patients in the control group.

From the abstract:

Of 50 patients treated with calcifediol [a form of Vitamin D], one required admission to the ICU (2%), while of 26 untreated patients, 13 required admission (50%) p-value X^2 Fischer test p < 0.001. Univariate Risk Estimate Odds Ratio for ICU in patients with Calcifediol treatment versus without Calcifediol treatment: 0.02 (95%CI 0.002-0.17).
comment by Milan_Griffes · 2020-09-04T19:19:21.753Z · score: 2 (1 votes) · EA(p) · GW(p)

Though perhaps the effect size they found is implausibly large...

Expressed as relative risk, vitamin D reduced the risk of ICU admission 25-fold. Put another way, it eliminated 96% of the risk of ICU admission. Expressed as an odds ratio, which is a less intuitive concept but is often used in statistics because it gives an estimate of the effect of the treatment that would be constant across scenarios with different levels of risk, vitamin D reduced the odds of ICU admission by 98%. Either way, vitamin D practically abolished the need for ICU admission.

Would be great if this replicates in a bigger study. In the meantime, supplementing Vitamin D is cheap & safe.

More Vitamin D discussion:

comment by Milan_Griffes · 2020-09-21T20:47:13.056Z · score: 2 (1 votes) · EA(p) · GW(p)

A good thread (a) summarizing a paper on our current understanding of coronavirus transmission dynamics.