A practical summary of the covid-19 situation

00:00:00.000 | So we kind of spent the weekend putting together some thoughts about some of the problems we

00:00:06.680 | saw and how people were thinking about this and tackling this and I was, I can't believe

00:00:12.880 | what happened.

00:00:14.040 | As you see over half a million people read this thing, which we've never reached that

00:00:19.920 | many people with anything before.

00:00:22.080 | So clearly there's kind of orders of magnitude more interest in what we have to say about

00:00:26.560 | COVID-19 in the community than deep learning, so should probably talk about it briefly.

00:00:33.560 | You know, one of the things we talked about in this, this post is now translated into

00:00:38.800 | like 15 languages, thanks to amazing volunteers.

00:00:42.080 | I did a little Twitter summary of it, I'll mention a couple of the things we said.

00:00:48.440 | Now that this is eight days old, a lot of this is pretty well understood, but for those

00:00:52.560 | people that aren't yet locked down like San Francisco is, you should probably be thinking,

00:00:57.360 | acting as if you are, certainly canceling events, getting together with more than 10

00:01:02.440 | people, is giving an opportunity to basically kill them and their families, because you

00:01:08.000 | don't know if you're infected.

00:01:09.800 | So work from home if you can, not everybody can, we'll talk a bit about sanitation, it's

00:01:14.520 | actually not that easy to do properly.

00:01:20.960 | Most importantly, a lot of people watching this are in a position of authority, and so

00:01:24.440 | I particularly want to talk to you and say, you know, give, provide sick leave, make meetings

00:01:29.120 | virtual, cancel events, so make them online.

00:01:32.160 | You're the folks who can save the most lives, perhaps other than the medical people.

00:01:38.240 | There are still people saying this is just like the flu, eight days ago, most people

00:01:42.880 | seem to be saying that, but the flu has a death rate of about 0.1%.

00:01:49.560 | The director of the Center for Communicable Diseases at Harvard estimates that for COVID-19

00:01:56.920 | it's like 1 or 2%.

00:01:59.440 | So 10 or 20 times more infectious, we don't know that, but this is a guess from somebody

00:02:04.800 | who knows what he's talking about, and it's not, you know, it's somewhat widely accepted

00:02:10.440 | amongst epidemiologists, some think it's higher, some think it's lower.

00:02:15.560 | Other modeling suggested it was 1.6% in China in February.

00:02:22.080 | So it's much, much more infectious, sorry, much, much more deadly than the flu.

00:02:29.920 | One of the really terrifying things also is it spreads incredibly quickly.

00:02:36.660 | So it's, there's nothing, you know, from an epidemiological point of view there's nothing

00:02:43.980 | that suggests anything you understand about the flu is going to help you understand COVID-19.

00:02:51.140 | A lot of people say they're basically, their responses don't panic.

00:02:56.960 | We find that a super unhelpful response.

00:03:00.280 | It can tend to lead to kind of apathy or hubris.

00:03:04.320 | In China, by the time they were where we US was when we wrote this eight days ago, tens

00:03:11.780 | of millions of people were already on lockdown and two from all these hospitals were built.

00:03:17.160 | China did this amazing, I mean, they did plenty of wrong things at the start in terms of not

00:03:22.640 | actually listening to the whistleblowers and in fact arresting them.

00:03:25.920 | But then they did an astonishing response, Italy, not so much.

00:03:32.880 | And at the time we were writing this, they had to go to the point of locking down 16

00:03:38.400 | million people too late, we'll talk about more about that in a moment.

00:03:42.840 | So we don't really know much about anything.

00:03:45.440 | And for scientists, one of the challenges here is we tend to not like to do things when

00:03:50.240 | we don't know anything.

00:03:52.120 | We demand high levels of evidence and, you know, this is not peer review level stuff.

00:03:58.600 | But we're not in that world anymore.

00:04:00.080 | We're in the world of like, what do we know?

00:04:02.760 | What don't we know?

00:04:04.480 | What's the kind of, from a Bayesian point of view, figure out the priors.

00:04:09.760 | How likely is it that this thing is true?

00:04:11.280 | How likely is it that this is false?

00:04:12.720 | What would be the utility impact of doing these two things?

00:04:16.480 | Kind of model that out.

00:04:18.820 | So we need to learn to behave in the, you know, in a kind of an optimal way under uncertainty,

00:04:27.880 | which is not easy and not very compatible with kind of usual scientific approaches.

00:04:36.000 | But we can see empirically, you know, what's actually happening in Italy, for example.

00:04:44.780 | So we can, we don't have to guess totally about where things will be in 10 days time,

00:04:50.280 | because we can see what's happening in countries that are 10 days ahead of the US or the UK.

00:04:57.460 | One of the horrific things about this we pointed out in our post is that in the US in particular,

00:05:04.880 | those impacted are those who can least afford to be.

00:05:09.240 | So the lowest 10% of wage earners in the US of that group only a third have paid sick leave.

00:05:16.880 | So one of the things that happens in a country like the US that doesn't have very strong,

00:05:20.600 | to say the least, social safety nets is we force people to go to work and, you know, financially.

00:05:27.560 | And as a result, we force them to spread their possible infection to everybody else at work,

00:05:34.080 | which is terrible for the economy and terrible for society.

00:05:38.280 | So these are some things to be aware of.

00:05:41.600 | In terms of like, what does this look like?

00:05:46.360 | It seems like it's something that people have trouble taking on board emotionally.

00:05:52.280 | Like this was a, and I'm definitely not hassling this person.

00:05:54.720 | I want to, you know, this person's been super brave, but I want to point out this guy's

00:05:58.120 | a doctor and he was tweeting just at the start of March, "Hey, don't worry about this.

00:06:03.960 | It's, you know, it's a cold virus.

00:06:07.640 | He's got to stop this fear mongering.

00:06:09.320 | This is insanity.

00:06:11.720 | Come on, guys.

00:06:12.720 | Look at this in perspective.

00:06:13.840 | You know, like it's so not deadly.

00:06:16.280 | Why aren't we panicking about other things?

00:06:21.200 | You know, look at all these other things which are much worse than COVID-19."

00:06:25.800 | And then, you know, good on him.

00:06:27.000 | He said, "Ah, my hospital asked physicians to form a response team.

00:06:30.640 | So I volunteered.

00:06:31.640 | Let's let you know how it goes."

00:06:33.960 | And then, you know, this is a terrifying bit.

00:06:37.280 | A few days later, oh, I've now seen this.

00:06:41.760 | And I can tell you in 18 years of medical practice, I've never seen anything like this.

00:06:45.800 | And if you now look at this, look at this guy's Twitter account, it's full of messages

00:06:49.300 | saying take this seriously.

00:06:51.040 | This is terrifying.

00:06:52.040 | I've never seen anything like it.

00:06:53.320 | So the different, you know, when you see people on the ground, how they're reacting, I find

00:06:58.400 | that pretty powerful.

00:07:03.880 | So, you know, the data, and I'm focusing on the US because I know most of our people watching

00:07:12.480 | this are in the US, but if you look at somewhere like Italy and you say, "What was Italy looking

00:07:20.060 | like in terms of the number of cases on the 23rd of February?"

00:07:25.080 | United States on the 5th of March had about the same number of cases.

00:07:30.240 | And so then if you go a day forward, Italy and United States a day later have a similar

00:07:34.760 | number of cases, and another day later they have a similar number of cases, and so forth.

00:07:39.480 | So this is just for Italy and the US, but actually if you draw this graph for most Western nations

00:07:47.120 | in kind of cold or temperate climates, it looks exactly the same.

00:07:52.120 | Everybody is following Italy just by a different number of days.

00:07:55.440 | And so that means that we know that if we respond in the same way as Italy, we're basically

00:08:00.920 | going to follow the same shape.

00:08:03.000 | So Italy is a very useful role model to understand.

00:08:07.440 | So you know, as data scientists, here's some great data.

00:08:13.960 | So turning that data into reality, here's what people were saying on the 28th of February.

00:08:20.880 | So the 28th of February is here, okay? Italy had 888 cases, the equivalent of the US on

00:08:28.800 | the 10th of March.

00:08:30.080 | So you know, tourists are hanging out, people are chilling.

00:08:35.160 | 28th of February.

00:08:37.440 | And then by yesterday, Italy was losing 368 lives in a single day.

00:08:48.520 | More than China ever lost in a single day.

00:08:53.440 | Many of those lives being lost in, you know, a fairly small number of locations.

00:08:57.960 | So what does that look like?

00:09:00.200 | I mean, you see it in so many ways.

00:09:02.640 | You see it as, oh, hi folks, I'm sorry, but our president of the medical group just died.

00:09:11.320 | The director of our training school just died.

00:09:14.360 | Doctors are dying quickly.

00:09:18.440 | These are pictures from Italy, it's, you know, everybody is just working all the time.

00:09:26.240 | There is not nearly enough beds and doctors are having to make horrific decisions where

00:09:33.080 | they have to make a judgment call about who is it worth them treating.

00:09:38.480 | Who is too old that it's going to die anyway, they just let them die.

00:09:44.160 | Who is, anybody under 65, last time I heard was basically not being looked at on the assumption

00:09:49.600 | that they should be, you know, that they might be fine.

00:09:53.120 | So this is scary.

00:10:02.080 | There are countries that look particularly scary, perhaps the most scary outside of Italy

00:10:08.400 | at the moment is the UK.

00:10:10.320 | The Imperial College just released their latest modeling, which actually had some, looks like

00:10:17.640 | some errors that might have underestimated the number of fatalities.

00:10:22.520 | But even with that, this red bar here represents the total hospital capacity in the UK.

00:10:32.680 | And these different lines represent the number of people that need hospitalization under

00:10:37.160 | the assumption of UK does nothing versus UK closes schools and universities versus UK isolates

00:10:45.680 | cases versus UK isolates cases and quarantines households versus UK isolates cases, quarantines

00:10:53.840 | at home and has social distancing.

00:10:57.240 | In every case the difference between the number of people required to be in hospital versus

00:11:03.320 | the amount of people that can be in hospital is vast.

00:11:07.600 | So what does that mean in the UK?

00:11:11.760 | This is basically showing what percentage of symptomatic cases require hospitalization.

00:11:19.760 | And in the over 50 group, it's well, you know, 1 in 10 to 1 in 4.

00:11:27.240 | And of that group, how many require critical care goes from 1 in 10 to 70%.

00:11:34.000 | Critical care basically means ventilation, oxygenation.

00:11:37.640 | If you don't get it, I don't know exactly what happens, but the answer seems to be very

00:11:44.920 | likely you die.

00:11:45.920 | I mean, you basically suffocate.

00:11:49.120 | So this is scary stuff and the UK obviously did nothing for too long.

00:11:58.040 | And countries like the Netherlands and Sweden have explicitly made a decision at this point

00:12:03.880 | to follow the UK's example, even though the UK is now, it was actually the Imperial College

00:12:11.120 | that did the original incorrect, totally incorrect modeling that caused the incorrect response

00:12:16.540 | from the UK.

00:12:17.540 | It's based on their new modeling.

00:12:19.960 | And so people who are still saying, let's do what the UK did heading in this direction

00:12:25.560 | and basing things off totally wrong information.

00:12:29.400 | Talking of information, it's very hard to be data scientists in this environment because

00:12:34.480 | the data is not there.

00:12:36.740 | In the United States, the number of tests we're doing per million people, 23 compared

00:12:42.360 | to South Korea, 3,700, things are starting to ramp up now, but it's a huge shortage.

00:12:49.840 | And just to orient on this graph, the blue circles are showing the number of tests and

00:12:54.560 | the per capita per million and the red circles are showing the population size.

00:13:01.360 | Thanks, Rachel.

00:13:02.900 | So one of the things that's really interesting is to look at South Korea and China because

00:13:11.020 | these are countries that have, you know, a very different experience as you'll see to

00:13:16.480 | places like the UK and the US and Italy.

00:13:20.040 | And so we'll talk about that in a moment.

00:13:22.800 | In fact, let's look at it now.

00:13:25.620 | This is from today and you can see we start basically at 0.0 for each country is when

00:13:35.320 | they have 10 deaths and then this is the cumulative number of deaths.

00:13:39.520 | So this is a very lagging indicator from a data point of view.

00:13:44.740 | This is something that happens along, you know, you've been infected, you've got sick,

00:13:49.480 | you've gone to hospital, there's hospital beds and eventually if things go badly you

00:13:54.640 | die.

00:13:55.640 | So on this very lagging indicator, you can see there's very, very different shapes, right?

00:14:01.480 | There's the Spain and UK shape, which is still early, but it's just, you can see these shapes

00:14:07.520 | once they are on a certain direction, they tend to kind of stick in that direction for

00:14:10.600 | quite a while, right?

00:14:13.880 | And then there's the, where's Italy, here we are, there's the Italy shape, which is pretty

00:14:19.760 | similar, not quite as bad, right?

00:14:22.740 | And we've already talked about where that shape's ended up.

00:14:27.120 | Iran's kind of similar to Italy.

00:14:29.760 | The US is a pretty big country, pretty spread out.

00:14:35.320 | So yes, Rachel?

00:14:36.320 | Oh, your cursor is showing offset from the curves you're talking about.

00:14:41.640 | Whoa!

00:14:42.640 | Yeah, it kind of just, it just goes a little slowly, okay.

00:14:49.840 | So I'm...

00:14:50.840 | So maybe just focus on the color of the curves you're highlighting?

00:14:55.960 | Yeah.

00:14:56.960 | I guess if I kind of sit there, yeah.

00:14:59.960 | Okay, so the pink one, US, I mean you guys can read which one it is, thanks Rachel.

00:15:05.760 | It's a very different geography, so it looks a little different.

00:15:14.440 | So there's kind of one set.

00:15:17.300 | On the other hand, look at South Korea and Japan.

00:15:23.320 | Now South Korea is super interesting because they haven't had a lockdown.

00:15:30.760 | So what's going on there?

00:15:32.560 | South Korea has not had a lockdown.

00:15:35.120 | So economically, they're keeping their economy going.

00:15:40.880 | And you know, for society, they're keeping their society going.

00:15:45.520 | So studying them seems like a great idea.

00:15:48.480 | And the things they're doing are very different to at least what the US is doing.

00:15:52.980 | In South Korea, they have massively invested in tests, in testing and in tests.

00:16:00.180 | So the US is good at investing in things and making things, but currently the US is not

00:16:06.960 | investing in that.

00:16:09.800 | They also are investing in masks.

00:16:13.560 | Everybody is wearing masks in South Korea and anybody who, you know, everybody's getting

00:16:19.280 | tested regularly and that anybody who's tested with the virus is getting quarantined.

00:16:24.760 | So South Korea is, you know, it's great to look at data, see outliers and figure out

00:16:30.080 | what the hell is going on.

00:16:31.800 | So South Korea is a great outlier.

00:16:33.960 | You can also see China.

00:16:36.000 | China has taken their curve and flattened it.

00:16:41.340 | And they did this by a very early lockdown and also what many would consider draconian

00:16:49.600 | measures.

00:16:51.740 | So there seems to be a couple of different models we can follow.

00:16:54.280 | There's the lots of masks, lots of tests model of South Korea or there's the kind of major

00:17:00.800 | lockdowns.

00:17:01.800 | But China is now at the point where they are, the disease is not really spreading anymore

00:17:09.800 | and they're starting to remove the lockdowns and wind the economy back up again.

00:17:18.440 | So the impact of these interventions is huge.

00:17:22.360 | We don't have to give up, right?

00:17:23.980 | R0, which is the measure of each person, how many people do they infect on average, was

00:17:29.400 | way up around 4 in Wuhan.

00:17:33.040 | And after the interventions in Wuhan, it went down to 0.3.

00:17:37.820 | So when you go down from one person infecting 4 to one person infecting 0.3, you're going

00:17:45.700 | from a disease which is growing to a disease which is shrinking.

00:17:49.200 | But we had a question about the previous graph.

00:17:54.440 | What's the significance of the 33% daily increase line?

00:17:59.520 | I'm not sure.

00:18:00.520 | I think my understanding is it was just like a kind of an estimate, it's kind of like an

00:18:05.160 | average of what seems to be happening in places that aren't really doing much around response.

00:18:11.100 | Somehow Spain and UK are even worse than that.

00:18:13.320 | But at the time they were adding this, it was before Spain and UK were appearing much.

00:18:18.000 | So it's kind of like, oh Iran, early China, Italy are all kind of, that seemed to be a

00:18:22.680 | kind of a default I guess that people were following.

00:18:31.080 | So early response, how important is it?

00:18:36.560 | Well these are two regions of Italy.

00:18:38.800 | And one region, Lodi, the green one, did a very early lockdown.

00:18:46.920 | And you can see what happened as a result, their number of cases has not increased exponentially.

00:18:56.160 | Bergamo, on the other hand, has.

00:19:00.300 | And the reason this matters is going back to the graph we saw earlier.

00:19:06.840 | If you can keep this going up slowly enough, you can try to keep it under the red line,

00:19:12.280 | which means people that need to be hospitalized get hospitalized.

00:19:16.480 | So it makes a huge difference to the number of deaths.

00:19:19.500 | And we saw that in the 1918 flu, where regions like, it was St. Louis, right, that responded

00:19:31.060 | early, Rachel?

00:19:34.240 | Responded early and did on the whole manage to get people into hospitals and ended up

00:19:38.640 | with a much lower death rate than places like Philadelphia that literally put on a parade

00:19:43.840 | for 250,000 people, just as the pandemic was spreading.

00:19:47.280 | And this was for the flu of 1918.

00:19:50.800 | 1918, that's right.

00:19:53.560 | So it's not like we don't have data or history to learn from, we do.

00:19:59.000 | One of the challenges is it's kind of, you know, in Rachel's work as the founding director

00:20:04.920 | of the Center for Applied Data Ethics, she talks a lot about disinformation and trust.

00:20:10.280 | It's very hard at the moment because we are in a low trust environment, for good reason.

00:20:16.440 | For example, the big UK newspaper, The Daily Mail, internally writes, "Official government

00:20:24.680 | advice is no longer adequate enough to safeguard our staff."

00:20:28.760 | At the same time that they write in the newspaper regarding Boris and his boffins, "We must

00:20:36.080 | trust their judgment and do as they say."

00:20:38.760 | So we're seeing, you know, this low trust environment being created where very clearly

00:20:46.040 | media and governments are not always telling the truth.

00:20:51.920 | It makes it very difficult to know where we get, you know, good information from.

00:20:58.440 | So an example of this is like masks.

00:21:01.280 | I mentioned that kind of from an empirical point of view, we can see that countries that

00:21:06.200 | are using masks widely are doing much better than those that aren't.

00:21:12.960 | Correlation is not causation, but as I mentioned at the start of this discussion, we are not

00:21:20.680 | looking for proof here.

00:21:22.360 | We're looking for like, what makes sense, right?

00:21:24.560 | And compare the cost of a mask or a bunch of masks to the cost of a lockdown.

00:21:32.880 | It's a big difference, right?

00:21:33.960 | If we can be South Korea, our economy is going to be much better, and of course our society

00:21:41.480 | is going to be much better than if we have a lockdown and even that turns out not to

00:21:48.080 | be enough.

00:21:49.080 | So we kind of, we have to go on the information we have.

00:21:51.480 | We can't look for proof.

00:21:52.920 | We can't look for there is no peer-reviewed, definitive answer, right?

00:21:58.840 | The fact is, when people cough, there's a lot of coughing involved in this disease.

00:22:05.720 | It's one of the most common symptoms as a dry cough.

00:22:10.840 | Droplets head out there somewhere up to eight meters.

00:22:14.040 | And if you've got a mask on it, it's your mask rather than, you know, hitting your,

00:22:18.440 | well, particularly if you've also got glasses or goggles rather than hitting your eyes and

00:22:21.640 | nose or whatever.

00:22:23.640 | So you know, that on the other hand, a lot of governments are saying they're not effective.

00:22:30.520 | Now we don't know that, as I mentioned, but there's a good reason for them to say this,

00:22:35.160 | which is there's a shortage of them.

00:22:37.000 | Yes, Rachel?

00:22:38.000 | I was just going to know, St. Upto Effect, she wrote an excellent op-ed on this in the

00:22:42.120 | New York Times today called "Why Telling People They Don't Need Mass Backfired?"

00:22:46.960 | Yeah, it's not great for all kinds of reasons.

00:22:50.000 | I mean, you know, she mentioned the obvious things like, hey, you're saying that we don't

00:22:53.240 | need masks, so you should stop buying them so that then doctors can have them because

00:22:57.840 | they need masks.

00:22:58.840 | It's like, what?

00:22:59.840 | You know, and she's kind of mentioning, oh, it's because you can't possibly know how to

00:23:03.680 | put them on.

00:23:04.680 | So come on, you know, it's a five minute YouTube video showing how to put on a mask.

00:23:10.920 | It's basically a government response where they don't want to do the stuff that's been

00:23:14.680 | happening in Southeast Asia, which is in Southeast Asia, they basically say, okay, the government

00:23:20.800 | has decided that this many masks have to be given to hospitals, so they've allocated them.

00:23:27.120 | And for whatever reason, we're not doing that in the West, on the whole.

00:23:31.480 | And also in China, they bought 38 million masks right at the start of this.

00:23:37.800 | The US is not investing in increasing capacity.

00:23:41.680 | So when you're not doing those things, it can seem like the right response is to tell

00:23:47.200 | people not to buy them, tell people they don't need them, but probably not a good long-term

00:23:53.000 | solution.

00:23:55.040 | So as I say, that the trust situation is such that doctors are saying they're learning more

00:24:03.720 | from Facebook and Twitter than peer-reviewed and even pre-pub medical research literature

00:24:10.040 | at the moment, which is, it's not a great situation, but let's accept that this is what

00:24:14.960 | it is, right? And it partly is because there's a great community response, you know, of doctors

00:24:21.400 | and virologists and epidemiologists saying like, okay, I don't have time to write a paper,

00:24:27.040 | but let me tell you, I just went to this place, I did this study, I found this thing, I talked

00:24:30.640 | to this person, you know, it's fast-moving information, you know, for, but we do have

00:24:38.000 | information about masks, not for COVID-19, but for other things that involve coughing

00:24:44.040 | and droplets moving around, so there is some research, not perfect, but can tell us a bit

00:24:52.760 | about what kind of masks work. And interestingly, this particular example showed that just simple

00:24:57.840 | surgical masks were just as useful as N95 masks for influenza virus. Virologists are

00:25:06.640 | telling us exposure dose matters and therefore masks help. And because when you have a mask

00:25:15.640 | it means, you know, less of those droplets end up in your nose and mouth and if that

00:25:19.920 | happens it keeps the peak viral load lower and so it does less damage, the immune system

00:25:24.480 | starts responding earlier and can flatten the curve. So, you know, there's a big difference

00:25:29.640 | between, at least in the West, the official information that's generally being passed

00:25:34.880 | on and the information that we can see in research papers and from, from experts. Having

00:25:41.400 | said all that, it's certainly true that there isn't enough masks for doctors. And so one

00:25:46.800 | of the things, you know, if people can find ways to acquire more masks, create more masks,

00:25:55.120 | figure out better ways to reuse masks, I don't know, like this is something everybody is

00:25:58.760 | trying to figure out right now because it could make a big difference, particularly

00:26:02.920 | because it's becoming increasingly clear that people who are infected, so most people who

00:26:11.040 | are infected don't have symptoms. But the problem is that it's, it's seeming pretty

00:26:15.560 | clear that those people are spreading the disease. So again, there's a lot of official

00:26:19.480 | advice saying, you know, if you've got symptoms, stay home, if you've got symptoms, self isolate,

00:26:26.640 | so forth. But the kind of the data we currently have does not show that to make sense. It

00:26:34.360 | seems actually like regardless of whether you have symptoms, if you have the virus,

00:26:40.480 | you need to be not spreading it. So that means again, more testing, more masks, more self

00:26:47.600 | isolation, particularly if we have testing like super wide testing. Another thing I wanted

00:26:55.680 | to talk about was young people. People under 50 probably won't die. Some do, the kinds

00:27:05.360 | of people who might, you probably know if it's you because they're kind of immunocompromised

00:27:09.880 | people and so forth. But you know, let's, let's be honest, you probably won't die if

00:27:15.040 | you're under 50. You might pass it on to your grandparents or parents and kill them, or

00:27:21.680 | your colleagues at work. But the other thing to mention is these kinds of diseases have

00:27:28.160 | tend to shown long term impacts for people when they get older. So less lung function,

00:27:37.080 | neurological symptoms, depression, you know, there's tended to be for this kind of class

00:27:43.040 | of viruses, there's tended to be pretty debilitating lifetime impacts even for young people. So

00:27:53.360 | you know, if it's not enough to say, hey, stay home so that you don't kill your community

00:27:57.960 | and family, then you know, maybe okay, we'll stay home so that you don't make your rest

00:28:03.680 | of your life much crappier than it otherwise would have been.

00:28:08.720 | I was also just going to add that you probably have more coworkers and acquaintances who

00:28:13.080 | have chronic illnesses that you're not aware of, because many people don't share this information

00:28:17.040 | out of a very reasonable fear of discrimination.

00:28:20.200 | Okay, there is something which is terrifyingly complicated, which is in the 1918 flu, the

00:28:29.720 | second and third seasons were much worse than the first. That's what a lot of people are

00:28:35.080 | worried might happen. I don't think anybody has a good idea really of what to do about

00:28:40.080 | this. It was because of this that the UK did the original disastrous response of, let's

00:28:51.600 | not close down events, let's not close down schools, and now there as we saw there in

00:28:56.600 | a lot of trouble. And indeed that same group, Imperial College, is now saying in the U.S.

00:29:06.360 | there's a chance that if you isolate cases and household quarantine and social distancing,

00:29:12.280 | we might be able to, for the initial period, we might be able to keep things under the

00:29:18.080 | red line. This is a super concerning, dangerous kind of approach, because if we don't close

00:29:31.840 | schools and universities and we don't calculate the exponentials right and nobody actually

00:29:36.880 | knows, then we won't find out that we failed to get this right until we're well up along

00:29:43.760 | kind of the black line, you know, particularly because we can't really do case isolation

00:29:49.120 | because we don't have testing capacity. So this kind of modeling that's being done to

00:29:53.120 | say like maybe we don't have to do too much right now is not based on reality of what

00:29:59.920 | we can do, and it's also kind of assuming data we don't have. And most importantly from

00:30:04.640 | a Bayesian point of view, we don't have confidence intervals here. So if you actually do this

00:30:09.400 | modeling with confidence intervals, you'll realize that kind of probability that each

00:30:13.680 | of these things is successful, the number of cases that we actually find through testing,

00:30:18.780 | the bounds we have on our understanding of what the death rate is and so forth, basically

00:30:24.000 | mean that the kind of confidence interval of this brown and green line kind of go from

00:30:29.000 | down here to way up here. And of course the utility of each direction varies a lot. So

00:30:37.680 | I kind of just wanted to mention that when you look at this kind of modeling, I haven't

00:30:44.000 | seen any done yet with that kind of full confidence interval approach and considering the utility

00:30:53.040 | of each of the possibilities. And so I'd say be very, very careful of this and that's what

00:30:58.320 | the UK got so very wrong so far. Finally, I wanted to talk about stuff that you can do.

00:31:07.240 | One thing you can do is not stress about the supply chain. There's plenty of food. There's

00:31:12.720 | plenty of toilet paper. The supply chain is not threatened. And by the way, I'm using a

00:31:20.880 | lot of slides here from other people without credit, which I normally wouldn't do, but

00:31:25.760 | I wanted to put together something that's very up-to-date and very quick and I apologize

00:31:31.120 | but I'm trying to put people's names at least including them where they were already on

00:31:35.960 | the slide. So this is from Michael Lin. So over the next few weeks as people settle down,

00:31:46.080 | you should see your shops. I mean, I'm still already starting to see shops starting to

00:31:51.120 | see goods again. This is an opportunity to take advantage of some casual racism. You'll

00:31:57.080 | find that Chinese Asian grocery shops are absolutely stacked right now. So Rachel and

00:32:02.880 | I went shopping at one the other day and got everything that we wanted straight away. So

00:32:07.920 | yeah, don't worry about that. It is hard to find hand sanitizer, which is super important.

00:32:16.520 | If you can find something with at least 60 or 70 percent, sorry, with something with

00:32:20.400 | at least 90 percent alcohol, you can mix it up with glycerin or aloe vera and create your

00:32:24.600 | own. There's actually an official WHO sanitizer recipe. Here's another one. It is a bit difficult

00:32:35.640 | to kind of handle things properly without hand sanitizer. I did post on Twitter a list

00:32:41.760 | of Singapore's official list of chemicals which kill the virus that you can find in

00:32:48.060 | basically household cleaners. And I found we had like five household cleaners at home

00:32:52.160 | of which four turned out to have the ingredients recommended. So you might be surprised at

00:32:57.720 | how many things you can kind of use. The tricky thing here is thinking about how to stay safe

00:33:09.240 | is Rachel and I kind of play this game. I think of this computer, think of it like a

00:33:14.800 | computer game. We call it the Code Red game, where somebody else has sat at a table and

00:33:20.440 | they were infected with COVID-19. And so you then sit at that table. So that table, we

00:33:27.400 | would say it's red, right? And you have to assume that everything that you haven't personally

00:33:33.320 | cleaned thoroughly is red because you don't know if it is or it isn't, right? So that's

00:33:38.720 | your red. So you have to assume that's red. So you go to a table at a restaurant, you

00:33:42.760 | sit down and your hand touches it. So now think of a computer game where like every

00:33:47.240 | time something touches a red thing, it becomes red as well. So your hand touches the table,

00:33:51.960 | it's now red, by which I mean potentially infected. You pick up your phone, bing, that's

00:33:59.080 | now red. You take the phone home, you put it on your desk, bing, that's now red. You

00:34:05.360 | pick it up off the desk and then you put your computer where it was sitting, bing, that's

00:34:09.240 | now red. Right? So that's kind of like how you have to think about this. And so the thing

00:34:13.120 | is now you then say like, oh, forgot to wash my hands after I went out to the restaurant.

00:34:20.440 | So you clean them and hands are now green. But this kind of now whole chain of things

00:34:26.160 | which touch things, which touch things, which touch things, they're all red. Right? So this

00:34:30.040 | is one of the reasons that kind of everything's taking us a long time at the moment to just

00:34:35.600 | get around the place is we have to make sure that everything red becomes green in a way

00:34:44.160 | that we don't make other things red in the process. And you know, sometimes we forget

00:34:48.200 | and then we realize there's a whole chain of things that happened and then we have to

00:34:51.040 | go and do all of those over again. But if you get good at this, then you can kind of

00:34:57.720 | like, as long as you stay distance from people, you can kind of go about your day to day life,

00:35:02.360 | you know, you can go and get takeout or whatever. But for this to work, you need hand sanitizer.

00:35:09.920 | It helps to have proper, what are they called? Night trial gloves. Night trial gloves. And

00:35:18.600 | of course, you know, cleaning fluids that you can just spray on a piece of paper on

00:35:22.400 | a piece of, you know, cleaning paper and wipe things down. So like when we get shopping,

00:35:29.720 | we just pop it on the front deck, wipe down everything with alcohol or bleach before we

00:35:35.600 | bring it in and so forth. So it's a hassle, but you know, it's, you can totally handle

00:35:44.640 | the situation this way. There's a lot of opportunities for data scientists to help. People have been

00:35:51.840 | reaching out to me and so this gentleman said like, hey Lombardi is, I mean Lombardi's really

00:35:59.080 | the place is nearly struggling the most. We founded a nonprofit group of data scientists

00:36:03.080 | where we're trying to help. So you can go to defeat covid19.org to find out what kind

00:36:09.760 | of help they need. There are ways to deal with, you know, a little bit some of the lack

00:36:16.680 | of testing to at least figure out a bit of what's going on, for example, by, so if you

00:36:23.040 | don't follow Eric Fagel-Ding, he has lots of great information about this. Folks like

00:36:29.840 | him have been looking at what's the kind of reporting of flu like symptoms based on just

00:36:35.720 | the regular reporting that we have coming out of hospitals and comparing that to the

00:36:40.320 | amount of flu diagnoses. And you can kind of get creative like that to figure out like,

00:36:45.720 | oh maybe the difference is because of covid19. And the bad news is based on that kind of

00:36:51.320 | back of the envelope analysis, things look much worse than the official figures. Pete

00:36:56.760 | Skomrock kind of had the idea, you know, as other people have had similar ideas, hey maybe

00:37:01.480 | things like the Google business system that tells you when it's busy or not could be kind

00:37:06.920 | of reused to try and help do some kind of pandemic social distance measurement stuff.

00:37:14.280 | This kind of stuff we've got to be super careful of creating some kind of dystopian surveillance

00:37:19.040 | society that we then can't untangle again at the end. But you know there's opportunities

00:37:25.120 | for data scientists and kind of software developers to think creatively about ways to help fill

00:37:31.560 | in these gaps, fill in the gaps around testing, around keeping data available.

00:37:48.720 | So one of the things to think about is how can we help improve testing. The data that's

00:37:55.920 | available is suggesting that, so this is from science, that the vast majority of infections

00:38:02.020 | were undocumented. And that's led folks like Jeff Dean to ask what's been the economic

00:38:08.760 | cost of delaying testing in the US. I reckon it might be hundreds of billions or trillions

00:38:13.240 | of dollars. You can see the difference between South Korea and United States that both saw

00:38:18.000 | their first cases within a day of each other. South Korea tested massively. United States

00:38:25.800 | just starting to test super late. So you know if you can help us find ways to do more tests

00:38:32.960 | that would be super helpful. Also I mean you know we're data scientists but there's people

00:38:43.600 | with all kinds of backgrounds here. If you've got a background in hardware or 3D printing

00:38:47.720 | people are saving lives here. These folks bought a 3D printer to a hospital in Italy

00:38:53.360 | and there are people now breathing who previously couldn't thanks to that 3D printing. The world

00:39:01.920 | being what it is these people have now been sued by a patent troll which I guess was probably

00:39:06.480 | predictable but yeah. I guess a lot of people are trying to get creative and sometimes working

00:39:12.760 | around regulations as necessary to save lives which hopefully will turn out to be a good

00:39:19.680 | thing not a bad thing. Ventilators is one thing a lot of people are trying to build

00:39:24.800 | at the moment so maybe you can help. So the director of the Johns Hopkins Center for Health

00:39:32.960 | Security is saying we need a wartime mobilization to make mass number of ventilators and to

00:39:39.560 | get enough oxygen. So kind of wartime mobilization means a lot of creativity and a lot of you

00:39:49.000 | know not being quite as precise about things as we might have used to have been but just

00:39:55.120 | trying things out and seeing what works. Mind you the first thing we could do in the US

00:40:01.000 | is just buy the capacity we already have. There's actually an opportunity to ramp up

00:40:05.560 | production fivefold right now but the US just hasn't actually ordered the product. So make

00:40:11.440 | of that what you will. Some of these folks reached out to me and asked for help finding

00:40:17.440 | people to help with open source medical supplies. They're sharing CAD files and there's now

00:40:23.800 | been responses all around the world to that and that's going pretty well. Okay so that

00:40:35.880 | is my little spiel about COVID-19. Rachel is there anything you wanted to add? I was

00:40:45.720 | planning to yeah. Okay so for those of you that aren't interested in that topic I apologize

00:40:52.160 | for taking up your time hopefully some of you found that useful. For those of you that

00:40:57.480 | did and are interested in contributing I have created a forum category for this. Here it

00:41:10.200 | is COVID-19. Now I think there's a lot of opportunities to matter I wouldn't say I think

00:41:17.440 | maybe there's opportunities to apply some of the stuff we're learning in this course

00:41:21.320 | to do projects that are related to COVID-19. It would be nice wouldn't it to spend these

00:41:26.680 | two months learning a new thing but also meeting other people working together on something

00:41:32.080 | that seems important. So if you're interested in doing that this is the place to put it

00:41:40.440 | because it's not a closed category like our course category is. If you do of course don't

00:41:45.840 | like say hey I've just learned this thing in the course to people who aren't in the

00:41:50.400 | course because they're going to feel jealous or whatever. So just remember you're talking

00:41:54.400 | to folks who aren't necessarily part of the course but yeah this is a great place to help

00:42:02.620 | write screen scrapers to put data sets together or set up automation of things or you know

00:42:10.520 | do a survival analysis that incorporates uncertainty in the way that I don't think I've seen done

00:42:15.360 | yet or you know so on and so forth. So hopefully that turns out to be useful to some people.

00:42:29.240 | So I'm back again and I just had a quick huddle with Rachel and we decided to put this thing

00:42:34.080 | online publicly not just for our course. So I figured I'd better just pop in again to

00:42:41.000 | folks that aren't part of the course and are watching this to fill in some gaps here. When

00:42:48.720 | I was talking about the forums I'm talking about this website here forums.fast.ai we normally

00:42:58.120 | use it to talk about deep learning not to talk about COVID-19 but we can talk about

00:43:03.720 | COVID-19 as well we've got a category for it there. If you're somebody who's not a data

00:43:09.920 | scientist but is interested in having a conversation about COVID-19 particularly if it's something

00:43:14.480 | that's more on the technical side or data driven or practical which is kind of our things.

00:43:22.840 | We would love to hear from you even if it's not a data science perspective if you're a

00:43:26.560 | 3D printing person if you're a chemist who knows good ways to create oxygen if you're

00:43:31.640 | somebody who knows a good source of reagents for testing whatever we would love to hear

00:43:39.600 | from you there. So thanks very much for listening to this extremely non-deep learning related

00:43:46.440 | video about COVID-19 and I guess the other thing to mention is you may have no idea who

00:43:53.760 | I am. I'm Jeremy Howard. This is my Twitter @jeremyphoward and Rachel is @mathratial_ratial.

00:44:06.480 | Don't forget the P in Jeremy P Howard. I'm going to be pretty busy trying to create this

00:44:12.360 | deep learning course over the next few days so I don't promise I'll be putting much on

00:44:16.200 | Twitter but I will do my best to share things which I think are of interest so thanks a

00:44:21.360 | lot for listening.

00:44:22.040 | [BLANK_AUDIO]

A practical summary of the covid-19 situation

Chapters