This analysis is difficult because it is related to testing capacity. We all know testing capacity was limited early on. It remains limited, but is growing. The faster capacity grows, the steeper the curve will be. Want to have a less steep curve? Lower the amount of testing.

In another comment, you claim looking at the curves leaves out politics completely, but it doesn't. Any politician can make the situation look worse on the curve, but better serve their people, by increasing testing. They can also make their curve look better, but do worse for their people, by claiming their state is unable to purchase tests. This can be a very political game.

They are more reliable than the simple number of cases but in the Italian province of Bergamo (and I suppose in neighbouring Brescia, too) lots of people have started dying before the ambulance even arrives at their homes. Those people will not show up on the official statistics of covid victims, at least not now.

A similar thing had started happening in Wuhan once the situation had become really dire, and the closest valid indicator people could find was to look at the number of incinerators and how busy they were (mostly looking at the number, because the incinerators were busy 24/7 anyway).

https://www.worldometers.info/coronavirus/country/us/

Here's a good article on estimating current number of cases from number of deaths:

https://medium.com/@tomaspueyo/coronavirus-act-today-or-peop...

As for that Medium article, it’s written by a tech exec with zero education or professional experience in anything related to medicine or the life sciences, so I would treat it with a heavy dose of skepticism. Many of the deluge of preprint articles from academic institutions written by epidemiologists or public health experts would likely be a far better source of info.

https://fivethirtyeight.com/features/infectious-disease-expe...

Responding on March 16 and 17 to the question, "how many cases will the US report by March 29", their median prediction was 19,000. As of this moment, early on the 21st, that figure has ALREADY been exceeded (per Worldometer). Based on current growth rates, it appears virtually certain that the actual figure on the 29th will exceed the "best-guess" prediction of every single expert surveyed, and the "high-end" prediction of all but one or two. They've completely whiffed a prediction just a few days out.

This is a novel (no pun intended) situation, even for epidemiologists. To understand it, critical thinking skills are at least as important as subject matter expertise. My read is that the Medium article represents some of the best critical thinking on the subject so far.

https://www.imperial.ac.uk/media/imperial-college/medicine/s...

The numbers may be wrong in that article but the math itself seems straightforward. It's certainly true that we need more tests to be sure of the numbers.

Here's a pretty interesting simulation from a research group at the University of Basel: https://neherlab.org/covid19/

I guess you're suggesting that when people die of Covid-19, we'd know it and count it?

That's not true.

First, CT scans (and other expensive tests) are not generally performed on people who have died of pneumonia, apparently due to the flu (or other causes). Second, the radiologists reading those studies would have to train to make the distinction.

Now, going forward, we have an increasing need to know whether the virus that caused the pneumonia that killed the patient was Covid-19 or not, so maybe such testing will become routine. (I doubt it will be CT scans though, because there will be much cheaper and reliable ways to do it.)

But so far we've been undercounting Covid-19 deaths for the same reason we've been undercounting cases: lack of testing.

But it's a good point that we might have more deaths than we've counted, and this might just mean we're further along that we realize. The rate of growth is probably not just an artifact of more testing; it's similar to the growth rates in other countries that haven't gotten control of the disease. And we definitely don't have it under control.

We can compare the usual number of people who die of pneumonia with the current one, and any significant difference can be attributed to this novel virus.

That works only at the beginning.

But not all symptomatic people receive tests.

Knowing those two number (number of symptomatic people who would have qualified for a test, were it available; number of test performed) could paint q better picture.

As you can imagine, there's more to epidemiology than this. I'm a big fan of letting the qualified people do their work, while we do ours: shut up and do not spread misinformation.

In fact, at the risk of being rude ill even go so far and say that it's quite arrogant to pretend to pontificate about this when your knowledge about this issue is 0, and also when one major problem in combating this pandemic is the "epidemic of misinformation".

I made it clear the limits of my knowledge - which may still not excuse me, however I have to ask if you are an epidemiologist and if so, please tell me how to model the data better. And I don't mean randomly google up some papers to dump here and pretend you are an expert (unless you are).

How far can I extrapolate e^x before it becomes useless? What are the confidence limits I can use in the extrapolation? How much can I approximate the logistic function with an exponential, when as now, the cases are currently a small fraction of the population? Why is this site http://nrg.cs.ucl.ac.uk/mjh/covid19/ done by experts using log axis any different from mine using curve fitting with exponentials? Why does http://nrg.cs.ucl.ac.uk/mjh/covid19/#w say the USA is 11.5 days behind Italy whereas my amateur attemp suggest ~9 days? (and yes, I wish I'd seen these graphs before I posted) What else have I missed? Criticise freely, with reason, don't just accuse me of misinformation because I posted my fears, which others have done too, only I did the maths.

Now please excuse me while I look at the logistics curve and compare its early stages with exponentiation.

There is no "actual data" in your comment above. There is a claim that you make that you have graphed some data, but no graphs and no data.

Also, it doesn't take an epidemiologist to know that a non-epidemiologist should not play at epidemiology. The other poster doesn't have any responsibility to tell you "how to model the data better".

There is in http://nrg.cs.ucl.ac.uk/mjh/covid19/#w which I said matches what I've found rather well.

> The other poster doesn't have any responsibility to tell you "how to model the data better".

Yes he does,

But I'll try to help. You're looking at a graph of confirmed cases, in other words, positive test results.

That may be due to an increase in the number of infected people; or because the number of tests being performed is increasing, and therefore the percentage of infected people reported as confirmed cases is increasing; or because some positive test results (even with confirmation) are false positives. And there may be other causes that I don't know, because I'm not an epidemiologist either, just a statistics grad student.

Figuring out the the contribution of each cause is a job for the professionals.

There may be data there, but if I understand you correctly you did not have it when you made your first post.

It is therefore inadmissible evidence as to whether your first post was a reasonable thing to write.

> Yes he does,

If you have decided you want to be an epidemiologist, or at least to learn their skillset, whose responsibility is it to fulfill that desire?

I don't think there is a reasonable answer to that other than "mine."

With this in mind I think we should stop nitpicking on non-specialists who try to raise awareness of this pandemic.

Yes there might be some differences from what a specialist would have done.

Due to the speed at which the situation evolves and the inability to test everyone daily there are lots of unknowns so even a specialist may not be able to give a more much precise figure, everything is going to be a rough estimate.

This is a very serious situation and it's better to err on the caution side, people's lives depend on everyone doing their part.

Non-constructive criticism of people who raise awareness will erode people's buy-in and people will die because of this.

May I ask, what area do you work in, numerical, discrete, logic, other?

Rather than admonish people to shut up, it would be more constructive to point out the flaws in their methods, so we could all learn and make better judgements.

Yes, but that's the thing isn't it? We're not informed, and fitting ke^x to data doesn't make us informed.

This reeks of Dunning-Kruger. I get that the intention is good, but we have to have some humility and not go all "hold on! I'll save the day!"

> and fitting ke^x to data doesn't make us informed

Well, yours is an interesting post but let me ask if curve fitting doesn't make us more informed then what does? There's an update to the data for the US in the past few hours and it doesn't improve things.

I very carefully stated out my limitations, and then pointed them out again when someone critiqued my approach (see edit on https://news.ycombinator.com/item?id=22645793). I have done my best to accept my limitations, but still I get people like you hammering me.

So the question is why are we even posting anything on HN? Bottom line is what is HN for - non-factual debate, or factual debate?

I strongly get the impression that much criticism isn't based on my admittedly trite analysis but because a lot of people just don't know how to handle really bad news and are trying to blank a nasty reality. Doubly so when I'm accused of of "hold on! I'll save the day!" which is purely your invention.

No, you just made a large number of flaws in your initial analysis, and it made people despair of going to the effort to point out where you went wrong.

E.g. https://news.ycombinator.com/item?id=22648730

You're producing a future estimate, assuming that all properties of your dataset are consistent (they aren't) and will continue to hold (they won't).

As a few examples: (1) varying ratio of confirmed cases to actual total, (2) % increase in testing per day, day-over-day, (3) varying time offset according to disease progression and symptoms presentation (aka testing).

You seem to be of the opinion that any information is better than no information. In scenarios like this, I'm of the counter opinion that bad information is actively harmful and inferior to no information.

Seeing as you don't have access to my analysis, you can't say that.

> You're producing a future estimate, assuming that all properties of your dataset are consistent (they aren't) and will continue to hold (they won't).

This is just vague stuff to put me down. I worked by confirmed cases in publicly available data.

> varying ratio of confirmed cases to actual total

I'm not sure what this mean but I went by the total confirmed cases, day by day. As for your 'increased testing', I'm fully aware the US is finally getting going.

> varying time offset according to disease progression and symptoms presentation (aka testing).

Why would the progress at different rates at different times in different individuals make any difference? I'm going by confirmed cases anyways so unless you know of a new strain with different characteristics that's starting to distort the figures (and there are possibly 2) then it makes no difference.

No, save your slapdowns for the people who push garlic as treatment. You are objecting to me for psychological reasons, not logical ones. You can go to the wiki page and see the growth day by day but you don't want to because it's bad news and you don't like that. Well, get used to it. Your life is going to change, nothing will stop that and the sooner you face it the safer you and your country will be. Hard times ahead. Good luck.

Like other people have said through this thread, it's rather simple: you don't know what you are doing, and at this time bad information is worse than no information, and it will hurt people. And reiterating my remarks on my first post: stop being so arrogant, this is really the only word to describe your attitude as far as I'm concerned.

And "I am a bloody genius [...] by plotting a bunch of points in a log scale" - I actually used the words 'admittedly trite' about what I did.

Whatever. This gets us nowhere. Your profile says you have a background in physics and logic/automated reasoning. I'm just a general purpose back-end dev. We're facing something that may kill more people than WW2, let's concentrate on that. Can we work together to do something useful? Thoughts?

Many doctors lack training in epidemiology. Many epidemiologists aren't actual doctors.

Healthcare in general is horrible with generating accurate raw data (JHU data comes often from scraping websites and the CDC itself has stopped publishing number of daily tests weeks ago). <-- this in fact is a technical problem not a medical one!

You don't have to be a Dr. med in order to see gaps and problems with statistics and a large part of the conjecturing actually comes from gaps and flaws in the sources and the way data is presented. A background in an isolated discipline (only med, only statistics, only Math, etc) is not enough to harness and make sense of complexity.

Finally research papers themselves are all pre-print and not peer-reviewed which is a trade-off in accuracy for speed.

We're all finding solutions as we go and the more eye-balls from different discipline this gets, the better the outcome!

Yes, and this is why doctors aren't being asked to do epidemiology and epidemiologists aren't being asked to care for patients.

And (likely) sofware engineers are not asked to do either.

I too am tracking the numbers. I seriously and honestly expect them to become unavailable- or at least very hard to get to, if things go very very bad.

I expect to see a bump in the "infection rate" in the next few days as testing becomes more widespread. In fact, I expected to see the start of that already, but if it's happened it's not obvious. The number of cases continues to double every two days and change.

throwaway_pdp09 isn't spreading unnecessary panic and their attitude has been anything but arrogant, pompous, or dogmatic. They are very open in admitting that they are not an epidemiologist. Saying f(x) = 9 and asking why their 'f' is wrong, and wanting to understanding its shortcomings and how to improve it, seems entirely fine to me. You are going to have to be far more specific on the harm caused by a lay-person projecting that we are fucked in 9 days, vs an expert's opinion that we are going to be fucked in 11.

If it helps, pretend this question is being presented and asked at TA office hours during Epidemiology 101 at university by someone who is considering going into epidemiology.

We are all here to learn.

Meanwhile, there's actual, harmful nonsense out there to rail against. Half-remembered gossip misinformation that spreads like an infectious disease with an R0 of 5. Complete crap, like holding your breath for 10 seconds will test for Covid-19, or that essential oils cure Covid-19 or that its eventual vaccine will cause autism. Fight that misinformation, not people who's latent interest in epidemiology has been awakened. Who knows, a reader here could be deciding to get a degree in epidemiology because of an interest sparked because an exp curve is too simple.

I'm just trying to make sense of things, same as anybody else. I don't think anybody's fucked either in 9 or 11 days, I actually have a great deal of faith in the United States and yes, I mean it, but bad decisions from high will cause a terrible amount of pain too soon. I just want to wish the US the best because they'll need it now more than any other time, probably in all it's history. I don't think WW2 stressed you as much as you are about to be now. Buckle up yanks and pull together, I believe in you! You will get through this.

Yeah I knew by Sunday that tomorrow we're hitting 30,000 cases in Spain. My assumptions worked. And it's just the tip, because I don't know anything about whether there's been randomized sampling, so who knows about the normalization.

I just can't believe this. No one listens and this is in-your-face obvious.

My key takeaway was that once you see the inflection point, you know that you’ll have roughly 2x the number infections by the time it’s done.

Can you clarify. I can modify the top post (mine) if you're saying I'm overstating things. Happy to be wrong here.

(what I thought you might have meant was a criticism of me misusing maths and you were getting pissed off telling people it was a sigmoid not an exponential).

(edit: good catch though)

(Of course this works as long as you aren't seeing any peaks)

oefrha, for some reason I can't reply to your comment just yet so I'll post it here: Of course. This is just my own criterion when I start with a bunch of data I know nothing about. Is there a complicated law behind them or maybe not? Am I seeing noise or there's a signal? You can fit anything smooth with a low order polynomial, so maybe there's a differentiable function right in front of you. It's just something I do and if you see no reason for it I might even agree with you on that. It never ends up in the model, though. It's just something I find useful at a preliminary stage.

OK, We're not being responsible doing this and I'm not qualified to do it. I just have been fitting the data and maybe I've been lucky and I'm just an idiot. Having said that, this is also my problem, I have right now a hotspot in the vicinity and I'm worried. I just want to have a quantitative idea of what's going on. Now it's your job as experts to explain what's going on to the politicians and make sure they hear and they can do the right thing. I wouldn't be worried if the messages had matched the numbers when this started, they didn't. At some point we need someone to tell us "we're at this point and tomorrow we'll be at that point". I shouldn't feel compelled to find out by myself what all this is about. I feel awful by posting all this, I'm perfectly aware that non-experts in a field spout nonsense, but this is a very nonsensical situation and I'm tired of hearing we've been doing fine.

I stopped trusting any of my rudimentary models by Sunday+incubation, because conditions changed then. Which data, why I trusted them and the set of assumptions taken to consider them representative for the shape of the time series/distribution, that's something I don't think I should share here at this point. In any case I'm used to extract some information from imperfect data, you approximate and estimate. I mean there was by then 12, then 16, then 30 positives in a nursery home next to my door, and no lockdown measures, now there's 100, an improvised hospital there and I'm not allowed go out except for buying groceries. Two weeks ago I needed some picture which I wasn't getting anywhere, the alternative was freaking out after comparing what the authorities were saying with what I knew about the situation. Now they are making sense and the measures in place too, so I expect it getting much better soon.

I don't want to go into epidemiology and I want to respect and trust blindly the epidemiologists, but in any field you expect the professionals having predictive skills working before your eyes, that's their job. Hearing that we should wait for the post-mortem to know where we are now, I mean, come on... And now I'll just shut up.

Scary that the US (especially some areas) and Switzerland are starting to track above Italy:

http://nrg.cs.ucl.ac.uk/mjh/covid19/20mar2020/covid-us-norm....

Clearly something completely different is going on there than in many other countries, and it seems they have an extremely effective and well organised testing system. They're testing about 160,000 people a week. In comparison, the UK to date has only tested around 50,000 people in total so no wonder our case numbers are far lower, even though we have double the number of deaths.

So Germany is an outlier, but it just goes to show even for other countries you can't just compare the curves and draw any definitive conclusions without also looking at the testing regime. This is where the experts with their detailed population models and simulation tools come in.

Also there is the purely anecdotal hypothesis of a cultural difference between for example Italy, Spain and Germany. In the southern countries it seems like older people have quite an active social life compared to Germany.

Take numbers from a region where testing is frequent, say the death rate is 2%. Then scale the numbers in a region with lack of testing to match the 2% death rate.

Not perfect, as there are external factors like hospital capacity/quality and existing population health.

However, it should give a closer approximation of total cases where politicians fail to provide adequate testing.

Do you have any statistics pointing to obesity as an increased risk factor for a respritory disease?

https://www.youtube.com/watch?v=0M4kbPDHGR0&feature=youtu.be... turn on google auto translate

TLDR: Any comorbidities present at the time of death are reported as the main culprit. In other news https://www.thelancet.com/action/showPdf?pii=S0140-6736(20)3... Table 2: Treatments and outcomes. Sepsis 100%, Respiratory failure 98%. There you go, nobody dies from the virus alone, problem solved, 20x less deaths than comparable Spain.

I keep asking people about this because I'd for sure would like to know if it were happening, but the only two sources I've seen sounded like speculation by Italian officials that at least as absolute as it is reported in other languages does not match the actual reporting here as far as I can tell.

Look for ILI (Influenza-like Illness) stats which were appreciably higher in January. The obvious inference is that many had Covid-19 before it was recognised as such.

This might be a good thing.

I know a person who back in January acquired something like pneumonia and a bad fever, went to the hospital for awhile and tests were negative for flu, etc. doctors just didn’t know what it was. The person has COPD however.

But anecdotally I’ve noticed quite a few more colleagues over the last few months out sick “with the flu or something”. No idea.

They’re still only testing people with symptoms. We need a randomizes test and a test for antibodies to have any idea how much it has already spread.

If it’s as contagious as they say and spreads in asymptotic people then why wouldn’t there be a significant population that has acquired and recovered from this? Many probably with very mild symptoms.

I could just be in denial right now as well.

The only places I've seen the first US infection being dated that early are conspiracies on Chinese social media that claim the origin of COVID19 to be the US instead of Wuhan. The only way that theory works is to date the first infection in the US and claim that Americans bought it into China during the Wuhan Military games. That theory is easy enough to debunk with common sense because if the US had COVID first the hospitals would be completed overwhelmed before China's was, but it's not the first time wishful thinking's buried common sense in Chinese history.

I don't know anyone that credibly believes it was in the US in November. We did, however, that something was amiss that early, and if the US had a pandemic response team who's job was to see something like this coming, we could have been getting testing ready since then.

Or the fatality and hospitalization rates have been overestimated due to severity bias.

We know this to be true, actually. We just don’t have good estimates of what the true rates are.

It wouldn’t be shocking at all to find out the virus is already widespread, and only slightly more dangerous than influenza.

Even in Italy, if there is widespread infection, the ~4k dead would reflect a low mortality rate.

In italy in hard hit city, they were unable to cremate people fast enough and army had to move bodies. Influenza does not cause that.

The error bars on CFR estimates run from fractions of a percent to a few percent. In countries that have tested large numbers of people who are not severely ill, the case fatality rates have been at the low end of that range.

This isn’t “shocking to epidemiologists”, it’s shocking only to people who dismiss basic facts and try to use irrelevant distractions (like cremation rates) to incite fear.

Epidemiologists do count with overwhelmed hospitals and services, excess deaths due that that, excess deaths on unrelated heath problems. It not just something they talk about, they have it in their models too. This does not happen with influenza, because it is easier to control. Irrelevant distraction there is Italy or Spain lately.

Epidemiologists do talk also about costs after epidemic ends, in treatment of people who will need care for long.

What amateurs do is to ignore all the above, pick up case fatality ration in well run non-overwhelmed health care system and pretend that is situation same as influenza to feel good.

Not that I have heard of. You’re going to have to cite some evidence here.

(They’re both respiratory viruses with similar R0, so your evidence needs to be pretty extraordinary to be convincing.)

Especially asymtomatic. You’d not be coughing or sneezing more than usual which must have some effect.

Just a thought. No one seems to have mentioned this idea.

https://www.nytimes.com/2020/03/10/us/coronavirus-testing-de...

Because all the effects are limited by the time it takes for them to happen. We know the speed of growth, it's much faster than the expected recovery times. From all known COVID-19 cases the number of recovered are still dominated by the recovered in China.

   China: reported cases: 81,008
          dead:       3,255 
          recovered: 71,740

   Whole world: reported cases: 282,395
                dead:      11,822
                recovered: 93,189

That’s exactly what happened to my dad in February.

What are you referring to? You don’t think the virus is already much more widespread in the US than currently reported numbers? I’m just making the observation that perhaps the large increases in US cases are more indicative of confirmation of already existing cases rather than necessarily being _new_ cases.

I don’t think that’s crazy at all. We had our first case sometime in January, and only until very very recently have we had any significant travel restrictions. It’s very contagious and for a significant number of people there are no real symptoms, or if there are, nothing life threatening. I don’t think it’s crazy at all that it’s been working it’s way around the US population for months.

Maybe to put it another way, if we could instantly test al 300+M Americans right this instant, would you be surprised if say 20M or so tested positive or had had it?

Yes, I would be very surprised. Given what we know about severity distribution that means 2M hospitalizations and ~200K-1M ICU cases that we somehow missed.

See table at bottom.

The problem isn't just that a lot of people will die because the health system is swamped. That's hardly good news, but when you have a high peak you have a lot of people ill at the same time. And that means you run the risk of losing a significant percentage of critical workers in essential services including food, utilities, logistics, and IT.

Some will die, some will "just" be very ill and unable to work for 2-4 weeks. If you're lucky the numbers of seriously ill will be low enough to keep things running. Beyond that you're on a scale from frightening inconvenience and breakdown in some locations to the economic equivalent of critical organ failure.

Pushing any country through this is a huge and utterly irresponsible risk.

The number of "cases" reflects at the moment the capabilities of the whole medical system in each country (it's dependent on who is selected to be tested at all) and the tests used. The number of deaths hides less. Also, even if the specific numbers can't be compared 1 to 1 across the countries, the characteristics of the curves can.

The growth is fast but it is comparable, and the measures take time to affect it.

Of course that’s assuming the labs doing the testing are doing it first in first out.

It's a risk, but the amount of hardship all the people who lose their jobs and business will endure if an indefinite lockdown is implemented is an absolute certainty. Poverty has lasting, lifelong negative health impacts.

What's more sure, in most of the countries in the world, is that if the tests are available at all, and the patients happen to be admitted to the hospitals with the symptoms, they will be tested, and the percentage of deaths will be comparable in all countries, and we should be able to assume that the hospitals will report the deaths of the patients tested positive.

In short, now that the numbers of deaths increase, these numbers reflect better the issues we are facing than just the reported cases.

We should estimate the number of "comparable" cases across the countries or areas with different capabilities from the numbers of deaths, at least for the countries and areas which have at least enough tests to test those admitted with symptoms and report the dead, and once these numbers aren't too low to be too noisy.

I've demonstrated some of my simple calculations in my other messages, so you can check these too.

How come? "Distinct outbreaks" just means the older state was underdiagnosed, not that the characteristics of the pandemics change.

Local saturation happens fairly quickly. Having 100 infected in one place is very much different to having 1 infection happening 100 different places when you look at it after a week

When you're looking at the "country" view, the "older state" gets lost in the transmissions happening from people crossing borders. So what I mean is, people have been flying back to the states with infections spread out over the entire country while cross-border spread has been much more localized in italy

I still don't see the difference? Why do you think that the cross-border movement of the disease should be different in Europe compared to the US? Why do you think that the US should be worse off, if that is your point? Do you think people more fly back to the US compared to the people moving from country to country recently in Europe in the last period (including Italy)? Are you aware of how Europe functioned up to now? Many people worked in other countries, and they typically don't even have to fly to achieve that, especially not across the ocean.

Obviously it's only one day so time will tell if that means anything or a pattern occurs. As others have said, the stats for number of people affected is so closely linked to testing rates which have been increasing that this could be a more indicative metric to follow.

[0] https://www.worldometers.info/coronavirus/country/us/

Then, I fear, St Patrick's day partiers will only be measured by next Wednesday; 5 days incubation + 5 days to have test results.

considering how the UK handles the crisis it won't be far behind the US either.

For the numbers to have meaning, a stochastic sample of the population should be tested at random, not people coming forward because they think they’re sick.