Hacker News Reader

Comment by atemerev | Hacker News Reader

Here are my observations as an actual epidemiologist:

1) Quarantine / non-pharmaceutical interventions effects start to be visible after about 2 weeks from their introduction, and are fully effective after 1.5 months.

2) Daily new deaths growth lags from daily new cases growth about a month (which is somewhat obvious, but needs to be reiterated).

So, unless some new mitigation measures are going to be introduced _right now_, this record-breaking growth will continue for quite more than a few weeks, followed by the corresponding growth of deaths a month later. We are far away from the saturation point and any resemblance of "population immunity".

View on HN

On number 2: that makes a lot of sense; yet in places that have had peaks (e.g. new york, italy, UK, spain) the peak observed fatality moment is barely shifted vs. the peak positive testing moment - maybe more like a week or so, normally? In all the data it's a little hard to spot trends, especially with testing rates scaling up, but still the graph peaks are not even close to a month apart, even where testing rates don't appear to be too variable.

One comparison: https://datagraver.com/corona/#/?regions=italy:hex005FA2,spa...

Massachusetts has fairly well defined peaks too: https://covidtracking.com/data/state/massachusetts

What's causing this discrepancy? Clearly the disease progression isn't that fast, so... what then? Maybe the more vulnerable groups tend to get hit more quickly, and as cases ramp up either die earlier or lock down more rigorously or both, such that the less-vulnerable form an infection peak after the death peak has passed? Or is it simply that people get tested late in the disease progression?

Given how little the MA testing rate has varied, and that the positive rate peaks also are much closer than a month apart from the fatality peak, I don't think testing changes seem likely to be a primary cause of the (to me) unexpectedly short period between case peak and death peak.

The big difference is testing capabilities. All places hit by big waves early on suffered from poor testing and public awareness, so testing happened in a burst as the deaths were starting to become problematic. And generally those getting tested at peaks were the symptomatic/hospitalized. NYC is estimated to have been around 20% infected, so the actual infection count was highly undereported. It is important to also recognize that every location hit hard was met with immediate lockdown, generally followed well by scared populations.

The situation is far different now, with a good chunk of people actively rejecting even basic distancing measures. If you look beyond positive test numbers at the % positive you can see the rates are rising even faster beyond our testing capabilities. And the idea that only 20-30 year olds are getting the disease is wishful thinking - the highest risk situation is from household transmission, at over 80% likely.

Combine high infections with a general disregard of safety and those older populations are bound to be infected as young spreaders interact with their families, at offices, supermarkets, etc. I think this is why the death spike lags a fair bit further than expected.

” And the idea that only 20-30 year olds are getting the disease is wishful thinking - the highest risk situation is from household transmission, at over 80% likely.

Three things:

1) It’s been over a month since case count started to grow in the southern states. Hospitalization and deaths by day are not growing at all in FL or AZ, and only hospitalizations are growing slowly in TX:

https://covid-19.direct/state/FL?tab=daily

https://covid-19.direct/state/TX?tab=daily

https://covid-19.direct/state/AZ?tab=daily

2) It isn’t “wishful thinking”; the age distributions of positive results out of Texas clearly show that most of the people testing positive are younger, and it seems to be true in FL as well:

https://txdshs.maps.arcgis.com/apps/opsdashboard/index.html#...

https://www.google.com/amp/s/www.news4jax.com/news/florida/2...

3) It’s not very common to see multi-generational households in the US.

Maybe it’s true that this surge will eventually make it’s way to the high-risk, but it’s been over a month already, and that doesn’t appear to be true yet. At some point, predictions of the future have to be tempered by current evidence.

Although treatment has definitely improved outcomes, IMO, at least some of this effect has to be due to the changing demographics of testing.

In the early days patients essentially had to be far along in the progress of the disease to be tested, we simply had very little testing capacity.

Now we are doing many times more tests against a much wider set of subjects. So it is inevitable that we are getting more of the milder cases and cases earlier in their development. This also accounts for the so-called “shift” toward many more young patients.

The disease hasn’t changed. Current testing is sampling a different distribution of patients than previous months had been.

https://coronavirus.jhu.edu/testing/individual-states the positive rate of testing is indeed much lower now, suggesting testing is meaningfully broader.

On the other hand, it's only a factor 3, which is certainly nothing to sneeeze at, but still remarkable that that's enough to so strongly decouple mortality from infection, compared to earlier.

Thanks for the info. Could you, or anyone else, please explain why quarantine takes 1.5 months to be fully effective?

Because there is an incubation period, where people already infected start to develop first symptoms. There is a development period (from the first symptoms to the point where they are severe enough for the person to seek medical help — only at this point the case is registered). The introduction of quarantine measures is not immediately effective — there is a transition period of at least few days. There are response / case registration delays. And finally, it takes some time for the new epidemic regime just to be visible on the curve (a perceptive delay until the trend is understandable).

Do you have an estimate of the number of days for each of the delays you mentioned?

How do you explain the "sudden drop" in cases in new York [1] after reaching some 25% infection rate? Do you think herd immunity (ish) can be reached at a lower saturation rate possibly?

[1] https://www.nytimes.com/2020/05/20/nyregion/hospitals-corona...

That seems to fit into the theory that quarantine lockdown measures are most effective after 1.5 months.

It also fits (perfectly) an alternate theory that the population exhibits a wide variation in susceptibility, which lowers the herd immunity threshold.

https://www.medrxiv.org/content/10.1101/2020.04.27.20081893v...

What is the basis for that hypothesis?

That it takes that long for this cycle to die out --> (infection, exhibiting symptoms, infecting close contacts(who start this same cycle again), becoming unable to infect others)

> So, unless some new mitigation measures are going to be introduced _right now_,

You will be able to claim that any changes made by local and state governments in the next few days/weeks “did the trick”. Nevertheless, this is a fair escape hatch when the following prediction does not come true:

> this record-breaking growth will continue for quite more than a few weeks, followed by the corresponding growth of deaths a month later. We are far away from the saturation point and any resemblance of "population immunity".

You will be proven wrong, both about the coming “wave of death”, and about how far we are from the saturation point. I’m quoting this and publicly disagreeing today, so that you might remember later on, and maybe even acknowledge that you got it wrong.

Hi. I read the essay "Pandemic Woo" which is linked in your profile.

Is seems like your position is that:

--COVID response strategy based on invalid "post-modern science",

--the conclusion that the disease death rate is order of one percent is incorrect

--and therefore, the policy response is grossly disproportionate and harmful.

It seems like you are arguing about facts not values, and you think the disease death rate is not really order of %1. Is this what you mean?

I'm also asking because I have heard many heated statements about "post-modern" science around here as of late that wherein the speaker seems to treat that concept as common knowledge, and then walks off without explaining themselves, but I don't really understand what they are on about.

The first point is characterized correctly. We are reaping the whirlwind from a pernicious post-modern academic culture that has been thoroughly corrupted, even in the traditionally "harder" sciences to a degree. Here is a much drier, but still excellent, argument along the same lines:

https://americanaffairsjournal.org/2020/05/science-without-v...

The second characterization is way off base. All IFR numbers are rough estimates at this point, and in fact the number is just a model parameter that does not correspond to anything in the real world, so it's a bit hard to talk about any "fact" there. The disease evolves, treatment protocols change (early venting was a terrible mistake), and susceptibility may vary widely from community to community. Still, the most recent CDC estimate for this model parameter is 0.26%.

The third is a prediction, not an established fact. Right now there is no way to know what the long-term effects of our policy response will be. I have cited many anecdotal "bad" effects which (together with others) I believe will outweigh the "good" effects, but we will have to wait and see.

To your third point, a recent study funded by the Gates Foundation estimates excess deaths over the next 6 months caused indirectly due to the virus response;

> While the COVID-19 pandemic will increase mortality due to the virus, it is also likely to increase mortality indirectly. In this study, we estimate the additional maternal and under-5 child deaths resulting from the potential disruption of health systems and decreased access to food.

At the low end they predict 250,000 excess child deaths and 12,000 excess maternal deaths, and at the high end 1,150,000 excess child deaths and 56,000 excess maternal deaths due to decreased access to health care and food.

The number of missed vaccines and cancer screenings alone multiplied worldwide turns into a massive hit to public health.

https://www.thelancet.com/journals/langlo/article/PIIS2214-1...

Sure, a fair point, I am not the all-seeing all-knowing authority. You have mentioned "a few days" — how about meeting here the next Friday and see who was right? There won't be any significant increase in the daily deaths (yet), but there should be another increase in the number of cases.

(I have (unfortunately) already won a few bets like that, but it's always a chance that I am wrong this time, of course).

In fact, there is already a failed prediction in your essay: "but we will fail to see the 'second wave' that was predicted". The second wave happened. Exactly in the way we, "postmodernist scientists", have predicted. I will be the first one to tell you that indeed, there are many problems in modern science. But this is not one of them.

> The second wave happened

I think there's an argument that what we're seeing in the US isn't actually a second wave at all; it's stepped first waves in different places. Rates in places like Florida never really went down all that much. The US probably shouldn't be considered a single entity for these purposes.

(To be clear, I would expect that there will be a real second wave in the US at some point, I just don't think that this is it)

To see how the real second wave looks like, look at Israel. The problem with the US is that this is the _beginning_ of the second wave.

We will see.

Well sure, but I don't disagree that lots more (both confirmed and probable) cases will be counted over the next week in these places. I'll argue that NYC will not have anything remotely close to Arizona's spike if you like.

How about we meet up in a month and look at the death numbers? I predict that the national 14-day average will continue to decrease, or maybe flatten out for a bit.

I'm back after 6 days. It looks like in the three states (CA/FL/TX) death rates are hitting new highs not previously seen in those states.

That was not the bet, we’re talking about the national number. I might still lose, it looks like there has been a statistically significant bump in the past week, and it may continue.

Some of it must be due to spread in the southern states, where (by no coincidence) people spend more time indoors during the summer.

Some of it may also be due to the “July effect” which I did not know about:

https://acphospitalist.org/archives/2011/04/coverstory.htm

https://en.m.wikipedia.org/wiki/July_effect

Agreed. I don’t see how there won’t be an increase in deaths in a month (save for some new miracle drug, which I doubt — antivirals are hard), so sure, see you in a month. My email is in my profile. (The bet is on nationwide increase).

Ahh I was hoping you'd go for an exponential increase - I might well lose this bet. But sounds good, see you then.

The confirmed deaths are a known undercount, and cases going up faster than testing also indicates that the confirmed death count will be more of an undercount, because fewer (proportionately) of the deaths due to the disease will be confirmed.

In a month you guys may be unable to reply, FYI.

I am pessimistic, but I am not _that_ pessimistic.

OH crap, no I meant, that hacker news will lock the reply feature!

On what basis do you stake your claim?

I'm commenting (as I always do) under my real name. I'm staking my personal reputation, and there is a very real chance that I'll be wrong.

I'm not asking about the stakes in that sense. Your claim is only interesting to the extent you can demonstrate a logical basis to believe it to be valid. Can you provide your reasoning?

edit: Ok, I see you are the one with polemic blog post. I see that you seem to think it is "just a cold" because it is a coronavirus.

It seems you are making a categorization error. Like saying because there are garden snakes, having a cobra in your house is no big deal--just a snake!

Why would the waves of death never come? So far more people have died than WWII. The deaths lag new infections, and new infections have spiked. I see you are interested in other folks admitting they are wrong--will you?

> Can you provide your reasoning?

Susceptibility varies widely between individuals, with four grades of antibody response, as well as at least one form of T-Cell-mediated response, already documented in the literature. There is enough evidence to suggest that this virus is airborne, and if everyone were completely susceptible the "true" R0 (this model parameter is a convenient fiction, of course) would be close to that of measles. Therefore it has already spread very far in any dense population that was susceptible, and we are basically already at herd immunity in NYC and many other places.

> Ok, I see you are the one with polemic blog post.

Yay, I'm a niche HN celebrity!

> I see that you seem to think it is "just a cold" because it is a coronavirus.

You are skipping over a critical detail, which is that any virus can be much more virulent after it jumps between species. But yes this one is basically a new cold, and in time it will present much like all the others (if it becomes endemic).

> So far more people have died than WWII.

What the fuck???

> I see you are interested in other folks admitting they are wrong--will you?

If half a million Americans die from this thing (not merely with it, and before it becomes just another cold, not over the next 20 years), you can count on it. This will not happen.

The attenuation you seem to be banking on only happens if there is selective pressure toward it. The selective pressure is usually a product of "too deadly or too strong a response prevents the host from spreading" so weaker strains that let the host spread more win out in a game theoretical sense by out-competing. This is not a likely force in this case--this is already spreading really really effectively, so I think the force behind attenuation is missing and your assumption is entirely faulty.

> So far more people have died than WWII.

> What the fuck???

You are right on this one, bad bad misquote on my part.

#2 isn’t obvious to me. Can you explain why?

The time for symptoms to show up is of roughly 2 weeks but the median time is at about 6-8 days. After symptoms show up, it takes a while before people get hospitalized, and then mortality seem to take roughly 21-24 days after hospitalization.

This means that from the time someone is infected until they die, a month and a half can have gone through.

By the time you take new isolation measures, you have to wait roughly two weeks to see if daily infection rates start to vary. But the lethality at this point, assuming no change in healthcare system overload, can be more or less statically predicted for weeks after the fact.

Those are called "lagging indicators"; the main thing you try to prevent are deaths, which you can lower by reducing the overload on the healthcare systems, which you can lower by controlling infection rates and keeping them low.

Due to the duration of the disease's evolution, you get to see if your policies really worked more than a month and a half after enacting them, which is extremely slow for a disease that propagates at exponential rates. So people look for proxies like infection rates that are still lagging, but far less so.

Do note that the one good metric you want in the end is going to be "excess deaths", which counts the impact of not just the diseases, but of all other side issues that the disease may have caused. This can usually take months up to years to properly account and analyze.

Yep. That was obvious.

Thanks!

I think gathering and reporting the metrics is an additional lag.

I don't understand. If it wasn't obvious before then it can't be obvious now. You've had it explained to you. If it was obvious then that would not have been necessary.

Sometimes I will look all over for my keys, only to find them in my pocket. This is the obvious place to check first, but I am not perfect, so sometimes I overlook it.

What would it mean if deaths did not spike by the end of July?

This is the money question, or rephrased "will deaths spike by end of July?". There is so much debate, so much messy data, in a week or two this debate will be over among reasonable people, in a month even extremists will be eating crow. I, like everyone, have my opinion, but soon enough we'll know for sure. People saying "deaths aren't tracking infections" are premature in their assessment by a couple weeks IMO because earlier and more testing is making this peak be shaped differently than the last one.

Sorry to hijack, your actual question is interesting too. I guess that would mean all the increased infection metrics are entirely from increases in testing and we're actually seeing that infection rates have always been way higher and the virus is way way less deadly than previously thought? Or maybe it has mutated and become less deadly?

It takes about two weeks for symptoms to arise and become severe, and then about another two weeks of severe symptoms that require ventilators before people die.

The pre symptomatic time wouldn’t be part of this gap unless the infected person got a test before symptom onset, which I believe is rare (most people who test positive get tested after they develop symptoms)

If you look at the graphs side-by side for areas with clearly defined peaks (I looked at MA, NY, Italy, Spain, UK, Netherlands) the peak fatality rate is no more than a week behind peak case numbers, and perhaps less. Testing rates in at least some of those areas were fairly constant around the peaks, so that doesn't seem a likely confounding factor.

I'm not sure what to make of the meaning of such a short gap, and of course people may now be testing earlier than they did in the early hot spots, so I'm not sure if this has any predictive value.

Testing in the beginning lagged significantly from the initial infection.

One caution about those numbers; many/most sites show cases based on the day that reporting was received and aggregated, _not_ on the day it was taken. This doesn't matter very much now, as turnaround on tests in most places is quick. But a couple of months ago delays of weeks were common.

People are far more likely to be tested before symptoms today than a few months ago; at least in Ireland when a case is detected everyone they've had contact with is tested, and some of those will be pre-symptomatic positives. And then anyone _they've_ had contact with will be tested and so on.

No, these measures affect new cases, not the already ill - so you've got to build in those 2 weeks to display symptoms for those people who would or wouldn't now be infected.

It’s new cases. Those are people that have tested positive. This usually happens days after the actual time they have been infected. At the time of infection, you’d test negative because it takes time for viral replication to increase viral load to levels that are over thresholds used in the PCR testing process.

But we're not talking testing negative, are we? We're talking those people showing up in the system (the > 2 week lag in time it takes to change course), or did I miss a level of questioning that threw this off course?

The original question was what the delay between “new cases” and “new deaths” charts was.

Bluntly put, it takes time to die from any disease.

It takes time to die from this disease and time for that death to show up in the official numbers.

Reply on news.ycombinator.com