Hacker News Reader

Comment by jwlake | Hacker News Reader

One of the super-interesting things here, is that apparently everyone was in the specific window where they test positive for the virus. This implies the population was recently infected, had not been previously infected, and it spread almost completely within a tight window.

This implies a shocking high R(effective) for that population. In 2 weeks we'll have super interesting data one way of the other on the CFR.

View on HN

A prison is very much like a cruise ship from a viral perspective. We know from the cruise ship data like the Diamond Princess that most people had no symptoms initially, but overtime most people became symptomatic [1].

Of course if we are really lucky and the prison was infected with a naturally attenuated strain we should make use of it [2].

1. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7078829/

2. https://www.tillett.info/2020/04/12/how-would-a-search-for-a...

I think that. depending on the air conditioning system, it might be quite a bit worse. The density of people is huge, air flows freely between cells, prisoners eat packed together at communal tables. I bet R is like 10-20.

On the Diamond Princess it was about around 14. I would expect prisons to be much the same.

Do you know how that broke down by room? Is there data on that? I wanted to make a map by room of infections, because that data point should be useful, e.g. was the rate higher or lower in rooms with outside access? Crew versus passengers? I just don't know of it already exists, or how to source the data.

We seemed to have early fluffed the data out of the Diamond Princess. We aren’t even tracking how many people died - it seems once everyone got off the ship all interest in it was lost.

I'd have thought that more airflow would help to reduce transmission, given that like other coronaviruses it's transmitted by droplets and not aerosol. Droplets would have trouble staying suspended with more air movement, and would be unlikely to make it through centralised air conditioning systems without becoming deposited on some surface within them.

The eating together at communal tables, and any other close-quarters communal activities, seem far more relevant to increased R than the air conditioning system and the free airflow.

Not really, not in a prison: Droplets exist in a continuum, and while most are large enough to drop out of the air, as they get smaller they stay in the air longer, and require less air current to stay aloft. [0]

As such, while the small particles may not exists in sufficient quantities when it is just one infected, or a few infected inmates, but with each infected inmate the concentrations of those small particles will increase. Let's say they're normally 0.5% of the total particle exhalations: 10 inmates get sick through direct contact, and their combined exhalations bring the quantity of smaller, longer lived particles up to an equivalent of 5%. More get sick through contact, bringing it to 20%. At some point, you hit a critical mass where there is a sufficient concentration to infect people, and creates a downward spiral from there.

[0] https://www.nytimes.com/2020/04/14/health/coronavirus-six-fe...

Prison demographics are very different than cruise ships, so might be interesting if different

To give evidence to this comment, the gp's source notes that ~75% of those that tested positive for covid were >60 years old. Whereas the BOP [0] reports a distinctly different age demographic. We know that covid dramatically affects people differently based on age, so I wouldn't suspect IFR to be similar to that of a cruise ship (where it is typical for the demographic to be older and retired)

[0] https://www.bop.gov/about/statistics/statistics_inmate_age.j...

> A prison is very much like a cruise ship from a viral perspective.

Prisons are much worse. Chow line with men almost heel to toe. Commissary line. Med line. Recycled air. Big crowds in small spaces.

That’s a cruise

I don't see them advertising it like that.

The table settings might be nicer, but it's still a chowline of people asses-to-elbows.

https://c8.alamy.com/comp/B5D6CE/breakfast-buffet-served-onb...

We could make the same dumb jokes about:

Popular restaurants, clubs, sporting events, concerts, movie theaters, stores on Black Friday, etc etc etc.

(Personally, I’ve been on 5 different cruises and they’ve each been among the best vacations of my life. But lol lots of people so they must be just like prisons amirite?)

I don’t live at/in a restaurant, club, sporting event, concert, movie, or store for a week or more.

Perhaps you do. But if you don’t, none of those things are similar to a week-long cruise at all, since you are there for hours and not days.

> I don’t live at/in a restaurant, club, sporting event, concert, movie, or store for a week or more.

I don't live asses-to-elbows in line with people for a full week on a cruise ship either. I have a room to myself. What point are you trying to make, exactly?

The room is only for sleeping. Everything else is massive crowded common spaces

I guess you guys will have to wait for your first time in jail to spot the differences, if you aren't going to learn from strangers on the Internet

Apart from the demographics, they seem to me pretty similar ;)

    > Prisons are much worse. Chow line with men almost
    > heel to toe. Commissary line. Med line. Recycled
    > air. Big crowds in small spaces.

How could the prison be much worse than the cruise ship you just described?

OP is describing a prison.

That's the joke.

Stress.

But in prisons you spend way more time isolated. There is less freedom of travel, and I'd assume they'd be less physical contact (although maybe more if you considered shared showers).

They share cells. They move around at set times. They are not isolated.

The soap is expensive, they don't have it and masks are not allowed.

I am communicating daily with a prison inmate and they are required to wear masks all the time. But yes, that won't help due to all the other aspects of prison.

> We know from the cruise ship data like the Diamond Princess that most people had no symptoms initially, but overtime most people became symptomatic [1].

Your wording is too strong. Data is 50% asymptomatic at time of testing, paper models that that number dropped to 20%.

> from paper on 19 February and 50.5% (320/634) on 20 February (Table). Soon after identification of the first infections, both symptomatic and asymptomatic cases were transported to designated medical facilities specialised in infectious diseases in Japan. However, these patients were treated as external (imported) cases, and a detailed description of their clinical progression is not publicly available.

We conducted statistical modelling to derive the delay-adjusted asymptomatic proportion of infections, along with the infections’ timeline. The estimated asymptomatic proportion was 17.9% (95% credible interval (CrI): 15.5–20.2%).

If you read in the discussion they reference other data sets to show that more asymptomatic become symptomatic. It is also what the Icelanders found with their data too where most people started out asymptomatic, but many later developed symptoms.

In a region in Italy they tested [1] people twice, 2 weeks apart, and found that 43% were completely asymptomatic over the course of the infection.

[1] https://www.medrxiv.org/content/10.1101/2020.04.17.20053157v...

Doesn’t that mean that 57% were symptomatic?

I didn't intend to contradict you. Granted it's totally valid to assume that many people who are tested positive and show no symptoms might develop some later, but I just thought it might be of interest to some people to have numbers where they tested patients twice over a long enough interval to conclude that they really were asymptomatic.

About 50% of the population in that study is over 50. I think it's a bad idea to draw generic conclusions about the disease from a diverse age range. You end up with conclusions that aren't relevant to either. Glossing over the tables I don't see any data about symptomatic over the two weeks by age. I'd be shocked if the rates of becoming symptomatic are not much higher for the elderly and much lower for the young. Didn't read it in depth though.

See: Simpson’s paradox, where trends in subgroups, taken together, can indicate a different trend.

https://en.wikipedia.org/wiki/Simpson%27s_paradox

Trends in subgroups can indicate a different trend. The above comment is not about a linear trend. It's about an average statistic for a population based on a sample that's not representative of the population age distribution for a process that probably isn't linear across ages. If the older half becomes symptomatic 100% of the time, and the younger half becomes symptomatic 0% of the time, it's very misleading, though not technically incorrect to say people become symptomatic 50% of the time. There's a huge impact on how to handle the problem.

Since the hospitalization and mortality rates vary so much with age, it would be surprising if the rate of asymptomatic cases didn't. Since prisons skew young we would expect a much higher proportion of asymptomatics to be found there than in the general population.

If you are going to say they skew young (or make any claim), please provide some figures. I did a search, from https://www.bop.gov/about/statistics/statistics_inmate_age.j...

  <18    Under 18     6         0.0%
  18    Ages 18-21    1,866     1.1%
  22    Ages 22-25    8,366     4.9%
  26    Ages 26-30    21,931    12.8%
  31    Ages 31-35    28,182    16.4%
  36    Ages 36-40    31,206    18.2%
  41    Ages 41-45    26,545    15.5%
  46    Ages 46-50    19,979    11.6%
  51    Ages 51-55    13,749    8.0%
  56    Ages 56-60    9,475     5.5%
  61    Ages 61-65    5,467     3.2%
  >65   Over 65       4,746     2.8%

Whether this backs you up or not depends on what 'young' is.

That's federal prison data. People who go to federal prison aren't necessarily the same cross-section of the population who go to state prison.

Anyway, those numbers are certainly much younger than the median age in the USA nationally, which is 39 years with over 65s being 15% of the population.

Skew isn't the right word. But relative to average this is definitely younger.

https://www.indexmundi.com/united_states/age_structure.html

Local demographics will vary wildly but it looks like the baseline would be 16% 65 and older, 13% in the range of 55-64. Since we're talking about prisons we should probably ignore the population under 18, so the normalized expected population over 65 is closer to 20%.

SARS, MERS, and now SARS-2 are all known for super-spreader events where occasionally one person just infects a ton of other people. And we know from testing that the amount of live virus in a person's phlegm can vary by many orders of magnitude so I guess that isn't surprising. There are a lot of people who infect just one other person but the overall R(effective) is substantially driven by the long tail.

Exactly! What are the chances of that, compared to a gross error in the testing, like someone processing the samples was infected and not very careful?

The chances of an antibody test kit not being sufficiently specific to SARS-CoV2 as opposed to endemic coronaviruses is quite high.

Now, what are the chances that some government institution acquired tons of crappy test kits, didn't validate them before use, and then proceeded to publish the results which just so happened to be exactly what they wanted them to be?

>> .. that apparently everyone was in the specific window where they test positive for the virus.

The data comes from 4 prisons. It is theoretically possible that all 4 of them happened to be in the same window after an initial infection, but it doesn't seem very likely. I guess the only way to be sure is to do follow up tests every week or so, hopefully that will happen.

I repeatedly point this out. There was the church choir practice where 45 out of 60 people were infected within what two hours? There is also a Korean call center where about 2/3 of the people in an open office were infected.

I've repeatedly linked to CDC study that estimates r0 in Wuhan before lockdown at between 3.8 and 8.7. Median 5.8.

Take away: Completely avoid being indoors with large groups of people.

Especially if they're singing or talking non-stop.

That's also shown in the recent outbreak in Singapore - almost all of the cases are migrant workers housed in tightly packed dormitories.

Covid-19 fatalities takes up to 2 months after infection, symptoms have a median of just over 5 days from first exposure.

Median of 10 days though.

Numbers I have seen where median 5.1 days from exposure to first symptoms and a median of 8 days from first symptoms to death in Italy. So, a median of ~2 weeks from exposure to death. https://hub.jhu.edu/2020/03/09/coronavirus-incubation-period... https://www.lastampa.it/esteri/la-stampa-in-english/2020/03/...

However, early estimates are going to be biased without the slowest fatalities. Further, that’s also population specific.

The South Korean data suggests there is a long tail on the death rate. It seems the very old and ill die quickly, but those that are healthier it can take a month or more to die.

Only if you develop a severe case in the first place, which you would know about within 2 weeks average. Incubation on average is 3-5 days, severe symptoms then start in the second week.

After you're hospitalized, yes you might or might not then have a month long battle for survival.

Keyword "up to".

The most common fatal case seems to be roughly: 1 week without symptoms, 1 week of mild symptoms, 1 week of severe symptoms, death. That's 3 weeks.

In rare (1/10000) cases, incubation time can reach 14 days, that's where the 14 days quarantine recommendation comes from. But the median is closer to 5 days.

Time to death can be much longer, the virus can cause all sorts of damage to the body, including secondary infections. And if the patient is strong enough to resist the primary infection but not enough to recover, it can take a long time.

> In rare (1/10000) cases, incubation time can reach 14 days

What's the cite for that number? I'd be very suspicious of anything claiming a 5-significant-figure result in a disease that has only (heh) 3M known cases.

In fact there's significant supposition that the incubation time and asymptomatic contagious state can last much longer than originally guessed. This would go a long way to explaining the difficulty of detecting early outbreaks (basically nowhere in the world was able to contain before community spread was happening) and the anomalously low rate of new case decline post-peak. But there's no good science on this, and probably won't be in time for it to be useful.

South Korea had a very rapid decline post peak. 1062 cases on 2/29 to 114 cases on 3/11. With an average of just under 10 cases per day for the last week.

Also, the number I have seen was ~1% of cases where asymptotic for 14 days. https://hub.jhu.edu/2020/03/09/coronavirus-incubation-period...

Do note that rapid decline in the day-to-day increase of case numbers is not necessarily indicative of rapid decline in transmission rate. Transmission rate could have declined gradually over the course of a week or more, and the rapid decline in day-to-day increase of case numbers could have just been detection catching up with existing infections.

Their daily testing was reasonably consistent over that time period. So, their is some wiggle room, but it’s at minimum a 80% drop over 2 weeks. Compared to the US where daily numbers are steady for weeks with the last two days having the greatest number of detected cases.

I've given up on the USA numbers. There is no way an underlying exponential process is going to deliver you a near constant 30K new infections for nearly 4 straight weeks.

Something else is driving those numbers, and it ain't the actual number of new infections. For me the obvious candidate is they hits some testing limit.

You shouldn't think of it as one underlying exponential process though. It could be many overlapping and time-delayed S-curve type graphs which are additively causing a constant rate of infections.

Although it's also most certainly true that we're vastly undercounting the cases in most (if not all) areas.

> What's the cite for that number? I'd be very suspicious of anything claiming a 5-significant-figure result in a disease that has only (heh) 3M known cases.

You are totally right about that. I misinterpreted the following paper https://www.ncbi.nlm.nih.gov/pubmed/32150748

The actual number is more like 1/100. The 1/10000 number is the probability for someone who has a 1/100 chance of being infected to develop symptoms after 14 days.

Uhm... too late to fix my original post, sorry.

The CFR will be an interesting (though tragic) data point, but I wouldn't generalize too much from it. The demographics of the inmates are probably not representative, and the health care that prisoners receive is absolutely not of the same quality that civilians may receive.

The Princess cruise was a good indicator of CFR for a specific demographic range when they tested early, often, and had access to quality care. If the CFR of this prison will tell us anything, it will be a counterpoint to such information showing what happens when quality of care is greatly diminished.

Are these tests covid-19 specific or are they based on IgM/IgG?

> The results of the tests for IgG, IgA, and IgM levels are usually evaluated together. Abnormal test results typically indicate that there is something affecting the immune system and suggest the need for further testing. Immunoglobulins testing is not diagnostic but can be a strong indicator of a disease or condition. There are a number of conditions that are associated with increased and decreased immunoglobulins.

I'm looking for this in the article and not seeing mention of a specific window. Can you help me find what you're referring to?

The sensitivity of PCR tests to SAR-COV-2 varies over time from time to infection. I think the the OP is overestimating how narrow this window is though. It's not like you need to test within a few days or a week to be positive - it's just that viral load is maximum for only a few days (roughly the +/-3 days around onset of symptoms).

Even after the viral load in the upper respiratory tract drops off, there's still lots of virus around. You'll likely test positive -1/+2 weeks around onset of symptoms.

Depending on the nature of the virus and how the PCR test works with it, I couldn't rule out that it could be much longer than that. Virii are not my area, but there are bacterial diseases that result in positive PCR years after infection. Disclaimer: microbiology is also not my area.

Yes, it all depends on how the agent infects the body and how the immune system is able to clear it. Our belief (based on how other coronaviruses behave) is that the body is pretty good at clearing it out, and that the virus doesn't really have a mechanism for forming a reservoir within the body.

The expectation is that PCR negative is roughly correlated with fighting off the disease in the short term, and basically perfectly correlated in the long term.

Is this test sensitive to just SARS-CoV-2? What is the possibility that other corona viruses (like the ones that cause many of our colds) could create a false positive? I haven't heard anything from a reliable source on that.

PCR is pretty selective for SARS-CoV-2

I'm trying to figure out what tests they used. PCR is very specific but only works reliably in the early stages and generally has a high false negative rate.

Antibody tests only work after several days of infection, but some of them are not very specific to SARS-CoV2 and therefore have a high false positive rate.

So you really need use both tests, but you can't just OR the results together, because then a faulty antibody test will massively skew the results upward.

> but only works reliably in the early stages

No, it only works if there is enough virus to be detected. If you are infected, you will have a high enough count to be detected. Each type of PCR test is highly selective. The point of them is that they can differentiate between many types of virus, depending on how its setup.

> Antibody tests only work after several days of infection

Yes. Depending on the type of test, and how its done. it can be up to a month before these tests are accurate.

In principle yes, but The PCR test is usually a throat swab, where detectable amounts often disappear after a few days, even in severely symptomatic cases.

This is the reason why China started to accept clinical diagnosis like chest scans, causing a sudden big spike in cases.

Is there a source that says they were using PCR testing? The article seems pretty light on technical details.

Do not know if it is in the article or not but I guess the reasoning from GP goes like this: huge percentage of people are pre-symptomatic. The median time for symptoms to appear is 4-5 days from infection time, so they were probably infected in a very short period of time recently.

You're probably right, but I think that's flawed. We still don't know how many people are completely asymptomatic. It's a very weak assumption to base this argument on.

We don't know if infection provides immunity, so they could be repeat infections, or even viruses that never cleared.

Why is this being downvoted?

Yesterday the World Health Organization said: "There is currently no evidence that people who have recovered from COVID-19 and have antibodies are protected from a second infection."

https://www.npr.org/sections/coronavirus-live-updates/2020/0...

We will also see if they are asymptomatic or presymptomatic.

Reply on news.ycombinator.com