Hacker News Reader

Comment by nkoren | Hacker News Reader

The primary point isn't wrong, but I am deeply deeply deeply skeptical of this, since the underlying data is of such varying and often miserable quality. A friend of mine in Oregon exhibited all the C19 symptoms, and went to get tested. It took 15 DAYS for her sample to be tested. The test came back negative, but the lab said they had no confidence in this result, given the age of the sample. This is apparently the default testing situation in Oregon.

Given those kind of inputs, how can one say with any confidence when Oregon will peak, and how large that peak will be?

Other states have much better testing and thus can be more confident in their predictions. Some have worse. For most, however, I get the impression that this kind of modelling is somewhere between a wild-assed guess and a crapshoot.

View on HN

These IHME predictions are actually based on deaths, not on tests. Also, please look at the wide error bars on the original data source: https://covid19.healthdata.org/projections

Of course there are still a lot of variables - one of the better criticisms is that it assumes that states lock down with the same strength, just at different dates. But it's the presentation of this Axios piece that is dangerous here: It strips out all the uncertainty estimates and presents exact numbers as if they were precise.

15 days ago was a different universe of testing capacity. Oregon had done less than 500 tests then and was still using the initial (semi-flawed) CDC tests. We've done almost 13,000 tests since then (I live in Oregon). Not to say we're doing enough yet, but it is a totally different situation today in Oregon and elsewhere:

https://www.deptofnumbers.com/covid19/oregon/

edit: that link is to a site I built to track testing capacity and new case growth.

They should be using death rates to estimate the peak, not confirmed cases. Death counts are very accurate if not perfect representation of the situation.

They're not. For a variety of reasons, many deaths are not being reported as caused by Covid-19, but rather other preexisting conditions which may have been complicated by Covid-19. Additionally, as I understand it, some number of deaths require post-mortem testing to confirm that the victim was indeed infected. Given the varied backlog of testing, the morgue backlog (or soon-to-be) and the lack of testing kits, I'd conservatively expect death counts to be lower than actual by anywhere from 10% to 150%. I wouldn't be surprised if a year from now we realize we were off by a factor of 2 or more.

You can compare normal death rates from previous years to determine if you are having excess deaths. Here is data from a town in Italy for example:

https://www.corriere.it/politica/20_marzo_26/the-real-death-...

“Nembro, one of the municipalities most affected by Covid-19, should have had - under normal conditions - about 35 deaths. 158 people were registered dead this year by the municipal offices. But the number of deaths officially attributed to Covid-19 is 31”

There are some caveats with the data of course - I.e. the town might not have any more deaths for the rest of the year since it’s burned through all the people that would have expired anyway, but it’s interesting reading.

The Corriere article has a major flaw that it's looking at deaths from Jan onwards only.

This year is unusual in that there was no flu season, really. Deaths up until the start of COVID-19 were way lower than normal all winter.

https://swprs.files.wordpress.com/2020/03/italia-mortalita-m...

So even a disease no worse than flu, you'd expect would cause a much higher death rate than normal for this time of year, simply by shifting the curve forward six months. And COVID-19 is likely to be worse than flu.

(I say likely because the data quality here is so poor we really have no idea how dangerous it is)

Can you tell me what your graph is supposed to show?

Bear in mind that Nembro has a population of 11,616 according to google. That would mean an excess death rate of approx 1000 out of 100,000 if we use the 123 excess deaths from the article. Compare that to 40.2 from here: https://pubmed.ncbi.nlm.nih.gov/31401203/

Now I’m not saying that Nembro is typical or even that all my data is apples to apples, it’s just the only data I’ve found where we can compare death rates in any way. The crude death rate for the whole of Italy here: https://knoema.com/atlas/Italy/Death-rate

It’s interesting that you call the data poor - this is exactly the data we should be looking for. I’d love to see this kind of data for the whole of Italy for the last 10 years to compare to the last few months.

The other data coming in is the dangerous stuff to me. We can’t tell if someone died of covid, or simply had it when they died. Every country has different testing regimes. Ultimately the death stats will give us the full picture.

Differing perspectives welcome.

The graph is supposed to show mortality vs the 'baseline' (good question how exactly that's calculated, looks like it's probably this one: https://www.euromomo.eu/methods/methods.html).

It shows the impact of prior flu seasons. Winter of 2016 was very bad. 2017 not quite as bad. 2018 about the same. 2019 .... gone. Trend line is below baseline and very significantly so up until the start of COVID-19. There was practically no flu season this year, which means a LOT of people survived the winter who in prior years wouldn't have done.

The problem is Corriere is ignoring that in their analysis. Deaths may be higher than expected now, but how much of that is because it was much lower than expected before?

I suspect in the next week or so you'll start seeing people doing integration on the graphs to try and figure out what the excess 'vulnerable population' was at the start of the epidemic; i.e. people who would in other years have been taken out by flu but survived this year.

The red trend line will shoot up of course. You can see the latest data point all the way at the top right, it's nearly falling off the image.

Lol, I didn’t even see that last point until you mentioned it.

It’s interesting data - where did it come from? I’d like to monitor it too.

Italian Ministry of Health

http://www.salute.gov.it/portale/caldo/SISMG_sintesi_ULTIMO....

via https://swprs.org/a-swiss-doctor-on-covid-19/ which has many links

NB: They've since updated the graph in the PDF with a slightly different one. Oddly, the wildly high data point at the end is now gone (wtf) and the COVID-19 period is shaded blue.

Italian death statistics show nothing out of the ordinary at this time. If it's true their infection levels are now stable I don't see how these figures can take the absolutely massive acceleration required to justify current measures. It's still less bad than the flu of 2017.

For a variety of reasons, many deaths are not being reported as caused by Covid-19, but rather other preexisting conditions which may have been complicated by Covid-19

Isn't it the other way around? That deaths are being recorded as COVID-19 even when there were severe underlying conditions and the person probably died of that? The latter seems to be the more common issue.

It makes death rates meaningless. What does it mean to say someone died of COVID-19 if they were 89 and had heart disease?

Deaths are partly unavoidable, but they're also partly avoidable because if hospitals have ICU /ventilator and bed capacity the death rate goes down. Once they're overrun then the death rate spikes.

Good point, may be they should use some custom heuristic that takes deaths + confirmed cases + fudge factor.

Testing will also peak out at the state's capacity to test, whereas people will have a hard time hiding dead bodies, so I think some numbers are better for forming a stable picture.

Number of new cases in Oregon is flat to slightly down in the past six days...

3/25 57

3/26 50

3/27 98

3/28 65

3/29 69

3/30 58

These are new case numbers. Data source: https://weather.com/coronavirus/l/3e056a178cd9fcb8c8efaa5819...

weather.com case numbers for oregon

I think the point is if testing is tightly constrained you can't know much about how many cases you really have.

We'd actually want some kind of randomized survey testing in an area to understand the prevalence there, and do it over time to understand the arc, but we aren't doing that -- we still can't test all the people who have a need for am immediate test.

Testing is not that constrained. Here you can see stats for both positive and negative: https://covidtracking.com/data/state/oregon/

Oregon is doing roughly 1400 tests/day, with ~5% positive rate. Places that are more testing constrained see 20+% positive.

Reply on news.ycombinator.com