[jsnell]

Sure, that model shows that there are multiple ways to fit the model parameters to account for the first 14 days of fatalities. But we have a lot more real-world data available that directly contradicts any scenario where half the population is already infected.

I'm just disgusted that the authors are now saying it was just an abstract demonstration of different scenarios, and pretending they didn't actually make the claims about the real world that they did.

[guscost]

...is all of that real-world data from Lombardy? If not, does it rely on circular logic? I’ve seen evidence in both categories, but not much else.

[jsnell]

Sure, here's some to start with.

1. Any situation where the testing isn't limited to just the most critically ill showing much lower infection percentages than this model would predict. A lot of these are tests of every person in a risky situation, whether they had symptoms or not:

E.g. the evacuation flights from Iran to China tested every passenger and showed a 3% infection rate. T The village of Vo testing their entire population twice soon after starting a Covid quarantine, with a 3% infection rate.

Others were testing large amounts of people with no particular reason they had Covid, but still with a skewed sample:

The Swedish sentinel testing of random people with any kind of flu symptoms (1.5% of people with flu symptoms testing as positive for Covid, vs. 30% testing positive for Influenza A/B). Iceland testing IIRC a volunteer 1.5% of their population whether they had symptoms or not, and having something like a 1% infection rate.

The thing all of these have in common is that they happened at a time in their relative epidemics where this model should have predicted the majority of the population was currently infected.

In fact, it's basically impossible to explain any testing results, since even when they're biased to cases where Covid is strongly suspected, the ratio of positives is so low. If we test the 1000 of people most suspected of having Covid right now, and get 10% positives, how can it possibly be the case that half the non-suspicious population are carriers at the time of that test?

2. If herd immunity kicks in 14 days after the first death as implied by this model, why haven't any of the epidemics died down by now? Italy is on what, day 30?

3. How does a super infectious but low mortality model explain the geographic clustering of deaths? Sure, the geographic clustering of known cases could be explained by testing bias. But deaths don't have that bias.

4. Observed high CFRs in limited populations where we know the infection rate was high. E.g. Diamond Princess at what 1.4%, and still another 2% in critical condition. How many top ranking Iranian leaders died in short order of Covid, and how does that fit in with a mortality rate of 0.01%? Or the cases where most of the patients of a health care facility or nursing home got infected?

[guscost]

1. Unless I'm mistaken, this (extreme scenario in the) Oxford model does not say that 50% of the population is infected all at the same time. Until there are serological surveys that indicate who was infected in the past, the current tests can't prove what you think. And any false negatives from the PCR test would add to this problem.

2. The progress of the disease takes time too, from infection to symptoms to detection/hospitalization/secondary infection. And efforts to "flatten the curve" will "slow the spread" too.

3. I'm not sure about this one, but shouldn't the infectiousness also vary quite a lot with different contributing factors (population density, air quality, etc)? The number used in the model is always just an "average" in a sense.

4. The cruise ship evidence is pretty significant, but it still has problems. Any passengers who had already recovered or did not show symptoms could have been missed. Plus I can name several factors in that situation that might increase mortality off the top of my head. It's not the best sample for drawing conclusions about people who are on average less old, not traveling, and so forth.

Admittedly, the extreme "50% infected" scenario has a risk factor (same as IFR?) of 0.001%, which my gut feeling says is too optimistic. But as far as I know none of the scenarios can be conclusively disproven (until they can do proper serological surveys).

[jsnell]

1. They have the virus go from like 5% infected to 70% in two weeks in that scenario. Given what we know about the incubation period and they assume about the infectious period, more than half the population has to be infected (and probably even infectious) somewhere along that curve.

2. Their model predicts that the peak of the epidemic in Italy should have been before March 5th (first death on February 22nd + 14 days, at which point easily more than half the population is infected). There should have been a sharp drop in new cases about a week later, as the virus burnt itself down. But here we are three weeks later, and it's still not entirely clear that the peak has been found.

Italy did not institute significant nation-wide measures until March 9th, so the "slowdown from measures" explanation makes no sense.

3. Agreed. But their entire model is predicated on treating the entire country as a single unit. That's probably a part of the reason why the results are so absurd. I don't think it's fair to excuse the model for regional differences, but require any criticism of the model to take them into account.

4. The difference between the model's prediction and apparent reality is likely to be about a factor of 200. Even assuming everybody on the ship was actually infected, that only cuts it to a factor of 40 difference.

And it's really not just that single case. Consider that infamous Washington state nursing home. 120 residents, 35 dead from Covid to date. Even if we assume that every single one of the 120 was infected despite not testing positive, that still an IFR of 30%. Sure, it's a high-risk segment. But it's also a large enough segment a 30% IFT for them makes it quite impossible for the population-wide IFR to be 0.01%.

(Re: your last point, they had two parameters. One for being at risk of becoming a serious case, and another of dying if serious. The two need to be multiplied to get their predicted IFR. They assumed that 0.1% of population were at risk to become severe cases, and 15% of the severe cases died. So about 0.01%).

[guscost]

So on the one hand, I think you're bringing up enough problems for me to agree that the extreme scenario is not likely to hold up.

On the other, I don't understand how after making so many (often reasonable) assumptions in your arguments, you can say that it's "quite impossible" for the total infected to be so high, or the population-wide IFR to be three orders of magnitude lower than a nursing home or any other special case that you do not fully understand.

Here are some of those assumptions, which again do not all seem unreasonable to me:

- A mild case would likely be detectable by a PCR test for 8+ days

- The PCR test does not have a high false-negative rate in mild cases (see https://www.researchsquare.com/article/rs-17319/v1)

- Italy has not already had a sharp drop in new infections/most new infections are being identified as cases within two weeks

- Italy did nothing to slow the rate of infection until the full lockdown was in effect (but slower spread would mean higher mortality, no matter the reason)

- COVID-19 was the only/main thing that contributed to mortality in the special cases