It isn't clear - that's a good point and would suggest that the upper bound might actually be higher than it appears at first glance. If we take 10% of infections being hospital based (which might not be accurate as that statistic is from South Korea and the above paper is in China outside Hubei) then 16% of the outside-the-home transmission might be hospital based.
I should say that only 284 of the 468 transmission events are included in either household and non-household. I don't know what the other 40% of cases were but I guess the researchers weren't able to identify the relationship from the public data that they were using. It does appear that this undefined 40% has a lower serial interval than either of the two defined groupings as the serial interval of all cases together is lower 3.96 [3.53, 4.39].
If initial viral load makes a difference one would expect to see shorter time from infection to diagnosis/hospitalisation in cases which are transmitted within households. There is suggestive evidence in this paper which includes data on the serial time for household (4.03 [3.12, 4.94]) and non-household (4.56 [3.85, 5.27]) secondary infections. The number in square brackets are the 95% CI.
This is fairly weak evidence that there is a difference and also gives some weak indication as to what the maximum effect of initial viral load might be.
The raw data from this paper, for example, might be used to give more information on this and also severity which is more what we're interested in - the Tianjin data appears to be fairly complete albeit with only 135 cases.
EDIT: added link to 2nd paper
I think even a few days has the potential to be extremely valuable if it can be pulled off. If worldwide reactions had happened a few days sooner then half of the cases could have been avoided. LW ringing an alarm bell a few days earlier might not have had an effect on policy but its important to note how big the potential gains are.
As you say in the OP, the next time any pandemic comes along the worldwide response is likely to be better. So my main question is how do we generalise this advice for other severe dangers.
To me one of the main issues if the speed at which things happen. Most things which happen gradually give enough time for people to react without disastrous consequences - COVID only gives a few days before your problem is doubled. This would be fairly high on my checklist specifically for a future pandemic - low doubling times - but for general alarm bell ringing speed of problem development should also be up there.
*insert obligatory FOOM comment here...*
Did you estimate how early using this would have caused an alarm to be raised for COVID-19?
I think the top 3 the harm questions were confirmed in this paper on 11th Feb but maybe there were other papers before this or we could have inferred from public data?
2,000 deaths was 18th Feb.
Escaping a lockdown attempt would probably be ~21st Feb in South Korea (the virus didn't really escape China lockdown - it had escaped before the lockdown)
Indirect transmissibility I'm not quite sure about a date?
Pre-symptomatic transmission again I'm not sure - from the papers in jimrandomh's post maybe early-mid Feb we had a good hint.
Yes, we definitely expect to see a lag between growth rates of cases and deaths, it is odd that even when this seems to be present it is only a couple of days to a week. I think this may be partly due to delays in diagnosis. 17.8 days is between onset of symptoms to death. However there is normally a lag between onset of symptoms and diagnosis (onset to hospitalisation I think is generally a bit less than a week) but even this still leaves a theoretical 10+ day lag.
That is all based on relative numbers within a country. Comparing CFR (case fatality rate) values between countries is notoriously unreliable due to testing capability. Looking at naive CFR I think the UK are about to overtake Italy as having the worst CFR in this set of 10 despite being earlier in their epidemic. This is either due to being worse at testing or better at diagnosing deaths as being COVID related (some countries aren't counting deaths which don't occur in hospital - source). CFR in the US is low compared to where other countries were at similar points in their epidemic so I guess it won't reach 10% but it is likely to reach 5%.
The ILO (international labour Organization, a UN agency) has a report on this.
Some key findings: Estimated increase in unemployment of 5-25 million - c.f. 22 million for 2008-9 crisis
These based on assumptions of 2-8% drop in global gdp
Value add from Chinese Industrial was down 13.5% in Jan/Feb
You might be interested in this post which explores similar territory.
South Korea, as always, are a treasure trove on information - they publish details every day which includes major outbreak clusters, some of which are hospitals. Of the non-cult related cases where they have managed to identify the source of the infection, hospital based infections account for 20%. If you include cases where they haven't identified the source then it's more like 10% which is probably a fairer reflection as hospital clusters probably mainly do get identified.
(They changed their reporting layout on March 25th and the new version doesn't quite contain as much information so I've based this on the 24th)
I think there's a decent amount of correlation with between lockdown dates and entering linear growth. Below are the lockdown dates and starts of the linear phase for some of the worst hit countries.
China 23rd Jan -> 5th Feb
S. Korea 20th Feb -> 1st March (This wasn't a mandated government lockdown but people did seem to stay inside in the worst hit areas)
March:
Italy 9th -> 21st
Spain 15th -> 26th
Germany 16th -> 27th
France 17th -> not yet linear (last 2 days have been high)
Switzerland 20th -> 21st
US 22nd (NY) -> not yet linear
UK 23rd -> approaching linear? Possibly already there
These are remarkably consistent at 10-14 days, apart from Switzerland (very fast) and France (looked like it had gone linear at about the normal time but has increased again).
This graph shows the same data but is annotated with containment steps taken by each country (it isn't averaged over 3 days so the exact numbers don't match up but the same pattern applies).
I share your rough estimates of IFR in your other comment here although I was concerned about how high IFR might be with overwhelmed hospitals.
Sampling bias at its worst here would mean that IFR is 3 times more than those calculations (i.e. 1.5-2%). If this is the worst case in Lombardy where the hospitals are overwhelmed then it is something of a relief to me that higher rates are unlikely.