Daily numbers, graphs, analysis thread 6

[Barracker]

Welcome to thread 6 of the daily updates.

Resource links:
Worldometer UK page
Financial Times Daily updates and graphs
HSJ Coronavirus updates
Johns Hopkins Coronavirus Resource Centre
NHS England stats, including breakdown by Hospital Trust
Covidly.com to filter graphs using selected data filters
ONS statistics for CV related deaths outside hospitals, released weekly each Tuesday

Thank you to all contributors for their factual, data driven, and civil discussions.

The Royal Society has a group working on alternative models to Imperial

We do urgently need this model created independently of Imperial, for an expert 2nd opinion on such a new kind of pendemic
AND
one created specifically for this virus, instead of being a tweaked flu model

Anyone know how far along the RS group is, possible delivery date of first results ?

An interesting commentary on models
peterattiamd.com/covid-19-whats-wrong-with-the-models/

It is naive to think that the old "normal" wrt money still exists

• although I am prepared to believe it does for anyone funded by the UK govt (any party)

"normal" has long been dumped because of the sheer cost of this pandemic

When every country is spaffing umpteen billions each month and economies are running at 50% or so,

then the cost / benefit analysis is to throw money at any and all possibilities of ending the disaster

We have been told that Germany is working flat out on all possible lines, money no object
I'm sure most other wealthy countries are too
Daft of the UK if that is genuinely different, if you are still in the "NICE" frame of mind

AFAIK the RS doesn't have a single group, the point of their project is to encourage individual groups in different places and provide a point of contact for them all. My dh is part of one such and has already produced a paper using the Imperial model and doing something Bayesian to it to work towards a better understanding of the likely IFR but I don't understand it myself so I haven't wanted to post about it here! Things are going up on the covid section of the medrxiv all the time - Medrxiv

"The IC based their model on the first case being end of January - it is looking like this is incorrect and there were cases before then."

==> Has any paper - other than the Oxford one - been published that states this for the UK ?

== Or has a reputable scientist outside that group stated publicly that they think it likely ?
(Not the contrarian US publicist Ioannidis, if he ever does comment on the UK)

It would probably be good news if the infection were around earlier,
but the UK total death stats from 1 January just don't show any significant effect from week to week, to back up that hypothesis

Yes, I don’t get the much earlier disease hypothesis.

If it were the case, surely there would be a concomitant increase in hospitalisationsand deaths?

I had read the RS had one group that was specifically going to create a new model, not based on Imperial ata ll

The Imperial one is all we have atm and it probably does need tweaking,
but that's not adequate for the independent 2nd model that we need, that should be specifically designed for this Coronavirus

Yes, I don’t get the much earlier disease hypothesis.

Indeed... It was an interesting hypothesis before the scale of the virus’ impact became an unmistakable reality, but it no longer has any credibility in my opinion.

I listened to a podcast from Neil Ferguson - chief architect of the Imperial model last night. He implied that there would be a new model, presumably updated to incorporate newer knowledge of how the disease behaves - in a few days.

Incidentally, I have been involved in a NICE guideline (for a completely different and unrelated condition, and it was a few years ago now) but back then, the rule of thumb was an intervention would be given the go-ahead if it cost less than £20k per person, per quality adjusted life year (QALY) that it gave. A bit of back-of-envelope maths for Covid gives an astonishing sum of money that, in principle, NICE would approve to spend on the disorder: In a worst case scenario, half a million could die in the UK, losing (if what i've seen on here is right) an average of 10 years of life. Let's reduce that to 5 years of quality adjusted life years. So, 500,000 people x 5 QALYs x £20,000 = £50000000000. That's a massive sum - £500 billion (??? I get lost with that many noughts).

I'm sure the calculations that have been made are nowhere near that simplistic but it does suggest that 'money no object' might not be far off.

Can I just ask if anyone has an opinion on whether the ONS data coming on Tuesday will be telling in terms of plateau/decrease in deaths?

Last weeks report covered up until 10 April so if we think the peak was 8th of April, the registered deaths in last weeks report will not have mopped up the ‘peak’ deaths will it? Has anyone put together/seen any models that are looking at ONS stats and when the peak happened or will happen?

Also, having got into a bit of a spat with someone online about prevalence on death certificates of Covid-19 what is the general consensus here? Are they encouraging CV as cause of death (the whole died ‘with’ or ‘of’ debate) or downplaying it and encouraging avoidance of attributing death to the disease? I was pretty confident that reported death rates were too low in the governments death charts but this woman was adamant that it is being overplayed and that thete is some global conspiracy to over play the pandemic, particularly in the media But the data seems conservative to me?

Surely the point of the Oxford study, was that it was modelling various scenarios based on making different assumptions about the spread of the virus and what percentage of cases required hospitalation or resulted in death? So if the proportion of people getting very sick and dying was very low, you would need a very large number of infections to occur before you started seeing any spike in hospital admissions or deaths. By extension if IFR and severe cases were low, mild cases would be much higher than the data being used for the Imperial model was suggesting and the virus could have been circulating undetected much longer than believed.

As I understand it, they never said that they thought that this scenario was necessarily correct, but that they were demonstrating the necessity for serological surveys to give us the data to answer the questions about R0 and IFR.

@Mustbetimeforachange that is a really good article, thanks.

@Mustbetimeforachange

Interesting article, though I think he seems to fall into the trap he’s trying to warn against, as I believe his assertions are premature as there is still too much we don’t know.

Regarding NYC, we can avoid some of the uncertainty over the antibody test results by focussing on the time that ‘lockdown’ measures were put into place. If the rise was at or close to exponential at that moment, it implies that relatively few NYers had been infected, as true exponential growth requires a ‘pristine’ population. As soon as you get much above 20% infections, growth should demonstrably no longer be exponential.

I’m assuming here that lockdown dramatically reduces the R0, and based on the fact that hosptialsations in NYC, it’s reasonable to conclude R0 is now below 1.

My conclusion is that, if NYC growth remained at or close to exponential at around the time of lockdown, immunity couldn’t have been much above the 20% in the recent antibody trial, and the fatality rate is close to 1%.

Bigchoc,

Is there an ‘Imperial’ model or just a standard epidemiological model ?

I would have thought that there would be a standard model with the 4 (I think) simple to state (though not to solve) differential equations.

I would have thought that sophisticated models would give granularity by modelling local clusters and linking them to the broader epidemic, for which you need a lot of computing power. Also modelling things like releasing different parts of the country at different times.

Regarding Salford

The dataset is here

www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/causesofdeath/datasets/deathregistrationsandoccurrencesbylocalauthorityandhealthboard

"Figures by place of death may differ to previously published figures (week 15) due to improvements in the way we code place of death.

These figures do not include deaths of those resident outside England and Wales or those records where the place of residence is either missing or not yet fully coded. For this reason counts may differ to published figures when summed. "

It is 'place of residence Salford', so not anyone who died in Salford Royal but was resident in Manchester.

There is a disproportionate number of deaths reported by Salford NHS Trust as opposed to Manchester NHS Trust (whose two main hospitals are both bigger than Salford Royal), which does suggest an outbreak in Salford.

There is something odd about the data in that Salford's data show 46 deaths in care homes and 84 in hospital, whereas Manchester shows 4 at home, 1 in a hospice, 10 in care homes, and 81 in hospital. Salford has half the population of Manchester so it's clearly a major outbreak in that disregarding the care home figures you'd expect 40 or so hospital-based deaths by population, not 84.

This is discussed in this local news site

salfordstar.com/article.asp?id=5568

The claim there from Salford City Council is that Salford City Council itself is responsible for the more zealous coding of deaths in Salford care homes as from covid-19 (!)

So to see if this is true we check the progression of the virus nationally. Here we can see that Week 10 ONS showed 2 deaths occurring (Date of Death, not registration), Week 11 38, Week 12 384, Week 13, 1774, Week 14, 4777, Week 15 6146.

So if we look at Salford & Manchester deaths they should be normal in Weeks 10 and 11, and possibly show spikes from Week 12.

The attached graph shows deaths spiking sooner in Salford hence the virus got there first and spread so the lockdown, which the government fucked up on very badly by implementing a week or so too late, saved more Mancunians than it did Salfordians. In particular we see week 12 deaths still normal in Manchester, barely elevated in week 13, but well-elevated at that point in Salford.

For care homes see the next chart. We can see, perhaps as expected, that the deaths spike later in care homes than other settings (i.e. hospitals), as the virus is brought there later. But then it does indeed rip through.

For weeks 1 -13 there is NO elevation in care home deaths in Manchester, though week 13 is somewhat high. For weeks 13-15 there are 68 deaths, as against the 10/week average for weeks 1-12, i.e. 38 extra deaths.

Meanwhile only 10 deaths were coded as covid-19 in Manchester homes for weeks 14 +15 (none for week 13). Obviously this is just total bollocks and not possibly true at all, not even a little bit. We can reasonably conclude that there were in fact 38 covid-19 deaths in those weeks.

Meanwhile Salford records 46 covid-19 deaths in weeks 14 + 15 (none for week 13, which is reasonable from the graph).

There were average 7.75/weekly deaths weeks 1-12 in care homes, and 138 (46 per week) in weeks 13-15.

So what we end up with is 115 total excess deaths week 13-15, of which 46 (40%) were coded as covid-19.

Meanwhile in Manchester there are 38 extra deaths of which 10 (26%) were coded as covid-19.

So are conclusions are:

1. Yes, covid-19 is ripping through care homes in Salford, more so than Manchester
2. This is because covid-19 is much worse in Salford than Manchester, but also in part because Salford is better (but still bad) at certifying
3. Everyone who says that loads of the covid-19 deaths are in fact cancer/diabetes (insert other illness here) deaths needs to STFU. It's the other way round - lots of missing deaths caused by covid-19 are not being coded as such.

The next graph shows the care home deaths by week for England & Wales. Week 13 it starts to rip through, though Week 12 is probably elevated.

For weeks 8-11 there are 9945 deaths total in care homes, and for weeks 12-15 there are 16,378, an excess of 6,433. Only 1643 (26%) were coded as covid-19. The other 74% are likely to be covid-19 as well.

For deaths at home over the same period there are 4,240 excess, of which 655 (15%) were coded as covid-19.

Interestingly there is NO SPIKE in hospice deaths.

1. It follows that we should probably not pay too much attention to the covid-19 death figures as they are not comparable between LA, but instead just use excess deaths.

@BigChocFrenzy
The Tim Spector research that I mentioned earlier today using the Covid tracker app suggests earlier cases. In case you don't know Tim Spector is Prof of Genetic Epidemiology at Kings, did all the Twins research and has published several books and is very reputable.

On the Oxford study @Reallybadidea is right- the whole point of the study was to demonstrate that actually some very extreme scenarios are actually possible and we have no idea because we haven't been testing. They didn't say they think that the 'circulating early' and lots infected model was correct, just that it is one of many possibilities because we know so little (and knew even less at that point in time). People (ie media) have been taking it out of context and focusing on the extreme of the models they suggested because it makes good clickbait. I've not come across any research or data suggesting there was circulation in December/early January although common sense looking at our flights from China suggests there may well have been a few isolated imported cases that were put down to flu. If it had circulated we would have seen the deaths start to rise earlier.

I agree about death certificates, particularly early on. When my mother died in hospital the death certificate was very accurate, listing 4 things that led to her death. When my father died in a care home the death certificate stated "chest infection". This was the final straw for his body but not really an accurate picture.

surely though we will never know about the earlier cases as there was no testing for covid until much later.
And assuming levels were low then it may not have shown up as a peak in deaths, especially if mixed in with the seasonal flu

@nettleTea totally agree.

All it does mean is that the 'starting point' for any epidemiological model could be way out.

@clarexbp I agree with the amounts of money spent calculations being huge, but compared with the cost to economy - not so - which is why lockdown will have to end sooner rather than later.

In addition bear in mind that a lot of this research is taking place in Universities which are facing mergers and closures, and departments are desperate to prove their worth. The most popular research politically will get given more staffing etc.

"Is there an ‘Imperial’ model or just a standard epidemiological model ?
I would have thought that there would be a standard model with the 4 (I think) simple to state (though not to solve) differential equations."

Larry A math model has some inbuilt assumptions, which we don't normally question, because they are in every such model in the field

At least this is the case in my Maths / Physics / engineering field:
we were aware on the very rare occasion that we needed to model a different kind of entity of situation,
that all the old assumptions need to be reviewed - and some might be missed

imo, a model built for flu epidemics may have different algorithms to one specifically designed for COVID, not just different values of input parameters,
because there are important differences, some of which may need to be in the model, some won't

Some possibilities:

• Before symptoms appear, COVID has an infection period of a week or sometimes longer
  Reduces the effectiveness of the classic test, track, isolate methods

• The lethality range is likely an order of magnitude above that of flu - which may just mean changing one parameter,
  or it may mean that e.g. the calculation / discretisation of an integral needs to have a different algorithm

• Similarly for the likely much higher R0 of COVID.
  Is the difference between ordinary and more severe flu modelled primarily as its lethality, or also with a significantly increased R0 and can this just be changed via a paramer value ?

• Flu deaths assume the presence of anti-viral treatments for serious cases
  Is this just a simple parameter value that can be modified ?

• COVID cases spend significantly longer in ICU if they go there
  So hospital resources, planning etc need to be different to keep within the curve of being able to treat all cases

"a lot of this research is taking place in Universities "

I admit I am more hopeful of results from better-funded institutions, private / commercial / govt outside the UK
Different attitudes

The decades-long penny-pinching mindset in the UK seems to remain - like Hancock - of "the British taxpayer getting their moneysworth
"Penny-wise and pound-foolish"
rather like running down the NHS for years

I hear "money is no object" from e.g. normally thrifty German officials and commentators
because when a $4 trillion economy (Germany) suddenly loses large chunks of GDP,
then it is fully worth investing $ billions in parallel research to shorten this crisis by even a couple of months

Here's a chart showing the excess deaths by region, by week.

If you take London out then the regional differences become clearer.

Thanks all, still following the thread with interest .
Thanks for taking time to post the brilliant graphs. Plus great discussion and links to articles and resources.
to you all

After Shoot's UK contour plots,
these are the case plots over Germany's 16 states, normalised / 100,000 pop

1. Total cases
2. New cases over the previous 7 days - clearly a few remaining hotspots

There is also a regional breakdown, which is probably not interesting to anyone here
but there is no ethnic breakdown I have found, only age & sex

The age group when normalised / 100,000 pop looks like a huge individual risk in the 90+ age group of catching COVID,
but these are only 2.8% of the total cases

The median age of those infected is 50