Friday night Covid Plotting #20

New Topic
This topic has been archived, and won't accept reply postings.
 wintertree 03 Apr 2021

Easter is here and change is in the air.  Or in the data.  After a few weeks of headline case numbers stubbornly holding level, we are seeing a welcome return to decay in all four home nations. 

In England, we know that a lot of the plateau is down to increased asymptomatic testing with LFDs associated with schools reopening to full in-person teaching, and that symptomatic PCR cases (which had been our barometer of infection for many months) were still falling - albeit very slowly.  Now, the number of LFD cases is plummeting and the number of tests conducted is reducing, as school is out for Easter.  We can only separate LFD cases from PCR cases for England, so for consistency I present total cases for all four nations, but I'll dig in to the split for England in a later post. 

I assume generally similar things are happening in the other 3 home nations, and there have been comments with experience along those lines from a couple of Scotland based posters over the last few weeks.

There exceptionally fast halving times seen in plot 9x are in the highly provisional window for this analysis, and are also I think largely reflecting the disseverance of the LFD signal - which was partly representing an additional fraction of infections being caught as cases, so I don't for a moment think that real infections are halving as fast as cases now appear to.  But I do think they're decreasing everywhere - by proxy from the more detailed England data.


OP wintertree 03 Apr 2021
In reply to wintertree:

It's a race to the bottom for all measures in England - everything continues to decay.  Ordinarily with the lag involved we'd expect cases to become negligible before deaths, but as plot 9e shows, deaths have been halving faster than both hospitalisations and cases since early February, as the vaccination program decouples deaths from infection.

The plateau I thought might be emerging in deaths looks less likely this week, the demographic data in a later post looks at this some.

Link to previous thread: https://www.ukhillwalking.com/forums/off_belay/friday_night_covid_plotting_19-...

Post edited at 21:10

OP wintertree 03 Apr 2021
In reply to wintertree:

Plots 18.1, 18.2

Cases are back to halving quickly in all regions except the North East. 

  • The decay of symptomatic testing PCR cases - the "barometer of infections" will be halving more slowly than this as this data includes the falloff of LFD cases with the end of the school term.  But, PCR cases are still falling fast - halving every 14 days at the National level, as I'll go through in a later post.
  • This plot shows cases as rising in the North East, but only in the last few days of the plot which, as always, are very provisional analysis.  Looking at the raw regional data, I think this is a particularly deep weekend low giving the impression of growth in the week that follows, and that this will be resolved with more data.

Hospitalisations - these are falling everywhere in all regions.  The characteristic times are looking noisier - as expected as the absolute numbers fall and fall towards zero

Deaths - these measurement are increasingly dominated by noise I think due to the very low numbers.   Which is great.  I might have to retire the regional analysis in the next couple of weeks.

Plots 22, 22r

Both measures of hospital occupancy remain in decay in all regions and are getting close to the origin - the start of a return to normality for healthcare.  I very much hope we don't have any more trips around this loop.

Post edited at 21:15

 Si dH 03 Apr 2021
In reply to wintertree:

I was wondering, have you changed anything in your deweekending algorithm in the last couple of weeks, with the all-case data? Because the LFTs peak on different days, the usual weekly trend has changed when all cases are viewed together (not when viewing PCR alone). This week the highest cases have been on Sunday 28th rather than Monday 29th, because of an LFT peak on the Sunday. If the algorithm is unchanged and moves cases from the Monday to the Sunday as per usual, then it will increase rather than decrease the error this week, which would bias your outputs where they use data up to those dates. The effect at the leading edge could be substantial?

Although of course for the next 1-2 weeks it will be different again due to Easter.

Post edited at 21:30
OP wintertree 03 Apr 2021
In reply to wintertree:

I'm looking at the PCR±LFD data and the demographics together.  Both are different views in to the same data - unfortunately it's not all separated out by test type and age.  The demographic data includes PCR and LFD results.

Plot x0 and x1 show that both PCR and LFD positives are falling over the days since schools closed.  It looks like the 3-week long spell of slow PCR case decay (grey 'x' markers stuck at a halving time of ~42 days, bottom right subplot of x1) is coming to an end, with the most recent 3 non-provisional days of data representing a halving time of 14-days (when looking at their decay from a week previously).

  • This near stagnation in PCR cases coincident with the schools period suggests to me that schools being open led to a similar increase in transmission to the "cold shoulder" in February. (partially offsetting the massive reductions from lockdown measures, but not quite enough to cause growth in symptomatic cases)
  • Children in schools?  Children on busses?  More parents and carers returning to the workplace?  Probably aspects of all of the above in various measures as well as other non-school related factors.
  • On the subject of busses:  I can't believe how many busses I am seeing out there with all the ventilator windows closed.  Seriously, the bus firms have had a whole god damned year to sort their crap out.  What do we get?  A few measly adverts on the radio and social media.  Just take all the fracking ventilator windows off and weld some grilles over the spaces.  

There's a really obvious island of orange in the bottom right of D1.c where cases grew and and now returning to decay in school ages.  It seems reasonable to assume most of this is LFD data.  I've added a second version of this plot with scribbles on.  The yellow line is an "iso-case" level I've drawn by eye where I'd expect cases to have decayed to a daily rate corresponding to the dark red colour.  There's a lot of bright red to the right of this indicating a failure of cases to decay, and in the youngest ages, growth.   There's a lot of almost white in the working aged adults around the corresponding yellow line on the rate constant plot showing the stagnation of decay.

This to me looks pretty compelling support for schools as the factor limiting the decay of cases - adults in the range most likely to be parents/carers are seeing the decay of cases slow to a crawl during school opening. 

We know some of these cases are "extra" LFD positives from the asymptomatic testing available to parents; but we also know PCR decay slowed to a crawl, and as this is where most cases are, it must be where most PCR cases are - which strongly suggests symptomatic PCR cases were anchored by the return of schools. 

I stress that I'm not saying schools are behind the transmission - a lot changes when they reopen as parents/carers suddenly enter a different world.

Plot D3 shows in or heading for decay in school ages - largely the collapsing LFD numbers - and the decay speeding up in all other ages except 90+.  With ~10 cases/day in the 90+ bracket the analysis becomes very sensitive to noise, and my continued failure to get meaningful errorbars on the exponential fits starts to raise its head...  Interestingly the move to faster decay is across all ages below 90 - so this includes those who clearly aren't parents of school children, so LFD results can safely be assumed not to be influencing what we see - suggesting that the changes are percolating through the ages.

Sooooo....  Perhaps I'm way of in my take on this, but it looks like schools take us right to the brink of growth - but not over in to it.  Hopefully cases will halve over the easter holidays, seeing England down to ~1,250/day by the return of schools.  Meanwhile, we're now seeing the vast majority of people over 50 having received their first vaccine dose, and I hope that will soon manifest as increased decay of infections and cases in those ages, which buys some more wiggle room so that cases keep falling appreciably when schools reopen.

I know I do keep banging on about falling cases; I know the English policy is now primarily about hospital levels, deaths, vaccines and variants, but one thing's for sure - the lower cases are, the lower the probability of a new nasty variant emerging locally, and the easier it will be to keep a lid on any variants with partial vaccine evading potential.   


OP wintertree 03 Apr 2021
In reply to Si dH:

> This week the highest cases have been on Sunday 28th rather than Monday 29th, because of an LFT peak on the Sunday. If the algorithm is unchanged and moves cases from the Monday to the Sunday as per usual, then it will increase rather than decrease the error this week, which would bias your outputs where they use data up to those dates.

Excellent observation, thanks.  I'll dive in to some details plots tomorrow or Monday and get back on that.  Briefly however, I'm not too concerned as:

  1. Cases are not re-assigned back in time "blind" from the Monday/Tuesday high, but based on an iterative attempt to converge data to a linear form over a short time window (much shorter than the characteristic time, so a good approximation). So, with the LFDs, Monday data just won't be moved back as much I think.  
  2. I always run select plots without the deweekending as well and compare them - they're a lot more jagged but any significant discrepancies should stand out, and none did.

> The effect at the leading edge could be substantial?

The leading edge of the exponential rate constant plots is always highly provisional mind.

> Although of course for the next 1-2 weeks it will be different again due to Easter.

Yup; I'll have a look at the testing numbers and might shunt a couple of days about once it's all resolved, but there's so much going on now anyway it'll probably be lost in the noise.

OP wintertree 03 Apr 2021
In reply to wintertree:

The demographic data shows how deaths continue to fall rapidly - and thankfully the ages were cases have been stagnating has little overlap with the higher fatality rates of the higher ages.  The deaths data for adults under 70 still looks like it might be plateauing somewhat - but it's low numbers and with the rates in other ages continuing to plummet, the overall trend is still down.


OP wintertree 03 Apr 2021
In reply to wintertree:

A grab bag of final plots.  

It's been a story of decay for the UK today.  Looking beyond our borders, a lot of places are going exponential by the looks of it, for example Brazil, India, France, Turkey, Germany and Iran.  Several of them look like it's going very wrong, very quickly.  When it comes to vaccine production and distribution, it's sobering to think of the difficulties getting enough vaccine produced fast enough in the UK and across the channel, and then to consider that our population is smaller than any of those countries.  There are more plants coming online in the UK and beyond in the coming months, and they're going to be very busy for the foreseeable.  

Plot 16  - UTLA Watch Plot - I set the "rebound from minimum" threshold to 1.5x this week and baselined it over a 5-week period.  Pretty arbitrary stuff, but this shows some of the areas doing a less convincing impression of decay amidst all the LFD additions over recent weeks.

North Lincolnshire is the only one to have exceeded its baselining level.  It does now look to be in decay but both that and North East Lincolnshire are still significant holdouts.

Vaccine Plot

There's a big switch to second doses now.  It looks to me like the second doses are probably larger than the first doses 12 weeks previously on the plot - although that week is not in the daily data.  I might do a noddy analysis to see how far ahead or behind we are on second doses and to project ahead on a rigid 12-week schedule.  That slim margin of pale blue going as first dose is slowly working its way down to my age range...

Variants Plot

No change from a few days ago - more here - https://www.ukhillwalking.com/forums/off_belay/friday_night_covid_plotting_19-....  This data is hard to understand as we don't know how much is from surge testing exploring and nailing down a network vs how much indicated genuine exponential growth. I'm kind of expecting to hear more about the new "red" variant soon enough however, it looks problematic if it's been growing exponentially for the last 5 weeks.   

  


 Misha 03 Apr 2021
In reply to wintertree:

> Sooooo....  Perhaps I'm way of in my take on this, but it looks like schools take us right to the brink of growth - but not over in to it.  Hopefully cases will halve over the easter holidays, seeing England down to ~1,250/day by the return of schools.  Meanwhile, we're now seeing the vast majority of people over 50 having received their first vaccine dose, and I hope that will soon manifest as increased decay of infections and cases in those ages, which buys some more wiggle room so that cases keep falling appreciably when schools reopen.

I think that’s spot on. The implication is that we’ll be back into growth post 12 April - the question is, to what extent. Hopefully it will be relatively slow exponential growth (I doubt it will be linear). The other factor is the weather - it should increasingly help to suppress case growth to some extent. The return of indoor meetings with friends and family and indoor hospitality on 17 May will be a bigger challenge I think. Perhaps they will push it back if cases are growing substantially by then but I doubt it, not least because cases aren’t one of the key factors which will be considered. Anyway, at this stage any decay we can bank is good...

> I know I do keep banging on about falling cases; I know the English policy is now primarily about hospital levels, deaths, vaccines and variants, but one thing's for sure - the lower cases are, the lower the probability of a new nasty variant emerging locally, and the easier it will be to keep a lid on any variants with partial vaccine evading potential.   

Also, with the vast majority of the under 50s not yet vaccinated, I don’t think we want to find out what naturally acquired herd immunity looks like in that demographic. 

 Misha 03 Apr 2021
In reply to wintertree:

Re second doses, we’re on about 10 weeks. That’s just comparing the number of second doses today vs the day when there was a similar number of first doses, which was 20-21 Jan. This seems about right as they presumably don’t want to run over 12 weeks and this gives them some margin. No doubt much of the margin will get used up given the two big weeks in the first half of Feb.

 Bobling 04 Apr 2021
In reply to wintertree:

> Easter is here and change is in the air.  Or in the data.  After a few weeks of headline case numbers stubbornly holding level, we are seeing a welcome return to decay in all four home nations. 

Amen brother!

 Toccata 04 Apr 2021
In reply to wintertree:

Still the most useful source of Covid data on the net. Chapeau.

1
In reply to wintertree:

> This data is hard to understand as we don't know how much is from surge testing exploring and nailing down a network vs how much indicated genuine exponential growth. I'm kind of expecting to hear more about the new "red" variant soon enough however, it looks problematic if it's been growing exponentially for the last 5 weeks.   

I'm having a hard time settling on a conclusion about how the proportional increase in cases sequenced as absolute cases reduces affects the meaning of this plot. It's possibly very hard to interpret. Possibly very easy. I keep running into brain fart and have to start again whenever I try to think it all the way through.

 Si dH 04 Apr 2021
In reply to wintertree:

Thanks again. Your deweekending algorithm is slightly more complex than I'd remembered.

Just a couple of observations. In plot D1c, the end of growth in 10-15 and 15-20 will be at least part due to the end of LFT testing for Easter like you said. However this can't be the case in the 5-10 age range, because primary schools aren't doing LFT mass testing. The growth in that age range rose only temporarily but it's a good thing. As per the discussion on last week's thread, ONS random sampling seems to suggest a levelling off in secondary ages too, independent of LFT effects.

Secondly I strongly suspect the weather is helping us again with the further turndown of PCR cases this week across the age range, given the strength of the effect it has had in the past. It's been warm and dry for almost a week now.

The whole of the north Lincolnshire and south Yorkshire/Humber area has now been falling slowly (or not at all) for long enough that despite having relatively very low rates in January, the region now has the highest prevalence in the country. Previously the behaviour was only visible with a lot of analysis, now all you have to do is look at the colours on the top level dashboard map...

 Si dH 04 Apr 2021
In reply to Longsufferingropeholder:

> I'm having a hard time settling on a conclusion about how the proportional increase in cases sequenced as absolute cases reduces affects the meaning of this plot. It's possibly very hard to interpret. Possibly very easy. I keep running into brain fart and have to start again whenever I try to think it all the way through.

It's also slightly confusing to me I think because it is plotting cumulative (sequenced) cases whereas all the other vaguely similar plots are plotting new cases per day/week. When there is a doubling time of 15 days in cumulative cases that's obviously not as bad as a doubling time of 15 days in the case rate. Without independent effort, I struggle to interpret whether the rate at which these variants are being found is increasing very much. It's confounded by other issues anyway of course like you and WT point out. But still worth tracking.

Post edited at 07:37
In reply to Si dH:

Well yeah, this is the thing. If it's actually in growth during a lockdown with ~50% vaccination then some people in a smoke filled room somewhere should be having a really good panic, and I should probably just quit the BMC and join the RHS.

If it's just an artefact of the compounding of a lot of relative numbers making a line go up then one day this might all get better.

 BusyLizzie 04 Apr 2021
In reply to wintertree:

Thank you again - I wait anxiously for your analysis every week, and whether the news is good or bad your objectivity is always comforting. Hope you are bearing up, and now able to get more exercise.

Here's to warm sunshine, for as long as we can get it.

OP wintertree 04 Apr 2021
In reply to Si dH:

> It's also slightly confusing to me I think because it is plotting cumulative (sequenced) cases

Correct.  The reason for this is that the data is released on an irregular time base coarser than 1 day, so plotting of new cases on a daily basis - like the other plots - would involve resampling the data, which isn’t really justified, which introduces issue and which isn’t actually needed...

> When there is a doubling time of 15 days in cumulative cases that's obviously not as bad as a doubling time of 15 days in the case rate. Without independent effort, I struggle to interpret whether the rate at which these variants are being found is increasing very much

Do you have a copy of Boas or Kreyszig dusting up on an old bookshelf? There’s a very interesting property to exponentials in the natural base... The derivative of e^x is e^x.  Applying the chain rule, the derivative of e^(kx) is ke^(kx), so the doubling time of the cumulative and daily measures is the same.  Something I should have explicitly noted with this plot!

I can’t sweep all the other confounding issues away you both note.

OP wintertree 04 Apr 2021
In reply to Si dH:

> Just a couple of observations. In plot D1c, the end of growth in 10-15 and 15-20 will be at least part due to the end of LFT testing for Easter like you said. However this can't be the case in the 5-10 age range, because primary schools aren't doing LFT mass testing. The growth in that age range rose only temporarily but it's a good thing. As per the discussion on last week's thread, ONS random sampling seems to suggest a levelling off in secondary ages too, independent of LFT effects.

Good points - thanks.  You can see on the right side of plot D1.c that the 5-10 age range turns to growth (orange) later - this first with the others first showing growth due to the switch-on of mass LFD testing picking up some existing infections in the age ranges 10-20, but it taking more time for additional, new, symptomatic infections to be created and then picked up in the 5-10 range. 

All the more mysterious than that the growth in cases plateaus sooner in that range.  

The levelling off implies either saturation is starting to kick in - which feels unlikely at this stage - or non-linear effects in control measures - they get better with more cases.  This seems unlikely but I've been pondering it.  Two possibilities I could think of that lead to rate limiting steps occurring as cases rise:

  • When a lot of infections are symptomatic in this age and many will test LFD negative, it will on average take multiple infections in a cohort/network/group before any are picked up.  If contact tracing then works out, some of the undetected, infected people will get quarantine orders.
  • Whole class suspensions have been happening in some locations as a result of test positives.

As you also say there's the weather and it's a stunning easter weekend, so let's hope that sees some more aggressive decay before schools return.

OP wintertree 04 Apr 2021
In reply to BusyLizzie:

Thanks.  Feeling like I have half an idea what's going on is very useful in terms of not stressing about it all.

> Here's to warm sunshine, for as long as we can get it.

Indeed - it's the lack of home schooling that's made the biggest difference and gives me some time to get out - how fast running fitness fades, and rather worryingly for the first time in my life, long walking fitness is not what it was - but the sunshine is a perfect bonus.

OP wintertree 04 Apr 2021
In reply to thread:

Last Wednesday is - or should now be - out of the provisional window for reporting.  The PCR only data showing a week-on-week decrease to about 62% in case numbers, corresponding to a halving time of about 10 days. 

It looks like the brakes are coming off the decay and we're going back to the "Rapid mode" seen after the February cold spell and before schools returned.

Never look a gift horse in the mouth.

To me, this data strongly that there are levers that can be pulled to control the transmission rate - given how uniform change in rate constants has been across working aged adults I think a lot of these levers apply beyond the confines of the classroom, even if they are associated with the return to school.   I still suspect a lot of it comes down to opening windows and ventilating workplaces and public transport.

Post edited at 17:12

 Wicamoi 04 Apr 2021
In reply to wintertree:

Thanks as always wintertree - your graphs and interpretations are very much appreciated chez Wicamoi. There's a very cold forecast for this week, especially in Scotland....

On a personal note: the fitness that you can lose quickly is the same fitness that you can quickly regain.

OP wintertree 04 Apr 2021
In reply to Wicamoi:

> There's a very cold forecast for this week, especially in Scotland....

Indeed; here on the fringes of Weardale, 18°C a few days ago is giving way to -4°C, 40 mph winds, snow and hail tonight apparently.  I fear for the buds on the fruit trees.  Still, the days generally still look nice enough and the cold is just a flash in the pan compared to February.

 3 Names 05 Apr 2021
In reply to wintertree:

Im in London, its snowing!

In reply to Si dH:

> Without independent effort, I struggle to interpret whether the rate at which these variants are being found is increasing very much. It's confounded by other issues anyway of course like you and WT point out. But still worth tracking.

So the briefing today pointed out the possibly obvious possibly plausible explanation that I'd missed, which is that this rise is driven by testing on arrival. If it is that, I'm less worried. But not much less.

OP wintertree 05 Apr 2021
In reply to Longsufferingropeholder:

> So the briefing today pointed out the possibly obvious possibly plausible explanation that I'd missed, which is that this rise is driven by testing on arrival. If it is that, I'm less worried. But not much less.

For the SA variant detection largely driven by travel, but for the other variant, which was first sequenced in the UK.... ?? 

But, we know cases leak out of MIQ in NZ which is far more robust than arrangements here, and we're not sequencing the majority of cases, so there's no guarantee that there's not some SA variant in common it transmission "below the radar".

In happier news, last Thursday has joined last Wednesday in having a big drop in cases week-on-week, again equivalent to a halving time of about 10 days.  Last Wednesday has stayed there, so it looks increasingly unlikely this big drop is due to reporting lags.  If that keeps up, it'll see us down to 1,000 cases/day in England by the return of schools.


In reply to wintertree:

Oh, yeah, no illusion.  It will be leaking out here but the exponential nature during a lockdown was intensely worrying, until the realisation that it's counting imported cases.

OP wintertree 05 Apr 2021
In reply to Longsufferingropeholder:

> Oh, yeah, no illusion.  It will be leaking out here but the exponential nature during a lockdown was intensely worrying, until the realisation that it's counting imported cases.

Did they say this was the case for VUI-21FEB-03 (B.1.525 ) in the briefing as well as the SA variant?

In reply to wintertree:

Not explicitly, no

 Misha 05 Apr 2021
In reply to wintertree:

Today’s press conference was unexpectedly cautious on the plans for cross border travel. That’s a good thing I think. I’m actually wondering half seriously whether we could kill two birds with one stone here - stay on top of Covid and make progress towards our green targets but destroying our air travel industry...

OP wintertree 06 Apr 2021
In reply to Longsufferingropeholder:

Thanks.  That didn’t touch in the other variant (still under investigation rather than of concern, the red one on my plot).  This is the one that’s likely being detected mostly locally I think.  The UK has reported the most sequences for this variant, and the first:

https://cov-lineages.org/global_report_B.1.525.html

It’s still not been declared “of concern” and there’s a lot of factors making the numbers hard to understand, but we’ll see where it goes soon enough I expect...

Post edited at 07:57
OP wintertree 06 Apr 2021
In reply to Longsufferingropeholder:

I only just came across this short letter from Alan McNally, from late February.  Good reading on the subject of variants, testing and immunity and accessible to a wide audience.

https://www.bmj.com/content/372/bmj.n504

OP wintertree 06 Apr 2021
In reply to wintertree:

PCR cases continue their ever more rapid fall; cases decreased equivalent to an 8-day halving time in the week up to last Friday, Aprilⁿᵈ.  

I think this is probably in part due to under-sampling on the bank holiday.  This plot is based on raw data without any filtering or smoothing; we might see some balance in the form of a longer halving time for this Saturday, then a similar split with a shorter halving time for the bank holiday Monday once it resolves, and a longer one for the Tuesday.

Even accounting for bank holiday effects, I think it continues to look great. 


In reply to wintertree:

If the VUI/VOCs really are in growth by local transmission, when overall cases are doing this, then I'm equal parts confused & worried. Must be mostly at the airport. Must be.

OP wintertree 06 Apr 2021
In reply to Longsufferingropeholder:

> If the VUI/VOCs really are in growth by local transmission, when overall cases are doing this

The UK variants table hasn't been updated since the fast decay really set in, and it's by reporting date not by specimen date, which will add some more lag - probably a few days on top of the P1/P2 reporting lag.

> Must be mostly at the airport. Must be.

What you need is a pair of peril sensitive sunglasses.  I'm pretty sure that'll hide the red curve on my plots from you. 

In reply to wintertree:

> What you need is a pair of peril sensitive sunglasses.  I'm pretty sure that'll hide the red curve on my plots from you. 

If that curve is real, then it's time to change the bulb.

 Toerag 06 Apr 2021
In reply to wintertree:

Have you seen this site? I think you'll like it.  Would be interesting to work out the causes of the sudden changes in relative prevalence of the variants.

https://covariants.org/per-country

OP wintertree 06 Apr 2021
In reply to Toerag:

That's great, thanks.

Mostly it looks like the variant first sighted in Kent is out-competing all other variants it comes in to contact with.   Interesting to look at that variant vs the SA one - in Austria, the SA one is not out competing it, but in Turkey it is - just as Turkey starts to have a big exponential rise.  It could be luck of the draw as to which is present at the start of an amplification phase; it's hard to tell without cross comparing to the absolute growth in cases for each place.

 minimike 06 Apr 2021
In reply to wintertree:

Although as they note the approach of sequencing in response to SGTF biases pretty strongly towards the Kent variant.

 3 Names 06 Apr 2021
In reply to wintertree:

> That's great, thanks.

> Mostly it looks like the variant first sighted in Kent is out-competing all other variants it comes in to contact with.

Does this show that Immunity is still effective against all variants?

 Misha 06 Apr 2021
In reply to wintertree:

The case numbers look a bit too good to be true - let’s see what comes through in the next 3 days in case there’s significant reporting lag. Also people seem to get tested less at weekends (PCR positives by specimen date are lower at weekends), so that might apply to BHs as well. Still, any reduction is a good reduction.

What do you make of the third wave modelling which was in the news yesterday? It strikes me as overly pessimistic re vaccine efficacy - clearly that’s prudent but I hope they’re wrong! My takeaway from it is the 12 April relaxation is ok, the 17 May one will be problematic but possibly manageable and the 21 June full unlock would be a disaster (even if the modelling is u duly prudent, it won’t be pretty). The obvious solution is to potentially push back and / or tighten the 17 May relaxation and then delay any further relaxation indefinitely - probably till next summer. Better to keep certain sectors on furlough for another year than have another lockdown. My concern is the government will place reliance on mass LFTs and miss the boat on control measures again.

I suspect either way measures will need to be tightened in early autumn (eg going back to the 12 April position) to get ready for a seasonal peak in winter. Whether they will actually be tightened

is another question. May be from this time next year we’ll be able to go back to more or less normal. 

1
 Misha 07 Apr 2021
In reply to wintertree:

If true, this suggests most under 50s won’t start getting vaccinated until mid June (c. 27m second jabs to do @ 2.7m a week). Which would mean ‘freedom’ on 21 June would be even more insane than it originally seemed.

https://www.theguardian.com/politics/2021/apr/06/england-covid-vaccine-prog...

1
OP wintertree 07 Apr 2021
In reply to minimike:

> Although as they note the approach of sequencing in response to SGTF biases pretty strongly towards the Kent variant.

Good point, thanks.  As cases wind down here I've wondered if we can't donate our excess sequencing capacity abroad, by processing samples air mailed in.  As well as useful horizon scanning for us, it would help other places understand and so deal with their outbreaks in a better way, which is to everyone's benefit.

In reply to 3 Names:

> Does this show that Immunity is still effective against all variants?

Sequencing data is really hard to interpret given how it's deployed "strategically" rather than random, and that website presents only ratios of sequences not absolute numbers.  I wouldn't want to read much in to it without studying the local situation in a lot more detail.

In reply to Misha:

> The case numbers look a bit too good to be true - let’s see what comes through in the next 3 days in case there’s significant reporting lag

Re: Friday's numbers - you can already see the provisional halving time is longer for Saturday and will I think go up more today - not lag but under-sampling on a bank holiday, as previously seen mid-winter.  The previous days are holding firm and are I think real however.

> What do you make of the third wave modelling which was in the news yesterday

I haven't read it yet or much other recent news (had my head in a book on the OpenBSD packet filter for light entertainment). 

Really noddy maths:  

  • We have about 15m under 18s and about 52m adults in the UK.
  • Assuming 85% of adults and 0% of under 18s are vaccinated by June 21 (the rest largely being refusals), we will have (0.85 x 52) / (52 + 15) = 63% of our population immunised.  Let's assume 20% of people have now been infected (generous assumption?) and have decent naturally acquired immunity, so 20% of the unvaccinated 37% are also immune, giving a total of 70% of people with some immunity.  The herd immunity threshold for the current predominant variant is probably ~90%.  So, once restrictions are gone, we will have uncontrolled spread and 30% of people are unprotected - so 20% of the population will get it before we hit the 90% herd immunity threshold and infection dies out naturally.   With 4% of that 20% going to hospital and 1% of them dying, it looks like 500,000 people are going to hospital and 130,000 people are going to die.
  • These are crude estimate and the gulf between them and reality is in...
    • The demographics of refusal - higher at younger ages?  Seems likely and will make reality less bad than my noddy maths.
    • How structured or random the non-immune sub-population will be.  The actual herd immunity threshold depends critically on how the sub-population is connected (its network graph), as does the shape of their hospitalisations and deaths peaks.  Perhaps they'll be sharp enough to overwhelm healthcare again, perhaps not - as the immune people will moderate and slow the spread if they're a significant fraction of the contacts the non-immune will have.  My worry is that as refusal is associated with distinct social groupings, the moderating effect on the immune on growth in cases in the non immune won't be great, and we'll see sub-week doubling times in the susceptible population.  Largely their choice, but the NHS' job to deal with the consequences - and there will be some immunocompromised individuals paying the price too.

Crude estimates like the above suggest there's a lot of cause for concern.  I haven't seen any demographic or social data on vaccine uptake/refusal, and without that a model is I think not that much more predictive than crude estimates.  I don't know how much of this information the modellers are getting?  June feels optimistic to me; hopefully the powers that be continue to study the data and make decisions accordingly, especially after the May step.

But, if we get a variant that evades immunity to the point of infection but not hospitalisation/death, and we drop controls, we're back to a sub-week doubling time and things go really south, again, with hospitalisations coming from the non-immune.  

> If true, this suggests most under 50s won’t start getting vaccinated until mid June (c. 27m second jabs to do @ 2.7m a week). Which would mean ‘freedom’ on 21 June would be even more insane than it originally seemed.

Yes.  Not a good situation for people in their 40s especially if it comes down to "put yourself at risk or loose your job".  A university not so far from the Wintertree bunker has it seems announced an extra week of term after restrictions are dropped to enhance the "Wider Student Experience".  I don't even know where to begin.  I'm thankful that I don't work in a pub or eatery frequented by students, a lot of people do.

Post edited at 13:23
 Toerag 07 Apr 2021
In reply to wintertree:

> Good point, thanks.  As cases wind down here I've wondered if we can't donate our excess sequencing capacity abroad, by processing samples air mailed in.  As well as useful horizon scanning for us, it would help other places understand and so deal with their outbreaks in a better way, which is to everyone's benefit.

Sequencing doesn't affect dealing with outbreaks as even the least infective variants require proper controls. Horizon scanning and immunity evasion are entirely valid reasons though, and capacity should be used for this in its own right.

> Really noddy maths:  

> We have about 15m under 18s and about 52m adults in the UK.

> Assuming 85% of adults and 0% of under 18s are vaccinated by June 21 (the rest largely being refusals), we will have (0.85 x 52) / (52 + 15) = 63% of our population immunised.  Let's assume 20% of people have now been infected (generous assumption?) and have decent naturally acquired immunity, so 20% of the unvaccinated 37% are also immune, giving a total of 70% of people with some immunity.  The herd immunity threshold for the current predominant variant is probably ~90%.  So, once restrictions are gone, we will have uncontrolled spread and 30% of people are unprotected - so 20% of the population will get it before we hit the 90% herd immunity threshold and infection dies out naturally.   With 4% of that 20% going to hospital and 1% of them dying, it looks like 500,000 people are going to hospital and 130,000 people are going to die.

> These are crude estimate and the gulf between them and reality is in...

> The demographics of refusal - higher at younger ages?  Seems likely and will make reality less bad than my noddy maths.

> How structured or random the non-immune sub-population will be.  The actual herd immunity threshold depends critically on how the sub-population is connected (its network graph), as does the shape of their hospitalisations and deaths peaks.  Perhaps they'll be sharp enough to overwhelm healthcare again, perhaps not - as the immune people will moderate and slow the spread if they're a significant fraction of the contacts the non-immune will have.  My worry is that as refusal is associated with distinct social groupings, the moderating effect on the immune on growth in cases in the non immune won't be great, and we'll see sub-week doubling times in the susceptible population.  Largely their choice, but the NHS' job to deal with the consequences - and there will be some immunocompromised individuals paying the price too.

There's also 'I've been vaccinated and won't die so why should I self-isolate or get tested' behaviour which WILL happen, especially as infections are likely to be asymptomatic or minor and people don't want to SI because of money or wanting to go on that holiday they booked.  As we've seen in the past, the virus spreads faster than governments controls can keep up with so it will end up in low doubling times.

> Crude estimates like the above suggest there's a lot of cause for concern.  I haven't seen any demographic or social data on vaccine uptake/refusal, and without that a model is I think not that much more predictive than crude estimates.  I don't know how much of this information the modellers are getting?  June feels optimistic to me; hopefully the powers that be continue to study the data and make decisions accordingly, especially after the May step.

> But, if we get a variant that evades immunity to the point of infection but not hospitalisation/death, and we drop controls, we're back to a sub-week doubling time and things go really south, again, with hospitalisations coming from the non-immune.  

....and of course the vaccines won't be able to be tweaked fast enough to prevent additional lockdowns either.

> > If true, this suggests most under 50s won’t start getting vaccinated until mid June (c. 27m second jabs to do @ 2.7m a week). Which would mean ‘freedom’ on 21 June would be even more insane than it originally seemed.

By your noddy maths there's a pool of ~13million infectables. Healthcare was virtually crashed by ~5k admissions a day at xmas which was caused by ~50k new cases a day the previous week, which was the result of locking down at ~40k new cases just before xmas.  Even starting from a point of 500 new cases a day (the low of last summer) it's only 6 doubling periods before healthcare is crashed again and only ~100k people out of the 13 million have been infected.  People have no idea how low the healthcare bar is compared to the infectable pool.  Even if vaccination reduces the transmissibility down to 25%, that means an R of 1+ instead of 4+ in an uncontrolled environment and we all know where an R of 1+ ended up at Xmas. There's going to be some very upset people this summer, either because they can't go on holiday or they're having their own lives ruined by covid or going to funerals.

 MB42 07 Apr 2021
In reply to wintertree:

I realise you pointed out its noddy maths but the susceptible 30% isn't just refusals, it includes all 15m <18 year olds for whom the risk of hospitilisation is significantly lower and who are possibly less likely to transmit the disease too, which is a pretty significant caveat. Also there is evidence that good weather slows transmission somewhat (unless you believe last summers lull was a result of 'world beating' test and trace, plus there are the accelerations in halving time during good weather in this lockdown) and by June hopefully the weather will have finally warmed up some. Last summer we had two and half months mostly unlocked - yes the kent variant has higher transmission rates but then also last summer no one was vaccinated, so assuming we can get the case burden down to a similar level at the start of the summer it doesn't seem unreasonable that we can avoid a summer wave whilst the remaining population is vaccinated.

I would still prefer to see caution in coming out from lockdown to ameliorate the risk of breeding vaccine dodging variants and complete release on 21st June seems unwise (certainly easy things like masks and distancing seem prudent to keep in place for the rest of the year regardless of case levels), but that modelling study is pretty pessimistic - and I'm a modeller by trade!

Post edited at 14:48
OP wintertree 07 Apr 2021
In reply to MB42:

You're right - children should be factored in for the spread, but out for the hospitalisation susceptibility; that drops it from 500,000 people going to hospital to 160,000 - still massive healthcare overload.  Perhaps tightening up several other demographic and social assumptions could whittle that down to safely below the point of disaster.   But... we could run with a pessimistic assumption on vaccination of under 40s and assuming many won't have been offered it yet, giving a much lower immune fraction than the 85% I used.  The underlying point is that however you divvy up estimates in an order of magnitude approach, it comes out nothing like far enough away from "everything will be fine" for comfort.

>  Also there is evidence that good weather slows transmission somewhat (unless you believe last summers lull was a result of 'world beating' test and trace, plus there are the accelerations in halving time during good weather in this lockdown)

Yes and no.  I've moved the trailing edge of my plot 9 on this week; if you go back to last week's plot [1], the most rapid growth in cases (fasted doubling time) since before the first lockdown was at the end of August 2020, when cases were doubling every 5 days in England.  It didn't stand out at the time because absolute numbers were low.  The weather is a modulating factor, but it's minor in comparison to the presence or absence of restrictions and based on last August I don't think for a moment that it's enough of a difference alone.

>  it doesn't seem unreasonable that we can avoid a summer wave whilst the remaining population is vaccinated.

I think we can delay it and stretch it out past the summer - I hope that the vaccinated fraction moderate the doubling time sufficiently that the rise is slow, and still ongoing out of summer.  But - will we vaccinate the remaining population as it comes down to refusals rather than waiting for the process to play out?  

> I would still prefer to see caution in coming out from lockdown to ameliorate the risk of breeding vaccine dodging variants and complete release on 21st June seems unwise (certainly easy things like masks and distancing seem prudent to keep in place for the rest of the year regardless of case levels),

I agree entirely.  

> but that modelling study is pretty pessimistic - and I'm a modeller by trade!

As a general observation, pessimism has worked much better than optimism in terms of predicting the near future with this pandemic.  

It's all spitballing really until we actually know how many adults are going to refuse vaccination, and presumably how many of those will reuse a more targeted, personalised round of contacts once the "first pass" of the roll out is complete, and if/when they start being limited by any sort of vaccine passport scheme.

[1] https://www.ukhillwalking.com/forums/off_belay/friday_night_covid_plotting_19-...

 Toerag 07 Apr 2021
In reply to MB42:

> I realise you pointed out its noddy maths but the susceptible 30% isn't just refusals, it includes all 15m <18 year olds for whom the risk of hospitilisation is significantly lower and who are possibly less likely to transmit the disease too.

If our experience here with our second wave is anything to go by kids quite happily seed the community with the Kent variant.

> Also there is evidence that good weather slows transmission somewhat (unless you believe last summers lull was a result of 'world beating' test and trace, plus there are the accelerations in halving time during good weather in this lockdown) and by June hopefully the weather will have finally warmed up some. Last summer we had two and half months mostly unlocked - yes the kent variant has higher transmission rates but then also last summer no one was vaccinated, so assuming we can get the case burden down to a similar level at the start of the summer it doesn't seem unreasonable that we can avoid a summer wave whilst the remaining population is vaccinated.

You still had no large gatherings & nightclubs though didn't you? Those large gatherings (festivals, weddings, funerals) will inevitably include significant 'indoor' gatherings amongst the people going to them (overnight stays, indoor socialising, car sharing, public transport) which you didn't have last summer. People will be even more hell-bent on having a good time than they were last summer.

> I would still prefer to see caution in coming out from lockdown to ameliorate the risk of breeding vaccine dodging variants and complete release on 21st June seems unwise (certainly easy things like masks and distancing seem prudent to keep in place for the rest of the year regardless of case levels)

Agreed.

Post edited at 15:44
 Misha 08 Apr 2021
In reply to wintertree:

Thanks, all good points. I’ve suggested a revised version below, taking into account a few nuances.

We have about 15m under 18s and about 52m adults in the UK.

Assuming 65% of adults and 0% of under 18s are vaccinated by the 21 June (the rest largely being people not yet immunised and refusals - note that 60% of adults have been vaccinated to date but this won’t increase much until the second half of June and even then it will take 2-3 weeks for a decent level of immunity to develop), we will have (0.65 x 52) / (52 + 15) = 50% of our population immunised.

Let's assume 20% of people have now been infected (generous assumption? - seems fair, 4.4m official cases x 2 for actual cases since June x somewhat under 2 for actual cases in total, so plausible to get to about 14m actual cases in total) and have decent naturally acquired immunity (this may be less certain...), so 20% of the unvaccinated 50% are also immune, giving a total of 60% of people with some immunity.

The herd immunity threshold for the current predominant variant is probably ~90% (at least among adults - young children seem to be less infectious so in practice the her immunity % may be lower for the population as a whole?).  So, once restrictions are gone, we will have uncontrolled spread and 40% of people are unprotected - so 30% of the population will get it before we hit the 90% herd immunity threshold and infection dies out naturally.  

40% of our 67m population is 27m. Of this 27m, 15m are children, who have negligible chances of hospitalisation or death. The at risk group is the remaining 12m adults. With 4% of that 12m going to hospital and 1% of them dying, it looks like 480,000 people are going to hospital and 120,000 people are going to die. However these %s are averages for the whole population and we know that 90%+ of the most vulnerable demographics have been vaccinated, so these numbers will be overstated. I lack the data to do a better estimate but I suspect the hospitalisations would still be too high for the NHS to cope with.

Interesting exercise, as instinctively I thought the numbers would be much lower. The numbers would be better if you ignore the under 18s and assume 85% of adults are vaccinated (i.e. ‘freedom’ from September say). Then you only need 5% for herd immunity (ignoring infection acquired immunity). 5% of 52m adults is 2.6m. Applying 4% and 1% (same caveat as above) suggests 104k and 26k. 100k admisssions is what we had from 1 Dec to 9 Jan, which got pretty close to breaking the NHS (admittedly starting with 13k more in hospital compared to now, but still). Of course the numbers could be significantly lower due to the caveat over the 4% and 1%, so it might be manageable. Then again, who knows what new variants might turn up in September, plus we’d be going into winter (just like we are this week, it seems 😕).

Note that around 10-25% of the vaccinated cohort might still get infected. However their likelihood of hospitalisation or death seems to be close to zero (eg  asked on the recent AZ US study). This is the but where I thought the modelling assumptions were overly prudent / rather pessimistic. Of course it’s early days yet and time will tell.

This ignores the impact of mass testing (I have my doubts - most people won’t bother to test and/or self isolate) and vaccine passports (not clear what will happen and any system won’t be ready till the autumn anyway so it’s all a bit pie in the sky).

 Misha 08 Apr 2021
In reply to MB42:

See my numbers re the 15m children. I thought it would make more of a difference but it doesn’t once you adjust the vaccinated % down as wintertree was overly ‘generous’ there.

Re last summer, it wasn’t really 2.5 months mostly unlocked. It was 2.5 months of shops, 2 months of pubs, just over a month of gyms and no mass events / spectator sports / theatres. So pretty different to ‘freedom’ on 21 June. As you say, prudence is called for. No point taking one step forward to then have to take 2 steps back.

 Michael Hood 08 Apr 2021
In reply to Misha:

As the number of people with immunity increases, the R factor will go down. Simply because on average, each infectious person will come into contact with fewer susceptible people.

So even if we're not yet at herd immunity, any increase should be slower than before (lower exponential rate). A longer flatter peak reduces the chances of NHS overrun.

However, in practice I suspect that what will happen is that we'll get lots of "localised" outbreaks of cases in communities (geographical or ethnic) with less immunity and many less cases outside of those communities. Of course, any such localisation makes modelling much more difficult.

 Si dH 08 Apr 2021
In reply to Misha and wintertree:

I think you are both over pessimistic about possibility of healthcare overload because of the time dependence (as Micheal has pointed out) and because you have used an adult-population-average 4% hospitalisation rate, which is very pessimistic for an unvaccinated population of 18-50 year olds, possibly by an order of magnitude. I also think your 20% immunity figure is an under estimate of how many people have been infected in these populations, although of course we don't know how long immunity lasts.

I don't really think healthcare overload is plausible if you correct the above points as long as the vaccine efficacy in over 50s is as good as thought. I'm much more worried about high case numbers leading to new variants that escape the current vaccines, so I agree with you that removing all measures on 21 June seems pretty unwise. I also think cases will be pretty high by then so it probably won't happen anyway, we'll see.

Edit- implausible is probably overstating it given how many vaccinated older people would still go to hospital. Still pretty unlikely though.

Post edited at 09:59
OP wintertree 08 Apr 2021
In reply to Si dH:

You could be right, let's hope so!   Hack a few factors of 2 out of my noddy maths and the overload problem changes to "very busy" - which is still a problem for the return of wider healthcare depending on the level of infection control required.  As Michael Hood says, the presence of many immune should stretch the peak, but how much depends on how structured/concentrated vaccine refusal is within communities.    If it does follow the localised outbreak model Michael suggests, the key is to jump aggressively on those outbreaks with local control measures, rather than allowing the outbreaks to coalesce into ever bigger regions as they did in late summer 2020.

> and because you have used an adult-population-average 4% hospitalisation rate, which is very pessimistic for an unvaccinated population of 18-50 year olds, possibly by an order of magnitude.

I think however it's over-optimistic by a factor of several for the older unvaccinated population however.  

> I also think your 20% immunity figure is an under estimate of how many people have been infected in these populations, although of course we don't know how long immunity lasts.

That's a good point - it's quite likely there's significant overlap between vaccine refusals and some of the communities that have been harder hit by Covid infection. 

> Edit- implausible is probably overstating it given how many vaccinated older people would still go to hospital. Still pretty unlikely though.

It's notable that deaths have consistently been halving much faster than hospitalisations, I think this interpreted with the demographics of the vaccine roll out says that for older people, the vaccination helps keep many of them them alive (and out of ICU?), but not out of hospital.   If it's acute and not critical admission, the overload threshold is hopefully better, and the wider population level immunity might start to change the level of infection control needed, removing some of the inefficiencies their necessity introduces.   Still, hopefully we'll have better therapeutics out of the trials pipeline by this winter...

> I'm much more worried about high case numbers leading to new variants that escape the current vaccines, so I agree with you that removing all measures on 21 June seems pretty unwise.

Yes; until we've got immunity across the adult spectrum it does't seem cautious enough.  

Post edited at 10:14
 Offwidth 08 Apr 2021
In reply to wintertree:

Don't forget long covid. I'm hopeful that will be much reduced by the vaccine but seeing so many people (especially the fit and healthy) reduced to struggling is a big ongoing public health problem, given the numbers and younger demographic estimated by ONS, even if deaths and ICU numbers drop to very low levels. The argument in the link Bob provided in the other thread is also important: vaccine passports could end up being counter-productive, aside from international travel use. More seriously from that link, it's a fact that Trace and Isolate is still not working anything like properly in England and very much needs to be working this summer. Variants that bypass immunity to a varying extent are the other really serious issue and our border controls are pretty lax.

I can see lots of local Tier based measures for the remainder of the year that will be stricter than they need to be, compared to under better prevention measures (with the associated additional economic and social damage and new variant risks).

This government just doesn't get what is necessary to fight such a pandemic, everything has gone disastrously wrong other than vaccination and financial support. Yet again on the news this am Hancock says "we have done everything necessary at every stage of the pandemic". He seems completely oblivious to upto 100,000 needless covid deaths; and many more hundreds of thousands needlessly suffering from long covid; plus all the control measures necessary, that damaged people and the economy; plus our gift to the world of the Kent variant (based on the probability of variants occurring when 'running too hot').

 MB42 08 Apr 2021
In reply to wintertree:

> Yes and no.  I've moved the trailing edge of my plot 9 on this week; if you go back to last week's plot [1], the most rapid growth in cases (fasted doubling time) since before the first lockdown was at the end of August 2020, when cases were doubling every 5 days in England. 

Thats pretty interesting. There was also something of a summer wave in Spain though conversely quite a few other european countries had in hindsight remarkably quiet summers.

There wasn’t any change in restrictions in England from 4th July till September sometime so what is the hypothesis for the change in doubling time at the end of august, or is it just really noisy at low numbers? A look at my climbing log suggests the weather was probably nice around then I remember modelling chat last year that the transmission had a high std. deviation (most people transmit it to no one and a few people transmit to a lot) compared to some other diseases. With reasonable case numbers this just gets averaged out and makes little difference, but possibly with a low numbers it only just clings on via the occasional super spreaders; then you have a confluence spreader events just happening to occur in one week to bring the baseload up and kick off the more averaged behaviour?

> The weather is a modulating factor, but it's minor in comparison to the presence or absence of restrictions and based on last August I don't think for a moment that it's enough of a difference alone.

One difference from last summer is that reductions in R even if they don’t bring it below 1 could still prevent the problem; last summer any R>1 would lead to catastrophy eventually whereas this summer if you keep the doubling time to say 20 days from June 21st then you have a good chance you will have finished the first jabs for adults and them become effective within 3 doubling times (assuming high enough uptake)

> It's all spitballing really until we actually know how many adults are going to refuse vaccination

ONS published data on uptake which was 90.2% for over 70s (up to 11 March, published 29th March) which is pretty good, though admittedly the assumption the problem is more with lower age groups would seem reasonable

> As a general observation, pessimism has worked much better than optimism in terms of predicting the near future with this pandemic. 

Whilst I concur in general, especially on how politicians and pundits have viewed matters, from a purely modelling point of view there has been quite a range. There were some pretty pessimistic predictions last year (NiH, Bristol, one version of the Imperial modelling) which turned out to be wrong, and equally some over optimistic ones (UCL, Oxford) and some (like the figures quoted by Chris Whitty about the autumn, Imperial I think) were more or less bang on (that was branded as pessimistic at the time but that seemed to be in relation to the public expectations rather than amongst other models where it was not exceptional). Of course if the modellers are doing their jobs properly then in a such an evolving and uncertain field this spread is to be expected and is a good sign…

Also  > You still had no large gatherings & nightclubs though didn't you?

Good point, I make the mistake of thinking of freedom in terms of the things I want to do and since I would be perfectly happy to never step in a nightclub again it wasn’t on my radar. I think there are already rumblings about large events being treated differently (e.g. covid-safe trials, vaccine passports) which would seem like a good thing, and nightclubs should probably be in the same bucket. But yes if you take 'all restrictions lifted' on June 21st as literal which it may well be then that is a worry...

OP wintertree 08 Apr 2021
In reply to MB42:

> Thats pretty interesting. There was also something of a summer wave in Spain though conversely quite a few other european countries had in hindsight remarkably quiet summers.

Yes, perhaps the real question given everything we now know is "why were some summers to calm last year?".

> There wasn’t any change in restrictions in England from 4th July till September sometime so what is the hypothesis for the change in doubling time at the end of august, or is it just really noisy at low numbers?

The numbers were about 1200 a day at the point the doubling time became really short; the fluctuation on the data was much less than the rise in exponential growth (shortened doubling times).

>  With reasonable case numbers this just gets averaged out and makes little difference, but possibly with a low numbers it only just clings on via the occasional super spreaders; then you have a confluence spreader events just happening to occur in one week to bring the baseload up and kick off the more averaged behaviour?

This was my sense at the time, the number of super spreader events in the North West and also in the North East (but less publicised) making the news really went up a notch.  It's also possible that there was some additional positive feedback if test/trace/isolate couldn't keep up with this sudden rise.

> One difference from last summer is that reductions in R even if they don’t bring it below 1 could still prevent the problem; 

Yup; hopefully the moderating effects of immunised people really stretch the peak out.  Depends how concentrated vaccine refusal is within communities.  The more concentrated, the worse it starts to look IMO.

> Whilst I concur in general, especially on how politicians and pundits have viewed matters, from a purely modelling point of view there has been quite a range.

Agreed; and the wide range in modelling outputs is something that isn't well understood I think beyond science circles, and is certainly exploited in attempts to discredit the role of modelling by deliberately latching on to an extremal model.

>  Of course if the modellers are doing their jobs properly then in a such an evolving and uncertain field this spread is to be expected and is a good sign…

Indeed - it's good to see a range of models and where they all sit.  There should be a secret modellers code against ever presenting a single model to political decision makers, as perhaps that's how you get to the point some misunderstand the predictive power of a single model vs its role as one of a sheaf of models giving some feel for the bounds on what lies ahead, and on the ability to fine tune an evolving situation vs understanding the precedence of parameters when it comes to sensitivity.  

 bruxist 08 Apr 2021
In reply to MB42:

> There wasn’t any change in restrictions in England from 4th July till September sometime

There was one change: Gtr. Manchester, Lancashire, West Yorkshire were placed under a local lockdown on the 30th July (no meetings between households).  The rates were rising rapidly enough - from 20 to 30 per 100k in 7 days - but after that, they didn't really start rising again until the second week of September.

 Si dH 08 Apr 2021
In reply to bruxist:

The biggest change in that period was not a restriction, but a discount: eat out to help out in August. Every restaurant and and pub serving food round here was packed to the rafters three days per week (with some level of spacing between tables of course, but everywhere pushed the rules are as far as they could because they were desperate for business.) Lo and behold we were one of the first areas to see cases rise significantly, with a really obvious uptick in case numbers as well as the rate constant just before the end of the month. I don't know if other areas were a bit less gung-ho about it or whether we were jus a bit unlucky to be one of the worst hit but it's certainly true that scousers love a bargain.

 bruxist 08 Apr 2021
In reply to Si dH:

Yes. It was utterly weird in Yorkshire. Couldn't meet another household, but we were suddenly being urged to go to the pub for a scotch egg. I was fully expecting the local Wetherspoons to start advertising Superspreader Sundays...

 Misha 08 Apr 2021
In reply to Michael Hood:

Yeah it will be more ‘localised’ based on area, job and vaccine take up. 

 Misha 09 Apr 2021
In reply to Si dH:

You may be right. The feeling I get from the models is that a lot of their hospitalisation and death numbers are based on assumptions about vaccines being significantly less that 100% effective at preventing hospitalisations and deaths (of course they won’t be 100% effective but the assumptions seem overly pessimistic). It would be interesting to see more details on their assumptions, including hospitalisation and death rates in the unvaccinated population.

Agree about variants.

My general take on things is there’s no point taking one step forward and two steps back. I think BoJo gets this in principle but we’ll have to see about the practical application of this.

 Misha 09 Apr 2021
In reply to MB42:

13 July nail bars etc

24 July face coverings in shops etc

25 July gyms and pools

15 Aug bowling alleys etc, indoor weddings with up to 30 people 

Local restrictions brought in over time - Leicester was first on 4 July

Cases started rising in early July, which makes sense as shops opened on 15 June. Then pubs on 4 July will have accelerated it. Nail bars, gyms, weddings and so on probably had a more limited effect due to involving fewer people overall. The other big factor was overseas importation and Spain, France and Italy were all a few weeks ahead of us. 

Post edited at 00:32
 McHeath 09 Apr 2021
In reply to wintertree:

Just flicking through the thread titles and, despite having seen this one umpteen times before, I read it as "Friday night Covid petting".

Must be going on all over the place  

 Offwidth 09 Apr 2021
In reply to Si dH:

I suspect you are potentially mistaking correlation for causation with eat out to help out. I'd lay strong odds people meeting indoors against the rules in the summer, and holiday returners, was a more likely cause of most of the rise. It was anecdotally very noticable how much things gradually slipped in indoor compliance in the summer as the threat receeded. By the time the rise was really obvious it was too late.

 Si dH 09 Apr 2021
In reply to Offwidth:

I'd bet a bet a lot of money against that where I live. Perhaps you don't live so close to as many restaurants and pubs as I do. It's not plausible that in-house transmission could match the number of large transmission events that would have been happening in those places, it was just ridiculous. That's how the disease spreads most.

 Offwidth 09 Apr 2021
In reply to Si dH:

I'm not denying the risk in that known area. I think you also need to take into account the risk that you didn't witness on summer holidays and in people's homes (ie increases in breaking the rules as overall case numbers looked low and restrictions had been going a long time), especially given the context of how your locals behaved where they could be seen by the public. I'd happily give you odds but we will never know. 

 Misha 09 Apr 2021
In reply to McHeath:

Ha, were you disappointed by the rather dry content of the thread?

 Punter_Pro 10 Apr 2021
In reply to wintertree:

JB-M at FT has posted a big international data thread on the contest between vaccine effects and variants. Quite an interesting read.

https://twitter.com/jburnmurdoch/status/1380512731456016385 (Click on the first tweet to view the full thread)

Also good to see what looks like Europe starting to get on top of things again, Germany issued 700k vaccines on Thursday and Spain wasn't far behind, great to see. If they can keep that up they will probably be able to just about catch us up come June.

Post edited at 05:02
1
 Si dH 10 Apr 2021
In reply to Punter_Pro:

Interesting, thanks. The 'vaccine effect' graphs are very over simplistic - case rates in different demographics behave differently anyway and how is dependent on lockdown measures - but I agree the gaps are large enough to show an effect.

I hadn't realised just how high excess deaths were across Latin America. The info on variants from Israel is also interesting and potentially very hopeful, but I don't know how much data they are actually working with and didn't come across the actual quantified results, just a headline article that he linked to - did you?

Post edited at 08:05
OP wintertree 10 Apr 2021
In reply to Punter_Pro:

Great; thanks for the link.  Interesting to see the graphs for France to show the effects working on a growing case rate.  It’s good that they’re starting to get their vaccination rates up - good for them and good for us.  I should have done some plots like that for cases; I’ve done them fo for deaths in past weeks.  

Also a good sign for the weeks to come given that vaccination is now landing in our working aged adults, where more of our cases are; so far vaccination has really cut in to deaths but soon it should really cut in to infection numbers and R as it takes hold on people who weren’t isolating so much and who mix in workplaces and vis their school children.

OP wintertree 10 Apr 2021
In reply to Si dH:

> I was wondering, have you changed anything in your deweekending algorithm in the last couple of weeks, with the all-case data

Sorry; only just digging in to this as I prepare the plots for #21.  I've attached the plots of raw/deweekended data with polynomial trendlines, and the residuals from the trendiness, with colour by day-of-week.

If you squint you can see that two Sundays switch from being under the trendline to over the trendline on the raw data.  I've circled their residuals on the annotated plot.  It looks to me like this shift towards a Sunday basically defeats the deweekending algorithm and leaves it little wiggle room to redistribute the Monday/Tuesday case surplus (as there isn't one, by it's local measure) so the data passes through closer to the raws.  I certainly don't think it's introducing any artefacts.

If I was to do this all over again, I'd switch to using the week-on-week measurements of change for each weekday as a way of measuring the exponential rate constant; it's purer than trying to fix the whacky sampling in the data then fitting an exponential function to it, and it automatically adapts to the whacky sampling as long as the biasing distribution doesn't change dramatically over 7 days.  We live and learn.  I don't see much data with a noise term dependent on the day of week in my post-Covid future however...  

Post edited at 17:14

 Si dH 10 Apr 2021
In reply to wintertree:

Thanks for looking at it. I guess we need to be extra careful looking at the leading edge whilst the effect of Saturday/Sunday/Monday noise isn't fixed in the usual way.

You have your work cut out interpreting the last couple of days' data this week!

 Offwidth 10 Apr 2021
In reply to Si dH:

The ONS findings of flat levels of infection for England in the previous two weeks.

https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/con...

 MB42 10 Apr 2021
In reply to Misha:

My memory is even worse than I thought Though the cases started increasing in July the case doubling time from 4th July was 51 days which was much longer than in the early Autumn despite the restrictions being more relaxed (albeit stricter than my seive of a brain remembered...), and this was the case in quite a few other european countries even e.g. Denmark where restrictions were definitely looser than here. I'd assumed this was to do with a weather effect but this thread has made me think its more likely related to case load; either the transmission dynamics or possibly test and trace actually being effective! Either way it seems like it could be an important factor in whether there is a summer wave or not.


New Topic
This topic has been archived, and won't accept reply postings.
Loading Notifications...