UKC

Plots are England only as other nations not reporting all data during the Christmas period.  I suspect there’s going to be a lot more reporting lag than normal for various reasons.

• Plots 1-4 - Missing as the latest ONS update is again lacking an estimate of new daily infections.
• Plot 5 - jitter on cases - case count.  This shows raw case counts and a trend line for England (left) and my attempt at re-distributing the Monday sampling spikes into the weekend sampling lows (“de-weekending).  This is an attempt as up-to-date as possible a look at the current situation.  Which speaks for itself.  I find the annotations on day-of-week make this a useful way of understanding what's going on amidst the lulls and spikes in reporting.  The most recent ~5 data points in this don't have complete counts due to reporting lag, and are not used to fit the trend line.
• Plot 6e, 7e, 8e - Hospitalisations has joined cases in clear exponential growth beyond the November lockdown level.  Deaths is more clearly levelled off now.  There is a broad distribution of times from when an individual is infected to when they're hospitalised etc.  This broad distribution "blurs out" features from one plot to the next, meaning that the drop in admissions was much smaller than that in cases, and there was almost no drop in deaths corresponding to the drop in admissions.  The total numbers of both still drop when cases drop, but the drop is blurred out over time so is less visible.  Deaths haven't clearly started growing yet unlike my expectation - part of this is demographic as well as the lag I think - more in a later post.

• Ploot 9e - The doubling time for England seems to have settled out at around 10 days; the right week or so of this plot is provisional however, and perhaps this will go up more - see the demographic breakdown for some hotspots in a later post.  Hospitalisations are doubling more slowly which I think is down to the demographics.  The next post covers this.

Most recent previous thread: https://www.ukclimbing.com/forums/off_belay/friday_night_covid_plotting_4-7...

This post looks at the demographics of the recent rise in cases over England.   

Plot D1

• Left side: Heat maps for infections/day for 5-year age bins.  T
• Right side: Heat maps is the corresponding exponential rate; +ve values mean cases are growing, bigger values mean faster.  The second plot shows the current exponential rate for the age bins, along with the corresponding doubling times which may be easier to interpret - this is the time taken for cases to double at that rate.
• Note, this is not normalised to population levels in each age bin but is actual case numbers.  A similar plot has recently appeared on the government dashboard when you click down to regions etc, theirs is normalised per 100k in each bin, mine is not.  Mine uses de-weekending instead of a 7-day rolling average.
• Note the giant spike around 09-28 to 10-05 in the 15-20 age bin (and a bit in the bin above).  This was the start of university terms and the resultant mass outbreaks.  You could be forgiven for thinking this spike pressages the broader demographic across the population.  I don’t think anything about that can be inferred from this plot beyond the possibility, but I look forward to reading the professional epidemiological studies around this in the future.
• The right side heat map has the darkest orange for ages 20-35, showing cases are doubling in those ages the fastest.  As a result of this, case numbers are now highest in these age ranges (left sub-plot).  As case numbers in these bins continue to out-grow those in older bins, they will come to dominate the national level doubling time.  So this is "good" news - cases are doubling fastest in ages least likely to be hospitalised or to die, and I think this is why hospitalisations and deaths are doubling more slowly than cases - so the situation is not as grim as the headline cases figures imply.  
  • Prevalence can't raise indefinitely without compromising shielding of the more elderly however.

Plot D2

• The same plot for all London UTLAs that have risen above their level on November 17th, which appears to be all of them now.
• Note that the colour bar on the right hand side plot goes to a higher exponential rate that for D1 - case are rising faster in London.
• Note the absence of a university spike - a couple of other posters have noted that the major universities in London have medical schools which influences a fair few things.  
• The same age group of 20-35 has the fastest doubling times.

Plot D3

• The most recent measures of exponential rates and doubling times for London and England.  
• Errorsbars are the mean and standard deviation of measurements for the most recent 5 days of data.
• The exponential rate is ~ 60% higher in London than in England.  This isn't so far from the LSHTM estimate of increased infectivity of the new variant.  I'll do a quantified measurement of the difference when I have time.
• Shielding of the elderly is clearly visible in the England data (the doubling times getting longer > 70 years of age on the red markers) but not in London.  This is concerning.
• The doubling times for the 25-30 age bin in London is 4 days, which is bonkers - that's right up there with the rates inferred for March/April.  I wonder how much is behavioural as well as this new variant?

But the current situation is perhaps best described as a “pandemic within a pandemic”, with the new variant having taken over driving cases around its epicentre, and presumably it’s rising exponentially to prominence in most or all other areas.  So, where London leads perhaps other areas will follow

Bringing several plots together from the last few weeks.  These plots show how the failure of lockdown has propagated outwards from the Thames Estuary area.  I say this referring to the effect propagating, not the cause, although there's enough evidence out there for me to be quite sure they are linked.

Plot X1  shows a heat map of the data on which cases started rising after the onset of the November lockdown (R>1).  There is a clear relationship between distance from the Thames estuary and the date (Plot X2). A linear least squares regression of this data gives a correlation coefficient of 0.78 (r-value). This has more than 6σ significance above a null hypothesis made from randomising the distance. (P<0.005, see previous thread).  Both the r-value and the significance imply very strong support for the statement that the failure of lockdown happens later with greater distance from the thames estuary area.

Plot X3 shows the correlation coefficient measured using every point on the map as the epicentre, as well as marking the epicentre as found by using an optimiser to find the location yielding the maximum value .  This moves around a bit as each new day’s data refines the estimate of lockdown failure.  The circle of probability is not very scientific, more indicative.  There are two other measures of lockdown failure I’d like to try following discussions with Si dH on the previous thread.

My interpretation of this is that the old variant - Covid-19 - was decaying under lockdown measures whilst the new variant (Covid-20?) remained at R>1 uni the Essex/Kent/Medway area for most of lockdown and so grew exponentially.  As this variant spread outwards from the estuary area, it started growing exponentially in most (all?) other regions, with the growth of this variant starting later further from London.  So, we have a pandemic within a pandemic and strong reason to believe that November lockdown measures were ineffective against it.  By extension, Tier 4 is likely to be ineffective too without an increased behavioural contribution, encouraged by clear messaging from leadership.

My interpretation of the lower PCR cycle threshold value noted in the early NERVTAG report is that infected people could perhaps be shedding more viral load, and the LSHTM pre-print analysis by model fitting suggests that this is down to increased transmissibility of the virus across all ages rather than reduced immunity.  So, it seems reasonable to expect most regions to head towards rapid growth of cases, then hospitalisations then deaths.  This is reflected in the stark forwards projections in table 1 in the LSHTM pre-print.   So, the importance of following or exceeding the guidance on e.g. distancing and ventilation has never been more important. 

I had said I was expecting an announcement by Boxing Day that the AstraZeneca vaccine is approved and a rapid vaccination program is to being imminently with armed forces assistance.  St John’s ambulance are also on a large recruitment and training drive for volunteer inoculators.  There’s good discussion on [1] from another poster on why the time scale may be closer to New Years Day.

[1] - LSHTM preprint thread - https://www.ukclimbing.com/forums/off_belay/lshtm_preprint_on_new_covid_str...

[2] - NERVTAG's first public document on this - https://app.box.com/s/3lkcbxepqixkg4mv640dpvvg978ixjtf/file/756963730457

Edit: The order of the photos gets messed up when I post, so I need to edit to fix them...

Plot 16

• A new addition is the bottom black lines - these show the minimum number of cases / 100,000 per day to have occurred between 17th November and the most recent day.   This can be used to eyeball where growth in cases is high or low.
• It could be easy to look at this and conclude Tier 2 is behind the difference in trajectories, but it's important to remember that the new variant emerged in an area very close to the London and other Tier 2 areas.
  • It is notable that the London UTLAs on the left side of the Tier 2 region on the plot, which had tier rules been applied at regional level despite their local prevalence (at T3 levels) at the time, have had by far the most growth in cases.  It's been claimed having tougher rules in some parts of London would not have worked due to the interconnectedness, but despite that interconnectedness these regions have remained "unique" over the last month.  This seems like a two way street to me, and so perhaps Tier 3 could have changed this.
• This plot looks like a dog's dinner - sorry - but it can be poured over to pull out a lot of detail.

Plot 17

• The "red" regions were so partitioned a few weeks ago when we were wondering why they were behaving so differently - at the time they were growing whilst "blue" ones were falling .  Well, now we know more about the difference beyond T2/T3.   Most "blue" regions are now growing too.

Plot 18

• We now have characteristic times plots for all 3 measures.  I haven't looked in to how to correspond NHS regions to regions yet so some areas are missing.
• The London and East of England doubling times are nudging 5 days, which is bonkers - this implies that cases are rising almost as fast as they were before we had control measures in place during March.  However, as the demographic breakdown above shows, this is currently being driven by ages unlikely to suffer so many effects or to be hospitalised, so the situation currently is not as bleak as these numbers suggest.  
• Deaths data is blanked before 2020-09-15, as the absolute numbers were low and so the plots are dominated by noise.  

Thanks as ever for all this analysis. Re younger people driving the case numbers, part of it could be behavioural but you also have to consider jobs and accommodation. More likely to be working in customer facing roles and more likely to be living in house/flat shares, especially in London. It would be interesting to see some studies on this. They should really collect that kind of data from people going for tests (perhaps they do?).

The concern is that people in their 20s and 30s won’t be entirely isolated from older people, especially with Xmas... and a lot of these younger people will be asymptomatic. So I fear the numbers in early January will be pretty bad. T4 and school / Uni holidays should help to mitigate of course. 

You’re right; “behaviour” is far too narrow a word for all the factors not directly related to the virus.

Its also the age group most represented in the exodus from London after the press conference, going off qualitative descriptions from the news reports.

At this point, I’ll be surprised if universities reconvene for term 2. 

Have you got a view on when we should see plot 8e rise if the variant behaves the same way as the base case, based on all the plotting of these curves you have done to date?

Given that London was worst affected in April/March, I think in October it was estimated that nearly 20% of Londoners had been infected and presumably more in this age group. From your data (thanks) there is no sign of any sort of "herd immunity" in this age group (and in particular those that can't /won't modify their behaviour), slowing the increase down at all? 

I bet you're weapon at Cluedo. 

Its hard to say without data from a parallel universe or without detailed longitudinal tracing of exposure.     

The proposed increase in transmissibility is really bad news for herd immunity; raising the immune levels needed to keep levels low without control measures from perhaps about 70% to perhaps about 85%.  

With those doubling times however, the fastest growing age bin isn’t far off that level of exposure to the virus.  Perhaps 5 weeks - but I’ll be surprised if things can be allowed to go that far.

I would have said any day now (well, did say in last week’s thread) but that’s based on the data from when cases were rising from a very low point - this time round there’s the falling deaths from the tail end of the November peak masking the rising deaths from this - I assume.  It’s frustrating that the longitudinal data the health service have isn’t well summarises publicly (from what I’ve seen), as this could answer a lot of questions.

So if the Oxford vaccine is "only" 60 to 70% effective it will mean control measures will be around for a long  while. Even at 90% efficacy it means 95% (as opposed to 78% previously) will need to have a jab. If we ignore those who possibly already have immunity by being exposed to the virus.

Probably, but we’re getting ahead of ourselves with regards what’s known about the new variant I think.  There’ve been a few different estimates from the downstream data but it’s very early days.    Given the apparent seasonality I still hope for a relaxed summer.  A pint and steak and chips in a beer garden and a visit to the curry house please.

This does feel like a big - but not unconquerable - setback.  I doubt evolution is done with this virus though and the world continues to give in ever more hosts to help it out.

Another variable into a sea of unknowns.

The crucial thing with the vaccine is what % it keeps out of hospital.....i don't honestly think we will ever get rid of it but if vaccination programmes keep it down to flu levels of pressure on the NHS I'm pretty sure that we would accept as a price worth paying to get back to normality. 

I agree, but that has now got a lot harder to achieve. If we don't get to herd immunity levels it would mean isolated outbreaks will spread without an effective track and trace, and even this would require local restrictions to protect local hospital capacity. This year hasn't been like a flu season with around 20% of the UK infected.

Well here we are, Boxing Day night and no national lockdown.. I was wrong! Do we think the extension of tier4 today is enough, or is it tardy stable door slamming, with the areas which would benefit most (north) left in lower tiers?

My vote is for the latter case, it’s weak minded lack of meaningful action. Optics over solid policy as usual..

I can’t set up a poll but maybe like this post if you agree with me or dislike otherwise?

My money in on a nationwide lockdown in England being announced as soon as the big spike in cases (post Xmas reporting / testing delays) manifests itself - some time between Thur 31 Dec and Sun 2 Jan. Initially for a month as the measures are meant to be reviewed by Parliament in early Feb anyway (2 Feb IIRC) but to be extended for another month to the start of March.

Yes sorry, I’m referring to England. Obviously other more enlightened nations have acted accordingly! I broadly agree with you but my fear is that the reporting delays over the holidays will muddy the picture so much they will delay until the spike from yesterday’s frivolities appears. A police officer I know says she arrested people at 4 separate house parties last night..

I had a go at a more robust answer to this.  I fit a model to the demographic data for England (cases/day in 5-year age bins) that predicts deaths based on an exponential model of case fatality rate (CFR) vs age feeding through a delay between detection and death that is a Gaussian distribution in time.  The exponential CFR model has a basis in the literature in some observations on IFR, but it does not incorporate the changing fraction of cases detected over time in the UK.  In theory the ONS survey could inform this, but recent data is missing.  The first plot shows the parameters of the best model fit.

The second plot shows deaths since the summer for England, and the results of the model fit.   I then use this model to predict future deaths over the next 3 weeks from the end of the actuals data (Dec 18th).  I do this for 3 scenarios each extrapolating the last week of demographic cases data - a constant cases, a linear extrapolation with time and an exponential extrapolation with time.  I think we will fall somewhere between the linear and exponential cases, so perhaps we'll see 500 deaths/day by Dec 27th which will be fully reported by Jan 3rd.   I clicked "Like" on minimike's 19:13 post/survey - I don't think there's much point in doing an extrapolation with falling case numbers in the immediate future.

This was a very quick and not very robust extrapolation - I can pick a bunch of holes in what the model doesn't include, and I've not used the demographic deaths data that I think is burred in some spreadsheets the NHS publishes, which would be an obvious way to improve the model. But I think it gives a reasonable idea of what we might expect to see.  

Edit: The IFR of this new variant is something of an open question I think right now, so that's another unknown in all this.

In case you missed it, some interesting discussion here:

https://ramp-forums.epcc.ed.ac.uk/t/react-1-round-7-interim-report-fall-in-...

about this:

https://spiral.imperial.ac.uk/bitstream/10044/1/84879/2/REACT1_r7_FINAL_14....

that pins a lot of the increase in London on early Dec on school age kids, which fits with the reports that the B.1.1.7 strain is spreading easily in schools.

They'll be more figures for you to crunch soon.

https://www.bbc.co.uk/news/uk-55428953  Looks like a new mutation this will soon add to the complications.

I knew it wouldn't be over by Xmas.  Now I'm not too sure which Christmas it will be.  

Happy number crunching!

It depends. Vaccines can be used to prevent death (the severest form of illness*), or reduce illness (and reduce healthcare demand), or even to reduce transmission. It is still not clear whether the current vaccines reduce onward transmission.

* Edit, I forgot, some illnesses can be worse than death, in terms of QALYs.

Thanks - I'll read that when I've not just spent several hours tidying up the post-Christmas mess...

Hmm.  The cases data doesn't show this unlike ONS and React, but with cases being based on symptomatic testing, children suffering less symptoms and ONS and React being random sampling, that's not necessarily a conflict.   I remarked a couple of weeks ago about the "demographic split" in cases where the highest prevalence shifted from early 20s pre-November to 15-20 and 25+ which would be compatible with a shift from child-free adults to school children and their parents.  This doesn't show the direction of causality though.

The LSHTM pre-print suggests strongly against increased susceptibility in children as being the root cause of the failure of lockdown, leaning towards increased transmissibility across the board.  My noddy analysis on the geographic spread of lockdown failure also isn't very compatible with schools as the prime mover - although they could be an amplifier.

We'll have a better idea from the behaviour of cases data, but I suspect that's going to be garbled until after New Years by the various reporting and sampling lags - so there's not going to be much time to make a decision on schools in January.

Thanks.

Sorry if I'm being dumb, it's late..in your curves, why do you cut over from data to model on about 17/12? Isn't there data on the dashboard to 22/12?

This is ultimately what I'm interested in looking for. If there is any sign it is different that could change the paradigm. But I don't usually follow deaths data very closely as the cases data is so much more 'live' so I'm not sure on the time constants usually involved.

As you mentioned earlier I think there is still some significant compensation going on for SE and London deaths, from reducing death figures in the North.

Because it's late and it's a bookkeeping muddle; the UTLA level data with demographics is good (in terms of lag and resolvable weekends) up to the 19th or so, and the API deaths data to the 21st.  Truncating to the common last date made things easier to code up.   

Yup.  The fortuitous deletion of one of the 3 primers used by the RT-PCR means that there should be enough information for the NHS to do a direct/longitudinal analysis on this, which is far better than trying to back it out of the top level data.  They don't seem to publish their longitudinal analyses much though?

Updated plot with the more recent days.  I also changed the constant case level to an average of the last 3 days - as it's rising so fast, an average of the last 7 days represented a big decrease.  The feint curves now show cases and their extrapolations.  I should probably swap the Y-axes labels around as the various extrapolations for one metric land on the axis of the other, which is confusing.

So it looks like the next 7 days are where we'll start to see what's going to happen.

No, it is stopping our hospitals being overwhelmed from admissions....

Same thing, effectively. You prevent some deaths directly and others (including cancer, cardiac, non-covid emergency... ...) by preventing healthcare collapse.

It is just React, so you've probably seen the paper but possibly not the discussion. 

It is tempting to infer that this variant has switched on the 'transmit at school' feature that has been (relatively) suppressed so far, but hard to tell if it could be explained by just more infectious across the board. Interesting to see if that manifests.  It would probably be hard to see in your plots at this stage given the short time with schools open and it ramping up, and the overwhelming numbers that would make it hard to pick out on the same scale. 

I want to think it's the latter. I'd bet 50p at this stage, possibly increasing to a fiver by next week, that schools will be moved up the vaccine priority list.

Small request, any chance of either squaring (or sqrting, whichever is easier), an axis on plot X2? Does virus spread work like diffusion?

I thought that to start with but it’s the wrong way round for diffusion - there the square displacement is proportional to time but here it’s more like the distance is the square root of time.  So it spreads faster than diffusion.  Which isn’t really a surprise as diffusion is a property of something that is conserved where-as infection numbers are anything but.

Not relevant right now but throwing this in here for the record because we don't have a suitable repository for interesting papers here. To be revisited whenever the "I'm not an anti-vaxxer but [something an anti-vaxxer would say]" folks pop up.

https://berthub.eu/articles/posts/reverse-engineering-source-code-of-the-bi...

Your threads always remind me of my favourite cartoon of the year!

That was a really great read, thanks.  

Whilst 2020 is getting panned as a disaster year, it's not a tenth of the disaster it would be if this coronavirus had emerged instead of classic SARS in 2002; both in terms of available technology and shoulders-of-giants stuff going on with technology in general and specific research started following on from SARS.  The article drives that home to me.

In reply to Robert Duran:

Too true; it also goes a long way to explaining why so many younger adults are having their productive capacity sucked dry by the credit "industry".  

Also, reassuringly, it indicates how rapidly the mRNA vaccines will (could?) be tweaked when (if?) the first vaccine escape mutation shows up. 

I still haven’t found a satisfactory answer to the required clinical trials and approval process for a ‘tweaked’ vaccine (of any type). Anyone know any more?

It seems like everyone in the industry and the media has quietly agreed that it's best not to ask that question until it answers itself

And where else they may be used. I’m 40 years on from studying this stuff, but I can see some possibilities for other vaccines using this route. 

That’s what I was worried about.. from both points of view.

I think it's a positive thing. No need to tie our shoelaces together over it now. If those decisions need to be taken again, like the ones made for the original approvals, they probably should be taken against the backdrop of current affairs. Otherwise whichever way you called it today, you'd end up in a situation where making the right choice elicits a load of moaning about a U-turn and hands a legitimate argument to the dickheads. 

Yeah, we've globally and collectively overlooked a phenomenal breakthrough for humanity here. That's forgivable. But yeah, first approved RNA vaccine will hopefully open the door to many more. Could turn out to be the best news of the decade/century.

Maybe from a public view you’re right, but I hope there’s been some private discussion with manufacturers of the regulators expectations, so trials etc can be expeditious if they are needed.

One would expect so. But I would understand why we wouldn't want that to play out in the media. 

People don't like it when scientists change their minds. And it's really hard to convey that that's the whole point of doing science.

I think both the mRNA technology and the delivery platform - medically functionalised vesicles - are going to change a lot of things.   The moral philosophers are more concerning themselves with CRISPR and humans at the moment, but targetable mRNA feels less like a sledgehammer to me and a damned sight safer for where we are now.  It can’t be long before some sports are using mRNA produced performance enhancing peptides as a way of evading some drug tests.  I don’t know how long it’ll be before synthetic vesicles can be injected in to the blood functionalised for uptake in to specific tissue types...  

Well, yeah, like any technology there will be plenty of disagreeable applications. And like any technology it'll be interesting to see what they are. 

Here you go:

https://p.dw.com/p/3n79W

Edit: fuller story here:

https://p.dw.com/p/3n6gv

I think it's a great technology but I think targeting will take a little while yet. The mRNA vaccines are an intramuscular injection which is fine for raising an immune response, but I believe if you inject into the blood the vesicles end up being cleared by the liver. Hence the existing applications tend to be for liver diseases. Before the pandemic the excitement around the technology was for cancer immunotherapy and I'm sure that will get another big push.

Good point on the kidney - that will limit access to other organs via the blood.  I’m particularly interested in vesicles as a delivery mechanism to the gut microbiome.

Hmm, I had a project about 15 yrs ago with a food multinational to try and use polymer vesicles to deliver nutriceuticals. Didn't work for a whole bunch of reasons (possibly ahead of its time).

I’d say 2021 is perhaps going to see a lot of nutraceutical activity...  Remarkably little know about the microbiome still, and much of it not amenable to culture in the lab.  

Interesting. The main difficulty we had was getting enough of the actives loaded in the vesicles for it to be economic in relatively low value products. RNA is actually easier to get in as it is a charged macromolecule. The trick the vaccine people use is to have a weakly cationic lipid whose charge state you can change with a pH switch so that it binds to the RNA. But there's a huge amount to learn and develop in the technology.

Not the context you were referring to, but the media have also been really quite quiet on the details of what this new strain means for the short term. 

The BBC are now running with the situation in London for 999 calls being as bad as the first time around - or worse.  I'm surprised they've not put the rest of the country in to Tier 4 to preserve hospital capacity for Londoners in two weeks time, let alone to slow the rise in cases in most of those other areas.

I think we'll see a ratcheting up in the media over the next few days.  Nothing that wasn't almost inevitable 2.5 weeks ago, but it's starting to feel like everyone is waiting till the time is ripe for a fear whipping campaign to set the ground for control measures, rather than calmly laying it all out a few weeks earlier.

Today's deaths data shows signs of an up-tick beyond the last few months and is I think the start of the rising exponential from the new strain breaking through in the national level figures.  I'll be surprised if we're not at > 1,000 deaths/day in a couple of weeks - with significantly more hospital occupancy than the first time round. 

https://www.bbc.co.uk/news/uk-england-london-55461390

Remember March? It was just as predictable then. Understanding exp() makes you Nostradamus in 2020.

I bet the microbiome metabolises all sorts of stuff as an interface between food and the gut.  Find the right precursor that's amenable to packaging...

 > But there's a huge amount to learn and develop in the technology.

I can imagine. Nature has to stuff a whole bunch of complex stuff in to the membranes to get them to hold together and to hold things when in a living system, it's really quite impressive how far create chemists and physicists are getting without proteins.  The vibe is very exciting in terms of being at the start of the end of the distinction between nature and man made - decades of work ahead but it's happening.

I got that this was beyond a lot of people's intuitive experience back in March.  

But for crying out loud, it's recent bloody history.  You don't have to understand the maths, you just have to look and see what happened last time.

Then again, as they say - the past is like a foreign country, and we sure as shit didn't learn from those back in March either.

On the positive side, it looks like some of the hardest hit areas in Wales have started seeing reductions in infection rates now (I don't *think* this is an artefact of Christmas reporting but it's possible.)  Medway, Kent and Redbridge UTLAs are also seeing a gradual flattening of the rate of increase (albeit with rates at a very high level and a lot of noise). This is positive because we can't be seeing the effects of Tier 4 yet.

In March we knew it was an exponential but we had no idea whether what number we were already at when we became truly conscious of the problem (is where on the curve), what exponent to expect in our environment, nor whether the problem was isolated to specific parts of the country. So even with an understanding of exponential behaviour the appropriate action was not obvious (although certain action/inaction was clearly wrong.)

Now, we have lots of good data and no excuse for getting it wrong.

I've been plotting Essex, Medway and Kent together to look for a tailing off - it seems rapid exponential growth is rarely sustained for long under any measure, possibly because it burns fast through the sub-populations at greatest risk of catching and transmitting the virus, and slower through everyone else.  

I swapped my plot to your regions - see below.  It does look like the exponential rate is backing off, as do the individual UTLA plots on the dashboard.  It's an olive branch, and I'll believe it if it holds up for another 7 days or so - I think there will be exceptional sample collection lag (4-day weekend, bank holidays, Christmas post) and maybe larger than normal reporting lag.    (This data comes from the bulk demographic download.  It stops 5 days in the past, and I chop the last 4 days off again as that can be provisional.  The rate backs of a lot more with 2 days chopped off instead).

This is something I find perplexing. I find the great majority of people understand accelerating growth one way or another. But there has seemed to be great reluctance to acknowledge it in this case. Might it be just simple fear and denial? 

That would go a long way to explaining how reassuring - but totally wrong - dismissals of the problem got so much traction at the time.

I was left doing an impression of a goldfish in late March when a colleague confidently told me "it's no worse than the flu" despite the press reports coming out of Northern Italy having no resemblance what-so-ever to any flu season in living memory.  Given that I worked in a place that's all about data, theories and evidence based thinking, this was rather more disappointing than seeing the same sort content on UKC being met with a lot of likes.

I only found this a couple of months ago.  It's prescient.   youtube.com/watch?v=bghbxemp4kQ&

I think the new cases in Wales do seem to have peaked. It seems to be to too close to the start of lockdown for that to be the full cause. I wonder whether the voluntary closure of many Welsh secondary schools a week early was responsible.

Anecdotal evidence, it's starting to feel like scary March/April again with the number of deaths I'm personally hearing about - mainly elderly (parents of friends or acquaintances) with co-morbidities (so whether Covid is the primary cause, a contributor, or just happens to be there is difficult to know) but it ties in with the nasty increase in case & hospital numbers.

I've wondered about the deaths not really taking off yet (hoping that it not yet mirroring the cases/hospital curves might indicate the new strain being less deadly), but I reckon the lag period might be over and once the distortions of the holiday period are over we're going to see some serious shit happening. Hold onto your seats ☹️

On another level, I'm starting to get a bit of survivors guilt complex (by proxy) because both my parents in their 90s are still here!!! Ridiculous, but this Covid is screwing with our lives in so many different ways.

We'll see another set of interventions. SAGE will be a shouting match already. 

Found a really good interview with Neil Ferguson a while ago (I'll try to find it. It's worth a listen). My takeaway was that his biggest regret was not being firm enough when telling politicians they needed to act. You'd think the argument would be an easy one today, but maybe the economic argument has become a harder one too beat. I don't know. Seems painfully obvious what happens next. 

It was this: The Life Scientific: Neil Ferguson on modelling Covid-19 http://open.live.bbc.co.uk/mediaselector/6/redir/version/2.0/mediaset/audio...

I'm a bit late catching this, but as someone who did a biochem degree a very long time ago, can I just say thanks very much for posting this. In amongst all this crap, this is one of the most fascinating and uplifting things I have read in a very long time. 

Cheers

Paul

I think you are probably right about the national death rate - it will rise fast soon unless there is a change in the IFR with this new variant.

Re,: the North West though (I see you are in Manchester) it is worth noting that case rates are still only at half of their October peak and rising substantially more slowly than they did through September, hospitalisations and occupancy are both also lower than November, although in some places not so much lower as case numbers because of the hangover from wave 2. Daily covid deaths in the North West have been falling consistently since mid November and continue to do so. In the national figures, this is still compensating for the rise in the South East.

I'm not advocating complacency - we need to minimise the spread of the more transmissible variant and even without that things will eventually get worse, fastest in Tier 2 areas - but if your personal experience of the pandemic is currently feeling more 'immediate', it is probably just down to chance. This happened to me at the end of August / early September when several things happened at once and it felt like the world was starting to fall in, even though rates were objectively still quite low.

The media ramping up continues with a new headline story on the BBC this morning

https://www.bbc.co.uk/news/uk-55462701

London seems to be hitting the same healthcare pressure points as in March/April; but then it hit them as cases were tailing off from all the actions employers and individuals took pre-lockdown and from the lockdown.  Now they’re rising exponentially and there’s a couple of weeks of hospitalisations already locked in.

Meanwhile the papers are almost all suggesting approval of the AZ vaccine and a mass vaccination campaign within a week.  Really difficult one this because the cabinet have to put across a nuanced message - please accept and follow this last minute panic driven lockdown even though the vaccine is going to totally save us.  I’ve not going any recent reports to update the one from early December about the production failures of the Oxford/AZ vaccine at the UK plant.

https://www.bbc.co.uk/news/blogs-the-papers-55462683

In reply to Michael Hood:

I’m not driving much at them moment but when I do my qualitative score of “have I seen an ambulance today?” has been replaced with “how many ambulances have I seen today?”.  I don’t know if it’s real or if it’s heightened awareness.  Re: deaths not taking off yet, the lag doesn’t just delay things but blurs them out as it’s a broad distribution in time - the recent dip in cases is so reduced in promenance in hospitalisations and almost gone from deaths.  That masks the early rise (its blurred out to fill the right side of the dip) but assuming there’s not much reporting lag (~450 reported) today’s update will show the start of an unambiguous rise.  

Just seen the front pages. The Express is my favourite one. I'd love it if they showed their working. All over by Feb? Sure. Of course it will be, The Express. Of course. When the vaccine takes a couple of days to weeks to take effect against an illness that takes about 3-4 weeks to run its course.... so.... the people catching it today will be the ones in the stats in Feb.... so..... you're going to, what? Travel back in time and vaccinate them? Brilliant.
What if any education do you need to become a journalist these days? Actually, don't answer that.

Several papers all saying the same sort of thing so someone has fed them February as a date.  

Which, as you note, seems a tad optimistic.

You're probably right about chance making it feel more personally immediate but my wife's extensive social tentacles 😁 - which is where the majority of my personal "news" comes through - extend a lot further than Manchester; a fair bit is from London.

She's pretty good at picking up how people are feeling, etc (great empathetic skills - we complement eachother because that kind of stuff goes straight over my head - woosh) so it's interesting to see how much her qualitative "feel" of the situation matches the quantitative data/analysis assessment (stuff which I'm okayish with but goes woosh over her head 😁).

I notice that some papers are "celebrating" 1m/week without thinking through the very, very basic maths that by end of June only approx 35-40% will be vaccinated by then.

As others have "said", standards of journalism - things that make you go hmmm.

I'm not clear whether it is meant to mean 1 million doses per week, which, with 2 doses, means only 1/2 million actually vaccinated per week, so only about 20% by June, or only about 6% by end of February which makes the headlines look like made up nonsense (no surprise of course).

Of course the idea of a return to normality by February is ridiculous, but nonetheless because of the steep age profile of risk the vaccinating the first few million will make a very significant impact on mortality and pressure on hospitals.  For example I've seen figures claiming that vaccinating the 2.8 m over 80s in England would reduce deaths by 60%.

Yes, I think a  huge amount will depend on the extent to which, once the chances of the NHS being overwhelmed is low, the extent to which government and individuals are prepared to accept the risk to lower risk groups. 

Mortality - for sure.

Peak healthcare pressure - not so sure, as this seems to be what government are using as their driver for policy, so they'll likely just run infections hotter until the same pressure point hits.  It will shorten that period of peak healthcare pressure but I don't think it will lower it.   Edit:  Although vaccination in age ranked decreasing order will build up insurance against the government getting it wrong and letting cases run beyond the point that will overwhelm healthcare.  Once it's significantly under way.

Yes, and as we've seen, the government's clear tendency has been to err on the side of overoptimism, driven by its view of what's good for the economy.  I do think that, for all the government has mishandled the situation in every other way, they did do a good job of procuring vaccines.  We've also been lucky that the vaccines so far seem to work, and since they've been pre-bought there's no reason not to use them, even for lower risk groups.  So things may look a lot better by the summer, but there'll certainly be bumps on the way. 

I think deaths is ultimately the barometer by which the government is measured though. If deaths really did drop substantially but hospital pressure was still high, I am not sure it is obvious what action would be taken. Maybe this is where the Nightingales could come in to their own? It would be interesting to see if the Government relaxed restrictions while also filling up their emergency hospitals (without enough staff) - risky but conceivable I think.

Edit to add, that LSHTM paper did predict high death rates right into late spring/early summer unless the vaccination rate increases substantially.

You would hope that they've learnt their lesson about running "hot", but maybe not..

The LSHTM paper predicted we could move to tier 2 at the back end of Feb with 2m jabs a week starting next and out of tier 2 in Mid-March.....it didn't say 'what into' though.

I would have liked to have seen a model with 1m jabs a week as this seem more realistic....

I don't see how we can't run it hot without shutting schools for months (SAGE reckoned them being open adds at least 0.2 to the R)......and that is not going to happen unless it looks like we can't avoid a disaster, kids education is too important and they have already had 9 months of disruption.....I think they might have to shut in January though.

Schools will definitely be shut, or in some way not fully open for the first 3 weeks of Jan.

I’m with you on this. I pay more attention to London/SE news than normal coz my daughter lives in Kent. London ambulances: Boxing Day was busiest day ever (the other one was last March).  At its worst, Wigan had second highest death rate in the UK, my work is a few hundred metres from the hospital and I work closely with Public Health and NHS. It felt utter shite last few months until deaths tailed off. National death rates will start to sky rocket - infection rates in SE nearly double those of Grter Manc and covering a population four times bigger. 

Perhaps not anything like as fast as you'd think from just looking at the cases plots though.   The demographic plots up at the top of the thread show that the growth in cases is being increasingly driven by the 20-35 age range, where the fatality rate is very low.   As long as that demographic separation holds, things are not as bad as they look from the headline numbers on cases.  

Some London hospitals reporting ‘wall to wall Covid’ so I expect deaths in SE to be as a minimum at least as bad as wave 1. Fingers crossed though. 

Yes, it’s not good.  I think it’s worse than last time for London as they’re not seeing the benefit of lockdown kicking in this time around.   But, hospitals fill up at a lower death rate than March/April due to improved treatments improving survival.  This effect can fill hospitals more than the change in death rate suggests by sometimes extending how long someone is in a hospital bed rather than the morgue.  So busy ambulances and hospitals aren’t the same indicator of deaths as before; although once hospital space runs out, all those decreases to the fatality rate start to go away.

Up here in the north east I hope we’ll have the benefit of nationwide control measures as London looses control.  Several regions are far enough behind London that they could get spared the worst of it for the same reason.  

If you read toilet paper...

Latest Independent SAGE recommendations.

https://www.theguardian.com/commentisfree/2020/dec/28/covid-vaccine-uk-rest...

Well it's not just us sat here wondering why things are continuing as they are then...

All seems jolly sensible.  

I see they've included "backwards tracing" - there's been a compelling case made for that for some time.   I wonder how the changed mechanics of the new variant change that...?

One thing missing from their list - as I understand it, one of the deletions in the new variant is one of the 3 primers used by the Lighthouse labs PCR tests, and the variant also has a lower threshold count so - so an RT-qPCR test outputs two pieces of information that can score an individual result for "new variant".  This score could and should - I think - be used to prioritise contact tracing to the new variant.  It's too late for that in London but doing so in more distant regions could buy a bit more time before it becomes established.  

Such eminently sensible points that they stand a precisely zero chance of being implemented, judging on past experience...

"This includes smaller class sizes (achieved through hiring extra teachers and teaching rooms)"

not sure where the extra teachers are going to appear from?

It does. However, all they've said is "good things are good" with absolutely no mention of any of the compromises, challenges or balances, or appreciation of why some of those things might be hard.

Yes, I noticed that as well. The extra classrooms conjured from nowhere don't seem very plausible either! Not very impressed with the common sense or pragmatism behind that piece of advice.

You’re right about them not mentioning the costs or challenges; then again they’ve also not gone in to the costs and challenges incurred by every day of inaction.  It’s hard to get in to one without the other and this keeps their suggestions neutral.  Some do need more to be credible (eg as raised by RobAJones) but others seem pretty self explanatory.

So, if universities don’t bring students in to residence in a few weeks, who is footing the bill for the unused private accommodation blocks - both the “independents” ones and the ones with nomination agreements?  Enquiring minds want to know...

From what I can tell the scoring and estimate of variant is happening for central data collection, but I’ve not seen any indication that the scoring is used to prioritise contact tracing or communicated to the testees.  I could be wrong.

With a bit of advanced notice (and the extra staff) that is less of a problem. When a neighbouring secondary was flooded they used our building, with an extended day, we taught in the morning they did the afternoon/evening. 

Linking to the above comments if when the new variant appeared in Eden it had been "back tracked" and isolated before it spread to Carlisle (definite) and Allerdale (probable based on latest numbers) schools in Copeland and Barrow might have been able to open in Jan. On another thread a poster used the founding of the NHS as an example to demonstrate that government's look further ahead than the next election. I'm not sure the current one are looking much further ahead than tomorrow with regard to their covid response. 

Looks like today's the day - 41k new cases, highest ever. OK, so there's no doubt some xmas delayed reporting / presentation in there, but it's not dropping. Live case increase is between 16-18k per day for the past week and seems to have plateaued from the previous week's meteoric rises. However, that's a lot of new cases to deal with and the resulting admissions are only just going to start appearing now.

Jersey had a similar outbreak resulting in a lockdown of sorts recently, with case levels dropping now the word on the street is that a significant number of people were avoiding being tested in the past few days to prevent their xmas plans being ruined.  I would expect a similar mindset in the UK.

Cardiff hospital ran out of A&E beds for 2 days last week and had to call for help according to BBC/Guardian.

20426 in English hospitals....the peak in April was 18976.

4957 in London alone, an increase of 47% in the last week.

Over 12000 new cases reported in London today with infection rate in those aged 60+ soaring... 

I predict the shit will hit the fan in London within the next week...

Was just trying to edit my post to that effect!

https://www.theguardian.com/society/2020/dec/28/nhs-hospitals-facing-unprec...

Today's admissions will be due to cases presented at the start of the plateau I described. There's going to be lots of ambulances on motorways and cancelled operations....

Indeed; I was talking more about the rise in deaths starting up.  You are right that it’s coming to crisis point in London.  
 
Interesting behavioural take on people delaying their tests, thanks.

my question is a few weeks back why did the deaths not drop but the cases did? is the new variant more deadly? Rising viral load doesn't sound good does it.

It’s an effect of deaths occurring at a range of times after infection. It’s like a smoothing out of the death times, so both dips and spikes are less pronounced.

I could see that going either way. Some people might have put off tests to avoid disruption to their plans. Others might have been extra eager to get tested to assuage their fears about meeting family. Hard to call a balance between the two.

One effect at play is shifting demographic distributions over time.  My plots D1-D3 at the top of the thread show the rise in cases is fastest (exponentially speaking) in ages with little death risk.  Most older infections will be detected as cases on admission to hospitals where as many younger infections will never be picked up, so the demographic separation in infections is probably stronger than in cases.  The public case data is not well broken down by the reason for the sample being taken so I can’t look for insight there.

Even so, the demographics aren’t enough and I’ve been digging in to that.  I found a bug this evening my the ultra-noddy model I did of age dependant CFR and Gaussian lag from demographic cases to deaths (up thread; the bug added a spike to the lag function at 0 days and pulled future deaths forwards). After I fixed the bug, there appears to be no solution with a time invariant, age dependant CFR and lagged deaths that gives such a long plateau in deaths as we’ve seen, and the models tend towards a maximally broad distribution of detection-death lags in an attempt to smooth out this valley in the cases that convert to “locked in” deaths.  Solutions all produce a gradual drop then rise in deaths instead of the plateau - shallower than hospitalisations valley, and lagging it, but not eliminated.  The only way to recreate the data we see now would I think be to increase CFR over time coincident with the failure of lockdown.  If you filter the hell out of deaths data you do see a small valley, but it’s well beyond the expected deviation from the deaths data of noise was Poissonian, and it’s shallower than any valley from a lag model (with lag confined to 0-28 days by the “within 28 days” reporting metric).

Two main interpretations:

1. CFR is increasing because the fraction of infections detected as cases is decreasing.  Not unexpected that testings scales sub-linear with infection, but test numbers are rising and there’s headroom in capacity.
2. CFR is increasing because IFR is increasing. 

Frustratingly we haven’t had an ONS random sampling update on daily infections for 3.5 weeks.  I flat out don’t understand the various explanations in their weekly reports.

No.  The lower “ct” value, the increased transmission during lockdown, the LSHTM pre-print suggesting increased demographic independent transmission - this all hints to more viral load at infection through transmission and perhaps reception coefficients.  With viral load having cropped up throughout the pandemic this doesn’t look good.

I’m hesitant to share much more of my modelling because it’s just not up to professional standards - either in terms of time spent on it, independent review or access to data - but the number of different things contributing to my pucker factor right now is exceptional, and so I’d be happier by far if leadership was erring well and truely on side of caution until these things are determined.

With the primer knockout and ct difference the state should be able to produce longitudinal demographic IFRs for the old and new variants rather than trying to back it out of summary level data. As far as I can tell (I may be wrong) this hasn’t previously been published.  It’s absolutely critical data and I hope someone is doing this - it involves probably retrospective analysis of lighthouse lab QC data - for Essex, Medway and Kent.

Doesn't sound right to me cases dropped for a month sharply but deaths stayed level, over a week or two that would make more sense to me.

I read some sensible news reports that said one of the features of the new variant is more viral load (ie Im not inferring that but perhaps could, but also not claiming its correct). Its possible its just replicating more in the upper respiratory track ofc which could be maybe better maybe

As I noted somewhere previously, the opposite was apparent in Liverpool and Sefton (and presumably other places with mass testing available) - 23/24 December saw several times the normal number of people turning up for a test before meeting family. Data on LFTs conducted per day is on the dashboard.

Yes; I recall you saying that and I take Luke90’s point, I just hadn’t considered that it’s an effect than can work either way.  On that note, it’s frustrating that LFT induced pillar 2 tests aren’t separately reported.  Cases data is getting ever more muddled.

I’m probably the worst person here to ask when it comes to considering behavioural aspects.  That’s why I tend to avoid people...

It does suggest if it's an orchestrated 'leak' (as opposed to a wined up dimwit minister off the record trying to impress a journalist half his age) that AZ's approval is imminant and there is a stockpile. Obviously reality and the inevitable bungled delivery are being overlooked but if that's the case we have a fighting chance of being rid of this without hundreds of thousands dead and before our credit line dries up. Bloody hell, I think I've caught a mild dose of optimism. 

Jk

Thanks for the reply, I bet previous comparisons with similar past positive test percentages can't explain that difference

35-40% dosed, done right, should be enough to let it rip without wiping out health care or other essential services. Not the ideal end to this phase of the pandemic but one it'll be very hard to avoid.

Jk

It’s a very timely question.

Forgive me for asking - is your background immunology?  I’m muddling a couple of posters in my head and contrary to popular opinion I don’t actually keep a black book of other posters.  Feel free not to answer!

Maybe I should take them time to write up my model fitting as a pre-print posing the question “where is the valley in deaths, and why?”.  I don’t have a clue how to do this in a way that adds constructively to the current situation, and the conclusions would largely be a commentary on why I think the model fitting failed and how to interpret that.  Not something that would ever actually get published but drawing attention to inconsistencies on the data.  Then again it feels slightly futile as some smart folks somewhere must be crunching mass, individual, longitudinal data.  I think a better approach is to look at CFR in the Essex/Medway/Kent area as they’re leaders in the new strain; I’ve started putting this together but it’s taken me down a rabbit hole of catchment areas for cases vs for individual NHS trusts.  Until two months ago I’d not appreciated how many incompatible geographic decisions there are in England used for different reporting.  It seems like a data nerd complaint but I think actually it cuts to the ability to make timely and accurate analyses in a crisis.  It should be fixed after the dust settles, the same as different reporting entities using different demographic bins, and the release of data with rolling averages pre-applied.

Nope, I just have a feel for things like this. My background is AI / Maths. You are far more statistically / applied science able than me. You know the covid variant is already thought to be 40% of scotland cases so 60% of somewhere in the SE isn't that much more is it.

It's already spraying about...         

This evening my daughter shared a post on her Facebook; daughter's plea first, then the post she shared - it's not sounding pretty.

It's really hard to know whether writing this message and sharing this will make any difference. But I have to try! I am hearing similar from more and more doctors and friends. Also working in the NHS myself I am well aware of how hospitals are struggling. Please everybody have some social responsibility. Make the correct choices even though it's hard. Covid hasn't gone away yet and it's not going to unless we all look after each other.

_________________________________________________________________________________________

From a Frontline Junior Doctor:

PAY ATTENTION.

I'm writing this from the middle of one of the most ridiculous night shifts. I'm really trying to not be hyperbolic when I say this, but the health service, specifically north London hospitals, is past breaking point. I cannot fully explain how much trouble we are in.

All of our intensive care units are full. Barnet, Royal Free, UCLH, North Middlesex. All of our hospital beds are full. We have patients on the verge of death in A&E department because there's simply nowhere to put them if they need a ventilator. I'm not talking about the elderly and frail. I mean people in their 40s, 50s, 60s. We do not have the space in our hospitals nor the staff to look after any more patients. Not only covid patients, I mean anything.

Please don't mix with other households. Not even your families. Not your friends. Not your mum or your dad. Not even if "I only see them" or "they don't see anyone else". You may think people are being careful but the reality is there's no such thing. Any human contact will spread this virus, especially the new mutant strain.

People are complaining that cancer services and routine operations are being put on hold because of lockdown and covid. This is so patently ridiculous a complaint but I need to address it. Hospitals are so overwhelmed dealing with emergency care right now that it is impossible to deal with anything else. It's not only covid emergency care, it's everything. This is the direct result and the logical conclusion of allowing the virus to rampantly spread through out communities as we have done.

I'm sorry if this sounds preachy, but it needs to be said. We need to take more responsibility for our own health and that of the rest of our local communities. We need to do the right thing. That's all this comes down to - doing the right thing. Not seeing other people. Not going into other people's houses. Don't invite people over. Not even to the garden. If you must see people, do it two metres away with masks. Forget whatever rules the government has made for us - it's clear they have largely not been based on what needs to be done to drive down infections. Just act as if everyone you know has just tested positive and you need to avoid them like - quite literally - the plague.

_________________________________________________________________________________________

Sobering stuff.

Deaths have dropped and still are doing where cases dropped the most. The answers above miss the regionality, which is crucial to answering your question because different regions have been driving the national level death data at different timee. The death rate hasn't actually been flat and plateaued anywhere after the dates on which a given place started seeing rising infections post summer. Go to the link below and cycle through the 9 English regions to see the graphs of deaths vs time (based on date of death.) You will see that in the SE, East and London rates have increased whereas in other areas of the country, particularly where infection rates were much higher for a period of time pre lockdown, the death rate has dropped a lot and is still heading down to the valley before it no-doubt goes back up. The trade off between regions gives the illusion of a plateau at national level, but it isn't one.

As for why death rates never got to a lockdown-2 peak and then dropped in those first three regions, it's to do with the very variable time to death as mentioned above, together with the fact that in these regions the infection increase pre-lockdown 2 was both small and late, ie there weren't many people who had already been infected for a few weeks come 05/11 and hence at risk of death, then of course only 2 weeks later (on average across the region) cases started rising due to the variant. So the variation in time to death is sufficient to obscure any inflection there might have been in a graph of infection dates for the people who eventually died.

Sorry, edited to add the relevant dashboard link as promised.

https://coronavirus.data.gov.uk/details/deaths?areaType=region&areaName...

Thanks.  My maths is pretty poor - I only look at things and see the answer in maths up to exponentials and convolutions, beyond that I’m out of my depth in terms of intuitive understanding.

3-4 weeks ago was a very different story though, and there are several small reporting blocks where it seems it was dominant and exponential, suggesting CFR data from deaths locked in then is a glimpse in to the future.

Edit: another interpretation compatible with the data, I think, is that the new variant has a shorter lag from detection to death but isn’t more lethal.  This doesn’t seem very biologically plausible though.

Do you have any stats to show regionality has a big effect on IFR? especially in the inner city areas where most the deaths are occuring.

Yeah that's true it will take a while for the new variant to gain a percentage advantage especially if there is cross immunity.

I think there is information in that illusion though.  The lag from cases to deaths can’t be more than 28 days under the reporting rules, yet a kernel longer than 28 days is needed to blur the valley in national cases out in to an apparent plateau (masked by noise) in national deaths - even after applying a demographic CFR model.

The problem is that both regional and demographic effects muddle things and there’s not an easy to access data source for regional, demographic cases and deaths.

That could be derived from the dashboard data at regional or local authority level fairly easily, but they don't provide such data obviously in graphical form on the website, so it would need someone like WT who is more skilled in data manipulation to actually produce the graph from the data download.

I'm not sure how much you would learn. You could compare an LA like Knowsley or parts of Teeside (is high deprivation) with somewhere like Kensington or other central London LAs and compare the CFR, but I suspect the effect of living conditions would be obscured by the effect of average resident ages so you'd have to know a lot of other information in order to back out what you wanted.

One day I hope the government will have to publish their science so those that want to can analyze it. Post factual society can't be the eventual outcome. 

I don't think lag can explain this. Regionality is a theory but there is nothing more to back it up than anything else.

The regionality effect I described in my original post above is explicit in the data, it isn't a theory. As to any regional variation in IFR I'm not sure (that's not what I had meant.)

To be honest I don't fully understand the question. I might be missing something but if the different regions just happen to be balancing each other out then to me that provides the answer?

However, it's worth noting that the lag is not a maximum 28 days. The dashboard reports cases (unless the data file is different from the website graphs) as number of people tested positive, using the date of their *first* test. It then reports deaths within 28 days of someone's *last* positive test. So for anyone who has more than one positive test the lag can be over 28 days.

It's there in the data, but I'm not sure it sufficiently explains the plateau in deaths.

Regional cases add linearly to national cases, and regional deaths add linearly to national deaths.  So, if the regional decor relation blurs out the valley in dearths deaths, it should blur out cases (minus the blurred lag on deaths).

But it doesn't.  UK level cases has a "valley" 32 days wide (27344 on 11 nov to 273536 on 13 dec).  This valley can't be removed by convolution with a kernel bounded to between +0 and +28 days (by the 28 day deaths reporting metric) containing only positive values.  Introducing an age-dependant CFR model roughly halves the size of the residual valley when running cases forwards through a broad, lagged time distribution to deaths but its still there.  (This happens because recent the recent up-swing in cases has a lower average age).  

The plot below is my the best fit model in my code (after fixing the bug I mentioned up thread) that uses a CFR that's exponential with age and now a skew-gaussian lag function, bounding by 0-to-28 days.   It drops off to the right due to the lack of future cases data.  

Staring at this, the y-axis variation on the daily deaths during the "plateau" is quite a bit more than would be expected for Poissonian statistics which is a big red flag for reading too much in to the data, such as the existence of a "plateau".  It almost looks to me like the actuals follow this model, but with a spike in deaths overlaid on the middle of the valley. 

One observation - if the noise is too bad at the national level to reach a definitive answer, breaking It down to regional levels or below is probably a waste of time, as the noise will be much worse. 

Regional IFR is one way to break the linear addition of cases and deaths that causes the apparent conundrum in this noddy modelling.  

Wow.  That's a really useful piece of information that I had missed, thanks.  So, the data is even harder to interpret than before.  The API now gives access to deaths within 60 days of a test as well and that doesn't significantly change the analysis so I don't think the 28-day cut-off is a major effect (10% maximum).  

Aside - the very small difference between 28-day and 60-day deaths provides a robust and compelling counterpoint to the idiots claiming deaths are mis-attributed due to false positives.  That hypothesis with have (60/28)x = 2.14x as many deaths in the later measure, not ~1.1x actually seen. 

Do the tabloids even need a source as solid as a pissed junior minister? I just assumed they were taking the most optimistic possible interpretation of the strictest scenario from the LSHTM preprint, which has us dropping back down to Tier 2 in February (but also has us all in Tier 4 already).

To a point. Overwhelming the health service is probably more important for most people...

Suspect you'll get your wish. Call me a cynic but London NHS being overwhelmed is going to have a much bigger impact on the politicians and the press than Manchester or Newcastle NHS being overwhelmed...

At a high level, I'm not sure the lack of 'deaths valley' is all that surprising. There are lots of factors in the mix - different case rates by region, with regions moving at different 'paces' in terms of cases, hospitalisations and deaths; different case rates by age band, with this changing over time; plus the fact that death rates take a lot longer to tail off than they do to ramp up. Mix it all up and it's hardly surprising that the dip in deaths has not been significant.

Bit of reading for those interested - PHE analysis of the new variant. In summary, seems to add about 0.6 to R but with no statistically meaningful impact on hospitalisation and death rates (however it's relatively early days to be drawing solid conclusions on hospitalisations and deaths). Which is similar to the LSHTM analysis.

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/...

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/...

Apologies if this has been posted already, I've lost track of all the links...

That comes under the heading disaster...  

DeltaR = 0.6... so not the 0.4 being publicised last week but nearer the (missing) central value of the range from the nervtag minutes. That’s really bad. I don’t think the raw numerical difference in the deltaR makes the point of how bad that actually is for the next few weeks.

Ok, I understand your issue now. If you want to understand the underlying reason a bit better I'd have thought it would be easiest to repeat your analysis on a regional basis. You can then work out whether this is a result of unexpected behaviour in the regions whose cases are reducing, in the regions whose cases are increasing, in all of them, or in some more specific regions. This would give you a clue as to possible causes. If there is no clear trends with region then I think you have to consider the assumptions you have had to make about demographic effects and distribution of time to death just don't fully account for all the behaviour we're seeing in reality.

(P.s. I think regional variation in deaths does explain the national plateau. I think that's inarguable from the data. However I can see from your analysis that the regional variation in cases does not seem to properly explain the regional variation in deaths and that it is the apparent national plateau in deaths that has put you on to this.)

Higher viral load? Low viral load infection takes a while to increase from a level which is detectable to death, whereas high viral load infection ramps up quicker due to the joys of unfettered exponential growth early on in the infection and subsequently hits harder?

Re. the valley issue - if different variants have differing 'times to death' then the change in variant prevalence will alter the spread in death dates and consequently potentially smooth out the valley.

I just did a very simplistic analysis of CFR for the 9 regions in England based on total cases and total deaths throughout the pandemic (using both deaths within 28 days of a positive test and secondly all deaths where covid was mentioned on the death certificate.) These figures are all high because of the lack of testing to pick up mild cases in the first wave.

EM - 3.2%/3.4%

EofE - 3.5%/3.6%

L - 2.5%/3.0%

NE - 3.3%/3.6%

NW - 3.2%/3.5%

SE - 3.1%/3.5%

SW - 3.0%/3.6%

WM - 3.6%/3.7%

Y&H - 3.0%/3.2%

There is nothing remarkable in these figures to me. Obviously London is the biggest outlier. This is sort of surprising because it was the worst hit area in the first wave when there was little testing, which should bias it's figure upwards. Presumably this is over compensated by the many recent cases that have not yet translated in to deaths (although the same influence is not obvious for SE and EofE.)

That is a good way of putting it.

 It’s unarguable that a clear valley is obliterated from the national level figures, yet one exists in the regional figures.   if CFR and lag are geographically invariant, cases so processed would obliterate the valley at the national level.

The more full analysis needed to test if it’s down to regional CFR differences is non-trivial as deaths is such a noisy signal you’d need to fold in hospitalisations and ICU levels to get any confidence in the results I think - at which point it’s a time consuming enough exercise in model generation and validation and in data gathering to be beyond my time.  Longitudinally crunching NHS figures with lighthouse labs scoring for variant is the clear unambiguous way forwards.  Wish I knew more about what’s happening there.  I will do a CFR vs time estimate for periods and regions dominated by old and new strain, but there’ll be geographic and demographic differences.

That seems plausible.

Yes, pulling forwards deaths from the new rising exponential could help level the valley.

I think at this point there’s too much correlated non-random noise in the national deaths data to use it as more than a starting point for thoughts when it comes to the last 6 weeks.  

I wish healthcare was publishing monthly, demographic, longitudinal CFRs along with segmentation by new variant scoring.  They’ve got the data to do this, they must do.

It's very worrying. I live in Kent and was talking to my sister who is an infection control nurse at Maidstone hospital. She was saying all the hospitals in this part of Kent are full and some have been sending A&E patients to other hospitals.
Her hospital is full and about 70% of people in there are covid patients.

I asked her about opening the Nightingales again, but she said there just isn't enough staff to man them.

The government need to address this immediately with a stricter lockdown than we had in lockdown 1, not just to slow the spread but also to help take the pressure off of A&E. Restrict driving, no driving for recreation, no dangerous sports, audit companies to ensure employees are working from home where feasibly possible. We all need to be doing as much as we can to avoid any hospital visits. 

I find it incredible that the government hasn't done any press conferences over Christmas to help drive the point about not seeing people and to urge everyone to do their bit.

I’ve just seen a post of 8 ambulance stacked up waiting outside the hospital in Southend.

What makes me most unhappy about how this crises has been handled is that we are again putting health workers into an entirely avoidable situation of physical and mental overload, and one which I don’t think I could ever handle.  There’s a lot of healthcare workers on here, and their relatives and friends.  I’m sorry that this is happening and my thoughts are with everyone working through this.

Absolutely, my sentiments exactly. My sister isn't frontline but even she has been doing 13 hours days for the last 4 weeks. She's classed as vulnerable too but has carried on working because there's nobody else who can do her job. 

She's now had the vaccine fortunately.

Guardian reporting ‘tier 5’ likely nationwide within days. Meaning t4 with schools closed.

only a week late!

"Figures seen by the BBC show at one London hospital on Sunday morning, ambulance crews were typically waiting nearly six hours to hand over patients to staff."

https://www.bbc.co.uk/news/uk-55462701

My daughter's GP in N London sent a video message to her patients - amongst what she said - apparently if you have a heart attack you may still be waiting 4 hours for an ambulance!

Question: if this new variant is more transmissible because the infected are shedding more virus then does this lead to (on average) people being infected with a higher viral load? And if so, assuming that there is a relationship between initial viral load and illness severity, doesn't this mean that the IFR will increase?

Have I missed something? 

That is a logical conclusion to make, yes. Unfortunately.

I think that's taking several pessimistic assumptions and combining them, so it gives you an estimate on the pessimistic side of things, but that's not a bad one to pick until more is known.

Thanks for both of those.  Very interesting things in there - the re-infection stats from the RT-qPCR testing are reassuring.  

The only surprise there is it's 'only' a week late. They're going to need Northern hospital capacity.

Jk

I think it is probably overly pessimistic given that everything I have seen from the scientific community seems intended to reassure that the progression of disease is similar. They don't have much good deaths data yet but they should have good hospitalisation data by now.

Conservative assumptions in predictions with safety consequences have an important place but being overly pessimistic often leads to the wrong action, poor understanding of the reality of situation, and/or losing people's confidence when the assumptions are later proved false. I think it's more appropriate that any action taken works on the basis of the available scientific advice, which currently is that the variant is more transmissible but doesn't lead to increased consequences to the individual. Of course z we should also keep interrogating the data.

I agree with all that you wrote; at an individual level I’ve no problem following a pessimistic assumption as my guide for risk, and I try to qualify when I think something is the pessimistic assumption, as I have above.

For an exponential risk, where there isn’t time to fix the wrong choice, erring on the side of caution seems prudent - and can be clearly explained as such, with rapid relaxation if it turns out to have been over-cautious, again explained as such.  Looked at as a control theory problem it’s the only viable control law that doesn’t risk catastrophic instabilities.  

Where as we’re in the situation where everyone it seems agrees tougher measures are needed it seems, and yet we wait - despite waiting only making things worse.  I’m not sure this comes down to optimism/pessimism at all. 

That's all fair enough.

Unfortunately I think the actions taken by Govt are too often beholden to politics, idealogy and to a fear that the public will react badly if they later think restrictions were over zealous. However, I do sometimes feel that overly negative discussion or predictions (I don't mean you here) can both "turn off" people who need to be persuaded of the problem and also cause unnecessary stress for people who don't need to be persuaded, so are generally unhelpful.

Indeed, and the lack of trust between the government and a lot of the people is a big hinderance to the kind of "Team of 5 million" rapport built between the NZ government and their people, which made it much less politically and socially fraught to sell what could easily be perceived as over-zealous restrictions.  I don't think the problems underlying this can be quickly or easily fixed, and without that trust, options are limited. 

Yup; I take your point on over-negative predictions and behavioural effects.

I take both those points; it's not a good situation all around.

Edit: On the subject of pessimists, today's reporting count passed 50k for the first time, and is close to a 30% rise on yesterday's record count.   This is whilst mostly reporting on Christmas Day and weekend samples which I'd expect to be lower than the ones around them.  I don't know how much of this spike was delayed reporting from the other nations, but it's not a happy number.

50k+ new cases today. Does anyone know the doubling period?

Unfortunately, we're not going to get back to a "normal" week with the standard "weekend effect" until the middle of next week. Hopefully the distortions up to that point won't cause the wrong decisions to be made or the timing to be delayed.

There's now 21,787 in hospital in England (April peak was 18,974), and the rate of increase has been increasing over the past few days - currently about 6% per day.

https://www.england.nhs.uk/statistics/wp-content/uploads/sites/2/2020/12/CO...

20th December - 12,103 free acute hospital beds. Covid patients 16,663 at that point. Assuming steady state of "other" admissions there's going to be about 7000 free beds now, which gives 5 days at current rates until we run out of beds. 

https://www.england.nhs.uk/statistics/wp-content/uploads/sites/2/2020/12/UE...

Note how many of the trusts in the SE were approaching capacity 9 days ago

Obviously I hope my assumptions are wrong.

This kind of thing, as well as the stacking ambulances outside A&E waiting hours to offload, has been the reality in many other parts of the country off and on for months. In fact in Wales it's often been like this for the past few years!

Very variable geographically and demographically.  Here's an all ages plot for the last two weeks (of data)  per English UTLA.

I can't decide on a colour map, so I included several.  One thing that stood out instantly - as Si dH noted earlier - things are levelling off in Kent, all-be-it at a high rate of infection.  But knowing that R can be returned to ~1 in an area dominated by the new variant is comforting.  (Although I don't know what the positivity is in Kent compared and if that raises any warnings over interpreting case counts - there's not much headroom left in the testing system nationally).

For some reason I found the first, orange/brown, map easiest to look at.

Tier 1 worked well for Hereford, IoW?

I know the populations are small but the different regions in Cumbria should be different colours and they would be quite distinct on your map  

It can be done from the MSOA data, but the download I used to use is gone...   It also doesn't work well as a static figure as you really need to zoom in to see the fine details, and one thing I don't have is the time to set up a website with a zoomable slippy map....  

Astounding.  At playing catch-up...

New PHE analysis of the variant published today, I came across it via BBC but not seen it mentioned on the forums.

Suggests there is no statistically significant difference between the new variant and others in terms of hospitalisation, mortality or the chances of re-infection (ie of getting this new strain if you've had a different one before.)  Also has some data on proportion of cases that are the variant in different areas, from specific labs.

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/...

P.S. as bad as things are, I think that doubling time map looks a bit lighter red than it would have done a few days ago? (I preferred the red/blue version)

Yup, mik82 posted it up thread from a different URL.  Made a nice change to see a chi squared value not a p value.   The reinfection data felt solid and very reassuring, the hospitalisation and mortality findings are couched in provisionals but are not bad news so far; I wonder about lag from infection to hospitalisation however...  If it changed the fraction of cases that are symptomatic or serious (for better or worse) that also wouldn’t be picked up with symptomatic testing.

I think you’re right but I’ve not done a time series of it yet.  Probably won’t get it done tomorrow as I’ve a day’s hard labour in the lab upgrading the interocitors. 

Red/blue is my favourite but the purple/orange corresponds to other plots where it’s used to differentiate from red/blue data of a different format.  Plotting would be way better if humans had 4 or 5 different orthogonal colour sensors in the eye.

For me, the third colour graph is terrible: the hot spots look intermediate. Either of the first two are fine, but I have a slight preference for the first.

AZ vaccine approved, and an announcement on the tier review this afternoon. I STILL don’t think they are going for a national lockdown. Sounds like more tinkering with tiers plus an announcement on schools (t5?)

Already hearing the ...to be administered in GP surgeries....phrase. So basically even with all this time available we haven't got a plan....just great. Shambles.

They've also taken a decision as previously speculated on ukc to prioritise getting a first dose of the vaccine to as many people as possible, delaying dose 2 until 12 weeks after the first.

Once things have settled a bit I'm hoping we'll see a full writeup of MHRAs findings about effectiveness etc. The news reports are saying that the data used to claim 90% adequacy with a half dose was inadequate and that the approval is for two full doses, which gave 62% protection in the trial... but that some unpublished data has also showed that two doses gives higher protection if taken further apart.

If one dose gives only, say, 50% protection then I think it's really important that's communicated to everyone who has it.

They need to start communicating that the nation isn't going to be out of the woods until most people are vaccinated - any relaxation of restrictions before then can still overwhelm the NHS because NHS capacity is so low in relation to the pool of infectable people.  It's overwhelmed now with ~400k live cases, that's just a tiny proportion of the population.   Many people seem to be of the opinion that as soon as the elderly have been vaccinated things will be OK, but that's patently untrue, they're going to need to get down to the 30yr olds I think.

Agree. But saying that would involve being honest with the population and disappointing those people who seem to think things will be back to normal by spring. Actually I don't know how many people really think that, perhaps not many now...

I keep reflecting on the 'return to the office'. I can't see a business as usual, 'full scale' return (recognising that few people will do 5 days a week anyway) until late 2021 at the earliest, as our office demographic is that most people are in their 20s and 30s - at 39, I'm one of the older ones. Offices with older workforces may be able to return earlier in the year.

Unlike our illustrious home secretary Pritti Patel, UKC is nearly always ahead of the curve.

Lots of new places being put in to Tier 4 from tonight - now most of the country. And some others bumped up to Tier 3. Not found a definitive list just yet.

No real notice which I think is sensible.

Edit: here it is: https://www.gov.uk/government/news/formal-tiering-review-update-30-december...

An unusual sense of urgency from them which hasn't been seen since March.

All eyes turn to Williamson now.

The BBC live feed includes this:

With the NHS now busier than it was in April, the Conservative chair of the Commons health committee, Jeremy Hunt, asks why the government is taking "huge risks" by allowing primary schools to reopen next week and universities to return.

That's the first mention I've seen of universities in recent weeks.  I consider it would be madness - utter madness - to bring them back in to residence until more is known, and until we found out of the regions beyond London and the South East are going to be able to avoid healthcare overload, let alone entering overload before London and the South East exit theirs. So, now I find myself agreeing with Jeremy Hunt.  Never thought that would happen.

From your URL:

Evidence shows the new strain of COVID-19 is increasing in the South West

From my Dec 22nd post on Plotting #4

The South West appears to have jumped over to a "London-like" doubling time curve. 

It's stacking up that regional analysis of the doubling time data is quite a powerful approach.

Today's reporting number are out.  UK level:

• Cases - just over 50k, despite mostly reporting on Christmas Day and the weekend where I'd expect numbers to be depressed by ~20%.
• Deaths - 981 which is a massive jump in daily reporting but looks to be largely filling in the reporting lag, much as expected.
• Hospital admissions - given for Dec 22nd - 2430.  This I think is an actuals number not affected by reporting lag and is growing at about 6% per day.  If that growth has continued unabated from the 22nd until today, around 3,800 people are going to hospital with Covid today.

Comparing the most recent Saturday (still provisional, may rise more) with the previous one, Saturday cases are up by ~52% in a week.   However, it looks like both Christmas Eve and Christmas Day had reduced samples taken, so some of these may be displaced in to the Saturday.  I don't think it's going to be possible to get a very accurate view of cases that is not biassed (one way or the other) by these effects and also by New Year's Day until Jan 7th or so.  Not that it really matters by this point.

Safe to say that this is "not good".......

Yes, agree with this, I've started taking less notice of the data this week as it has become impossible to work out whether the numbers are biased up, down or neither due to reporting and specimen date lags. A week in to the new year we should also be seeing the effect of Tier 4 in half the country so that's when it might be interesting and worthwhile again.

I’m actually unsure what ‘switching over to a London like curve’ signifies in plot 18.

As you know I’ve been trying to apply sigmoidal fitting to this rate constant data (with some success in fixing the new variant onset dates more precisely, but at the cost of significant added complexity).

Anyway, Its clear every region experienced a shift from a falling rate constant to a rising one. The underlying (pre lockdown 2) infection rates and rates of change varied somewhat between regions, as did the point when the new variant took hold. However, it’s clear to me that by the time the south west curve ‘joined’ SE and London, it was already well into the ‘new variant’ phase (as was every region). So what does it mean? Well I guess it means the SW somehow caught up with the SE, in other words the rate of change (of the rate of change) was higher for a few days in the SW than anywhere else had experienced.
 

Why? I have no idea.

Again some of the Tier registrations look 'barking mad' compared to the most recent area data. As ever, including some 'old faithfuls'.

Herefordshire up 69% to 364 per hundred thousand. Up to only Tier 3?

Rutland up 70% to 189 and completely surrounded by Tier 4, up to only Tier 3

I look forwards to seeing it when you're happy with it.  I think there's a paper in this in terms of how to spot an emerging variant in a future pandemic - if I'd been updating my time-series maps at the time, someone might have put this together a week before the variant was announced.  It could still be important to countries less far along the exponential curve for the new variant, when they have less sequencing than us.

Wolly words aren't they.  A brief increased gradient in the doubling time for the SW led to it changing vertical offset from the "blue" grouping in that plot to the "red" grouping.  

Yup.  Then that "acceleration" term stopped when the SW joined the other higher-growth regions.

 Indeed.  I've often had the impression that the exponential rate is highest when cases per 100k are quite low, and then drops off, as if cases have a moderating effect (on a faster timescale than policy) or as if they tear through a small but more susceptible sub-population first.  

There's a decade of studies to be done by people in the fields after this bloody thing is boxed up.  That's some off-topic reading for me for some time.

I've seen this too - it has definitely happened in quite a few places, but generally in a time/place when it was confounded by a change in regulations so it's been difficult to say with certainty that things were plateauing anyway. Kent at the moment is a good example - I think it is too early to be tier 4 taking an effect but can't be sure.

If you want my totally unsubstantiated speculation (from March) on this point, I think there is a small sub population which is highly susceptible and a larger fraction who are less susceptible.. possibly due to partial cross immunity from other coronaviruses (common colds). The less susceptible population is the majority but can still be infected (albeit with lower probability at the same viral load, or requiring higher viral load for similar infection rates) so once the small zero immunity sub population is saturated, there is a step change in R0. This is observed in the first weeks of every country from March/April where cases race up to typically a few 10k and then the log case count gradient changes.

As I say, totally evidence free speculation.

Unfortunately I think the solution I’ve got is specific to the UK conditions of a short lockdown over the period when the new variant took hold. I suspect it wouldn’t easily translate.. the real power is in looking at the rate constant, not the cases. That’s all your work. My contribution (if you can call it that) is just in the noise of trying to specify the kick off time variation. FYI, R^2 for critical time vs distance (from Medway) improved from 0.74 (your analysis) to 0.89, so something but not earth shattering. Certainly didn’t change the conclusion.

It wasn’t the words I was meaning. The actual dynamics is really difficult to explain. 
 

there must be a super susceptible sub population that got exposed very rapidly. 

see my other post above

aye, not for me though.. ;-p

A plot and run as other things to do...  Super quick plot that doesn't specifically note periods of control measures etc. 

For every date between 2020-08-01 and 2020-12-15, for every English UTLA, fit an exponential to ±7 days of case data to measure the exponential rate, and measure the average number of cases in that ±7 day window, and normalise this to the population.

• Scatter plot, coloured by date (left)
• Density plot (right).  Pseudo-log colour map.

The far left (green, August or so) is not very reliable I think as the numbers were so low that there was a lot of noise, meaning that measurements are fuzzed up and down vertically broadening the "underlying" pattern.

There looks to me to be a pretty clear diagonal from (x-0, y=0.1) descending down to (x=20, y=0), implying that, for R>1 (exponential constant +ve), cases rise fastest when case rates are lowest.  This is the "most common" line and is blue in the right plot, the same is then visible in red for higher values, but happening less often. 

• Sub-populations?
• Saturation of local testing?
• Something else?

It looks a bit like the "envelope" on the plot expands recently.  It would be interesting to segment this in to two graphs based on when each UTLA fell to the new variant.  

In reply to minimike:

Gotcha.  Yup, really intriguing dynamics.

More than halving the misfit is another way to look at it.

As ever, your data analysis and plotting is convincing in showing there is an effect. I don’t quite follow your comment about the envelope widening recently..

I wonder if the sub populations would break down by age demographics or more controversially ethnicity (genetics) or socioeconomics (housing density?)?

Behaviour change? Restrictions introduced/increased? Scary local news stories?

Wait, is this currently active cases/100k, or cumulative since start of pandemic?

either way an axis max of 100/100k seems too low..

edit: I re read your post and it’s the 7 day case load, so neither of the above. In that case I really don’t understand as it’s the current number of cases which apparently moderates the infection rate, not the cumulative total, which you’d expect to be more relevant in terms of herd immunity in sub groups. Or have I missed the point?

On first pass, it looked to me like there were more pink points than white points in the region where the product (cases/100k)x(exponential rate) is most positive - the region of high absolute case growth (I should put absolute growth rate on the plot as contours...).  Now I'm not sure if I believe that or not.  It needs a proper test.

I think it's a mix of everything; it comes down to the product of their risk of catching it and of transmitting it I suppose, which are probably quite similar, making the risk proportional to the square of how much interaction they have.  Employment type being a large factor which has links to ethnicity; there's a lot to be unpicked here in terms of the clear bias in the outcome of this pandemic.  

I tend to think of it as graph theory / a network effect and I think that goes a long way to explaining why large exponential rates are only seen with low case rates.  The network that the virus is transmitted along is not homogenous - some people have many more connections, and people with more connections tend to cluster (social, demographic, employment).  The R number is higher for more connected sub-graphs.  When prevalence is high, there is transmission is both high and low density parts of the graph, and some sort of average R and exponential rate is seen.  When prevalence is low, the outcome is more probabilistic, and one part of the graph or another may drive the overall value - that is, if there's an outbreak in a highly connected part of the graph but not in other lower connected parts, it's not "lost in the noise".

That sounds plausible, but I'm not going to make a model to test it - too many free parameters to be credible, certainly if done by myself + a lot of google searches....

I clipped the axes a bit to zoom in on the bulk of the data. Expanded plot below.  It's the daily average cases for a 14 day window (±7 days), so it's lowballing the highest rates in the current growing phase and it stops a bit in the past.  It's the same 14-day sliding window used to do the exponential measurement.   I plot the average daily load and the government dashboard does a 7-day sum.  So, multiply by 7 for compatibility with that.

That's the point - it's an immediate effect it seems, which strongly implies against immunity.  I think attack surface / graph theory and prominence of a large attack surface only being possible in low case numbers may be part of it.  

I hadn't considered doing total cases vs rate - that's really interesting.  Cases data misses so much early on though; I might do it for UTLA level hospitalisations as a more accurate proxy.

In reply to  Longsufferingropeholder:

I don't think so - or not exclusively, at least. 

• The periods of very fast doubling (~4-5 days) rarely last more than 1-2 weeks, and they often relax before restrictions change.  They tend to happen at low absolute case rates when behavioural levers aren't being pulled because policy and messaging is driven by high absolute numbers, not low number high growth.
• I've made of movie of R and direction of R (growing/falling) around the failure of lockdown, and the wave of failure heading out from the Thames estuary shows each region turn to R>1, R growing and then R>1, R falling as the initial very fast doubling time passes.  This all happened during T2 and with no obvious trigger for behavioural change.

ok yes I see now. I agree the effect you see in the latest plot can’t be immunity driven on that logic. 

This is what I was thinking of in my earlier ‘speculative’ post. I suspect you’d see a very strong effect looking at early data in this way. I don’t think it matters that a lot is missed, provided there’s no dramatic change in testing during those early weeks (generally there wasn’t, and you’d expect increased testing to have the opposite effect of better detection efficiency anyway).

hospitalisations is I think the best proxy data but it’s not available at UTLA level for March iirc..

There was a reporting error on the Guardian map Herefordshire is up 69% on the week to 183 ( the number of cases was used in error.... now corrected) ... that's good news as cases there are at last holding steady.

:

Numbers in hospital are more up to date and ultimately more important than admissions but similar story there. Steady rise since the start of the month. Up by 1,000 on 27th and 1,200 on 28th. Then if you look at the details by Nation you get a preview for the 29th - up by 1,300 in England only.

It's always worth checking the data in guardian map against the gov.uk dashboard if somewhere looks surprisingly high/low. It takes data from the dashboard but often contains errors where one UTLAs cases are transferred to another nearby and other similar issues. Useful visualisation though.

It's worth noting that the weekly average case figures have almost all dropped today vs yesterday, because they now run up to and include 25/12, on which very few positive specimens were recorded. They will go up a lot over the next two days.

I would speculate that this is more due to socio-economic factors such as type of employment, housing density and so on.

You appear to be correct; it just doesn't seem to be available at UTLA level, which shouldn't have surprised me because the NHS trusts don't align with the UTLA boundaries...

This sent me back to re-reading the API documentation.  There's a wealth of stuff in there I'd overlooked including case figures by gender and apparently cumulative hospitalisations by age which can always de-cumulated.

Here's a plot from the API data of hospital occupancy - total vs ventilator beds.  Markers are raw data coloured by date.  The grey line "joins the dots" and has been filtered to smooth it.

You can see "loops" formed by the first and second waves, with the second wave having a much lower ITU occupancy as previously discussed.   The lag from hospital admission to ITU admission causes the lissajous figure style loops.   

Look how close both got to 0 in the summer.

What's really interesting is that the 3rd wave is forming a shallower angle than the second - ITU admissions are a lower percentage of admissions (or are going to happen later) than for the recently passed second wave.  

This suggests that the third wave is currently proving less lethal.  I assume there has been so sudden improvement to early stage treatments around Dec 6th when the angle of this line seems to have changed.  I think this tallies with the demographic shift in the cases data.  At this point I think I hit a wall in what I can do, but the key question it raises "is this due to demographic shifts or a reduced lethality associated with the new strain?".  

Yes, it was an awful choice.  I picked a stepped one because the steps let you read off actual values quite well, where-as that's really hard to do for a graduated map.  God knows what I was thinking picking a stepped one with green in the middle...

Better treatment compared to spring - more use of non invasive ventilation not requiring ICU. That wouldn't explain more recent shifts.

It's all rather gloomy but just imagine if we didn't have the prospect of a mass vaccine roll out in the coming weeks...

Nice plot.

Is the angle shallower? To me it looks like negligible difference really, I think it just starts slightly lower because of the decline in ITU cases from the 2nd wave. It's early to say either way though. Worth monitoring in your weekly update if it's easy to do so? It's certainly not steeper which is good

The thing about it that I don't understand, which is less important now, is why when the second wave started in summer, it looped upwards and back over rather than under. I think this suggests a higher proportion of people affected early in wave 2 were needing ventilation than had been who were left over in hospital at the end of wave 1?

Edit, the 3rd wave is obviously less steep than the first wave.

Perhaps we are looking at the same behaviour in a different way. Essentially, the start of the second wave was convex in shape rather concave, like the 1st and 3rd wave, if you understand what I mean, which means the very beginning was steeper but only for a short period of time. I don't really get that - it seems intuitively sensible to me for ITU capacity to rise initially more slowly in a new wave as people come in and then get transferred to ITU after a short period of time, with an equilibrium being reached after the early admissions in the wave have all got to ITU if they are going to? I suppose this depends on assumptions about how people progress through hospital.

Re: third wave less lethal. I regularly compare local authority areas for Grter Manc, Merseyside and SE/London. Whilst there are reporting issues over Xmas etc (Wigan just reported 20 hospital deaths - most likely due to reporting issues rather than a spike in deaths given our rates etc), it does look like proportionately fewer people are dying in the SE. I doubt treatments are better in the SE then in the NW last November, so this points to the new variant being less lethal. Early days but given SE cases started going up over 4 weeks ago I would have expected deaths to start going up last week. 
It may be due to health inequalities (eg Wigan has historically had high rates of respiratory diseases due to industry) and poverty/health factors, but age does appear to be the key factor in death rates. I get that in SE high % younger people have covid, but that was also the case in the NW (perhaps to a lesser degree?)  

Wigan is now very close to having double the deaths during this wave than the first. Never thought that would have happened. 

Zoomed plot below with some linear regressions.  It's quite a bit shallower, but there's not much data to estimate it from - but the regressions fall pretty close to their raw data points.

Its comparing apples and oranges though - the tail end of wave 2 with the start of wave 3 so you are right to say we need to wait for more data.

Totally agree that it's to early.  It's much easier to add things to the daily updates now I'm using the API rather than manual downloads...

Agreed - and it's complicated by the demographic changing over time, with the probability of going in to ITU presumably increasing with age.  I'm going to expand my noddy demographic model to go cases > admissions > ITU as a way of getting my head around it all.

in reply to Mick Taylor:

Interesting digging through regional stats - thanks.  

One interpretation is that factors (social, new variant) are causing proportionally more spread in younger people, which means that fewer of the hospitalisations are "serious", so the variant could still be as lethal for a given age, but it's opened up spread in less lethal ages.  There's a difference between individual/specific lethality and net/all encompassing (across the demographic) lethality.   I am absolutely 100% not levelling what I say next at you Mick but I've seen other other people quite deliberately misrepresenting the virus as having become less lethal in periods when it was largely confined to younger people.  So it's important I think to be clear on if the virus itself is less lethal or if the effects of adding the virus to the demographic mix are less lethal.  It's nice (and unusual!) to see a hint of something positive in the data but a lot of this will come down to demographics I think.

Spot on. Care homes are much better at dealing with the virus now and the behaviour of their occupants hasn't changed due to changes in restrictions - once they're in the home they're in (relatively speaking). Conversely, the rest of the population have been relaxing their behaviour as time has gone by, especially in low prevalence areas like Cornwall. This relaxed behaviour also explains the initial high rate exponential rise in cases there - as soon as the local gossip / social media mill hears of cases people modify their behaviour, especially when it's relatively close to them. Knowing someone infected one or two degrees of separation away is a much bigger behaviour modifier than reading about cases in the next town / county.  I've witnessed this firsthand, and seen the effect on Jersey's social media with their outbreak before xmas.  (This outbreak was caused by private parties around halloween time, both amongst adults and teenagers). Parents hearing about 3 infections in a sixth form class were then panicking (relatively) because their little Johnny was in year 7 at the same school and modifying their family's behaviour. Not extreme modification, but probably enough to change the rate of spread in my opinion.

Your proposed cases>admissions>deaths stats by demographic group would be interesting to see if its not too onerous to collect. The effects of tiering on the working population would show quite strongly I suspect.

I think perhaps I've been underestimating the behavioural effects of local rising cases - as you say the grapevine is strong, especially in the highly connected communities where transmission will happen faster.  So, the burst of the highest doubling time is self-limiting in that it relies on people initially taking a relaxed attitude to the rules, which tightens in response to the local situation.

I've just had a go at pulling the data down from the dashboard API.  FFS it uses different demographic bins than the cases data - not just coarser (which is fine) but with some boundaries mis-aligned.  That has to be bodged around to join the two up.  More places for bugs to hide, more free parameters.

If I do pull a model together, it can either be cases>deaths or cases>hospitalisations>ITUs but not including deaths, as quite a few people die outside of hospitals and the data doesn't appear to be broken down publicly in a way that separates that.

Plot below of the demographic admissions data for England.  The main thing to stand out to me is that the "85+" age bin is doubling faster than the others and will soon be the highest in absolute numbers/day.  

If you think you can fudge some sort of model that would show trends it's worthwhile I think, even if it doesn't quite match up with boundaries. Unless population centres are in one set of stats and not another then they should be similar enough.

Looks like the prevalence has got too high to shield them effectively - xmas family gatherings aren't included in those stats . Looks like we can expect a hike in death rate soon as a result.

I'm hoping large numbers toned down xmas and most of the real idiots will have been infected earlier in the wave and recovered.

Latest (from Guardian live feed)

UK surpasses 73,000 deaths

The UK government said a further 964 people have died within 28 days of testing positive for Covid-19 as of Thursday, bringing the UK total to 73,512.

Separate figures published by the UK’s statistics agencies for deaths where Covid-19 has been mentioned on the death certificate, together with additional data on deaths that have occurred in recent days, show there have now been 89,000 deaths involving Covid-19 in the UK.

The government said that, as of 9am on Thursday, there had been a further 55,892 lab-confirmed cases of coronavirus in the UK.

It brings the total number of cases in the UK to 2,488,780.

It's really bad. However, my reading of the data today (latest weekly averages plus looking at the full cases graphs for some specific locations like Basildon, Havering etc) is that the worst hit areas of London and Essex are following Kent - they are turning the corner. I think this could be Tier 4 starting to take effect now, or certainly the London change to Tier 3 on 16/12. However, I have had a drink so I'll look again in 24 hours. Happy New Year folks!

I'll be happier when the local data includes the rises in the last few days.

Its nice to see Herefordshire level off on the Guardian map for Tim and the other locals.

Happy new year data voyeurs and data sceptics.

Do you think this is going to be boxed up anytime soon? I suspect we are going to be looking at a virus that quickly mutates around the vaccines, I wouldn't be surprised if this virus is a sporadically reoccurring issue. Its another reason why we are going to need to vaccinate everyone we can of all ages, as even if children or young adults very rarely have complications they will allow mutations to occur for it to re-escape.

All this "spike protein" talk - it appears that the virus is effective because of the spike protein - or at least it needs it to get inside our cells. The vaccines are targeted on the spike protein (if I understand correctly) so virus mutations that still need the spike protein should still be combatted by the vaccines.

If a mutation occurs that doesn't need the spike protein but still leaves the virus as "effective" against us (by finding another way), then the vaccines are much less likely to work and we might almost end up back in square 1 - not a pleasant thought.

You are right in that the more virus that is out there, the more mutations will occur, so the likelihood increases that one of those mutations will be untroubled by the vaccines but still as nasty to us, so yes we should aim for vaccination of as many people as possible (globally) as quickly as possible.

You’d need one of the biologists here to give a proper answer.  

Mutations tend to gradually accumulate, and each one at worst usually just weakens immunity but doesn’t break it. It seems likely this new variant broke one particular epitope or “attack location” for neutralising antibodies but leaves many more intact.  So I hope we have - worst case - a gradual failure of immunity.  This means people gradually get re-infected and there isn’t a repeat of the problem where everyone is susceptible and so it spreads like wildefire.  I’m more worried about mutations that make it more transmissible or more lethal.

Of course, it could swap its spike with another variant.  This I think can happen.  That’s a bigger problem and perhaps how this one came to be.

The less of it thats out there, the less likely it is to get worse.  I doubt it can be eliminated globally now so we’ll just have to learn to live with it.

I’m hopeful things hit pretty normal by may and then stay good in to winter next year, but we just have to wait and see.

Happy new year all!

Here's a paper where they have mutated the virus to be completely resistant to one person's antibodies to the original one. It took 80 days of continuous incubation and 3 mutations . They then tested it against other people's antibodies and 8/20 still provided a response.

https://www.biorxiv.org/content/10.1101/2020.12.28.424451v1

Aside from being slightly terrified and hoping that this lab's biosecurity is up to scratch (!),  it does suggest it doesn't mutate too quickly/easily and even if it did escape, a significant proportion of people would still have a degree of immunity

This is the compelling reason why you need everyone vaccinated, and over the shortest timeframe possible.
Unfortunately anti-vaxers are almost by definition too thick to understand any of that paper.

Also illustrates that it will happen. It can't not. Which is what I always end up trying to explain after someone says "when things are back to normal in spring" or "when coronavirus is over next year".

Really good read that. Thanks for posting.
Just went back round for a proper full read. My take-away is similar to yours. It's when, not if, but the vaccines should present multiple targets so some will be lucky and growth should be slower. Also if there is a highly successful strain we might not sit around holding our dicks while it ramps up for quite so long, and might have time to get a tweaked vaccine out.
And so the cycle repeats until it starts to feel a lot like flu 2.0....

My understanding is that this was established a while ago from following the genome sequencing data. Which is rather lucky really. The last thing we need is Covid to mutate at the rate the flu virus does (although they seem to be able to tweak the vaccines accordingly every year).

We can all be grateful that Trump is getting kicked out of the White House because unless we get the US and EU giving serious funding for the WHO to procure vaccines for all the poor countries of the world it will have plenty of hosts to mutate in.  

Pinch of salt alert - anyone reading this post, don't read too much in to it.  Modelling can be used as a way of trying to understand data and that's what I'm doing here, rather than trying to predict.  Even then, it's a quick, non-robust and non-reviewed piece of work, so it has no conclusions only tentative observations. 

I did a model with a hospitalisation rate exponential with age (as with published observations on deaths, not sure how valid it is here), and with a lag from test to hospitalisation that's a skew gaussian, with the skew gaussian parameters being a 3rd order polynomial with age (to allow different ages to have different lags, but to maintain a continuous behaviour).  This is applied across the 5-year age bins from the demographic cases data, and then re-binned to the irregular age bins of the demographic hospitalisation data. Model parameters are optimised with a Nelder-Meade simplex to minimise the RMS error between actual and modelled hospitalisations.

Mainly it showed me that even a simple model can produce what-ever interpretation you want with a few tweaks to the structure here and there, so I'm not going to share the full results as it's navel gazing rather than science.

My main observation is that any variant on the model I tried fit to data up to 20 days ago then, when run forwards, predicts that hospitalisation start rising ~ 10 days later than in the actuals, although they rise at the same rate.  Lots of different interpretations to this:

• My models are bollocks
• Artefact of fewer and fewer infections being detected as cases as a consequence of the current exponential phase
• Reduced time-to-hospitalisation but not increased hospitalisation probabilities for the new variant.

So I don't think it tells us anything about the changing situation, but it does hint at what questions to ask of the longitudinal healthcare data.  

The other observation is that the older you are, the sooner you're going to hospital from a +ve test, and that for younger people the window of test > hospitalisation is several weeks long.  This tallies with qualitative stuff I've read elsewhere.

One internal part of the model is the cases multiplied by their hospitalisation probabilities across the 5-year demographic bands before the lag function is applied to give final hospital admissions.  It's a speculative way of increasing the vertical resolution on the coarse age resolution hospitalisation data I plotted above.

Another 53k reported cases today. But.. Specimen dated positive cases from the 29th.. 64k(still incomplete)!

Yes, not good.  Although I’m guessing it has an exceptional reporting spike in it, falling after a 4-day weekend.  

Another report is out on the new variant - https://www.bbc.co.uk/news/health-55507012

The Imperial College study suggests transmission of the new variant tripled during England's November lockdown while the previous version was reduced by a third.

Full link for the Imperial College study here:

https://www.imperial.ac.uk/media/imperial-college/medicine/mrc-gida/2020-12...

They give ratios for the increased R as well as absolute values - so estimates 1.4-1.8x the R of the non new-variant.

Interesting but not happy reading, isn’t it.  It’s good to see the LSHTM model fitting analysis and the independent basic-but-robust (*) data analysis approach from ICL coming to basically the same conclusion - this gives me high confidence that SAGE are working this problem from a well informed position.

Figure S1 looks to correspond well to my epicentre analysis.  They have a much more scientific way of saying what I did on the PPE thread about the power of spatiotemporal correlation over many locations effectively ruling out other reasons for the new variant rising to prominence.

I’m still pondering the narrative that both additive and multiplicative factors to change R have mechanistic interpretations compatible with it being more transmissible; I think that has to rely on the demographic shift to effectively being new links in to play by crossing a threshold. 

(*) meant in the best possible way.

Edit: Figure S1 doesn’t seem to support implicating the new variant in Wales’ recent period of sustained exponential growth in cases.

I agree, the R factor thing is complex. It’s clear enough what it means as a pure R0, but if the change is in time to infection, rather than infections per generation, it’s not captured in R at all. Hence in a current ongoing pandemic (with countermeasures, partial exposure, etc..) the exponential rate constant (in time) is a better measure I think. I would be interested to know what the numerical impact on that is. From your analysis so far, at least +0.03 per day, which is a LOT.

Figure S1 in their paper gives me a basis to pick comparator regions and date windows to look at the demographic ratio of exponential rate constants between the variants.  It's on the list... Eyeballing it suggests a factor of about 1.6x.