UKC

Another happy update with little changed in terms of what's happening from last week's update.

Cases, hospital admissions, hospital occupancy and deaths all continue to fall in England.  The last few days of the cases curve in particular on plot 9e are provisional as always.  Before that, it looks like cases have been halving every 15 to 20 days.   A few observations - with some speculation on my part:

• An encouraging sign is that deaths are reaching a similarly short halving time to cases.  They really are plummeting fast in absolute terms right now.  It's notable that they're apparently halving faster than hospitalisations; this could be because some of the more elderly care home patients were not being admitted to hospital, but that's speculation on my behalf.
• We can see that hospital occupancy is decaying much more slowly than any of the other measures.  I think this is going to be a slower process than in April 2020, as the big lessons learnt in clinical care since wave 1 mean that more people are not dying and are therefore needing more time in hospital to recover.  
• Looking at the cases data, perhaps it's heading for a slower exponential rate in the last few days, meaning the decay is slacking of a bit.  The data in plot 9 is fit to a ±7 day window so isn't very affect by just the last few data points yet.  Sometimes this is just the noise in the data, sometimes it turns out to be real.  We'll know next week, but...
  • Something different is happening with pillar 1 + 2 testing - normally testing is demand driven and so should drop when cases drop, but although cases are down ~3x from their recent peak, testing is up and still rising.  There is normally a gap between actual infections and detected cases of perhaps 2x - 2.5x.  The ONS haven’t updated their estimates of daily infection since November for reasons I don’t really understand.  There’s a lot going on at the moment including mass testing with LFTs which I think funnels people in to the pillar 1+ 2 testing route on a positive LFT, and mass population-wide PCR testing is being rolled out in areas where the South African variant has cropped up.  Perhaps a bunch of symptom-free people who wouldn’t normally be caught by this testing are being funnelled in to cases; if that’s going on, it means cases aren’t falling as fast as infections, which is good news as it means more infections are being caught and funnelled in to test/trace/isolate which will go on to help infections fall faster
  • Another interpretation could be that the chains of infection clearly broken by lockdown have now run their course and we're left with the chains that are much harder to break, meaning that we've moved from a fast decay to a slow decay.  I'd have naively expected this to have hit about 2 weeks before now if it was the case.

Link to the last thread: https://www.ukclimbing.com/forums/off_belay/friday_night_covid_plotting_10-...

In reply to wintertree:

The four nations cases and characteristic times plots.

The decay looks to be really slowing down in the last few days of data in Scotland and Northern Ireland in plots 6n and 6s.  There are corresponding rises in the exponential rate (or increases to the halving time) in plot 9x, but that's in the highly provisional zone.  

From what I can see, the pillar 1 and pillar 2 data is not broken down by nation, so I can't use the data to understand where increased mass testing could be closing the gap between cases and infections.

As with the English cases, we'll have to see how it develops over the next week.

In reply to wintertree:

The regional plots all show the continued decay of all measures in all regions.  Deaths in the South West have now entered clear decay to join all the other regions.

In general cases are halving faster (in fewer days) in the "red" regions that were hit earlier by the new variant.  That's quite interesting to think about.  

There's a lot of "wobble" visible in the halving times for cases on the right hand side of plot 18.1.  This is largely residual "weekend effect" not fully fixed by my filtering.  How much of the current slowdown in decay is "real" and how much is residual from the weekend effect is unclear.   This will resolve by next week, but then we'll be left asking the same questions about next week's update.

Still, everything continues to move in the right direction.  This is the first update where all regions have been in decay in all measures which is a nice update to post.

In reply to wintertree:

That’s disappointing that Scotland’s decay seems to be slowing. Anecdotally I get the impression significant numbers are giving up on the lockdown rules. Glasgow has been under heavy restrictions, initially promised as short term, for over three months now. I am convinced the Scottish government needs a new plan (no idea what...) that recognises that locking down ever harder isn’t going to fix things. 
Or maybe I shouldn’t be too down about it- maybe it’s just a shoulder on a general downward trend 

In reply to wintertree:

A slight variation on Plot 16:

• Plot 16.1 - UTLAs are ranked horizontally by their case rates when we entered lockdown on Jan 5th
• Plot 16.2 - UTLAs are ranked horizontally by their case rates two weeks before the date of the plot

Plot 16.2 gives insight in to more recent changes in the data.

A reminder of the annotations that not UTLAs showing concerning trends:

• ▲ - Cases have risen more than 1.1x above their minimum since the ranking date, suggesting that cases could be on the rise - or it could be noise.  The minimum level is shown by a horizontal line under the data marker.
• ▲▲ - Cases have risen above their level on the ranking date
• ▼ - Cases have fallen by less than 0.1x their level on the ranking date

Last week I noted that cases dropped disproportionately fast in regions that had had the highest rates; I wondered last week if they would now drop in lock step and that seems to be happening.  That really is fascinating.  If the regions with high case rates were somehow preventing the fall of those with lower case rates through cross communication, then I'd expect the  UTLAs on the far right of plot 16.1 to start falling more, now that the other UTLAs  are down to their level. So, perhaps I'm not so worried by the UTLAs on the far right of plot 16.1 throwing up warning signals. 

So, Stockton on Tees, Nottingham, North Tyneside and Rutland present as the main worries.  I looked at each of these on the government dashboard data, which includes the provisional window which I don't use for these plots, and which doesn't have my deweekending.  

• Stockton on Tees looks to be dropping.  It looks like it's had a big weekend effect which is mucking up the analysis.
• Nottingham looks like it could be going in to growth.
• North Tyneside looks like it's definitely growing
• Rutland looks to be growing slightly from my data; the provisional data (which I don't use) has a tripling of cases [1].  The absolute number of cases in Rutland is low, but this is way, way beyond statistical noise.  AFAIK there's no enhanced mass testing going on in the area.  The absolute change in number of cases (about 60 more in total, so far) is sufficiently low that this could be one super spreader event.  If it isn't, it's deeply concerning.  One to watch very carefully.

[1] https://coronavirus.data.gov.uk/details/cases?areaType=ltla&areaName=Ru...

In reply to wintertree:

The demographic data on English cases.

Plot D1 - the demographic breakdown of cases and their exponential rates for England.  

• All demographic bins remain in decay
• The decay rate has backed off (longer halving time) for all but the youngest age bin over the last few days.  This is visible as the intense blue becoming paler.  The right most measurements are highly provisional and will change with more data; they may be affected by a strong "weekend effect" in the cases data - or they may be backing off.  Time will tell.
• Some of the most intense blue (fastest exponential decay) is in the most recent week of the data for the oldest age bins - could be a sign of the vaccination program.

Plot D3 - rates.  This plot has been overhauled to show change:

• The data markers (black circles) show the average rate over the most recent 7 days of data
• The vertical "tails" show the change from the same average over the 7 days previous to those used for the data markers.  Blue tails show things getting better, red worse.
• Changes in the last week
  • The biggest improvement (more negative exponential rate, shorter halving time), is in the oldest 3 age bins.  Very encouraging given the vaccination roll out; this suggests to me that it's visible in the data.
  • Most other age bins have a slightly worse rate (more positive, slower halving).  I wouldn't like to say if this is real or partly weekend effect.  But, compared to these increases, the decreases in the oldest ages seem all the more noticeable.

Plot D4 - this is not a prediction. It extrapolates the current situation forwards to March 8th, the notional day for reopening English schools.  It's not a prediction as things will change between now and then - vaccination, the weather, variants, behaviour.  It's done as a way of understanding changes to the current situation, by asking "if this goes on as it is...".  Michael Hood offered this alternative explanation which I like:

• For each date ("from date" - "to date") on the x-axis, the "point" shows the number of cases there would be on 8/3/21 (y-axis value) using the latest data to determine the exponential decay rates at that date (on the x-axis) and maintaining those rates through to 8/3/21 - so every day the curve will approach one "point" closer to 8/3/21.

The last few days of update to this are not what I'd hoped for.  Whilst cases are halving faster in the oldest age ranges, these contribute a very small fraction of cases (but a much larger fraction of hospitalisations and deaths, which is why the vaccine is targeted there).  This worse extrapolation reflects what I've said about the decay in case rates appearing to slacken off in various plots.

 But...

• Projecting an exponential forwards in time is very sensitive to noise and there could be weekend related issues.
• There's speculation on my behalf that case rates are including more asymptomatic cases due to enhanced testing.  If that's the case, this data will go up which looks "bad" but is actually "good" .  There's no "touchstone of truth" for this however, and cases are an inherently limited measure, but the data is what it is.

We'll just have to see what another week brings.  In the mean time, more and more people are getting their vaccines and the evidence is coming in that this reduces transmission a lot.  Right now, everything is still in decay, which is great.

In reply to wintertree:

Just wanted to say thanks.

In reply to wintertree:

The decay of cases in intensive care is now clear at a national level and is there for some regions.  There's some very interesting regional differences going on with the phasing of the decay of one vs the other.  

At a national level it looks like ITU occupancy decaying much more slowly compared to general occupancy than in April (gradient of the line of best fit is much shallower), but looking at the regional plots, that is far from uniform.  There's hints to me that it's not unrelated to the new variant and my "red/blue" region assignment, but it's far from certain.  I hope that those with access to the data are crunching this on a longitudinal basis including the "SGTF" scoring for which variant each person has.

Big picture - things are getting better.   But, to contextualise where they're getting better from - perhaps another 10 days of data (8 days actual) are needed before hospital occupancy is down to the level seen in the first wave.  As talk invariably turns to opening up it's important to keep in mind the sheer number of people in hospital beds and the much larger number of people who've been over worked and continuously exposed for the last month to a level of patient care that brought healthcare to its knees in April 2020.   The emotional come down from this when the pressure backs off is going to be forceful; I hope that someone somewhere is planning the support for this.

In reply to Wainers44:

You're welcome.  Hopefully in a few months I'll be looking for a new hobby - and able to get out and about to do it!

In reply to wintertree:

There is something apparently going on in Lincolnshire. At UTLA level it doesn't show because rates are falling well in Lincoln itself.  If you view the map at LTLA level there are a few areas currently rising on a 7-day average basis. The worst is East Lindsay - up 40% week on week - and Boston is also up 25% week on week despite being an area with a higher absolute rate (Lindsay is still fairly low.) I hadn't previously picked up the issue you found with Rutland or Nottingham. Looking around the map now I see that Corby and Kettering (LTLAs, parts of Northamptonshire at UTLA level but very close to Rutland) have also flattened and are currently slightly rising on a seven day average basis.

This is all a bit worrying because these areas all fairly close in geographical terms, ie same part of the country. It certainly looks like any significant decay has paused in large parts of that region, and not in a way that is explained by how the low prevalence areas are affected by high prevalence areas as you discussed above. Hopefully this is all an artefact of increased testing in which case it should start falling again once rates of testing flatten.

In reply to jonny taylor:

It looked very similar two weeks ago - before resolving itself back in to decay.  The cases data is really messed up with big lows over the weekends and highs on a Monday - and the amount of messed-up-ness is very variable week to week.  You can see it apparently emerge and then get overcome in the last two updates - and then you can see the "step" in plot 6s on this thread.  

To quantify "messed up", the standard deviation about a polynomial trendline is ~ 20x what you'd expect for Poissonian noise for UK level data on cases.  It's about 2x that for hospitalisations and deaths.  I'm 95% convinced this is because the date given is the data a sample enters the lighthouse labs, and not the data written on the postal sample container.   This is maddening, because if fixed it would massively improve the data quality and the ability of people making decisions to understand what's going on now - lags being one of the main drivers of instability of any attempt to control case rates.

https://www.ukclimbing.com/forums/off_belay/friday_night_covid_plotting_9-7...

https://www.ukclimbing.com/forums/off_belay/friday_night_covid_plotting_10-...

I'm convinced that better messaging would improve compliance.  Simple things like setting milestones in terms of case rates so people can feel the end in sight, rather than notional dates.  I'm afraid I've not paid much attention to messaging north of the border, as things got worse I got more and more myopic on my local area.  

I'm very much hoping it's just a "shoulder" in the data.  Perhaps I should train up a recurrent neural network on it all to predict ahead....  I kid, I kid.  I wouldn't be able to live with myself.

In reply to wintertree:

Glass half full - this is what we would expect to happen as a result of vaccinating the oldest first, right? As a much bigger proportion of hospitalisations is made up of still in vaccinated groups, that will be less effected by vaccination. If deaths are dropping as fast as cases at the leading edge, I'd say that's a very promising sign. Normally they drop much more slowly.

Edit - I meant to say a much bigger proportion of hospitalisations is made up of still in UN vaccinated groups

In reply to Si dH:

A truism in the absence of Covid... Thanks for the comments on LTLAs within Lincolnshire.  Despite being a UTLA, Rutland is closer to an LTLA in terms of population and so it tallies with your observations - thanks for those.

I need to bite the bullet and figure out the LTLA part of the API.  

That's a good point - there's only so much that increased testing can raise cases before it ceilings out on detectible infections.  I'm not aware of any drive for increased mass testing in Lincolnshire though?

One point about rapid mess testing - the number of "live" infections in an area is perhaps 20x the daily rate of new infections, so a very fast mass screening program could detect an awful lot of new cases vs symptomatic screening in just a few days.  More than ever I think it would be useful if P2 cases that originated from mass screening (LFT, boots on the ground response to new variants, etc.) were separately reported.

Indeed.  If the trend doesn't stop very soon, it's one to escalate for attention by other channels where possible.  

In reply to Si dH:

Could be, could be.  It's been a long time since I trawled through the various NHS spreadsheets and did some modelling (to overcome their woefully coarse age brackets) that made me think a lot of > 85s weren't being hospitalised.

It looks like there is demographic deaths data now available from the dashboard, so that's something that could illustrate this to help understand better.

In reply to wintertree:

Interesting tidbit for next week (and as I'm focused on my local area) - Sefton has been dropping fairly continuously well since lockdown was introduced but from the date when the discovery of the SA variant was announced (ie in the days for which data is incomplete) the case rates have turned sharply upwards almost overnight, I hope and assume focused on the Southport area where door to door PCR testing is happening. I expect this will only last a few days while they do all the local community testing. Will be interesting to see how it looks in a week, here and in other areas that have had similar testing done.

In reply to jonny taylor:

It is disappointing but maybe not surprising.  To compare with the England graph remember that the population of England is slightly more than 10x the population of Scotland.  So if you normalise for population by dividing the English numbers by 10 Scotland is levelling off at about 900 where England is falling fast but is still at about 1500.   The England graph may well level off at a similar level, it just hasn't got there yet.

One thing is for sure, if the levelling off continues then talk about opening up again is premature.  It's more about what to tweak to make the lockdown more effective or how to shift the vaccination program to reduce infections.

In reply to tom_in_edinburgh:

I think “levelling off” is a bit of a stretch for the data we have now.  Look at the raw data markers on plot 6.1s around 01-16 for context.  They had the kind of dispersion we see at the current leading edge.

Perhaps the exponential rate is getting lower but I’d be very surprised if what we see now translates in to a levelling off.

As the easier chains of infection are closed out, we get left with the more resilient ones - critical workers, care workers, high dependency individuals, rule breakers.  So there will be some case rate where the decay gets slower when all the “easy wins” are won.  But there are counter-mechanism; lower case rates has non linear benefits, for example in the responsiveness of test and trace, and in helping infection control in hospitals and care homes.  

In reply to wintertree:

The Rutland data looks a lot like one big outbreak, no? Trending slightly downwards and then a massive spike. That doesn't seem like Rutland folk just being a bit over-keen on dinner parties...

A quick Google finds a local news story about an unspecified number of cases being detected in HMP Stocken (Category C, 950 inmates in total), which had previously been Covid free. That seems like the sort of thing that might account for it? Not a great situation, but probably less worrying than an overall upwards trend...

In reply to my faithful and spiteful disliker:

Hello!   I missed you last week.  Feel free to use your words to explain your problem.  

You’re slipping by the way, the time was you were mashing that dislike button within minutes of me making a post.

In reply to Ramblin dave:

Hard to say as it’s in the provisional window.  If it’s one big outbreak, that should be clear in 2-3 days.

I agree; it’s what you’d expect if something suddenly clued the authorities up to test a specific sub population.  

You’re better at googling than me.  A prison of that size could well generate a big spike; as I noted up thread, Rutland is a UTLA with a very small population (more LTLA level), so a single super spreader event (outbreak) could be unusually dominant on its numbers.

I ageee; it would be a lot less worrying.  The counterpoint is Si dH’s commentary/analysis at the LTLA level in neighbouring Lincolnshire.  I will see if diving in to the demographic data on the two areas offers any more insight.  It looks like the gov dashboard api also offers cases by gender; gender has significant correlation with incarceration so that could be another way to dig in to this.

Thanks for the thoughtful comments and for looking in to it.

In reply to wintertree:

Thanks as ever. Is it right that occupancy is falling more slowly than in April? Looking at the curve on the UK graph on the gov dashboard, the fall actually seems steeper, albeit from a much higher level. I suspect there will be a long tail, as there was back in spring / early summer, once the numbers reduce - being the proportion of patients who take a long time to be discharged or indeed to die...

In reply to wintertree:

Just to underline this point:

https://www.theguardian.com/books/2021/feb/06/ive-been-called-satan-dr-rach...

In reply to wintertree:

https://app.powerbi.com/view?r=eyJrIjoiMWViOGNkYjUtYTExMi00NzBlLTk4MmEtYzRj...

Can't see any real issue in Nottingham and nothing on the local NHS grapevine. Bassetlaw is the only obvious concern in the county but looks flat with some weird Pillar 1 peaks that look like precautionary testing for  targeted potential outbreaks. The Guardian On the Rise map shows the 'Lincolnshire' arc of growth well, stretching from Boston to Doncaster with East Lindley the worst hit.

In reply to thread:

Preliminary results are being reported from studying the Oxford/AstroZeneca vaccine when it comes to people and the SA variant.  It seems it reduces incidence of severe symptoms, but does not prevent mild infection:

https://www.bbc.co.uk/news/uk-55967767

It seems likely the vaccine won't do much to reduce transmission of the SA variant, unlike with existing and "Kent" variants.

So, there is an urgency to find and eliminate all branches infection from the SA variant, or the benefit of vaccination to lowering cases will be lost until new vaccines hopefully come on line in the autumn.

Fingers crossed.  I wonder how many of the countries we have non-quarantined travel with have this variant circulating below their radars?

There's a noddy plot below of the case numbers reported and gathered by the Wikipedia page.  Some of the most recent reports were more community cases.  It seems likely that there are quite a few chains of infection out there.  

In reply to wintertree:

Thanks again for another update, it is amazing how quickly things have turned around compared to just a couple of weeks ago.

With regards to the Oxford/SA variant news, an official press statement is now out.

https://www.ox.ac.uk/news/2021-02-07-chadox1-ncov-19-provides-minimal-prote...

So basically, not enough data at this current stage to say either way as the sample size was 2000 people with a median age of 31. Not very useful info at the current time, hopefully there will be some more concrete data soon, in the mean time, let's hope that the community testing keeps it in check....

 

In reply to Ramblin dave:

I've been watching an MSOA near where my parents have been going shopping. The last couple of weeks had a sudden rise, putting it at 1000+/100k, although small absolute numbers (80 ish). The latest weekly figures have cases dropping by 85%. Local news sites have some barbed comments about illegal shooting meets...

In reply to Punter_Pro:

Ditto all. To my naked eye, the rate of decrease in deaths during this wave looks similar, or slightly quicker, than in the first wave. The graph only goes to 31st Jan and deaths appear to have come down at similar rate since then.  Hopefully we will start to see the effect of vaccines and continue to see deaths fall at this rate. Fingers crossed. 

In reply to mick taylor:

Quantifying your graph by zooming in on the dashboard - the 7 day average has dropped from a peak of 1338 on 19/01 to 838 on 31/01 - an average drop of 500 deaths per day in 12 days. As we left the first peak, deaths dropped from a 7 day average peak of 974 on 10/04 to 718 on the 22/04 - a drop of 256 deaths per day in 12 days.

So the drop now is almost exactly twice as fast as it was when we came off the April peak. I'm sure there are various confounding factors but in terms of the vaccination effect on death rates with the current variants I think things are looking very positive now.

In reply to Si dH:

Yes, things looking positive. TBH, my gut feeling was with this current version of lockdown light, and my own observations was that things would come down at a slower rate than the first wave. Glad I was wrong. Shows how treatments have improved and hopefully immunity kicking in. Just skimmed an ONS article that states a quarter of people aged over 80 have antibodies - which is good. 

In reply to mick taylor:

The other aspect helping out now I think is naturally acquired immunity.  I think there's quite a lot of that floating about given the numbers believed to have had the virus, and the growing evidence base over persistence of such immunity.  I think that  it'll be concentrated in the sub-populations most likely to catch and receive the virus, especially those at risk of it during lockdowns; so I think more of the pre-lockdown transmission was outside of those sub-populations, making lockdown all the more effective.  This is another reason the SA variant worries me; it could re-energise those subpopulations (at risk of catching and transmitting the virus) like a lightning bolt spreading out tentacles in to every part of the population.  I'm glad to see the government treating this variant so seriously.  

It also shows that despite all the talk of lockdown fatigue, I think most people are sticking with it.  Enough to make a hell of a difference.  

This theory of higher naturally acquired immunity in the sub populations most likely to transmit the virus, combined with the provisional findings over the AstroZeneca vaccine and the SA variant makes for some interesting pondering.   As we come in to autumn, hopefully the current vaccines have done their best at protecting the population from risk of severe consequences from the SA variant; would it then make sense to use the vaccines modified for that variant on those most likely to transmit the new variant instead?  Critical workers etc.   It's far too early to know, with a lot more to be learnt about the variant - and whatever emerges next.  

In reply to wintertree:

It's giant fire breathing lizards next. We've been over this.

In reply to Punter_Pro:

Thanks for that.

Indeed.  It might also have a good effect on lowering general transmission in those areas as well.  

In reply to wintertree:

There seems a possibility of the same transmissibility problem with the BioNTech/Pfizer vaccine. NDR are reporting the county council in Osnabrück are saying they've had 14 infections in a care home in Belm, 13 days after vaccination with the second dose: https://www.ndr.de/nachrichten/niedersachsen/Belm-14-geimpfte-Seniorenheim-.... Not enough detail, and no indication of when these seniors might have been infected, but I'd assume they tested negative before their second dose. It's the Kent variant rather than the SA one they've tested positive for, with mild or no symptoms.

In reply to wintertree:

Going off on a bit of a tangent here - have you had any luck with your stuff being received by a wider or more important audience than UKC? - I know you were sending your work elsewhere.

I had a thought (see below, this bit's just preamble) - if it's not getting anywhere within the scientific community then presumably it's because people are thinking that you're operating outside your specialist area, when in fact what you're doing is being a data scientist and the fact that it happens to be pandemics/virology is almost irrelevant to your work having validity (even though it's specific).

I presume that you can demonstrate that what you were doing showed certain things happening (like the spread from Kent) before that knowledge was generally out there (although of course SAGE and other expert areas may have known). Why not try approaching media outlets - news channels etc - trying to "sell" what you've done and what your monitoring can quickly show in the future as something that may lead them to possible news stories before others get there. Would only take a couple of scoops and then a lot of relevant people would be taking notice.

In reply to wintertree:

I have that sinking feeling again. I would be amazed if SA variant isn't widespread already, and test and trace is still too poor, and cases too high to be able keep them nailed down. I have a horrible feeling it'll be back to square one if (when?) SA variant becomes dominant. Hope I'm wrong.

In reply to bridgstarr:

If people follow the covid rules it won't spread. Where people can't due to essential work etc precautions will need to improve. It's very likely current vaccines will offer significant protection against hospitalisations and death for the SA variant and new vaccines tailored for it will be available much quicker than it took to get the first vaccines (less than a year). Be careful but don't worry excessively.

In reply to bridgstarr:

It's not quite "back to square one".

• Depending on how much naturally acquired immunity in higher transmission risk subpopulations is contributing to the effectiveness of lockdown, the current lockdown should still be suppressing SA cases - but that is suppressing them on average with local flareups.  There's a lot more local resource being used with contact tracing than before, and testing seems to be flowing in to contact tracing much faster - so the chances of keeping on top of the SA variant are much better than they would have been last March.
• Also everything Offwidth says in the post above this one.

Still, all eyes now turn to Hancock's briefing...

In reply to wintertree:

Rutland cases now doubled in a week ( to 386 cases per 100,000 on the Guardian on the rise map)

https://www.theguardian.com/world/2021/feb/07/covid-uk-coronavirus-cases-de...

In reply to Offwidth:

The large majority of new cases in Rutland are in a single MSOA and mostly in a couple of days, I think it must be the prison outbreak be Stretton that you(?) mentioned the other day.

https://www.leicestermercury.co.uk/news/local-news/prison-around-950-inmate...

Hope it is well contained. There are modest rises in some of the local surroundings so presumably not completely.

The concerns I had on Saturday night about that area of the country up through Lincolnshire and down to Corby/Kettering don't look so well founded now, most of those areas have put in better numbers the couple of days since and will be back to falling again in a day or two, albeit slowly. I think basically it's just noisy and not on a consistent 7-day pattern.

In reply to wintertree:

I dunno, if they were serious they'd lock down travel properly. If it's in the UK then it's in Holland and Germany, both nations with reasonably strong ties to SA. Then you have people travelling between SA and UK via indirect routes.  There's also the undocumented variants everywhere else around the world the UK could be letting in every day and not knowing about it until its too late.

In reply to Toerag:

I’m sure it’s everywhere and just like with Italy the first time around, we won’t know it’s a problem in an intermediate country X until that country has seeded many cases here.  We’re not going to find out it’s a problem in X until much later on that last time for many countries, because it’s masked by the old variants.

But we don’t have much travel and what we do have at least has self enforced quarantine.  It’s not great, but it’s a lot better than early March 2020.  

I’d much rather we had hard, enforced MIQ on all travel and only returning residents allowed in to that.  

I’m a bit unsettled by all the talk along the lines of “it’ll be okay, we can update the vaccine by autumn for this variant”.  What is this variant going to change in to by late summer if we let it loose or import cases from places letting it run loose?  What we should do reminds me of an old key ring dad got given at a garage in the 1980s.  “Look ahead.  Think ahead.  Drive ahead”.  We’re still not doing that far enough ahead.  Perhaps we’ll luck out this time...

In reply to wintertree:

None of our deaths in the first wave were in hospital, they were all care homes.  25 of Jersey's deaths were in care homes compared to 40 in hospital.  People 85+ years old often simply aren't worth sending to hospital because they're too weak to take invasive treatment, or have significant co-morbidities and they often would rather die in familiar surroundings with familiar people.

In reply to wintertree:

As ever, I am so grateful for all this.

Could someone please remind me what UTLA and LTLA stand for? I expect it's in here somewhere, sorry ...

In reply to wintertree:

You do have enough though, and I believe people are pretty much ignoring the UK self-isolation rules.

True.  I don't think it's going to be enough, especially as lockdown is exited - it'll be a repetition of the first wave with mass seeding events all over the country of more infectious variants that can infect enough people to cause a problem.

In reply to BusyLizzie:

I don’t think it’s ever been mentioned.  Upper and Lower tier local authorities, as used for geographic reporting of cases.  Civic subdivisions like counties and city councils etc.  

Deaths are also reported by UTLA.

MSOA is a weird reporting unit (middle super output area) used for fine grained reporting of hospitalisations.  Hospital data is organised by NHS Region (different to the English regions), NHS trust and MSOA.

Hospital data is also reported in different demographic bins to cases and deaths data.

UTLA and LTLA boundaries depend on the year of reference.

I don’t think the reporting units are well set up in terms of rapidly identifying links in real time in a pandemic.  To be fair, they were never designed for that and now is not the time to reorganise it all!  

In reply to Misha:

I’ll do a plot of exponential rates next week.  It looks like the fall is currently at the maximum exponential rate seen last time, but last time it wasn’t seen until the numbers were much lower.  So the absolute fall is faster right now.

The tail will be interesting; it may be worse this time due to the much improved survival chances from the lessons learnt.  I think for at least another month the consequences of a new variant breaking through control measures as with November could be catastrophic.

In reply to wintertree:

Yes, that’s what I was wondering - the drop is steeper in absolute terms and the curve looks steeper as well (sorry I’m being very simplistic here). So where’s the tail? As you say, no doubt there will be a tail. I guess what we’re seeing is less serious cases being discharged relatively quickly and so the tail hasn’t appeared yet. Why didn’t this happen last time though? Perhaps care has improved, with the use of dexamethasone and whatever else they’re using now. Perhaps it’s early days yet. Either way, the relatively fast drop is great to see. Still a long way to go of course but better than a slow drop...
 

At current rates, around 20k patients by the end of Feb and around 10k by the end of Mar seems feasible, perhaps a bit higher depending on how soon the tail kicks in, perhaps a bit lower if admissions drop off significantly as the impact of the vaccines makes itself felt. So my uneducated guess is around those levels. 10k was roughly the level when things started to open up in England in mid May, though of course retail and hospitality had to wait another 4-6 weeks and that was in the summer. If things are opened up from Easter, it will be the infections vs vaccinations race... Let’s hope for some nice weather to help keep things on the right side of the line.

What strikes me is that despite the looser lockdown adherence and the more infections Kent variant, cases have reduced significantly (the ONS data is less positive but what we really need is an estimate of new infections, which is missing as you’ve pointed out). Still too high for contact tracing to manage though. Let’s hope the opening up of schools from the 8th (which seems likely given where the numbers are heading) won’t upset things too much.

As others have pointed out, I’m a bit perplexed by how blaze Vam Tamm seemed about the SA variant. Not his usual approach so perhaps he really isn’t too concerned. My simplistic thinking is I get that it’s not more transmissible than the Kent variant before taking vaccines into account but what happens when a significant number of people have been vaccinated? Inevitably the SA variant will start to take over. Perhaps they’re hoping that the numbers are low enough and the vaccines are still effective to an extent, so it will take a long time for it to get out of control again and by then the vaccines will have been rejigged and readministered to the most vulnerable. Seems a bit of a gamble though... I’d like to see some modelling of this.

The real concern is what if there is something highly infectious and relatively vaccine resistant out there...

In reply to Si dH:

BBC are following us again

BBC News - Covid: HMP Stocken outbreak 'behind Rutland's surge'

https://www.bbc.co.uk/news/uk-england-leicestershire-55988076

In reply to Toerag:

Sure there is a risk here, but I think your assessment is over pessimistic. The reason I think this is that in this country we already have our cases dominated by a significantly more infectious variant than that in most places. So to come in to this country and spread significantly, a new variant from abroad has to have made a big jump in transmissibility. We are already seeing that the SA variant doesn't have that, according to modelling referenced by VanTam yesterday which is presumably not published.

Of course, later in summer enough people should have been vaccinated against the Kent and original strains that a new one that was vaccine resistant would take over anyway. But by definition that can only happen once there are relatively few people left who are susceptible to the Kent variant, by which point, with the combination of vaccination, the current lockdown and then subsequent summer, I think we can be confident that cases will be very low. What we shouldn't have is a situation in which new variant(s) are taking over from the Kent variant while infection rates are high. Once rates are low the testing system needs to be able to cope with them and find new variants faster. The record is terrible but the system does seem to be getting better.

Edit to add, given the increasing numbers of cases of thenKent variant being found to have added the E484k mutation - it's Manchester today - I suspect that domestic mutation is the greater risk, because we have a large pool of the already more transmissible variety as a starting point.

In reply to wintertree:

Thanks so much for the explanation of the acronyms! I would never have guessed MSOA.

Hope you are doing ok wintertree, and getting enough exercise to keep you sane.

In reply to wintertree:

Given the transmission advantage of the now dominant strain, and winter, and fatigue, it's impressive that we can hit the same rate. Antyone ready to posit that vaccinations are having an effect???? 

In reply to Longsufferingropeholder:

On the death rate, definitely yes as of this week, as per comments higher up the thread.  It's falling relatively faster than hospital admissions, more like cases. Previously it has always lagged admissions and followed the trend of hospital occupancy fairly closely but with more noise. I suppose it's possible there is some short term reporting issue happening but I'm 90% sure what we see is a vaccination effect unless the trend changes dramatically this week. It's further corroborated by the slower national decline in ITU occupancy than hospital occupancy - this is also a vaccination indicator because a higher proportion of ITU occupants are in un-vaccinated (upper middle age) populations than overall hospital occupancy (because the oldest don't get out on a ventilator as often).

In reply to Si dH:

I think there are things to worry about.

We have high levels of infection and poor control of borders. We are likely to import or grow our own mutants and there is likely to be a lag in spotting them. I suspect we will fold the lockdown with incomplete suppression (tthe libertarian argument will be that the heat id off healthcare).

Furthermore the combination of individuals with high levels of immunity (illness / effective vaccine) and a weakr immunity (partially effective vaccines, waning immunity - presumably part of the Manaus problem) seems an excellent environment to drive mutations.

Manaus remains a problem, though the hopes for herd immunity to emerge seem to have rather low % levels required - a bit of boosterism? (I suppose you could say India is a hope...).

On a different topic, now there are likely to be significant levels of population immunity emerging (10-15%? )  how does this affect the calculation of the governments R number...or is it not really 'R' anymore?

In reply to HardenClimber:

I agree with all this (assuming that by incomplete suppression you mean we won't go for elimination.) For the reasons given in my above post, I think domestic mutation is more likely to be the problem than imported, at least in the short term, beyond which (through summer with low case rates) we should hopefully be able to pick things up before they spread far unless we are particularly unlucky. If you mean that you think we will release restrictions with case rates still too high to allow any effective identification of clusters and variants before they spread too far by t&t- well, we will see, this is what I had always expected to happen until these important mutations occurred because I don't see middle age illness as justification for lockdown; with these variants occurring the picture has changed and the Govt stance now seems to be much more cautious. Another point VanTam made at the press conference yesterday was that they would like to get down to the number of cases at which they can sequence 100%. He isn't in charge of course.

It hasn't been R for quite a while, it's just a number they use to provide a message to the population. Their calculation isn't public afaik but they are open that it considers hospitalisation and death rates as well as cases. Best ignored for any intelligent analysis.

In reply to Topic:

I thought that JVT was a voice of calm last night, had quite a good chat also with my partner last night and the murmurings around the Hospital also echo what he has said, a couple of points to provide a bit of optimism
 

• The data for the AZ trial that everyone is panicking over was using a dosing interval of just 4 weeks apart, we know now due to the more recent trial data that spacing it out longer provides the optimal stronger anti body response. As already mentioned, the study was so small with such wide CI levels, it is basically meaningless anyway at this current time. 
   
• We are only  7 weeks away from April, which means (hopefully) warmer days and more sunlight, more people outdoors etc. It would be very very unfortunate in that short time span for this strain to become dominant with all the current restrictions in place, the R rate is believed to be below 1 still. 
  
  This should hopefully buy us enough time to get things under control, once summer is over new boosters will be available (Maybe sooner). Cases/deaths have been plummeting in SA since January if you look at the data, that is without direct competition from the Kent variant which could actually help us slow it down, the SA variant is not currently believed to be more transmissible than the Kent.
   
• Astrazenca are still confident that this will have an effect on severe disease, all it has to do is prevent hospitalisations and deaths and it is still a massive win. Trials in SA using the J&J vaccine showed that it did stop moderate-severe infection and hospitalisation, the AZ data showed a similar immune response to that vaccine, all whilst using the incorrect dosing regime.

This is all optimistic I know, and no one knows for sure, but the general mood is quite positive that things will continue to head in the right direction and we will be able to keep on top of this.

In reply to Punter_Pro:

I generally agree with your post.

This is only competition if the Kent variant produces cross immunity.  I imagine like the vaccines apparently are it will be weaker than immunity for its own strain.  Perhaps it has an advantage over vaccine induced cross immunity given the shared change on the RBD?

There seem to have been a lot of changes-for-the-better under the hood recently.  More commitment to control measures, clearer messaging, making more use of local public health infrastructure around the new variant, lots of reasons to feel positive.  The latency in dashboard data is way down and I think that corresponds to entry of cases in to test and trace; Hancock gave some numbers on test and trace yesterday that suggest it really is sorted out in terms of speed of response. 

But... we’re playing with fire and hoping a softly softly approach will work.  It might - so long as we get cases low and hold them low through to the autumn and winter.  We’ve already had a poster on here advocating for weaker control measures now because cases are dropping, and that’s a push that’s going to keep gathering momentum.  As ever I think releasing them too soon dooms us to net total worse restrictions in the medium term.

In reply to all

New study that  (unsurprisingly) finds worst affected areas correlated with ONS deprivation index.

https://www.lcp.uk.com/our-viewpoint/2021/02/an-unequal-second-wave-variati...

In reply to wintertree:

Just for fun I was trying to get my head around how I think the different strains should interact. Even without cross-immunity, I think there should be some "behavioural competition", shouldn't there? *If* a significant amount of spread is due to people/workplaces/etc that are magnets for infection, then I'm imagining that the more virulent strain is likely to hit first, and take people out of the pool via self-isolation/illness/etc (at least temporarily), before the less virulent strain can spread in that environment as much as it would have in the absence of the more virulent one.

Of course, that makes some assumptions about people, not least that they would self-isolate - or at least that the people in that environment would change their behaviour in response to an outbreak. But it seems to me that there should be some sort of effect there?

In reply to Longsufferingropeholder:

Also, about 3 million confirmed cases since 1st October should have some immunity.  My fag packet maths says:  over 20% of the country has some immunity, vast majority of these in the adult and vulnerable population. 

In reply to jonny taylor:

Also, big picture, when one strain is highly prevalent we have lockdown or other more stringent restrictions, which will mean it takes much longer for another strain to grow noticeablely from one or a few initial infections than if restrictions were lax at the time.

In reply to Misha:

My fag packet maths agree with yours: we had 17k hospital admissions in last 7 days, 21.7k week before. Currently 30k in hospital. So 20k end of Feb. But I reckon vaccinations and natural immunity and few more weeks of restrictions will (hopefully!) mean less of a tail than last spring. As long as new variants don’t go ballistic we could get to 10k in hospital mid March, probs a bit later. 
 

This weather will slow up vaccination rate, but hopefully nothing too major. 20 mill vaccinated by end Feb, well over 3 mill naturally acquired a degree of immunity.  So a huge % of over 50s and vulnerable a with some immunity.

I also the the govt knows that as we reach end Feb lockdown fatigue will be seriously creeping in so on 22nd Feb (or whenever it is) he will announce easing of restrictions and pray like mad on the vaccinations. 

In reply to jonny taylor:

The way I see it (in extremis) Is to think of them as separate diseases/pandemics.  Then, the more we shut down to control one disease the more we control the other - that seems to be the case with flu this year.  When an instance of a given disease gets in to a local, high-spread environment, there is locally increased control measures for a period of time. When an instanced of each disease makes it in to a high-spread environment is probabilistic and independent; so the more of one we have, the more the other is controlled.  

I think as well there must be some "matrix" formulation of co-immunity - with one axis being the strain that causes immunity and the other being the strain against which that immunity holds.  I'd imagine that is has a strong diagonal, and if the axes are ranked by phylogenetic distance (in some nebulous way) that the values get weaker further from that diagonal.   The values are probably a set of polynomial constants to describe the strength of immunity vs the strength of exposure and vs time since exposure....  So the chances of ever measuring it all are basically zero.  But in this noddy mental model of mine, perhaps exposure to the Kent strain is more beneficial that the old ones in terms of protection against the SA one.  Perhaps.

In reply to wintertree:

Yeah good point, it makes sense.
 

Definitely agree here, we have come too far now to just throw it all away with a premature relaxation of measures, I think for everyone's sake, this needs to be the final time we do this so we need to get it right....

I've also just come across this twitter thread which makes for further positive reading with regards to AZ and the SA variant. Apparently a ''Furry'' account originally that is actually run by a Molecular Biologist, specialising in viral pathogenesis and host immunity.

https://twitter.com/sailorrooscout/status/1358869462414467085/photo/1

 

In reply to Punter_Pro:

Woman on BBC news said she just flew in from Johannesburg and was through to arrivals in 10 minutes with zero checks for her and none apparent for others and that she is really worried checks were so lax (it took an hour to pass checks on the way in to SA). She also said on the underground link they were packed in like sardines and that seemed the most likely place in her journey to spread the virus. This is the current reality behind the government messaging. It's hardly like Heathrow are busy at present.

In reply to Punter_Pro:

I agree with the first bit. I'm not confident 'we'* will manage the second...

* by 'we', I mean the government.

In reply to Punter_Pro:

Only sensible to understand and guard against the Galloping Furry Rot.

Quantitative data on T-Cells - I could get drawn in to that, thanks for the link.  As I understand it from the resident immunologists, the T-cell responses are broader being based on shorter and un-folded portion segments meaning that point mutations break far fewer of them.  However, they come in later in the process, tagging infected cells for destruction, where as the B-cell produced antibodies can bind and tag the protein before it enters cells, and critically some can block the binding of the spike protein to a cell (and perhaps block operation of the membrane fusion protein, for naturally acquired antibodies?).  So T-cells can make a big difference to the course of an infection, but antibodies can prevent it.  Given where the SA variant has changes, it would fit that infection is no longer presented, but illness is more mild as that would tally with neutralising antibodies being rendered ineffective but T-cells largely doing their thing.  I could really get in to immunology.   

In reply to Michael Hood:

I mainly wanted to get my write up of the spatiotemporal spread of the failure of lockdown out there widely at the time in December, because it seemed to genuinely add something important to the situation and flew in the face off the various microbiologists commenting "there's no evidence this strain is more transmissive" when what they really meant was "no evidence from microbiology".  I did get that write up to some very interesting and relevant places.  As it since transpired, people within the response had similar information not from analysing case numbers but from the "SGTF" failure data from the lighthouse labs, from a kind of "pre-discovery" re-analysis of their data.

It's tempting; I've wondered if the Byline Times would be interested in running a data column.  It's a big investment of time attempting to set it up however, and it would tie me down on the delivery which brings a lot of pressure I don't really have the space for in life right now.  Also, I suck at selling myself - very much a back room operator...

What I'd be much more interested in doing is having a meeting with someone high up at PHE to talk about how they currently analyse and present data, and constructive approaches to automation and visualisation.

This has turned in to a funny sort of analysis - it's pitched more towards data and science types I think, although it is hopefully generally quite accessible; certainly doesn't get a free pass on any of the methods from the UKC audience!  Several of whom have contributed significantly through questioning and suggestions - thanks all.

In reply to BusyLizzie:

I'd never heard of an MSOA until this pandemic; sorry it's used for fine grained reporting of cases, not hospitalisations. 

I'm far too busy to have a chance to go insane thanks . I look forwards immensely to schools re-opening and not just because Jr + Jr 2 will be so happy...

I'm not getting enough exercise to keep on top of my waist line and I've gained a decade since I last accidentally got the podge; I'm not sure I'm up to casually running for 4 hours every Sunday to fix that in the way I used to.  Might have to cut out the tins of condensed milk...

In reply to wintertree:

Strange, my mental image of you is a Cassius.  

In reply to wintertree:

At some point, demographic deaths data has appeared on the government dashboard in the same age bins as cases data.  I’ve made a version of my D1 plot on demographic daily numbers and exponential rate constants using deaths data.  This plot is now split in to D1.c for cases and D1.d for deaths.  

Both plots fit exponential functions to a sliding window to measure the exponential rate constant.  The deaths data has a lot more statistical noise due to the much lower numbers, hence I fit to a larger number of days of data to reduce the noise in the measurement (at the expense of decreased locality), using a ±8 day window for cases and a ±14 day window for deaths.  The window truncates by necessity towards the edges of the datasets; this means that measurements within the most recent 8/14 days of data are likely to change as future data comes in.  A further change for the deaths plot is that the measured rate is set to zero for deaths on days where there are less than 30 deaths (total) within the ±14 day window, as when the absolute numbers are very low, the result of measuring anything from them is basically just noise that makes the plot look awful and detracts from interpretation.  This "masking" makes the right hand side plot on D1.d look a bit like it’s been cut out with scissors.

Flip-booking between D1.c and D1.d shows both just how shifted deaths are towards the highest ages vs cases, and also the lag.  The big drop in cases in November (blue vertical bar) barely appears in deaths as the valley in cases is blurred out by the dispersion statistics on the lag from cases to deaths.

I’ve tried a few different ways of visualising the relationships and have settled on a Lissajous figure of the exponential rate for cases (x-axis) vs deaths (y-axis).  This kind of xy plot is a great way of visualising phase relationships - that is two signals from a time evolving system where one signal lags behind another.  By looking at exponential rate constants and not case numbers, we “normalise” all the different age bins to the same axis ranges, making comparison much easier, by focusing on the growth/decay mechanic of the pandemic and not absolute numbers (which include further information on the history of how we got to that point and on the demographic makeup of the population).

I’ve started this plot with data from 2020-11-30, which is around when deaths started to turn positive after the failure of the November lockdown.  Here we see that things get broadly worse with age - the case rates are positive for all ages and larger for older ages meaning faster growth.  Deaths are in the last days of decay for younger ages (a long halving time and a small, -ve rate constant) and have started growing for older ages (a long and shortening doubling time, a small and growing +ve rate constant).  From here, the rate constants increase, first for cases then deaths, before decaying for cases, then deaths.  This staggered relationship between the measurements as lockdown kicks in to cases, then propagates to deaths, "walks" the curves around in an ellipse that ends in the bottom left quadrant where both cases and deaths are in decay.  It’s notable that the curves are quite tight where they cross y=0 (deaths going in to decay)

To my eye, it's notable that the oldest ages are striking out further to the left at the most recent point of this plot, having come from a tight cluster when crossing the axis.  This means cases are in the fastest decay (exponentially speaking, which reflects the disease's spreading mechanic independent to the population size at a given age) and that the rate of increase of decay is fastest in these oldest ages.  This seems like a promising sign that the immunisation program is starting to feed through in to cases; the timing is right for that given the lag of a few weeks from immunisation to significant protection; if this is the case I hope to see these oldest ages move further left to faster decay of cases, and soon enough for their deaths rates to become more negative as well.

These plots end on Feb 2ⁿᵈ, which is day I could do cases data to last Saturday.  Currently Saturdays seem to be the best point in the week to get an up-to-date view of cases that is minimal biassed by the weekend effect.  I look forwards to updating this plot to Feb 9ᵗʰ this Saturday.

Edit:  To get Plot D1.x looking reasonable for interpretation I had to heavily filter the individual curves; due to the smaller numbers the noise is a lot worse in a single age range than population wide.  I used a 21-day, 1ˢᵗ order polynomial filter.  This moderates the peak rates significantly, meaning the doubling and halving times seen in this plot are less than those seen in the regional and national plots - the peaks in those curves have been broadened and lowered by the filtering added to combat the noise.  For comparisom, I've added an unfiltered version of D1.x.

In reply to David Alcock:

If you refer to Gaius, not a bad guess before I ate all the pies.  Hopefully I'll get out running, swimming and cycling a lot once schools are back and recent trends will be reversed.

If you refer to Peter Henderson, I was too young to be a KGB mole at the time, and these days the KGB. are long gone.  Well, their name is, anyhow...

In reply to wintertree:

The ‘Weekly Influenza and COVID-19 Surveillance graphs’ from PHE are now available to look at here

https://assets.publishing.service.gov.uk/government/uploads/system/uploads/...
 
Just like last week the graphs in the ‘COVID-19 Vaccine Impact on Surveillance Indicators’ section (pages 55 to 61) seem to show vaccinations have had little impact so far with trends for older age groups mirroring those for younger age groups.

Also, following your suggestions on earlier posts I’ve done my own quick analysis on the age demographic of hospital admissions for England using the data available here 

https://coronavirus.data.gov.uk/details/healthcare?areaType=nation&area...

I’ve looked at how the ratio of hospital admissions for the two older age groups to the 18 to 64 age group has changed on a daily basis since the start of the year. I had thought vaccinations might have made a noticeable change in the last week or so but unfortunately apart from some small fluctuations the ratio has been fairly steady. Maybe just a slight decline for the +85 age group just appearing but not that convincing yet. Of course there’s no way of knowing from the published data if people going to hospital have been vaccinated or not.

Interesting article by Professor Paul Hunter on the AZ delayed second dose strategy so I’m still hopeful things will work out.

https://theconversation.com/amp/astrazeneca-vaccine-delaying-the-second-dos... 

In reply to AdJS:

I think it is still to early to be seeing the true effect of the vaccines, the data reported on the 11th of January is showing that only 2.4 million people in the priority groups had received their first dose, it is only in the past couple of weeks that the vaccine drive has really ramped up to the 12 million figure we are now seeing.

https://coronavirus.data.gov.uk/details/vaccinations

Looking at the latest Yellowcard report, up to the 31st of January nearly twice as many Pfizer jabs had been given out compared to AZ, Pfizer takes 21 days for protection to be at its strongest from the first dose, so going off that it will be another 2-3 weeks before we truly start seeing the effects, there is definitely some hints of it starting to show however.

https://www.gov.uk/government/publications/coronavirus-covid-19-vaccine-adv...
 

In reply to wintertree:

I've had to spend some time staring at it, but that's a very interesting graph. One thought - maybe more relevant for a few weeks further out, would be adding another date point, perhaps in the upper left quadrant or near one of the axis crossing points, to help give context to when the middle of the lines are?

In terms of interpretation, it will be interesting to see if that break out for the older ages continues. I'm trying to work out if it's a recent thing or actually it's just a persistently different gradient from soon after cases flipped to decay - they've been close together for a while but perhaps only in the drawn out process of crossing - younger ages have a tighter arc which is now starting to really stick out.

Curious that they cross the line more or less together w.r.t cases (I assume it's all at slightly different dates). I'm not sure I've rationalised in my head whether that's a consequence of the earlier pattern and its interaction with the age related mortality, or just a coincidence...

In reply to wintertree:

Quick (visual) look at those shows a 2-3 week lag between cases and deaths - is that about right?

The other thing those new plots really show (by you not being able to meaningfully process the data) is confirmation that deaths in the lower age bands are statistically insignificant.

Nice work 👍

Edit: has anyone confirmed a reason why morbidity is higher with age beyond there being a higher prevalence of co-morbidities. And even if that's the only reason, why do some diseases not behave similarly? (E.g. worse for the young)

An unconnected point, my wife's physio has quite a few long Covid patients with a huge range of symptoms. She feels it's almost as if long Covid finds each person's weak points and attacks that.

And another, is there any way do you think, where we can help get all government agencies to report using the same geographic and demographic boundaries? Pressure from the ONS maybe.

In reply to AJM:

Yes, I have started at it quite a lot.  I still don't think I understand everything that's going on there.

Yes, I think that would be useful and simple to do.  I'd put it in the upper left quadrant, as I was thinking about breaking out the axis crossing times for each crossing as a separate plot.  I need to then get on to quantifying the uncertainty in the exponential fits; whilst it would clutter plot D1.x massively, it would be useful on the axis crossing plot.

There are a lot of curious details emerging from the data - the tendency of an area to rush through high cases/100k when absolute numbers are low and for this to then lower, the way areas of vastly different cases/100k pre lockdown all sank at different exponential rates so that cases/100k ended up broadly similar, the apparent clumping for the axis crossing here.  It all hints to me that the complexity and behavioural and geographical interlinks to the mechanic of the pandemic going on under the covers.

In reply to AdJS:

Thanks; It's been a couple of weeks since I looked through the weekly surveillance report from PHE - it seems to have a massively expanded focus on data for pre-school, school and university ages and outbreaks.  A sign about what is being discussed in the most detail behind closed doors?

I think that a cumulative plot is a really bad way of looking for these changes as you end up looking for small changes in gradients on lines that are well separated.  It'll be interesting to see what tomorrow's update to D1.x shows - I think the best way of visualising the data to look for these effects is to look at a normalised form of growth rates, be it exponential rates as with mine or with the % week-on-week changes favoured elsewhere.

In reply to wintertree:

Interesting thread on David Spiegelhalter‘s Twitter feed that supports the idea that deaths are falling more quickly in the +85 age group.

https://mobile.twitter.com/d_spiegel/status/1359941056842891267

And a thread from Chris Giles (FT economics editor) that the fall in cases for the +80 age group is very slight compared with the under 80s.

https://mobile.twitter.com/ChrisGiles_/status/1359549725247238145
 

In reply to AdJS:

Interesting.  It's possible that for the oldest people the vaccines don't prevent infection but do reduce the severity of infection and so death.  It's so much staring at digital chicken bones right now, and where I'd caution on both those interpretations is that there have always been demographic differences in the exponential rates for cases and for deaths, and so the appearance of another small demographic difference doesn't really stand out - yes; they're in the direction we expect from vaccines but it's still needing an optimistic take on things!

That being said, Spiegelhalter's plot looks quite convincing. It'll be interesting to see how the new D1.x plot looks tomorrow.

In reply to wintertree:

Confess that to me eye, the uncertainty in your data for D1X appears to be greater than the difference you can see between age groups.

Do you not think the overall faster decline in death rate now than last Spring is a good indication of a vaccination effect? If not, what do you think are the main alternative explanations? The difference is quite big.

In reply to Si dH:

Could be, could be.  It’s a broad filter though; we’ll find out tomorrow...

I don’t think so because:

1. The rate of decay got quite aggressive too soon on deaths for the vaccines to have kicked in.  It appears to get more negative on the RHS of Plot 9.e upthread; but that’s in the highly provisional zone.
2. There’s little sign of this in the demographic deaths data up to last Friday - that may well change with this week’s data (as with the Twitter post up thread) but that’s now, not two weeks ago.

> If not, what do you think are the main alternative explanations? The difference is quite big.

I need to finish my plot of the exponential rate constant on deaths across the whole pandemic.  It did get to a similarly negative value last time but only months in when cases and hospitalisations were low.  My main thoughts:

• We don’t know how fast cases fell last time which is critical missing information to contextualise the fall in deaths
• We’re a lot better at lockdown now - more people adapted to WFH, much improved transmission control measures in shops l, supermarkets and takeaways, much more online shopping / delivery capacity for essentials including food.  
• The significant improvements to clinical care - eg Dex and priming - mean the course through hospital is very different this time around.  I don’t know enough to understand how this affects things but we’re clearly not comparing like with like.

I’d expect to see cases in the oldest ages to continue to move further left in D1.x tomorrow and maybe their deaths to move down a bit, and then to move down more in another week.  It’s all in the phasing really - although critical information not published to understand this is the distribution of times from detection to hospitalisation and deaths with respect to age; some noddy modelling suggests to me those times are shorter for older people, so the latency to deaths may be a fair bit less here.  Who knows....  The NHS and PHE know.  They have all the longitudinal data on every aspect of this...

In reply to wintertree:

The other thing I meant to mention was that your graphs last week showed deaths dropping faster than hospitalisations, whereas previously they have always lagged.

I have been assuming throughout that the vaccine will have a small if any effect on case rates at the upper end of the age band and that reduced deaths would be the first indication we get (or hospitalisations, if you had the data in enough age bands.) Obviously if cases do drop more that'd be great but I don't think we have any data to suggest we should get high efficacy in that sense for older patients, especially after one shot. Case rates per 100k population are still high in the most elderly as seen in the second chart here:

https://coronavirus.data.gov.uk/details/cases?areaType=nation&areaName=...

In reply to Si dH:

I think you did mention it?  It's certainly notable.

I wonder if I've been dunderheaded in thinking this through; you could well be right given reduced immune function.  Conflated by a lot of care home staff now being immunised reducing transmission routes in though.

I got the latest demographic data from today and re-binned it into coarser bins to reduce the statistical noise, and dropped the exponential fitting window from ±14 days to ±7 days (tapering down to just -7 days on the most recent point).

The measurements are pretty noisy in the last week but you could convince yourself the curves are separating - as with D1.X they're oddly aligned during the falling phase of the most recent wave, despite having had less coherent behaviour up to the peak.  

I might have to concede I was wrong to be looking for a shift in cases first.

In reply to Michael Hood:

It's about 2 weeks.  What I should do is measure the zero-crossing on rate plots demographically and do a plot of the lags between zero crossings.

Agreeing geographic boundaries as well would be good.  I've not seen grumbling about this elsewhere; it's one that seems like an obvious "easy win" for helping the data to flow however.  It's the sort of thing I'd expect a competent government advisor with an interest in big data and a professionally versed sidekick to have taken a keen interest in.  

As it does with society.  It makes more sense if you think of it as a vascular disease with respiratory consequences than as a respiratory disease.

In reply to wintertree:

.

I did some own analysis this morning, just taking raw total deaths data in 5 year age bins from 50 upwards and plotting it Vs date, looking at both the first and second waves. (I am not capable of all the data processing you do to smooth out the noise and generate rate constants).

Coming off the second wave it's clear deaths have been dropping fastest in the 80+ group, as also indicated by the tail end of that exponential constant plot you just put up. In fact, in the last couple of days data (6/7th Feb) the total deaths in the 80-84 and 85-89 age brackets briefly crossed over (ie for a day were lower than) the total deaths for the 75-79 age bracket, and the 90+ bracket is not far behind. The brief crossing is caused by noise but there has generally been a more significant gap between total deaths per day in the75-79 and 80-84, 85-89 age bands in the past. Deaths in the oldest age bands dropped far more slowly after the first wave, where the drop in deaths in under 80s was much more similar to what it is now. However, the difference might not all be attributable to the vaccination effect as I'm sure cases dragged on at high rates in care homes as we left the first wave peak. Still, it's definitely positive that the current fast drop in death rates is looking like it is driven by over 80s. Another piece of partial evidence supporting a claim the vaccination programme is working.

Edit here are a couple of basic graphs plotted on the same scale for first and second wave. Apologies for Excel, I will try not to pollute your threads with it again

In reply to Si dH:

I think your plots would be clearer if they were normalised by total deaths per period. i.e. showing the demographic fractions. The absolute values are hard to compare, if you are trying to look at how death rates per demographic range are changing.

In reply to captain paranoia:

That's a good idea. I'll have a look if it's easy to do later on.

In reply to Si dH:

Just had a go but with all the noise it's not very easy to see the trends. There is some reduction in fraction of total deaths attributable to the top few age brackets in the last two weeks, but it's a small effect (from just over 60% through January, to about 55% a week into February). I find the graphs above more compelling. Will be interested to see WT's update this week to see if some of the other possible indicators of vaccine effect from last week are still looking good.

In reply to Si dH:

I think it's important to separate the general decay rate of deaths and the demographic differences.  Those two plots you did really show how much faster deaths are decaying across all ages; I think this is probably largely driven by a faster reduction in cases from there being significant immunity this time around in the sub-populations that drive transmission during lockdown and from the massive improvements to transmission control measures for the people still exposed to significant risk of transmission during lockdown.  All just spitballing however.   

It looks on your second plot like the 90+ deaths were significantly higher at peak and are now narrowing the gap to the next age bin down, which suggests a faster decay.  It comes out however you look at the data in the last week. 

There's several things I'm looking forwards to revisiting with this evening's update, although the evening feels like a long time away.