UKC

Good updates this week.  Pretty much universally good news and I think as optimistic as one could reasonably have hoped.

Plots 6e,7e,8e - all looking at England

• Cases  continue to fall really well
• Hospital admissions were turning the corner last week, now they’re clearly falling
• Hospital occupancy was turning the corner last week, now it's clearly falling
• Deaths were levelling off last week, now they’re clearly falling

Plot 9e

• All measures are now in to negative exponential rates indicating decay.   Cases - which lead the other measures are now halving less than every 10 days.
• Remember that the last week of this plot is provisional and could change for the worse as more data comes in, but it looks to me like halving times are going to get shorter yet for all measures.

The elephant in the room - 🐘

• As reported on the BBC and elsewhere, the ONS Pilot Infection Survey has not showed much drop in infection levels in the last two weeks. 
• This was discussed on plotting #9 towards the end.  (Search the thread for "ONS")
• I think this comes down to the difference between "cases" data here mostly measuring new infections as a result of symptomatic testing, and the ONS random sampling  measuring "live" cases which could be old.  As an infection remains "live" for some time, there is effectively a lag where the "cases" data leads random sampling data during a decaying phase.  
• So, the ONS data should soon start to show decay.
• I believe the drop in cases because the exponential rates in plot 9 are following in sequence for hospitalisations and deaths. 

So, things are going in a good direction.  I hope that helps people struggling with the impact of the restrictions (I don't ever want to home school again...) to have some faith that it is working and to stick with it.  I also hope it means that the realities of working in healthcare right now are improving, and are going to continue to improve.

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

The 4 nations plots.

• Cases continue falling in all 4 nations. 
• Wales has the most impressing halving time of < 10 days - although that’s the most recent value where these plots are considered provisional and will change with the next week of data.
• There's a definite wobble in the Scotland cases data about a week before the leading edge of the plot; this is where it looked to be levelling off in a previous update. 

I've said before I think lockdown is more powerful when everywhere does it at once - it reduces the cross-seeding of cases compared to areas going out of sync with each other.   Whatever is happening to make this lockdown work so well, I am grateful.

• Cases - decay in all regions, and the halving times all look to be getting shorter still (better).  The best regions are heading for ~ 8 days.
• Admissions - decay in all regions and the halving times are getting shorter, lagging those of cases.  They may never get as short due to the demographic effects where cases initially fell fastest in younger adults, where hospitalisation is very rate. 
• Deaths - these are now decaying in all regions except the South West.  This is clearer on Plot 18.2 than 18.1; the rate constant curve for deaths isn't lagging admissions and cases like in the others.  The South West has stood out on these plots since the November lockdown started failing, where it's rate constant for cases jump to a "red region" like trajectory, going off the colours I assigned to regions in Plot 17. 
  • I hope this standout behaviour is getting close attention from PHE.

The rebooted English UTLA level plot 16

• UTLAs that meet a triggering condition linked to things not going well, as given in [1], are annotated.
• Over the last week almost all UTLAs have fallen out of the zone of concern.  
• Barnsley is still noted, but it has fallen since last week - and having a small number of cases it's subject to more noise than other regions.  Hopefully it will continue to fall.
• Hartlepool is annotated has being on the rise from it's post-lockdown minimum.  Looking at the provisional data on the government dashboard (which is not used for these plots), I think cases were have at least levelled off.

The data points on this plot are really interesting; their value now has very little relationship to their value at lockdown (which varied by a factor ~4x across the nation), and all UTLAs now have a broadly similar case rate.  

• Plot 16x1 shows what fraction of the lockdown level case rate each UTLA now has - it's lower on the left, which is the UTLAs with higher case levels at lockdown
• Plot 16x2 shows this ratio vs the level at lockdown - there's a clear inverse relationship, with the highest levels having dropped the most. 

All sorts of things could be going on here.  I thought about it, and my preferred theory is that cross-communication between UTLAs links their rates, and that this is more important under lockdown as essential transport continues but local spreading is slowed.  This would mean that all cases would tend towards a uniform floor - with higher regions pulling lower ones up, and vice-versa.  This echoes why I think it's important for all the nations to lock down together.  If this is a vaguely right theory, we've seen cases reach a common level across England and now they'll fall in lockstep.  There are plenty of other interpretations though!

Anyway, don't over think it Wintertree, it's a happy plot.

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

Demographic plots D1-D3

• All age ranges are in decay in London and in England. 

• The weakening of decay (less blue, more orange) in the bottom right hand side of D1 and D2 last week has not continued to develop, it looks more like a “blip”.  I had been worried that the open fraction of schools and open nurseries was starting to reverse the trend here, but that’s not continued.  Good. 

• It’s hard to know if we’re seeing an effect of the vaccine or not; the oldest years were the last to turn to decay which is factors locked in before vaccination could have had an effect; looking at D3 they now have similar decay rates to other ages; in London the oldest bracket (90+) is now the fastest exponential decay, but that’s low numbers of cases, so not great data. 

•  D3 is now much more uniform across all ages than it had been  so I think we’ll see the decay rates for hospitalisations come close to that for cases as there’s no longer “demographic masking” where cases were falling much faster in young adults that weren’t likely to be hospitalised.
   

Plot D4

• Introduced in Plotting #9
• This is not a forecast or prediction
• This takes the demographic data from Plot D1 and for each date in the X-axis range, the case numbers at that point in each age bin are projected forwards to 8th March 2021 using the exponential rate constants at that point.
• This is done individually for each age bin so that it can adapt to, e.g. the demographically dependant effects of the vaccine.
• It's asking the question "What will it look like on March 8th if the current situation continues" - where the current situation embodies case levels and their exponential behaviour.  
  • The reason it's not a prediction is that vaccination and weather should hopefully have beneficial effects on the exponential decay rates, and other factors (variants, social changes) could have detrimental effects.
  • So, what's the point?  It's a simple way of collapsing all the information in plot D1 into something we can use to see if the future is looking better or worse as time passes.  Right now it's looking better.
• Residual "weekend effect" in the cases data introduces some false structure in to this plot I think.  To be fair, compare only values from the same week day between weeks, e.g. using the vertical grid lines which are on Mondays; my preferred day.
• The last 7 data points in this are provisional, as new cases data comes in the measurements of exponential rate on those days will improve. 

Last plots for today.  These show that both acute and ITU bed occupancy is in decline nationally and in most regions. The NHS regions of Midlands and East of England are not yet seeing ITU levels in decay, but I hope that will change in the next half week.

It's all very upbeat plots today, but the best way I would describe our situation is "totally bonkers" and "pulling back from the brink".  Things are in no way normal, and a rational look at the absolute values of every measure presented is "very bad", but the direction of travel is almost universally in the right direction.  

Assuming vaccination brings the extrapolated case level for March 8th down significantly from 2,500 / day, we will have a level of cases not seen since perhaps last August, and we will have that at the onset of spring when the apparent seasonal effects of this pandemic will start tipping in to our favour.  My feeling is that with more vaccinations ahead of that, if test, trace and isolate can be turned into a highly responsive, effective process focused on a nationwide level of < 1,500 cases / day then with judicious use of responsive, local restrictions to contain super-spreader events and boots-on-the-ground contact tracers working "up" to find the spreaders and then down and out, we could be able to release national restrictions progressively throughout spring whilst keeping the exponential growth rate down and succeed where we failed last August.  My personal take is that allowing cases to rise again now we have evidence on over variants and vaccine evasion would be a false economy.  We need cases low both to prevent an immune evading variant from evolving, and because if one does evolve or get imported, our only chance of containing it is with low case rates.  If we loose control of the pandemic again, we have no chance to fight a more lethal pandemic-within-a-pandemic.

Thanks. Great news that those awkward looking local areas from the last two weeks seem to be resolving themselves.

I want to see some good news about vaccinations in the data showing that the oldest age groups have a similar exponential decay rate to younger ages despite having had lower peak cases... but doing so is probably a bit too optimistic.

Edit to say, your graphs showing hospitalisations are dropping significantly more than ITU could also show a vaccine effect, because of the difference in demographic profiles of the hospital and ITU numbers. I wouldn't be confident without another week's data.

It would be amazing if we got rates that low. I wouldn't have believed that was possible a month ago. The political pressure to open up everything too fast will be absolutely immense if that really happens though. It'll be really interesting to see how the Govt amend their approach to unlocking vs last summer.

Love this, as always. Thanks.

Tune in again next week when we find out whether we can screw our eyes up and stand close enough to pretend we can see the vaccine working...

Agreed.

Sadly, I have little faith that our government can manage to understand any of that, let alone achieve it...

Thanks for this.

Vaccinations: 488k reported today. Even if they keep this weekly average going it means about 6 million vaccinated by mid Feb, which is hitting the target. Which is bloody good!!

Any increase (probable) means surpassing the target.

Interesting point - thanks.  I was pondering how demographic the effect will be in the cases date; it's possible that for the most elderly - where there isn't much data from the trials - it reduces the severity of covid more than preventing it, so it may be that it doesn't show in cases much.  I'll dig out the hospitalisations demographic code for next week.  With carers and healthcare staff also being vaccinated, some of the highest risk people across all adult demographics are hopefully going to stop becoming numbers in these plots, and I'd expect that to have a disproportionally powerful effect which will however obscure trying to read digital tea leaves in the demographic data.

In reply to Si dH  ...

... and Capitan Paranoia:

I think things are changing, along with the departure of one advisor and apparently the rising advisory capacity  of Neil O'Brien given his involvement in https://www.covidfaq.co

I suspect the global situation could change a lot - for the worse in many places - over the next 2 months with recent and future variants becoming prominent.  The long term future for the UK depends I think strongly on how many different variations on the receptor binding domain emerge, and if that set of reachable changes is small enough to be controlled without an endless arms race between the virus and vaccines.  Right now I think we just don't know how that's going to pan out.  The severity of the potential problems ahead seems clearer to more of the important people now.  

I hope we see a big scale up of vaccine technology in the UK and beyond this year, particularly on scaling up production of the vaccines that don't use a carrier virus, and on funding doses for the developing world as this pandemic isn't over until it's over globally.  I hope there is some effort going in to future vaccines that target the other key surface portion of the virus as well to better guard against encouraging immune evading variants.

We have still got a month before any suggestion of reopening (even of schools) though, which is good - things have come down a lot in Jan.

I think we have the potential to go for elimination in conjunction with vaccines.  Not sure they'll have the guts though.

Boris has regularly said things would be better by Easter. More recently he has emphasised that decisions will be made based on data at the time and his tone has been far more cautious... but if rates are relatively very low by March I think he will want to make good on his old promises.

Unless of course he gets all excited about being the one that gets to "save Easter" and lets everything unlock all at once again (a la last summer). There's tiny bit of me that thinks he may have finally learnt but I'm not holding my breath! 

All looking so much better - keep up the good work - looks like this wave (is this 2 or 3 or 2 part B?) is on the way out. My main medium term worry is that we open up too much, too soon and go off into wave 4. Long term of course the worry is vaccine evading mutations.

Plot D4 I feel needs a bit of better explanation - IIRC there were questions about this on week 9's thread and at first glance I (like I think many others) wondered why the line didn't go up to 8th March. Here's my stab at explaining - hopefully you'll understand what I mean and then phrase it properly...

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 key point being that it's a plot of a series of daily extrapolations with the leading edge going forward one day at a time.

Regardless of that, I think even though D4 is not a prediction, it does usefully give a pretty good (qualitative) confidence that cases/day are going to be an order of magnitude lower by beginning of March - which is a good thing.

I really hope we are able to go for elimination and I really hope that becomes the plan.  The graphs that cover the whole pandemic make me gutted looking at the chance we had over the summer and the subsequent shit show. 

I'd be beyond gutted if it's f*cked again from a strong position.

Potential positives happening at the moment though so here's to hope .

I have expanded on the axis labels but it's clearly not enough, nor is my description...

Spot on - if you don't mind I'll use that for future posts.

I think the graph is inherently very confusion because it looks like a case number evolving over time, and no amount of text and annotations are going to stop the automatic/intuitive part of the viewer's brain which is highly trained in looking at value vs time and cases vs time plots - it's current form sets people with automatic but wrong pre-conceptions.

I was wondering about swapping the 'x' and 'y' axes as then it won't resemble the kind of plot it isn't.

Exactly, and it will also show if things are going the wrong way.  As we get closed to March 8th, it becomes more and more predictive as there's less and less scope for dramatic change from the current course.

You're welcome

I was thinking about this, and how else it could be tackled, and I couldn't come up with anything better. Only idea I had was to use today's cases number on the RH axis and overplot a decay from cases today (greyed out up to current date) down to meet a dotted horizontal from latest extrapolation point, crossing on the target date. That would sort of illustrate it in a more accessible way, perhaps, but I can't convince myself that's any less confusing without an accompanying paragraph.

Also that would make it look even more like what it isn't

I'm not sure if I understood you right or not.  What I think you suggest is close to the internal figure I should have made anyway as part of verifying my methods.

I've made the plot now - it shows the daily case values as circle data markers, and the exponential projections from them as lines, coloured by a rainbow map (old is blue, recent is red).   The value of these curves on 03-08 is the content of plot D4.  I think the pair of plots will make it easier to talk people through this.

Making the supplemental methods plot showed me a bug in my code for this - I was extrapolating forwards from the exponential rate constant from the x-axis date, but using the case rate from the most recent date in the data.  This means that the older values in Plot D4 above were off; the net effect of fixing this is that D4 now feels - qualitatively - like it's going to get better than it had looked.  The most recent projection is unchanged.

Folks, this is why you should always make plots of your methods in action, even if you don't normally share them...

Maybe x-axis labelling like this...

"on [insert x-axis date] if we carry on like we're doing so far, this is the number of cases we'll have on 08/03/2021"

(I wanted to use greater/less than arrows rather than square brackets but they get stripped out - presumably something in the posting is seeing them as xml/html type stuff)

That works, and basically yes. You've gone above and beyond.

To satisfy curiosity, what I had imagined was the latest of the decay plots superimposed on the D4 plot, and also a horizontal dashed line from the last point on the D4 plot running across to meet the red line and decaying line at the same point.

So, put simply, what I had in my head was brown line onto plot D4. (could have all the coloured lines but would look v busy)

Xlabel could be as simple as "if nothing changed after". The title should be enough to cover the rest.

The left hand rainbow plot is far better. Until I saw that I did not understand the right hand plot.

Many thanks for your time and effort to extract and present the info.

One problem with the intuitive understanding of the D4 plot is that the "confidence limits" of the extrapolation ought to narrow through time (I understand why it appears to do the opposite presently, based on the +/- 10% thing). I think you suggested a partial solution would be to switch to a fixed value of the exponential rate constant (ERC) in #9. However, given that an error based on the ERC, whether a percentage or a fixed rate, remains simply a made up error rather than an error estimate, wouldn't it be better to base the confidence limits on the time remaining before 8th March?

Might I also suggest a title like "Evolution of the estimate of the number of cases expected on 08/03/21 "..... with details of how it is estimated following

In reply to Michael Hood, Longsufferingropeholder, elsewhere & Wicamoi:

Lots of great feedback there people, thanks.  It's great to get feedback from outside my own head (where it all obviously makes sense) and particularly from someone who didn't get it, and then did.

The "confidence limits" - I think I have to bite the bullet, model the noise in cases data and propagate this through to a genuine errorbar on the exponential rates, and use that errrobar and the modelling derived errrobar on the cases data to propagate that through to a genuine uncertainty on the final value.  As the noise on cases is correlated, non-gaussian (well large number poissonian), this is going to involve some bootstrapping that I'll feel mildly bad about.

Also - the ±10% shading will narrow as we approach 03-08, but for now its in the "far field" - the spread of dates on the plot with data is << their mean distance to 03-08, so to a crude approximation the spread isn't changing as a function of x-axis position, only of predicted case rate.

This - nails it ✔

Sorry if I've missed the initial explanation for you D4 plot, but is there a reason you picked a fixed target date rather than a rolling "2 months into the future" (from each x datapoint)? I can imagine pros and cons to both.

In terms of modelling the uncertainties, I'm struggling to picture a logical middle ground between a naive fitting-based error (covariance matrix from scipy.optimize.curve_fit) and a full PHE-grade epidemiological/demographic/sociological model. Interested to see what you come up with. From the outside, it seems to me that it would be very hard to know which of the so many complexities "matter" the most for your error model. For all its limitations, I'd have thought a naive error bound obtained from curve_fit covariance might be quite interesting to try. 

I think I noted the reason for the date buried at the end of the last thread.  I should put it clearly here - off your geographic radar I think...  it’s the nominal date for English schools to reopen, hence fixing it in time on the plot.

Noise - all good comments, thanks.  Sticking to my guns on extrapolation vs prediction, I think the only “fair” approach is to consider the noise in the cases data.  I’m assuming it’s awful characteristics make use of the covariance data a bit dodgy.  But then so is any approach...  One to continue thinking about.

Thanks for the latest update.

Things are definitely heading in the right direction, so good to see hospital admissions and deaths falling across the board.

It's been a bit of a slow week for Vaccines but they seem to have caught up over the weekend. A total of 7.7 Million people have now had a first dose in England alone which is amazing, still a long way to go I know but the NHS are not messing about.

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

https://www.england.nhs.uk/statistics/wp-content/uploads/sites/2/2021/01/CO...

Am I reading it right? 600k first doses on Saturday?!? That's...... That's a lot

Yes, I was three minutes quicker than you

Just shy of 9 million have been given the first dose. 

Yup! Incredible Stuff.

I was quoting England's figures but yes for the whole of the UK it is 8.9 Million!

Sorry, didn’t read your post properly. Bloody impressive really (that’s vaccines, not my post reading !)

Yup.

9 million with their first jab and another 18 million/month if the best rate yet seen was sustained.  I don't think the rate has finished rising yet.  It'll be interesting to see how capacity is moved from first to second dose, and how much we end up with, but the speed of this is making me very happy.  It's feels unlikely but not impossible that the whole nation could have been offered a first dose by the end of April.  

Here's a go at a different vision of Plot D4 updated with today's data.  It shows the "rainbow plot" and joins a couple of extrapolations from that to the results plot.  The second version using a log-Y axis; this makes a kind of sense as we face ever decreasing returns in the case rate in terms of absolute numbers as the situation improves... The best possible goal of order 1 case/day doesn't look impossible on this variant, but the more I think about it the more complicated it all gets to interpret...

I like it but I'm not sure the log scale does any favours for the accessibility agenda. Will most likely add value in making it clearer as time goes on though.

Edit: will it? Should be a flattish line at 2500 from now on, right??

Re vaccines, I think it's done the rounds before but have a play with this:

https://www.omnicalculator.com/health/vaccine-queue-uk

Also their reopening calculator is a fun one...

It never does, but I've a suspicion there's more than a few scientific types reading these threads...   I think shifting the x- and y-axis ranges on the linear plot is the way to go for a wider audience.  

Good old SIR model.  Not exactly fit for purpose with viral loads, varying susceptibility, variants and all that jazz.  I still see modelling as walking a tightrope between broad-brush estimates and things that are so detailed they can never be accurately constrained.  

I think one of the more interesting things about this graph is why it has changed in the way it has. I.e. why has the (demographically calculated) exponential constant varied sufficiently through time over the last few weeks for the 08/03 extrapolation to keep moving so much, and why has it not reached an equilibrium. I actually almost commented on the fact that the first version of the graph (edit: in this thread, not week 9) had a flat line for the last few days, which I thought maybe indicated the behaviour of the pandemic in lockdown had reached some sort of stable trajectory. However, Wintertree then found an error with the graph that changed it again, so I was glad I hadn't.

Personally I don't think the scale matters too much on this graph, it can be read easily with either. If the predictions dropped another order of magnitude you'd be better off with log, but I don't think that's credible.

It's not fit for purpose at all if the purpose is prediction, but it's a good cartoon. It might just be the site with all the sliders I was after last week.

and Si dH:

Yeah, that's what I was getting at; surely we don't see it going down to 100s, great as that would be.

Sorry; I missed that!  I’m hopeful that vaccination and non linear effects in transmission mean it keeps dropping over the next few weeks.  But who knows!  (Non linear being stuff like cases getting low enough for T&T to work better)

In reply to Si dH:

Trust your first instincts when you look at a plot before you think too much about it... If you look at plots as much as me, anyhow.

Halving times are still getting shorter; they don’t seem to have reached their minimum yet, and mostly vaccination is not yet showing through in these figures.  I think we’re going to need the log plot.  Could be wrong...

Edit: Refreshing to be pondering how low something may go.

I'd place a small bet that the scale needn't go down to 1 though

If we were betting, I’d say 250 cases/day when we actually get there.  

 But taking the scale to one shows the full range of possible outcomes.

https://www.bbc.co.uk/news/health-55889391

eep!

On the positive side, at least the waiting game is over.  

That article says 5% of cases are being "tested" - I assume they mean "sequenced".  So, with only 50% of infections being detected as cases up by P1/P2 testing, and only 5% of this being sequenced, there's ballpark 80 cases of the SA various out there right now.  

Happy days.   Underlines the imperative to bring total cases down, down, down, so that the SA variant can get the attention it deserves and not be buried in the noise.

yes....get cases down.

Sorting out travel restrictions and the exceptions. (soon!)

When things are changed it should be done with a mind to epidemiology and monitoring.

eg returning to school...if we all do it in the same way and one area starts increasing up it means you look for deeper causes, rather than decide it is just opening the schools.

Boris will see the window for his snap election sliding away (the uptick he seems to havve got from the vaccination program). Presumably there is a risk of odd decisions.

(presumably it would be possible to design a screening pcr which 'fails' to detect the SA variant...a bit like the current test only comes up with 2/3 of the targets for the 'uk variant').

Looking forward to a Summer of wack-a-mole.

Yes - does look like they are pushing up testing in the affected areas - hope it’s enough although I guess the bomber always gets through....

That's hardly surprising, is it? If anyone believed that the virus, having been brought in by international travellers not required to quarantine, would not be transmitted to anyone else in the UK, they'd be daft. Oh. That will explain the belated government response.. Again. The horse has bolted and is in the next county.

Let's hope the moles are more effectively whacked this year than they were(n't) last.

It’s more surprising that it’s taken this long for community transmission to kick off.

This is about "primer" design.  

Primers are templates used by the PCR goop to match against the DNA of interest (which has previously been made from the viral RNA in a reverse transcriptase step). One of the 3 primers used by our tests no longer matches the relevant bit of RNA from a "Kent" variant - a massive stroke of lock as it allowed "pre-discovery" analysis of the spread of the variant.  

To an outsider like me, primer design is part science and part arcane, voodoo level skill and luck - some primers work better than others.  So I don't think it's guaranteed that a primer can be made that

• either matches or fails against the SA variant but has the opposite behaviour for existing ones
• and gives good false negative and false positive rates 

I imagine people with that arcane skill are hard at work on the problem right now.

In reply to Neil Williams:

Well, this year they are sending people in to the new outbreak area to do "boots-on-the-ground" contact tracing and mass testing which is a far cry from March last year when contact tracing was abandoned due to lack of capacity.  That's in the "pro" column.  In the "con" is that unlike last year, every sample is going to need sequencing and I presume some amount of expert time starting at sequencing results to identify the new variant, and it's a currently tiny number of cases against the background of the old pandemic.

In reply to Dr.S at work:

It took a while for community detection to kick off enough to be detected the first time around, with the first importation events apparently being identified and contained.

Still, today's update to the dashboard shows everything continuing to fall, except for deaths in the South West.

They probably aren't trying to sequence every case, just making sure that they catch as many cases as possible. Logic will be something like if you have covid and you live in an area that has a lot of the SA variant, we want to make damn sure you get picked up and even surer you isolate.

It's clearly out there already, so now it's about buying a few more doubling times while they're cheap.

On the news on the radio tonight, Hancock said every case is being sequenced, although over what areas it wasn’t so clear.

They did say today that they want to eradicate it, that they wanted everyone in the affected postcodes to get a PCR test asap and that they would be sequencing every positive test from those postcodes. How realistic that is in useful timescales I don't know. Apparently one of the SA cases identified today was originally tested in December.

On the positive side, although there must be lots more cases out there that haven't been sequenced, it did strike me that only 10-15 of the just over 100 sequenced SA variant cases announced today were ones without known links to SA. Or in other words, 90% of known SA variant cases are still people with known links to travellers. Which suggests it hasn't spread that far yet.

Yep, just heard that and came back here to correct myself.

They’re not mucking about either.  I wonder if there’s data to tie up MSOAs and postcodes?  It should be possible to see if the mass PCR testing causes a spike in covered MSOAs; if so that informs how much regionally targeted and staggered mass PCR testing could help drive infection down faster for the old variants.

Notable that they’re using PCR not LFT for the mass tests...  Why do we suppose that is?....

I hope we find out what fraction of people engage with the mass testing.

MSOA-->postcode doesn't really work, at least in many areas. My postcode is the majority postcode for 3 MSOAs...none of which is mine, in which most people are a different postcode!

It might correlate better in some areas though. PR9 (northern Southport, one of those with identified cases) seems to correlate to about 4 whole MSOAs and half of two others.

Similar thoughts.

I was hopefully assuming (naively assuming??) that it's to do with being easier to jump from a positive PCR to sequencing...

TBH they should be doing what they're doing while at the same time throwing LFTs out from a moving van like candy at a pantomime.

Looking at the spread of locations they're interested in, I think it's too late, the cat's already out of the bag. Hopefully they're correct that the SA variant isn't more lethal and that the vaccine still protects - at least to the extent of stopping people becoming seriously ill and/or dying.

It does however look like there's a possibility that the government (they still haven't earned the capital G yet) has finally learnt that every day you delay a variant with increased transmittability in the early stages of exponential increase gives you several days breathing space to marshal defences for when the numbers get big.

Also, it's another race against vaccination - similar to the pre-xmas breakout from Kent, only this time there's a chance of winning that race; i.e. having done enough vaccinations to greatly suppress the numbers needing hospitalisation.

I think it probably suggests they haven't been picked up yet, or identified as the SA variant.

Shame they allowed so many cases to get into the country...

Have we got quarantining in place yet, or are they still 'thinking about it'...?

Not really. Obviously as I said there will be many other cases if they have only sequenced 5% of recent cases. However, if the % of sequenced cases with this variant are people with a traceable link to South African travel, as is seemingly the case, then you would expect that% to extrapolate out to all the other cases you haven't found as well, wouldn't you? Which then does imply that it has not yet spread very far. Equally, it also implies our restrictions on international travel have been a catastrophic failure, at least during the period in which these infections occurred.

As for geographical spread of these cases, I think the implication is that these are mostly or all from separate entry points to the country. Or maybe a shared plane.

I expect we'll find out how serious the spread is fairly soon anyway.

Edit to say: the figures are that since 18 December about 5% of all cases in the UK have been sequenced. 94 cases have been found with a known link to travel to SA and 11 cases have been found without such a link. The total of 105 cases, even scaled up by a factor of twenty due to only having 5% sequencing coverage, would be approximately 2000 cases nationally in that period, of which the vast majority would be linked to travel. 

Since 18/12 there been approx 1.75 million cases recorded in the UK, so 2000 is ~ 0.1% of cases in that period.

Thanks for the info on postcodes and MSOAs - as disparaging as it is for further analysis.  I hope we look at our reporting systems at some point and ask how they can be made more aligned in the future.

In reply to Longsufferingropeholder:

That's a good point.  The Oxford Nanopore stuff integrates it all in to one workflow I believe, for example.  The higher sensitivity is also important though.

In reply to Captain Paranoia:

This is my view- the lack of proper managed isolation and quarantine on inbound travellers from most countries just puts this strain a couple of extra hops away.  One way or another, we'll know soon enough and I'm very much hoping that we can box it up, but 11 community cases without an identified link to travellers, and ~ 100 cases with links to travellers is not comforting.  There's probably as many cases again out there not yet found.  There's also a selection effect that someone who is travelling internationally right now, and who has caught Covid whilst doing so, and who has transmitted it to others after their return is probably not someone who is going out of their way not to transmit the virus, nor are the people they're giving it to.

Edit: In reply to Si dH:

I'm hoping the enhanced contact tracing means it's much closer to 200 cases than 2,000...

1)The false negative rate for Lateral Flow Devices / antigen detection systems is such that it would render the screening almost pointless (from a contact tracing point of view) in this context. The goal is to break transmission.

2) To move to sequencing you either need an original specimen (ie a positive screeing pcr and return to the original) or depending on the state of development (not yet for covid I suspect) and where the mutaion lies probe the product of the pcr or do limited sequencing. Thus starting with a sample suitable for molecular analysis speeds things up and reduces likelihood of organisational 'dropouts'.

If it really has been out of the box since 18/12 it isn't good looking... (and we haven't sorted out travel yet).

This is the sort of thing the nanopore sequencing could be really useful for.

Quite.  If the variant is spreading in the UK, it’s probably spreading in every country that we have travel links with that hasn’t hard hard MIQ on the borders.  As with March 2019, we can close links to the origin of the virus but it just comes in via somewhere else, and enough cases can come in before that somewhere else is overwhelmed.  Most countries we exchange people with have less sequencing going on than us.

Its notable that recent “leaks” from MIQ in Australia and NZ have involved the SA variant.

We need to shut travel down except for absolutely essential purposes and repatriation of residents and nationals, and all return needs to be via MIQ, no exceptions.  If not, we don’t stand a chance of keeping this variant out; but this variant isn’t the biggest threat as it still has some cross immunity.  It’s what it evolves in to next. 

I think for the next few months we should go as far as switching accompanied freight to unaccompanied freight by shunting on to and off of the ferries.  Yes, it’s slower and more expensive, but it’s a damned sight cheaper than an additional 3 months of lockdown, another 50,000 dead and a reboot of the vaccination program.

I don’t have high hopes of seeing this happen, so I just have to hope I’m proved resounding wrong in my fears over the coming months.  Perhaps luck is finally on our side.

Such a phenomenally cool machine.  A digital tape reader for DNA. At some point I’ll have enough justification for one in the lab...

Israel closed Ben Gurion (main airport) a week or so ago and is keeping it closed until at least mid-Feb. That's the way to close borders.

Yup.  Shut it down.  Shut it all down.  

Israel is in an educational position for the rest of us; storming ahead with vaccine roll out but allegedly not for those in the occupied territories.

The pandemic isn’t over until it’s over globally.  Having two large populations in close proximity - one vaccinated, one unvaccinated - gives the optimal conditions to bread then spread a vaccine resistant strain.

I watch with interest.

Not sure how I haven't seen this before:

https://www.cogconsortium.uk/wp-content/uploads/2021/01/Report-2_COG-UK_SAR...

It's it just me or have we not linked to that yet?

Thanks, I don't think so (I've seen their previous reports that were entirely focused on B.1.1.7 but that one has a lot more information.)

Via one of the links in the pdf you also get to a website showing maps and graphs of the growth in number of cases, although obviously the data is not very refined and will be very incomplete:

https://cov-lineages.org/global_report.html

The 'data' page on their site (where I found that pdf) has loads of other good stuff. It's made it onto my obsessive refresh list.

Edit: here https://www.cogconsortium.uk/data/

I thought it timely to remind people what happened in Manaus recently as Brazil have no real pandemic control and this is what 'let it rip' looks like. More scarily Manaus was said to have reached herd immunity in the summer after absolute carnage in the spring...... yet this second chaotic outbreak happened and the hospital system collapsed again. 

https://www.theguardian.com/world/2021/jan/24/brazil-covid-coronavirus-deat...

The Guardian Science Weekly blog looks at the concerning implications of all this. Herd immunity may simply prove impossible as the virus mutates.

https://www.theguardian.com/science/audio/2021/feb/02/covid-19-what-can-we-...

Thanks, I assume the article was based on this report.

https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(21)00183-5/...

Wintertree's graphs are currently much better viewing than the ones in it.

Yes, it’s not looking happy in Brazil.

Given today’s news on another mutation seen in the Kent strain that’s looking convergent to the SA one, cases can’t drop fast enough.  We need the clarity to be able to go hard for elimination on these variants. 

I am very happy that we are trying to jump on them, despite the real possibility that we just delay the inevitable rather than stop it.  This is a big change from this time last year, and will improve the outcome for everyone.

Thanks for doing these, interesting as ever and good to be the right side of the curve again.

Plot 17, I can't quite figure out the Y axis. Is it absolute distance from the black line indicates case numbers, the length of the black bar underscoring the Y-axis legend indicates the distance from the axis line for 100k cases?

jk

I think I saw a headline where they've agreed to give some vaccine to Palestine, didn't read the article though.

^^This. If not, then it'll be an endless sequence of mutations in one part of the world defeating the previously immune in another part, or permanent hard borders.  The crazy thing is that the virus can be defeated, yet governments are choosing not to .

Yes, it will be interesting. For me the most interesting thing to watch will be the mutation / vaccine arms race - how long before the UK's great vaccination scheme is rendered obsolete? It looks like the other great hope isn't so great either:- https://www.theguardian.com/world/2021/feb/02/monoclonal-antibodies-great-h...

The length of the black line on plot 17 shows the y-axis space covered by 10,000 cases/day. 

The main factors are prevention of hospital overload during outbreaks and keeping unnecessary deaths to a minimum alongside economic damage.  The speed of vaccine production offers hope as does the way some countries have responded whilst retaining normal life most of the time: using border controls, swift TTI action and strict lockdowns as required. Still our UK government regard saving the economy as a balance against response... it's a completely false dichotomy: economic damage anti-correlates with tightness of state prevention measures.

Do all governments have a choice? 

Perhaps I'm being pessimistic, but the idea that globally we can lock things down tight enough to get R<1 for long enough for it to go away feels fanciful.

Some places, sure - demonstrably with enough willpower to create and maintain hard borders, ideally with a bit of geographic advantage, some places have, and I'm sure that if the will was there many more places could, especially those with decent healthcare, reasonably strong civil societies (i.e. a state that can enforce rule of law fairly consistently, particularly in terms of control of borders and ability to enforce lockdown measures) and some form of social security net to protect people whilst the lockdown is in place.

But there are a lot of places that have one or more of poor healthcare, weak civil societies with porus borders, and lack of a social security net or a large informal sector that can't access it. There are plenty of poor states, there are plenty of failed states and there are plenty of states that are just "small" (in terms of a small government with limited levers to pull to control the economy, not physical size).

With enough will, I'm sure you could create a covid free British Isles (we have obvious geographic advantages!). Given some of the previous porus borders to migration in eastern Europe, could you create a "Fortress Europe" or similar? Perhaps. Could you wipe out covid in Syria, or Yemen, or any of the parts of Africa where ex-colonial ruler-straight borders mean little on the ground and government writ runs weak (Mali? Congo? Libya?)? Of that I am doubtful. And as you say, it's not over til it's over everywhere.

I feel we're going to win via a mix of harder borders and the application of massive amounts of science rather than by global elimination. But obviously I'd love to be wrong!

If only Darwin and Gulick could be here to see this.....

https://www.bbc.co.uk/news/health-55900625

Looking in to this on unrelated project I just found out that scipy.optimize.curve_fit has gained a "bootstrap" option, and I have a noise model for the cases data...  You can tell where this is going...  It also puts the amount of work required to get a somewhat acceptable output back in the "hobby project" range...  (Edit: Because I think to justify using the covariance with the non-gaussian noise would need some hard sums (hard for me, anyhow...) to justify it and to scale it accordingly). 

https://stackoverflow.com/questions/14581358/getting-standard-errors-on-fit...

The good news is that the noise is going to start getting a lot worse - proportionally speaking - with the way the cases continue to plummet.

Nice - thanks. I too will definitely have to look into this for unrelated projects!

A mid-week update to the hospital occupancy plots.

It's come a long way from 4 days ago but we're still at about 1.65x the peak hospital occupancy from the first wave, and ITUs are perhaps ten days away from getting down to their peak value from the first wave.

All regions have now finished "turning the corner" and both occupancy measures are in decay in all regions.

Does the type of test used matter for working out the trend based on case numbers?

I noticed that round about the 18th of January the ratio of LFT to PCR tests increased quite suddenly, and over the last few days LFT accounts for between a third and half of all tests.

I'm heartened by the national trend you describe, but don't see the same pattern in my own LA. I know it could be an outlier. But hospital occupancy in our local trusts has been steadily increasing week on week over the last four weeks (as of last Thursday's NHS weekly stats report) whereas our 7-day rolling average of cases hasn't changed for the last three weeks, and the peak of our second wave of cases on the 6 Jan wasn't much higher than it is now. I'm puzzled, and sure I'm missing something: the increase in mass testing is my only theory so far, and I'd very much like that theory to be faulty!

Si dH will hopefully be along with more accurate commentary than me shortly - I think they've read a lot more about the LFTs and their place in the system.

As I understand it, a +ve LFT result is confirmed by sending another sample from the person for RT-qPCR testing under Pillar 2, and if that P2 test comes back positive, it is reported in with all the other cases from P1+P2 in the daily number.   So, if an area has mass LFT testing going on, more symptom free infected people will go in to P2 than normal boosting case numbers.

Are you looking at upper tier or lower tier LA data?  If its upper tier, I'm guessing you're somewhere on the right hand side of plot 16 (link below).  The UTLAs with the lowest case rates at the start of lockdown have seen a disproportionately small drop in cases (even exponentially speaking).  I'd hoped someone would have some observations to explain what I pulled out of this plot, but so far no...  My hope is that there was some "anchoring" effect to higher prevalence areas, and now that they've all equalised, they'll all drop together including the ones on the RHS.  This is somewhat wooly headed thinking mind you!

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

It's important to keep in mind (as you clearly do) that the relationship between infections and detected cases is flexible; hospital admissions are much more of a touchstone of truth - although with a significant lag.   You can use the dashboard to get daily updates at the individual NHS Trust level by going in to "Healthcare" and then clicking on "United Kingdom" at the top.

FWIW (not much) I think that makes sense. If there's supply of infectious people around it's clear there's going to come a point where imports and their associated seeding events will prop up case numbers. Sticking a pin in when that should happen numerically and deciding whether that's realistic for those areas for the numbers now..... Can't help with that.

In areas with a lot of lateral flow tests taking place, it certainly makes a difference. In Liverpool the number of positive LFT tests has generally been 20-30% of the number of total positive cases since the mass testing started. Liverpool council publish data very clearly on their own website showing the number of positive cases from each type of test, which is how I extract this information, but most councils don't.

Yes.

Firstly it's worth saying that not all individual trust are showing a decline in occupancy yet. Some of those that had an infection peak in early January and haven't submitted a very recent data update are still showing occupancy at the peak (Liverpool is a case in point again.) However the admissions data should be showing a fall by now - then occupancy can only follow. If not, see Wintertree's post about relatively low case areas.

Re LFT contribution to the totals, it would be worth checking the council website where you live because the government dashboard data is confusing on LFT data. If that's unhelpful than you should be able to work it out from the government dashboard, but it's not very clear unfortunately.

- The 'cases' page has a table at the bottom which shows total positive cases from both types of test (it is explicit about this) over the last seven days. As Wintertree said, it's certainly true that for the purposes of mass testing throughout the Liverpool region, all positive LFT cases are prompted to get a confirmatory PCR test. However I don't know for certain that this is the policy for all applications of LFT in all regions.

- The 'testing'  page on the dashboard at regional or local authority level has a graph showing both the total number of PCR positive cases and the PCR positivity rate over the last seven days. From the description, I think these are calculated in such a way that you should just be able to multiply the two in order to back out the number of positive PCR cases over the last 7 days (and hence calculate how many of the total seen above are from LFT.) 

So in theory doing this you can work out how many of the cases reported for your area are from LFT. However, (1) you don't know for certain whether these same cases are also captured in the PCR numbers, and (2) when I did the calculation quickly just now for a couple of sample areas, the results didn't make sense, so I'm left wondering whether the data is actually what the description implies. Maybe I messed up the calc as I only just had my coffee.

If you give up on all the above you could infer something about the significance of LFT in your region from the Liverpool data:

https://liverpool.gov.uk/communities-and-safety/emergency-planning/coronavi...

For example from the gov.uk dashboard last week there were an average of around 3-5000 LFT tests per day conducted, let's call it roughly 30000 per week. The population is about 500,000 so that's about 6% of the population being tested in a week (although a reasonable proportion might be repeated tests for the same people.)  From the Liverpool council website we know that currently 19% of positive cases are found using LFT and therefore you could assume (ignoring anyone who would have got a PCR test if LFT weren't available, I'm sure they exist) that these 19% of cases are "extra" cases that would not have been found before mass testing began. If you looked at the number of LFT tests (of all outcomes) on the gov.uk dashboard in your local authority as a percentage of the local population, then you could extrapolate a very rough estimate from the Liverpool data of how many of the local cases LFT is likely to be finding. It's very rough but in principle it should be independent of local infection rate as that affects both PCR and LFT positivity.

Not a great answer but hopefully has some useful pointers.

​​​Edit, sorry for poor writing, I was on my tablet while trying to make my son eat his breakfast. Failed miserably.

Unless I am missing it (possible!) the tracks of data by region aren't figuring much this time in the media briefing? I guess the data is still there, wonder why they don't seem to want to signpost it this time?

Another thing that bends the glass a little is that a reasonable chunk of those people taking LFTs will be the (now mostly vaccinated) NHS/care staff who, for the most part, each provide 2 negative ones every week. Not sure whether they're counted in the local figures or just the national ones, and I don't see how that could affect your thought process, but in some of the directions one could go next it might.

Aberdeen study looking at international ravel influence on covid death rates.

https://www.abdn.ac.uk/news/14653/

Whilst we might all say "that's rather obvious" it's good to have it backed up by some research.

However, the main point is, will it be listened to by our "great and good" (sarcasm alert) leaders when the successor to Covid rears its head sometime in the future?

Regarding LFTs.  That date might correspond to workplaces using them. My wife’s school - all full time staff twice a week, into their second or third week. I was trained the other week, and the email invite to the training had LOTS of other organisations invited. 

The "Testing" data on the dashboard shows that PCR testing rates are higher now than when cases peaked a few weeks ago.  If it was purely driven by symptomatic demand we might expect it to have fallen.  As well as the LFT mass testing in some areas, as mick notes some employers are now using them - my local university is, for example.  

If all positive LFTs lead to a PCR test, this could be funnelling quite a few more people in to cases meaning that the gap between infections and cases is closing.

Likewise, all the people in the hotspot areas for the SA variant are getting PCR testing - this will close the gap in those areas as well.

Lots of speculation and inference in there but it's good to see the testing capacity being used. , and it it's right, infection is falling even faster than cases suggest which given what cases are doing - at least at a regional level and above - is great.

Simplicity I think - the prefer a simple message and reinforcement.  Who knows though...

How's D4 looking now? Might (only just...) need that log scale after all.

Dominated by the unresolved weekend effect right now, but I’ve got highly hopes for the Saturday update that it will still be on a decreasing trend.

I think the top right corner of D1 is looking a lot bluer, and I’ve revised the form of D3 to show change as well as value.  The weekend effect won’t resolve in the demographic data until Saturday though.

Not yet - but it will be straightforward to derive the impact of SA variant testing on case rates in local areas from the dashboard map in a week or two. Should be interesting.

469k first vaccinations just reported. The last four days have nearly 100k/day more than equivalent days last week. Incredible. 

Yup.  There's a defiantly weekly pattern emerging to the numbers isn't there...  High hopes for this Saturday... 

Also a record number of RT-qPCR tests done yesterday, circa 802,000 / day.  The data in the provisional window makes it look like the exponential rate of decreases might be slacking off a little bit this week, i.e. things are falling a bit more slowly (exponentially speaking).  I'm hoping these aren't unrelated, as discussed up thread. 

Not that incredible. We can't afford to have large numbers of unvaccinated vulnerable people, with higher risk factors, living in substandard accomodation, where social distancing is difficult, if we wish to avoid future hospital pressure and many more needless deaths. That this problem was coming was obvious months ago. What's been done to coordinate an appropriate response  (all we have is a few excellent locally driven initiatives).

https://inews.co.uk/news/uk/ethnic-minorities-far-more-hesitant-coronavirus...

Certainly good news about the number of vaccinations being done but if you have a look at pages 55 to 61 on ‘COVID-19 Vaccine Impact on Surveillance Indicators’ in the latest PHE report there does not seem to be much evidence yet that UK vaccinations are having much effect. Perhaps it’s too early to say yet, or I’m missing something, but the fall in COVID case rates/hospitalisations doesn’t seem to be any greater in the age groups already vaccinated than in the general population as a result of the lockdown. 

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

Don’t loose the faith:

• The kind of change you’re looking for is hard to spot in cumulative plots until it builds for a few weeks
• The vaccine takes a couple of weeks to generate significant immunity.  It takes a few weeks from infection to hospitalisation.  Chain those together and it takes 4-6 weeks from reaching mass immunisation to a clear reduction in hospitalisations.

The first place to look for an effect is in demographic cases data.  I’ll be updating my take on that this Saturday.  It’s about time the effect became visible - but vaccines are also being given to careers and health workers of all ages; these people are a likely conductive bridge between the elderly and other demographics and vaccinating them is going to have a disproportionate effect across all ages.  So I don’t know how much it’s fair to expect to see in the demographic data; as longs as everything keeps going down fast I’ll be happy.

I’m not clever enough to understand that report, but I am clever enough to know that it takes a while to develop enough immunity after being vaccinated and up to the 10th jan only 2.4 million people had to their first dose. So I agree with you: it’s too early. 
 

But it’s never to early for a Gangnam Style !!

Interesting study here:

https://github.com/hrossman/Patterns-of-covid-19-pandemic-dynamics-followin...

on what’s happened in Israel.

I would be worried if there aren’t similar UK studies published in the next week or two showing the effect of the UK vaccination effort. 

Where do you find this sort of info?  Not something most folk would stumble across. 

All via Twitter! Seems to be more up to date than the mainstream media.

I don’t use Twitter, or anything else, but I’m getting more and more frustrated by mainstream media * so might have to give it a bash. 
 

* the beeb has just informed the viewers that there is no gravity on the moon, ironically during an article about how far a golf ball traveled......