UKC

The government dashboard often makes retrospective changes to the data, listed here [1].  Quite a significant one this week, removing 850 cases from the North East and Yorkshire between 23rd and 25th March.  I don't know how much of a difference it's made as I haven't been looking at the data much this week.  Still, 850 is quite a big number for a region or two given where we are now.

As with the last few weeks, we have a masking effect on the English data from the sharp uptick in asymptomatic LFD testing associated with schools returning.  This data is only separated out at the national level and then only for England - not in demographic or regional data; so I start with a quick look at what insight we can get.  Then it's just inference in interpretation in later posts.

Plot x0 shows cases in England from symptomatic (PCR) and all (PCR + LFDs - u:unconfirmed, c:PCR confirmed).  There is significantly weekly jitter in the data from day-of-week or weekend effects.  I process a lot of this out, but it doesn't all go away.  PCR cases are actually still falling, but the fall is small compared to even the residual weekly jitter making it hard to see in the top of plot x0.  It's visible in that the characteristic time for PCR cases remain a halving time meaning cases are still falling.  For the LFD included data, cases are hovering right on the edge of growth.

The consistent fall is more visible in plot x1.  This is a bit of a technical spod plot for peeking at daily numbers.  The scatter plot on the bottom right shows the fraction by which cases fall over a 7-day period for every day.  The days for which the fraction is calculated are shown by the x-axis - further right is more recent.  The key think is that the day-of-week sampling jitter seems quite consistent and to not translate in to a decay fraction between the same days of two separate weeks.   

So, it seems that PCR cases, indicative mostly of symptomatic infections, are still falling - but we won't always see falling cases in future plots as the LFD data is lumped in.

Linked to last week's thread - https://www.ukclimbing.com/forums/off_belay/friday_night_covid_plotting_17-... 

[1] - https://coronavirus.data.gov.uk/details/whats-new#removal_of_cases_incorrec...

The four nations cases plots and characteristic times.  Very little change from last week - all are muddling on with cases about level (a most unnatural circumstance for an exponential mechanic) and any changes from that are likely within the noise.

I'm assuming similar LFD effects are going on in the other nations, but it's all anecdotal.  One poster reported PCR being used in Scotland for random testing within their employment sector which muddies the waters further.  

Still, in the absence of anything clearly growing exponentially, and with the vaccine roll out continuing, things aren't getting worse - and in England at least we know they're still getting better in terms of detected symptomatic infections.

I don't normally include these in the thread - the deaths plots for each nation.  We're pretty much at a watershed moment now as most nations head for 0/day.  Wales and Northern Ireland have both had days of 0 deaths/day, and Scotland is close to having its first.

Deaths in England look to be levelling off away from 0 death/day; I'll talk a bit more on that in the next post.

The plots for England. 

The fall in both admissions and deaths looks to be almost plateauing off; perhaps not surprising given that PCR cases are close to plateauing over the last 3 weeks in Plot x0.  This is a very recent effect over the last few days so won't show much in plot 9 yet which fits over a ±8 day window to mitigate noise. By next week we'll see if these signals to track the near levelling off of the PCR cases signal.

This may or may not be happening in the other home nations, but with their much smaller populations and deaths, the numbers are so low that it's not really possible to tell for the noise on them.  

So, the big question is why did the PCR cases stop falling a lot 3 weeks ago?  There doesn’t seem to be a strong geographic signal to explain this failure of lockdown to keep R<1, and unlike the proceeding “cold shoulder” there’s no obvious correlation with the weather.   The failure of decay set in noticeably before the return of schools.  Perhaps this is the much discussed gradual lack of commitment to lockdown emerging - schools reopening is certainly a big mindset shift.  The return of many more parents to the workplace probably contributes more, as well as some spread associated with schools.

The other big question is, does it matter that PCR cases are no longer falling fast?  

• I am very uncomfortable to have R close to 1.  Every time this has been allowed to persist in the past, it has resulting in exponential growth in hospitalisations until we are forced in to lockdown. 
• But, now we have the vaccine.  It’s wound back the death rate much faster than we could have expected without it and is likely winding back hospitalisations significantly.  It’s apparently moderating case growth in the oldest people, and should be even more effective at that in younger people.  The first dose is now in a lot of working aged adults and is starting to go in to the older parents/carers of secondary aged school children.  It’s preventing hospitalisation and deaths.  Healthcare is gradually returning to normal.
• So, unless hospitalisations start to rise again, at a societal level, "does it matter" revolves around some intangibles over the risk of having a rising prevalence of cases.  
  • The risks to my view are based around variants and the selective pressure applied to those when they butt up against a partially immunised population. 
  • The benefits of recent changes are clear - it’s been awful for children (and no cake walk for many parents/carers) to have schools closed.  It would be intolerable if the alternatives were not so much worse. 

Now, it all changes again.  Schools start closing for the Easter vacation and the "stay at home" order becomes "stay local" on Monday.  Perhaps having people able to get out and about outside more freely will lead to fewer people meeting cross-household indoors and will help.  I fear for the countryside however; it's a shame that there's not been a well organised effort to prepare for the opening of the floodgates - there could be plenty of work involved in stewarding parking to preserve fragile local relations in beauty spots etc, and to help more people enjoy the countryside we have with less stress.  Anyhow, I digress from the matter at hand. We might walk in to Bollihope next week if the ice cream van is there.  Sighting reports are welcomed so we don't make the trip in vein.

The English regional plots.

Plots 18.1, 18.2

• Last week’s plot 18 was correct, but looking back it wasn’t as up-to-date as it could have been, so there’s more like two weeks of updates being revealed today.  Doh.
• Some regions have been flirting with growth in cases but this is due in large part to the additional LFD data coming on line in addition to the PCR data, and does not necessarily reflect an underlying stagnation/growth in infection - or at least not as much as these plots imply.
• Yorkshire and the humber is the only region to have had a clear growth signal.  I can't say how much is in the symptomatic PCR and how much is down to the LFD testing but it seems reasonable to assume that without the LFDs, it would still be in decay for cases.
• All regions remain in decay for hospitalisations and deaths.
• Hospitalisations perhaps are starting to see faster halving times - more effects of the vaccination program?
• The deaths data was becoming very noisy - as the absolute numbers become low, the exponential fitting was basically measuring noise.  I’ve expanded the window for exponential fitting to ±11 days as noted in the titles.  The means the peaks and troughs are wider and lower (more blurred out by the filtering effect of delocalised measurements). 
• Deaths are still falling in all regions, generally with fast having times.  These halving times are all likely to tend towards infinity soon (the point between growth and decay) - equivalent to an exponential rate constant of zero.  That doesn’t in present circumstances  indicate that the decay in deaths is slacking off, more that as deaths reach zero, there is no exponential growth and so the rate constant is zero and the characteristic time is infinity - balanced between growth and decay.  At that point the regional deaths plot gets closed out, and hopefully hospitalisations soon after.

If you zoom out or sit really far away from plot 18 and squint a bit, you can see a single diagonal line running between the orange and purple as cases tip over to decay first in the red regions of cases, then the lower blue regions, and then likewise in hospitalisations and deaths.  

Plots 22, 22r 

• Hospitals continue to empty out of Covid patients in all regions and in both measures.  I hope to the gods that we do't go around this loop again.

Plots D1.c and D3 are critical ones for interpreting what's going on with PCR and LFD cases.  Rate constants remain negative at most ages but positive for the three age buckets in the range 5-20.  Plot D3 shows the average rate constants for the last week, and their change from the same period two weeks prior to that.  Rate constants are still positive (cases in growth) and rising (faster exponential growth) for 5-10 and 10-15.  The age range 15-20 is growing at a roughly constant rate.  It seems fair to ascribe most of this to asymptomatic LFD testing.  It's notable that the LFD positives and positivity has risen over the last two weeks suggesting that there is rising cases in thee ages and therefore likely school associated transmission.  

All adult ages are in decay (negative rate constants), but for those under 80 yeas of age, the rate constants continue to head towards y=0 and then growth.  Si dH noted on plotting #17 that many parents are now doing LFD tests and so we may have some symptomatic LFD positives here masking the behaviour of symptomatic PCR cases.  This makes it really hard to interpret IMO.  Still, as long as total cases are falling, it could be a lot worse.  

It's notable that the oldest ages see their rate constants getting more negative - their cases are still halving even faster than last week.  If this is an effect of the vaccine as immunity goes on to develop, let's hope to see this effect spread leftwards on this plot. It's notable that the slowest halving times - those closest to growth - are the younger ranges.

The D5 plots show deaths in demographic bins.  I've expanded the youngest one significant.  The log plot also normalises all data to the maximum value of the smoothed curves.  

• Deaths for those ages 70+ continue to plummet.
• Deaths for those under 70 look to have plateaued - it's hard to tell how much is noise over a short time scale.
• As the deaths in older ages - which previously had by far the largest numbers - continue to fall, they will become lower than the deaths in the younger ages, and the younger ages will go on to dominate the national level curves for deaths and their characteristic times.  Likewise for cases.
  • I think we're seeing the near-plateau in PCR cases for England translate in to a plateau in the death rate for ages where vaccination hasn't really started to feed through in to the deaths data given the various lags involved.  
  • The next logical step is that if PCR cases start to rise, hospital admissions and deaths are going to start rising, too.  Right now, with where the vaccination is, there are likely enough unvaccinated people susceptible to hospitalisation to break healthcare.
  • So, alarm bells are going off in the Wintertree bunker.  As long as vaccination -including first doses - continues to roll out at pace, and as long as PCR cases do not start to double too quickly, then the vaccine can still roll out in time down to the perhaps 45 years needed to prevent healthcare overload if control of cases is lost. But if there are any significant disruptions to first doses over the next few weeks, it may be that we have to retain control measures for longer.
  • As uncomfortable as I am with the situation, it is the best situation we have been in for a long time and it is better than that faced by many people across the world.  I'm sure most of us have family, friends or colleagues in the various countries that are facing a troubled spring.   There is hope in how well the vaccinations are working, and I hope that a lot of the production capacity still coming on line and part funded by UK orders can end up being exported soon enough.   Likewise, I hope to see a couple of promising therapeutics coming out of Phase 3 clinical trials in the coming months which could potentially do a lot for the worsening situation in various other places. It's bad news for everyone, us included, when things go from bad to worse anywhere.

The final plots of the night

UTLA level analysis.  This was lit up everywhere tonight with the LFD data bundled in there.  It makes interpreting the plot really quiet subjective IMO.

I've wound it back to only show about 20 of the most concerning regions; this is done by baselining over a 4-week period and showing:

• ▲ - Values that have rebounding more than 2x their minimum during the period since baselining.
• ▲▲ - Values that are currently above their level at baselining.
• ▼ - Values that have fallen by less than 10% since baselining. 

As usual there's some familiar names in there; my sense is that this plot has largely been showing the UTLAs with the most sluggardly decay and more recently those subject to individual mass outbreaks.  Something still looks bad in the North Lincolnshire area.

Vaccine plot - second doses are now ramping up to large numbers; total doses this week are running only just behind last week's record levels.

One final errata - I called this week thread #19 when it should have been #18 - although I'll retcon this to restoring continuity to the timeline as I had two weeks in one of the early threads...

Thanks again. One thing I don't really get from the data here - you suggest the England death rate is beginning to flatten because lower age groups are starting to become dominant due to vaccination effects, but from the linear version of D5 it still looks like more deaths are occurring in the over 80s, so this effect shouldn't be very significant yet. I can't see the proportions very well from that graph due to the scale at the bottom end.

I suspect the current apparent flattening in England death rate fit (graph 8e) may be partly a result of the noise in the data as there was what looks like a noise peak in the raw data on the 19th March, although it's impossible to be sure yet. Regardless of this, I think if the trend you describe towards younger ages becoming more dominant were real, it would not be a bad thing because we know any residual delayed vaccination effect in the 60-80s will catch up the over 80s very soon. Using the dashboard, total deaths in the under 60s are still much lower than in the over 60s (11% of the total were under 60 in the last week, vs around 7% back in January), and I assume most of those under-60 deaths are people in their 50s, who are also now being vaccinated (I saw a figure this morning morning suggesting 75% of 50-54s had now been done). So I think the result of all this is that we should not expect the death rate to plateau at all for any significant period. It should keep approaching zero but obviously at an ever slower rate and with noise effects this might start to look like a very low plateau, but not yet.

Good question.  I've done a crop and zoom of the actuals (not normalised) data on a linear plot below.   Deaths in under and over 70s are approaching a similar magnitude, and the ever-faster exponential decay is limited to the 70+ ranges currently - it should move down as the vaccines already given take effect.   There is a lot of noise, but the <70s has bottomed out and maybe even rebounded a bit - which is enough to level off total deaths over the last few days, although there is as you say a lot of noise.

The effects at work I think are:

• Deaths are falling in proportion to PCR cases with a ~3 week lag in <70s,
• Deaths are falling faster than PCR cases in >70s due to the vaccine.

As PCR cases have been almost level for 3 weeks, deaths are now starting to level off in the group that hasn't yet been decoupled from cases by the vaccine.  There's a total lag of more like 4-6 weeks from injection to the leading edge I think of this as it includes the combined time for a vaccine to take effect, or not and for a persons exposure to bounce of the vaccine, or to get them ill enought to eventually be hospitalised.  So the pertinent ages for vaccination are likely those happening 4-6 weeks ago.  

I agree - but I think this is showing that trends in cases are not yet fully decoupled from trends in deaths - and presumably therefore hospitalisations (different, coarser demographic bins on that data...) - and that we're depending on the last couple of weeks of vaccination and the next few weeks to further that decoupling - so as long as there's no significant disruption to the vaccination program, hopefully any plateau in the data won't last for long.   But 

It's all too close to the wire for comfort, but really it's a better situation than I could reasonably have hoped for 3 months ago, and than many other places are in.  

Thanks.

So it seems that transmission within schools is taking off but isn’t (yet) translating into transmission to adults. I’d have thought that by now we will definitely have seen the impact on parents. Perhaps it’s there but hidden by a fall within the childless demographic? As I’ve said before, it looks like we’ve got away with the return to school but we need another week’s data to confirm this. Hopefully we’ll have another 3 weeks of ‘boring’ data with the case numbers slowly declining, before the impact of back to school and the wider reopening starts to feed through into the numbers for w/c 19 April.

I don’t think the relaxation from this Monday will have much of an impact as the focus is on outdoor gatherings. We could do with a longer spell of warm weather though, as it won’t last till the bank holiday weekend. Retrospectively, I think the criticism last spring of people going to parks and beauty spots to enjoy the nice weather was misplaced. Some people will want to meet anyway but with nicer weather and small outdoor gatherings being allowed, they are more likely to meet outdoors, as you have pointed out. Plus people are more likely to ventilate indoor places, as you’ve also pointed out.

Hopefully a full week of deaths in the double digits this week. As you say, these numbers are plateauing but that was to be expected. The vaccines will continue to do their good work at driving this down but with vaccination rates relatively low in some demographics and the mid April opening up, I think we will hit a fairly hard plateau in May. I don’t know how low it will be - at a guess, around 20 a day on average (may be I’m being too pessimistic here). This should decline but very slowly, as long as we don’t get hit hard by new variants...

With regard to the situation in Scotland, I'm not sure how much extra testing is being done across work places and how much of the plateau this would explain. However, I thought that the UK wide survey which is done once a week has also shown levelling off/increase in Scotland the past few weeks. If I understand the system correctly this should be independent of any increase in asymptomatic surveying related to schools/workplaces. I've not dug into any of this in detail though so may have misunderstood how that is calculated.

Fascinating as always.

In Scotland, the case rate has been fairly static for a good few weeks, and the test positivity rate is typically hovering around the 2-3% rate, with about 20k test a day.  

I know that in healthcare, oil and gas/marine (I believe) and through some government schemes there is *fair amount* of asymptomatic PCR testing carried out. I'm unsure about the exact percentage, and struggling to find any stats on this.

Should we be getting concerned that we're plateauing due to hitting a level of false positive inclusions? Over on UKB I linked this: https://arxiv.org/ftp/arxiv/papers/2102/2102.11612.pdf and a poster replied that there was a change to 1 gene due to the UK variant, but I still find it odd that more than 50% of positive cases are now "single gene".

Any thoughts on how low case rates, asymptomatic testing and false positives could skew the numbers?

Quick reply for now as I'm about to Go Local for the first time in months; hoping to find an ice-cream van or two.  I'll reply to everyone later properly.

False positives - they're not really "false positives".  A genuine false positive in RT-qPCR should be exceptionally rare and will be QCd for internally with control negative plates.

The preprint you link is mostly talking about "over-sensitivity" where a test might detect an infection that has recently passed and is no longer infections.  In terms of the ONS random sampling survey they refer to, this follows reality in a different way to the Pillar 1/2 testing, as the later are "gated" by the onset of symptoms, where-as the ONS one is random.  The ONS is something of a convolution of an "virus is detectible following infection" function with the symptomatic data I think.   That's often reflected in the shape of the ONS curve lagging the Pillar 1/2 data. 

The paper you've posted looks perilously close to pointing at numbers and saying "Look, numbers!" without making a scientific case.  We know we expect 1 fewer gene now the "Kent" variant is dominant as this has had a change on the S-gene.  

There seem to be two plausible hypotheses IMO

1. Detection of virus for a longer - but still finite - period post infection leads to a reducing average number of genes being detected in a decay phase of the virus
2. The rise to prominence of the Kent variant with the SGTF means the mean number of genes detected has been falling as that variant rises.

It's an interesting possibility but the pre-print doesn't really dig in to it IMO.  I don't have much patience for a pre-print saying "look there could be a problem" without taking the time to dig in to the details to examine their own question, e.g...

• If the data he uses to examine the period from October 2020 to now is available for last summer, he could look at the mea number of genes detected then - does it fall likewise?
• Data is available on the rise of the "Kent" variant from COG-UK.  He could plot this against the fall in genes detected.  Does it look like there's a link?  Can he frame a plausible hypothesis test around the two datasets?
• Given that "Kent" is perhaps associated with producing a higher viral load, it could well be that it leaves more virus in the body for longer, leading do a doubly effect of having one gene missing from test results, and extending the post-infection detection window when used with random sampling (e.g. ONS).  

Confirmatory signals that make me think the plateau is real:

• Deaths for unvaccinated ages in England are tentatively plateauing a couple of weeks after the decay in cases really slowed down, this is an emerging confirmatory signal for cases near-plateauing.  Perhaps it's noise and will go away.
• Rising positivity in asymptomatic LFDs in England - so inflection is likely rising in the younger ages; demographics shows a convincing fall in the most vaccinated ages.  Add it together for stagnation in cases - for now... 

There is a question discussed a couple of times on here as to how much the symptomatic sampling is "gated" and how much it is "random" in modality given the non-specificity of the "official" symptoms.

Regardless - decisions aren't officially being made on cases any more, but on hospitalisations and deaths.  So it's a bit of an academic discussion.

Edit: False positives in LFD data are a different issue; regardless they don't related to rising positivity in the English LFD data - the only data published on a daily basis by test type AFAIK.

Yes, it all seems a bit opaque - more comments from Alasdair on that front.  I hope people making decisions have a better understanding and more separated datasets...

Thanks.  Yes, the ONS random sampling survey has levelled off everywhere and is now rising slightly in Scotland.  The asymptomatic testing doesn't feed in to this data - it's a fully separate survey to everything else.  I don't think it includes prisons or student accommodation blocks - both of which have had significant contributions to cases on a regional level at various times. The ONS data is on "live" infections and normally lags cases date on "new" infections by a couple of weeks, so it's a bit strange the disparity between the two for Scotland now.  I'm assuming the difference is down to the low numbers and the noise this brings.  Could be wrong... 

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

Thanks. I found that paper through searching various terms to try to understand how false positives, or just plain old over-representation of the prevalence were dealt with.

It may be "academic" in England, but the Scottish policy hangs of getting case numbers below 50/100k and my concern was that when true case rates get very low (i.e. now -  we only have 264 in hospital, no deaths yesterday and 11 in ICU), then the impact of any type of overestimation could be another 2, 3, 4 weeks in a higher level of lockdown that required. 

I'm hoping that'll change to be based on more useful measures of hospitalisation and ICU.

I hadn't looked at the release criteria for Scotland; I appreciate now that as you say it isn't "academic" for your car.

In terms of false positives over-estimation, at the UK level...

• Last June, we had ~130,000 PCR tests/day in pillar 1+2, and around 620 positives a day, giving a positive rate of around 0.5%
• Currently we have ~280,000 PCR tests/day in pillar 1+2 testing, and around 5,500 positives, giving around a 2% positive rate.

The false positive rate must be less than or equal to the positive rate of the tests.  So, unless the false positive rate has incurred ~4x since last summer (0.5% to 2%), it's clearly not responsible for a significant factor of the UK wide PCR positives.   As it was, I think the 0.5% returns last summer were real predominantly real positives, considering all the aspects of the data.  There was a big misinformation bruhahaha over false positives that started last summer from a discredited blog post from the CEBM - from authors who have produced several flawed analyses all flawed in one direction.... [1].  

The possibility that the Kent variant has a more persistent viral load and so has enough material persists for a post-infectious PCR detection feels much more credible to me than the joke blog out of CEBM -  but I'm spitballing and don't know.  However, I would expect this to affect the ONS survey far more than a symptomatically gated cases - and it may not be a "false" positive as it would likely correspond directly to those infected people having a more persistent viral load and remaining infectious for longer.  Complicated stuff way beyond my ken.  

More data on test numbers undertaken here - https://coronavirus.data.gov.uk/details/testing

I think the issue is that "true" cases are being decoupled from hospitalisation and death by the vaccination - in England, cases are falling in older people where much hospitalisation and death occurs, and cases are rising or heading for growth in younger people.  I have't seen demographics for Scotland but it doesn't seem a stretch to infer a similar effect.  So - cases can grow for a while without impacting hospitalisation, but that growth remains exponential and until 90% of people are vaccinated down to age 45 or so, that exponential growth can still present a risk.  

So, the question for me isn't so much about false positives (not I think a big issue) but about the level of uncertain risk around rising cases in the young that is acceptable.  I'm on the very risk averse side of the fence on that but with the rate vaccination is going the risk is hopefully reducing significantly by the day.

LFD false positives are a different issue, but with LFD positivity rising in England, it's one that is secondary.

[1] https://www.covidfaq.co/Carl-Heneghan-and-Tom-Jefferson-Oxford-CEBM-eb9a2ce...

Just so long as it's forwards looking so that action is taken before things go too wrong; the problem with using them as the main measure is that by the time they're bad, worse is locked in.

Thanks for the reply! If I'm thinking about it correctly, I'm not sure that a more persistent detectable viral load would cause a plateau if cases were falling. Instead wouldn't you expect a delayed drop off from the peak?

I agree low numbers mean that stochastic sampling effects come into play more for the ONS measurements. From what I can gather the most recent ONS Scottish figures are based on around 15,000 people. Last week 53 tested positive, this week 63. 

Indeed - I don't think a more persistent viral load is behind the plateau, but it could be behind the increasing number of PCR results apparently matching fewer genes/primers as per the pre-print Alasdair linked.  Perhaps.  The timescale for the change observed in that document is since last October, which corresponds to the time the new strain has been rising to prominence.  It's why it's a shame the author didn't look at data on both.

A more persistent viral load could cause a plateau during falling cases if the variant(s) with more persistent load increased in prevalence during the falling phase, but that's a pretty specific set of circumstances, and as the Kent variant has already risen to prominence some time ago  I don't think it's worth considering now.  I just think the fall in cases has really slowed down in the P1/P2 data.

Yup.  Comparing over timescales of a week becomes a bit fraught, better to look at long term trends and to look for corresponding signals in hospitalisation and deaths - but without demographics that's basically impossible as changes from the vaccine effects are far larger than recent changes to the daily case number.

As per my post on our lockdown, our cases continued to rise post lockdown for a month instead of the 2 weeks that most lockdowns seem to create.  Our cases were all Kent variant, so I suspect the increased infectivity reduces the rate of decline. With the UK's relatively light-touch lockdown I suspect the less infective variants have dropped out of circulation quicker than the Kent ones and now the rate of decline is reflecting the relatively high prevalence of the Kent variant.* It could well be that the UK's lockdown rules aren't enough to cause the Kent variant to decline at all and it's reached a 'steady state' constant level of infection.  One thing you could look at is the test count and infection rates - are they increasing/decreasing in proportionate step?  Or is the rate of 'infection rise' outstripping the rate of 'tests performed' rise. Obviously it's hard to tell at a time of inflexion in the graphs, but it's something to look at for clues.

*Which means unlocking further could get bad quickly, depending on vaccination progress. Does the UK government have the balls to deny people their summer holidays? Probably not, so enjoy your icecreams while you can!

Having said I don’t think the data looks like it’s anywhere near a false positive floor, and I still think that to be the case, I think it’s something we’re going to hear more about given the very concerning article below, following up on some reporting from last year.

https://www.bbc.co.uk/news/uk-56556806

I hope the QC fails are relatively uncommon (several examples from 18 shifts), and as positivity is also low, the effects should be quite limited - but that’s no excuse for what is just awful lab practice that rightly appears to undermine confidence in this particular lab.  This is an absolute own goal against the misinformation pushers quite aside from being a serious problem in its own right.

Someone seems to be setting targets that encourage bad results over voiding bad samples.  That’s just not science.

It is really weird the way the cases curve has gone almost flat but way above 0.  That's the sort of thing you'd expect with a system in control and a feedback mechanism but I don't see any control mechanism here (nobody is changing lockdown rules according to infection level) and the basic infection process is exponential.

Flat is also something that might happen when someone is cheating.

In the context of interpreting things like this, I think it's important to recognise the cases curve for any one type of test is not flat. PCR cases are still declining, but with a relatively long halving time. The effect of adding in results from LFT is that the graph looks very flat (although, there is still a very small decline in the total which can be seen in either the dashboard cases graph or from WT's exponentiate still being very slightly negative.)

The other reason that I don't think we need worry specifically about the implications of a flat curve, though, is that the pandemic is not a simplistic exponential process. Of course it's easy to describe it as one mathematically to make sense of trends. But in practice infectious people can only infect people close to them and the infections in different areas are largely decoupled. I like to think of it as many, many different local outbreaks that in practice only connect with each other sporadically, especially during a period of lockdown when travel is restricted. So I don't think it is any more surprising that the overall summation of cases from all these different areas is flat, than it would be that it was any other particular gradient.

It's not made clear in the article what targets are actually used. Obviously throughput rate of these centres is critical for the national response so it's appropriate there is a target. It has to be a repeatable, fast but high quality industrial process, not like an experimental science lab at all. I could imagine that as a starting point you might end up with a target on rate and a target on number of voids. However I agree that could also lead to problems if there is a lack of oversight. It looks like what might be needed is tighter oversight and inspection, occasionally unannounced, from people who themselves are not subject to the targets and who have some power to intervene if necessary (said like a true regulator.) 

PS on the specific cross contamination risks highlighted in the article, it also looks like the equipment was designed poorly (maybe in a rush or on a restrictive budget)? It wouldn't be hard to design it such that the robot moved the samples to the test plate without traversing the pipettes above the other test tubes. Eliminating a particular risk is always better than trying to control it.

You should watch the Panorama relating to that article you linked, seems like plenty of scope for cross-contamination, false positives etc. The working conditions looked cramped and frankly awful compared to any other business. They stated they had multiple outbreaks in the lab, samples leaking, poorly trained and careless staff etc which could all cause false positives far in excess of what I imagined could occur simple through the inaccuracy of the testing itself. Certainly demonstrably less well run and more hazardous to public health than the vast majority of places that have been forced to close for months.

I'll admit, the first thought I had was "that wintertree guys assessment of the inconsequence of false positives seems off". This is certainly a complex situation and pretended the science/data is totally conclusive, or even that the way we are reacting to this situation is totally scientific and without other influence eg. political/financial is right. I worry that people have a little too much faith in our systems sometimes.

The way I see this kind of false positive and the data is that it doesn’t create positives out of thin air, it gives a number of false positives proportional to the number of real positives (*), so it multiplies the cases curve but a factor.   A factor not much more than 1.  The real information in cases is not their level - which is always related to reality by an unknown factor that changes over time - but their trajectory  - rising or falling, and the exponential rate at which they do so.  It’s their trajectory coupled with the trajectory and actual values of the other measures that has the most useful information in it.

So, in terms of cases as a source of data on the situation, and how well control measures are working, it doesn’t change how I interpret them.

As a multiplicative factor, it also can’t create a “floor”, or constant level, below which cases can’t fall whilst infections keep falling.

Given the comments in the article about this getting worse during the day, hopefully it was related to the workload during the peak period, and as the number of people going for tests backs off now, so does the problem.  Hopefully.  

None of which excuses running a lab like that - as you say the apparent hazard to staff and the irony of that compared to closed businesses, and the individual consequences of false positives for the people on the receiving end.

(*) it’s a bit more complicated than that due to sampling effects  but that’s the gist.

Caaes have a lot of problems on top of this - a false negative rate, some people with and many people without symptoms not going for tests, and the fraction of people going for tests changing with prevalence of the common cold and level of “overload” from it all.  So I’ve always been careful to use cases to interpret the other much more concrete measures of hospitalisation and death (which often involve confirmatory PCR testing on admission to hospital). And that data has been really quite unequivocal right, down to crude measure like “how many ambulances go past my house in a day”.  

But I think the reality of this is that it’s going to undermine faith for many in “the data” far beyond what it should - and that’s the biggest failing of the lab IMO as the scope for consequences is far, far wider.  Hopefully the vaccination program renders the point moot.

Agreed, but if you push that target down to the lab people over and above a commitment to quality, it’s not going to work.

I think it’s repurposed from a pretty standard lab robot - but you don’t normally have snot or cotton buds in your samples...

Agreed.  I also wasn’t happy to see hands working around a robot like that.  For fixing units in the field, perhaps it would be easy enough to attach to the part that moves horizontally but not vertically to fit a “snot guard” plate which rides above the “at risk” palates.  It could also be given a simple machine vision system which stops the robot if there’s a cotton bud or stringer.  But until then they should not be afraid of voiding samples, and perhaps shortening the shifts.

PCR cases are still falling in England with a pretty consistent but slow halving time; it seems to be a weird coincidence that the additional asymptomatic testing is filling this in to a fixed level.  (Data by test type is only out there for England AFAIK.)

I agree that if PCR cases were level it would imply a good control system - one we’ve never been able to achieve.

I just read something about false positives from Vaccine, is this a thing and could go some way to explaining this?

Another point worth noting about all the above is that if there were a significant number of false positives due to cross contamination, they wouldn't cause a plateau in cases anyway because the number would be a function of the number of true cases (or rather positivity) because with few true cases going through the labs there would be few sources of cross contamination. Number of false positives from contamination would be a function of number of true positives, rather than a direct function of total number of tests. So they might cause a proportionate change in the reported numbers of cases but would not stop the numbers of cases dropping to zero. This is different than any false positive rate that is some inherent function of the test itself when given a negative sample with no contamination.

I keep thinking about starting a thread on vaccines, symptoms (should you go for a test?  Are you told not to for a fixed time period?) and PCR positivity after each of the various vaccines.   There’s a lot I don’t know to understand if it could have an effect or not.  It’s an excellent question.

I was just thinking this through and came to that conclusion. But then had a what if moment.
Is this necessarily true? What if the source of contamination was some control samples or something in the lab? Then you'd get a steady level of false positives, but whether that would be as a proportion of tests or as a constant function of how many calibration samples they run or whatever isn't easy to find out. I don't know much about procedures in labs like these, but I can naively contrive a mechanism that would stop cases ever going to zero.

Edit: obviously doesn't come close to accounting for the current plateau, that would be insane, but it's a mechanism for a non-proportional number of false positives that crossed my mind.

Well yes, it will be in some way proportional. However looking at the handling practices of the lab featured in the article/episode, it seemed quite possible with sufficiently shoddy practices/staff/equipment to contaminate multiple covid free-samples with 1 infected sample. Not to mention the close quarters nature, lack of social distancing and live outbreaks amongst the staff conducting the testing. I can't see how I avoid reaching the conclusion that false positive rates are higher than previously assumed.

It does make sense to me that any in house testing at hospitals or labs that were already established before covid would have higher standards than this, so I'm not trying to write off all testing. 

It has occurred in other countries too however; have a look at the case curves in Denmark and Belgian both of which have had winter peaks then subsequent long lockdown with R ~1 but cases well above 0. I was pondering whether you might get some of these dynamics with much smaller subpopulations of key workers where R > 1 but which exist in a background population of R < 1; the latter will restrain the R at below one but the former will keep it close and the superposition of lots of subpopulations makes it look flat...I need to set up a model to play with it though as I'm not convinced thats what will happen.

If you have x% false positives it doesn’t really matter as long as that % stays constant. If it’s systemic issues, you’d expect that to be the case.

Your assertion of the risk profile of a testing lab compared to say a heaving pub seems a bit odd.

Exactly, some key workers etc plus some people who totally ignore the rules means there’s always going to be a minimum number of cases which will take a long time to reduce. 

Interesting that case rates among secondary school pupils are definitely going up whereas among adults they are steady or falling. Would love to know if that’s because kids don’t transmit as much or because the adults aren’t getting tested for whatever reason (perhaps due to being asymptomatic) or due to some other reason.

Looks like the nice weather is going to last a bit longer than initially expected. I reckon that plus the school holidays will see us through to mid April with a further gradual decline in PCRs. The lower we can get the case numbers before 12 April, the better...

Very briefly. Opt in bubble lfts for people back at school. We've started getting slack when the kids aren't about. We've been as good as good all the way through, but we've skipped a couple. Once a week instead of twice. If I'm doing that, and my partner is doing that, you can be sure it's a small number of those LFTs being done. The percentage I'd like to know is how many results are sent in without the test being done.

But this is just pissing around the margins. What lockdown? I've not seen a meaningful one since last spring. 

I guess somebody decides how many test kits to issue and that could be a control mechanism i.e. slow down the amount of linear flow kits you issue so you don't end up increasing the overall positive tests.

As a general principle I am suspicious of coincidences which result in administratively convenient constants.   It's like the accountants / tax man flagging suspiciously round numbers on company results / tax returns as a reason to dig deeper.

I didn't say heaving pub, you did. We don't really have heaving pubs anymore. I'm comparing my experience of pubs/gyms and my own workplace during the pandemic vs the lab shown on the Panorama episode. The lab was by far less clinical and well run from a covid prevention standpoint than the aforementioned businesses. Have you watched the episode? Have you been to any leisure/hospitality venues in the last year? Have you gone into work?

The lab shown stated something like 3 outbreaks, about 10 employees each time. The places I'm comparing to have had 0 outbreaks and no cases.

I am asserting that having businesses closed in part due to data contributed to by labs that are not only falsely inflating numbers through poor standards and practices, but also by spreading the disease amongst themselves through ignoring even the most basic distancing practices is tragically ironic.

I'm irritated enough just by having my usual routines disrupted, if my business or income was negatively affected throughout I'd be fuming at this revelation.

Completely off topic but this sort of statement really winds me up. This has been a real lockdown. We've had it easy with the ability to work from home etc but my family are still at the end of their tether. (It's better for me, having climbing as a release again from this week.) For many people it's far worse, many have lost their jobs, some had to home school multiple kids. WTF are you expecting? If it's easy for you then just be grateful and keep your mouth shut.

Rant over.

I haven’t done a plot on it, but the data is on the dashboard.  LFD positivity has been rising as the test number remains roughly constant (integrated over a week).  

PCR cases in England look to be going in to slightly faster exponential decay.  With a grace period in school related transmission over Easter perhaps we’ll see cases drop in all 4 nations by this time next week.

You could break out Benford’s Law on the daily figures...  I don’t see any methodological cause for concern in the data; just concern that cases aren’t falling as fast as I’d like - but every other measure continues it’s rapid decay and they’re the more actual and more important readouts.  

I have a bit of a bee in my bonnet about pubs. Some may be well run, most are not from what I’ve seen.

By their nature, the labs will have relatively few staff and presumably they operate distinct shifts. Of course they should take the necessary measures but any outbreak would be relatively small and easy to trace.

Climbing walls have been pretty busy at times. Not so bad round here as can normally find a quiet area and the last hour of the day is fine anyway. 

Exactly, the fact that hospitalisations are now dropping much more slowly suggests that cases are plateauing (considering that vaccination will gradually cause the hospitalisations to decay even with similar case numbers - which should then decay in due course as well).

It sounds like your addressing your own personal perceived gripes about pubs and haven't confirmed you've even watched the exposé. Lets not bother talking passed each other any further.

Does the PCR test look for genetic code in the virus that the vaccines also contain? If not, then there shouldn't be any effect.  If it does, then it's probably possible to make the PCR also look for something in the vaccine that isn't in the virus so it knows if the 'virus positives' are actually vaccine positives. The vaccines could also be deliberately genetically tagged so each can be identified and thus PHE could work out efficacy of each vaccine against variants.

Vaccines alter LFT results. The reduced viral load created by the vaccine means that some people infected are less likely to test positive on a LFT. Thus our LFT surveillance testing is being wound down as vaccination winds up.

Interesting article here

https://bylinetimes.com/2021/03/30/gavin-williamson-re-opened-schools-with-...

about the role of UsforThem in the reopening of schools last September.

Looks like the government listened more to a group of dubious right wing lobbyists than it did to warnings from SAGE about what might happen if schools reopened without adequate safety measures.

I'll read that later - thanks.  I have a lot of time for the work Byline Times are doing re: Covid and more generally.

I think you've got the wrong end of the stick. Who on earth said it's been 'easy' ffs?

My point about the difference between last spring's lockdown and all the subsequent ones since the summer is that in my locality and my bubble's locality people are constantly in and out of each others houses, cars, etc. I don't see any lockdown worth the name being observed in the areas I'm familiar with. 

Someone on Reddit sharing their views on the group’s Scottish Facebook page.  Some of what the allege may have a familiar ring to it with regards the many covid related pop up accounts on here.

https://www.reddit.com/r/Scotland/comments/i5zn5q/my_deep_dive_look_at_the_...

Some mid-week updates:

Plot x1 - this is the week-on-week fractional change in case numbers - which is less muddled by day-of-week effects in the sampling. This shows the hopefully not provisional data for PCR cases over the 7 days ending on days from Thursday to Sunday last week falling ever faster in exponential terms, heading back towards the ~14 day halving times we had before schools returned.

• There have been two period of slowed decay this year -the first is the "cold shoulder" around mid-Feb and there seems to be reasonable agreement that the cold spell was at fault.  If this trend to faster having times continues,s the second period corresponds quite closely to the period from schools reopening until the onset of the Easter vacation.  However, it's a bit too good a fit in that the decay of cases really set in on the opening day of schools, but those infections would have happened over the proceeding few days.  
  • Really this is clear speculation on my behalf as we've only a couple of days of faster decay in the data, and frankly who knows what's going to happen next. 
  • Still, if the decay holds, we'll see half the number of PCR cases when schools go back after the holidays - not to be sneezed at, and brining us close to the minimum level from last summer.
• LFD cases are falling of a cliff.  The number f tests being conducted has done likewise with the end of term, so I'd not read too much in to it directly.  However....
  • Just as we have seen the fall of symptomatic PCR cases in England masked by LFD results, now I expect we'll see the fall boosted by their disappearance in the cases data.  So, whilst PCR cases are halving faster than recently, we're not as out of the woods as the emerging trend in headline numbers might suggest over the next week.
  • Will this translate in to headline numbers for the other home nations?

Plot x0 - PCR cases and PCR + LFD cases for England.  This has my algorithm applied to mitigate day-of-week sampling effects although some are still visible as periodic wobble in the data.  PCR cases look to be accelerating their decay on this plot too.

Vaccines plot - second doses are really kicking up in the last few days, but they're still managing 200,000 first doses/day.  The NHS are quoted as saying they've now jabbed 85% of people aged 50 to 54 [2].  This should really start cutting in to transmission rates as it takes effect - if you look at the right side of the cases part of plot D1.c up thread, you see the cases start becoming large in number under about 55-60 years age, so the program is now reaching the point where it's prevented most of the future deaths and is hopefully going to start making serious inroads to transmission.

Variants plot - pucker factor: high

• Big warning that this data is based on surge testing and targeted, enhanced contact tracing so is following a very different process to general PCR cases.  It's a lot sketchier to tie trends in this data up to assumptions on country-wide prevalence.  Still, if it's rising, it's not good.
• Data is source from the government websites [3].   There's been a big change in the data but not the plots - a lot of "Genomically probable cases" have disappeared and the number of "Genomically confirmed cases" has risen by more than the recent trend; I plot the sum of these.  I assume there's been a big clearcut of probables either by either binning them or converting them to confirmed.  I don't know how many were rejected. 
• The SA variant continues to follow a reasonably exponential trend, doubling around every 40 days.
  •  I get the impression a lot of these are importation events however, so the trend may reflect more of the trend on the other end of travel links than in local transmission.  
• VUI-FEB-03 was first sequenced in the UK.  This is not conclusive proof of origin, especially given how much sequencing is going on in the UK.
  •  It's a VUI (Variant Under Investigation) not a VOC (Variant of Concern) for now, but perhaps we'll see that change soon enough.
  • It's doubling pretty consistently every ~15 days, despite nationwide PCR cases falling over the 40 days this has been getting tracked.  
  • If this indicates that it's more transmissible (inherently, or through reduced co-immunity in vaccinated and/or infected+recovered individuals), that's bad news whilst some vulnerable people are unvaccinated - here I am assuming that the vaccine will at worse not prevent infection by this variant but still significantly moderate illness.  I have't see any evidence of that yet but it seems reasonable given other studies on other variants including E484K.  I could be wrong..  At the rate it's currently doubling,  my sense is that barring change for the worst, it won't pose a major problem before the vaccine has taken effect in everyone aged > 45.  Which would be nice.
  • Or the rise could indicate that enhanced contact tracing is spreading out through a pre-existing network of infected people.  The spread would be exponential if not resource constrained.  This can only yield exponential growth in cases for so long however, and I don't think it could contribute for more than 2-3 weeks.
• Notably, the Brazil variants are not rising in a very concerning way IMO within the UK.  

On the subject of Brazil, Offwidth has shared several articles recently there.  This looks like it's going from bad to worse to disaster.   Deaths are rising in a way not predicted by cases, it's almost certain their diagnostic capability is overwhelmed.  Deaths are hitting 3,000 a day and there's another 3 weeks of growth in cases locked in, and the news reports on the political side are not happy reading.  It's simply awful, and I fear it's going to get a lot worse.  

[1] https://coronavirus.data.gov.uk/details/testing?areaType=nation&areaNam...

[2] https://www.bbc.co.uk/news/health-56598217

[3] https://www.gov.uk/government/publications/covid-19-variants-genomically-co...

[4] https://www.worldometers.info/coronavirus/country/brazil/

Thanks. There seems to be a lot of good fairly good news in terms of basic cases, I noticed today's ONS update is reporting that the increase in secondary school age cases that was seen in the week to 20/03 had petered out in the week to 27/03. I don't really understand what mechanism could have caused that, other than there only being a small pool of them without pre-existing immunity when they all went back, which seems very unlikely.

Maybe the mass testing of school kids is working, Keeping positive cases away from school and stopping the spread?

Re school cases petering house. IIRC before schools fully went back, Jenny Harris said that would happen. 12 months ago I gave her a fair bit of flack, but she sussed this one. 

Yer, I guess I don't understand why that would lead to cases initially rising then flattening though.  Unless for some reason performance of the system has improved through the first couple of weeks - I suppose that's possible and if so it's got to be a good thing. I don't know how though, testing rates were always reasonably high. I suppose the kids might have got gradually better at sticking the swab up their noses!

I like Jenny Harries. Always thought she seemed smart and well-intentioned in what she planned to say at the press conferences, but not quite so media-aware or as nimble under questioning as Whitty and Van-Tam. Did she give a reason for what she said about schools?

was the return to school staggered for secondary schools? 

I’m busy googling to try and jog my memory.   Think it was during one of the no. 10 conferences. 

Notionally they all started back the same week, but I understand many schools staggered years across a few days in that first week for practical reasons.

Yes, it was over three days for year groups at my lads' school. The first week they did three LFTs at school, and since then they are supposed to do them themselves. As I said upthread, I'm doubtful about the honesty of the reported data from home testing. This article today gave me pause as well:

https://www.theguardian.com/world/2021/mar/31/fewer-quarter-covid-symptoms-...

Nationally, for secondary schools attendance for Monday 8th was only 30% this gradually increased during the week to about 80% by the Friday,  even then some schools were just getting the students to do a test and then sending them home. The following Monday attendance was about 90% and as far as I know everyone was back to a full timetable.

Edit. Those figures are for England

Yes it’s a bit of a mystery. As you say, improved self testing ability and staggered return could be some of the reasons. As well as the fact that LFDs aren’t very accurate so they might have missed cases first or second time round. Plus declining compliance with testing / reporting over time. All of the above I suspect - but that wouldn’t explain the ONS state showing a levelling off as well. 

Great to see PCR numbers dropping as well. The school children positives don’t seem to be driving positives in their families (which the ONS also shows) - another mystery...

Hopefully another couple of weeks of declines, then it starts creeping up again...

Yeah, that plot is looking worrying...

This ONS survey on long covid is also worrying....  https://www.theguardian.com/society/2021/apr/01/long-covid-snapshot-poll-fi...

Good discussion piece here on the investigations into rising cases in Kent in Nov/Dec. Lots of interesting tidbits for those of us who were following things through the data at the time, about both the timeline of events and the way these things work (and apolitical, before someone comments about me sharing the Guardian again.)

https://www.theguardian.com/world/2021/apr/03/has-everyone-in-kent-gone-to-...?

Thanks, now I’m wondering how I’m going to get the 15 minutes of peace and quiet that needs to properly read.  A lot of lessons in there on what to look for and how going forwards.  SAGE advice from the WHO: “We won’t get fooled again”.   

In other news, PCR cases look like they’re back to halving on a 2-week timescale...  So it looks like we can just hold the line with schools open. Hopefully when they reopen after Easter, cases are halved from when they closed and with the vaccine taking hold in the 50-59 bracket and starting to cut in to workplace transmission.  

Serious long covid impacts on NHS staff

https://www.theguardian.com/society/2021/apr/03/nhs-feels-strain-tens-thous...