More Covid bants

New Topic
This topic has been archived, and won't accept reply postings.
Removed User 06 Nov 2020

My Dad keeps sending me links to alternative Covid viewpoints from scientists. I think you have all done a solid job of discrediting Yeadon's position - the idea that false positives rates are essentially constant means that we should see matched trend in cases vs tests carried out if it is all being driven by false positives. Of course that is not the case.

Now could you have a go with this chap Hutchinson...

youtube.com/watch?v=nEFXRnDCYkc&

I can see a few gaps in the 3 points he makes especially with regards to influenza, misdiagnosis and the PCR tests e.g. no mention of the probable impact of social distancing on flu cases.

What about the idea of virus is naturally becoming less virulent? Not borne out in the death data I would say.

What about masks - I can see this is not so clear cut. Has anyone done a meta-analysis on the studies yet (maybe there aren't enough)?

Finally can someone explain why the death rate is rolling off over the last week? Is this the effect of tier 3 restrictions on Northern urban areas?

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

Thanks in advance - I am both busy and lazy.

Post edited at 09:55
2
 wintertree 06 Nov 2020
In reply to Removed User:

I just can’t cope with YouTube videos.  Even at 2x it’s like having teeth pulled being drip fed content and I can’t engage with them as my downtime / thinking time activity.

> Finally can someone explain why the death rate is rolling off over the last week?

I always ignore the most recent 4-5 days in deaths by actual date as they’re subject to significant reporting lag and jitter.  Daily updates can add and rearrange historic deaths over more than 30 days but most updates are confined to this recent window.  Please excuse the crude annotated plot from your link, attached below.   Looked at excluding the provisional window, it’s not slacking off.  It’ll be nice if it does though.  Give it another 4-5 days then look at the data for the same fixed time period to decide if it really slowed down or not.

Doing some simple projections of hospital occupancy (below) suggests that is starting to rise at a slower rate pushing the hospital overload point further off than “end of next week” so that’s calming if it continues.  I haven’t done any geographic breakdown of deaths but some data supports a tailing off of case rates in some regions with currently high numbers which could be translating through to deaths.  But, cases data is close to garbage right now with a high positivity well past guidelines and lots of spikes from university outbreaks masking the slower but consistent rise in the rest of the areas.

I would say that there is an awful lot of uncertainty in and between the various data sources right now.  So, as ever, a time to err on the side for caution and hope for the best.

Post edited at 10:02

mick taylor 06 Nov 2020
In reply to Removed User:

Death rate rolling off?  Where do you get that from?

Removed User 06 Nov 2020
In reply to mick taylor:

Look at my last link - government data (Wintertree's point about lag/jitter nonwithstanding).

 mik82 06 Nov 2020
In reply to Removed User:

I don't think there's any evidence it's becoming less virulent. The large numbers of cases over the summer were in countries with much younger populations.

There were about 200 deaths reported per day last week in England, Given the lag from infection to death of 3 weeks, this would correspond to a period where there was an estimated 27,900 cases per day (ONS) - so an infection fatality rate of 0.7%. Actual deaths with covid on the certificate are higher than the government stats - so this is very similar to the 1% IFR that was initially quoted, even with the infections being biased towards the young in our second wave so far.

There has been a meta-analysis of masks - probable large reduction in risk, but with low certainty.  (Lancet systematic review).

It's too early to tell about the death rate due to reporting delays.

 Neil Williams 06 Nov 2020
In reply to wintertree:

Fewer deaths is caused by improved treatments (e.g. antihistamines and anticoagulants) not by it becoming less virulent.

This, paradoxically, may be causing hospitals to get fuller, because people aren't dying soon after admission, but instead need to be there (and ventilated, potentially) for longer until they recover.

Post edited at 10:32
 Offwidth 06 Nov 2020
In reply to Removed User:

A useful link on why some prominent scientists initially didn't understand the science behind masks and were mistaken. Any hold-outs on the benefit of masks are just talking rubbish.

https://www.wired.com/story/the-face-mask-debate-reveals-a-scientific-doubl...

Plus another on why they work

https://www.ucsf.edu/news/2020/06/417906/still-confused-about-masks-heres-s...

Post edited at 11:03
Removed User 06 Nov 2020
In reply to Offwidth:

The Lancet meta-analysis (mentioned above) on social distancing and masks makes for interesting reading (well at least the discussion - the stats methods are rather beyond my limited skill set in this area).

https://www.sciencedirect.com/science/article/pii/S0140673620311429

Not certain on masks due to a paucity of evidence, very certain on social distancing. Randomised trials really needed.

Post edited at 11:04
 wintertree 06 Nov 2020
In reply to Neil Williams:

> This, paradoxically, may be causing hospitals to get fuller, because people aren't dying soon after admission, but instead need to be there (and ventilated, potentially) for longer until they recover.

Yes, a point often missed when I’ve been having real life discussions about the looming hospital occupancy crisis.  I don’t know what sum total effect these treatments have on occupancy but it’s far from clear cut that they’ll reduce it.

 Offwidth 06 Nov 2020
In reply to Removed User:

Just read the links I posted.

Removed User 06 Nov 2020
In reply to Offwidth:

I have - they are saying evidence is based on studies showing the efficacy of masks for the limitation of particle spread as measured in a laboratory setting. This approach appealing to me (as an engineer) but is still a jump to showing that masks deployed en masse in society have a large impact on virus propagation beyond that already provided for by social distancing. There are also more complex psychological factors at play that a laboratory based analysis of the physics cannot account for e.g. "could mask wearing result in a more lax approach to social distancing."

I am not anti-mask by the way and agree with erring on the side of caution with these things.

Post edited at 11:45
cp123 06 Nov 2020
In reply to Removed User:

Interesting sounding chap, clearly a wealth of experiences in dealing with respiratory illnesses and the pathology behind it like, but as always judge the argument, not the person making it.

> What about the idea of virus is naturally becoming less virulent? Not borne out in the death data I would say.

I think he was talking about the global sense, viruses do mutate, and in the case of the Spanish flu it mutated into something less virulent.  Clearly the timescales are different as Spanish flu took in the order of years into to do this whereas we are still in the order of months Here is the website he mentioned: https://nextstrain.org/ncov/global

> What about masks - I can see this is not so clear cut. Has anyone done a meta-analysis on the studies yet (maybe there aren't enough)?

Mask wearing and its use in pandemic situations were known issues before the current pandemic. Remember at the start Whitty and the government were saying their is no evidence that the population donning masks made a difference? They were basing that on studies done in healthcare prior to COVID-19 which didn't produce any evidence of an effect. A mask, when used properly clearly helps with aerosol spread, but this requires disposing/washing of masks on a very frequent basis (like every use) and not touching it after fitting.  How many members of the public are doing that rather than having a mask that gets repeated use and adjustment when on and washed once every few days, if at all? If Covid and other pathogens accumulate on the outside, the taking on and off and adjustment when on may make things worse. [I wear my mask btw and always do when needed.]

> Finally can someone explain why the death rate is rolling off over the last week? Is this the effect of tier 3 restrictions on Northern urban areas?

Its a good question, but as wintertree says, you need to wait a few days for the final count. Zoe COVID are wondering if we are past the peak: https://covid.joinzoe.com/post/over-the-second-wave

Post edited at 11:56
 Neil Williams 06 Nov 2020
In reply to cp123:

The "smoothed" 7 day average graph on the Government site is showing daily cases *slowly* coming down.  Hopefully the lockdown-lite (!) can make that a bit steeper.

 oldie 06 Nov 2020
In reply to Neil Williams:

> Fewer deaths is caused by improved treatments (e.g. antihistamines and anticoagulants) not by it becoming less virulent.  This, paradoxically, may be causing hospitals to get fuller, because people aren't dying soon after admission, but instead need to be there (and ventilated, potentially) for longer until they recover. <

Might not the fewer deaths be due to a lower initial viral load in many cases, due possibly to increased restrictions/precautions? Also do those improved treatments also potentially shorten hospital stays and aren't the majority of deaths some time after admission? I'm no expert but just curious.

cp123 06 Nov 2020
In reply to Neil Williams:

Whilst I'm not a fan of the nationwide lockdown, I too hope it has a positive and noticeable effect. 

 Neil Williams 06 Nov 2020
In reply to oldie:

> Might not the fewer deaths be due to a lower initial viral load in many cases, due possibly to increased restrictions/precautions?

There's certainly an argument that says masks are causing more cases (because they make people risk compensate and not distance properly) but less severe cases (because they reduce viral load).  It seems plausible.

Not sure about death timings.

 wintertree 06 Nov 2020
In reply to Neil Williams:

> The "smoothed" 7 day average graph on the Government site is showing daily cases *slowly* coming down.  Hopefully the lockdown-lite (!) can make that a bit steeper.

I am very cautious of the testing data - look at the plot for tests processed per day, it has a recent dip corresponding to the lower results.

Today's ONS update has landed and this is promising in line with the cases however with the  North West, Yorkshire and the Humber, the North East and London all showing a levelling off.  These can disappear by the next week's update with the smoothing the ONS use, but the North East in particular has been level for longer tan is needed to guarantee against that happening, and the correspondence to the cases data you cite is comforting.

So, it starts to look like the Tier 2/3 levels have fed through in to cases; that's the good news.  The bad news is that the other regions are generally rising and would have taken over driving total cases, hospitalisations and deaths up.  I say "would" as hopefully lockdown will bring the benefits of T2/T3 and more to those regions.

As I said upthread, hospitalisations seem to be slowing as well - this fits with the lag from the ONS data's most recent period (> 1 week old) to where we are now.  So, there's every reason to hope that deaths start levelling off but right now it's a bit soon, and the reporting lag must be considered.

Let's hope that the lockdown measures are sufficient to tip all regions over in to level or decreasing incidence so that the looming hospitalisation crisis isn't just deferred but avoided.

mick taylor 06 Nov 2020
In reply to Neil Williams:

> Fewer deaths is caused by improved treatments (e.g. antihistamines and anticoagulants) not by it becoming less virulent.

> This, paradoxically, may be causing hospitals to get fuller, because people aren't dying soon after admission, but instead need to be there (and ventilated, potentially) for longer until they recover.

Chatted to someone in a local ICU who said that. 

mick taylor 06 Nov 2020
In reply to Removed User:

I think over a period of time, reporting lag etc kinda evens itself out (enough for me to get a rough grasp on the situation anyway).  All I know is the death rate keeps going up, but at a slower rate than say end March into April. The death rate in Grter Manc keeps increasing. Wigan had 100 deaths October and looks set to have more in November (few days this week we’ve had ten or so deaths).  Grter Manc has seen a slowing or if the rate of increase (even coming down) but school half term would be a factor  

My concern is the death rate will come down at a very slow rate. 

Edit:  yesterday, Grter Manc reported the 4th highest covid hospital deaths since the start of the pandemic.

Post edited at 13:44
mick taylor 06 Nov 2020
In reply to wintertree:

> I don’t know what sum total effect these treatments have on occupancy but it’s far from clear cut that they’ll reduce it.

From someone working in a Grter Manc hospital (not Wigan), treatments are meaning people are staying in a lot longer, but lots more people are leaving alive than April/May.

 Neil Williams 06 Nov 2020
In reply to wintertree:

> I am very cautious of the testing data - look at the plot for tests processed per day, it has a recent dip corresponding to the lower results.

That isn't necessarily bad; testing is only presently done (with exceptions) if you have symptoms, so fewer people with symptoms is likely to = fewer cases.

 wintertree 06 Nov 2020
In reply to Neil Williams:

I agree.  It depends on which one is leading the other - but it makes me more cautious of the figures alone.  Their agreement with various random sampling surveys is encouraging though.

 Neil Williams 06 Nov 2020
In reply to wintertree:

The less "happy" interpretation is that the dip was caused by half term, i.e. schools are basically the only thing that makes any considerable difference.

Post edited at 15:06
 wintertree 06 Nov 2020
In reply to Neil Williams:

That's a good point for taking the pessimistic view. The other pessimistic interpretation is that the various university outbreaks cased a large, transient spike in pillar 2 data, and that this masked the underlying, unabated exponential rise in the rest of the population.  That's certainly been the case for my county and some others with big university outbreaks.

This data is from [1] and it's not clear if the dates given are the start, middle or end of the week being reported.  So the plots are between 1 and 2 weeks stale.  With these you can see how the bigger university outbreaks give a confusing picture in the total data which does not reflect what the wider population is doing.  Still, as I understand it the ONS survey does not sample halls of residence, and also shows a levelling off in some of the regions containing the outbreaks in the plots below.  So, that's promising... 

But there are lots things that could be lending false comfort to the data.  My hunch is that things are getting better (in that they’re not getting worse so quickly), but I wouldn’t want to translate that in to relaxing any policy changes just yet.

[1] https://coronavirus.data.gov.uk/details/about-data#cases-by-age

Post edited at 15:25

 wintertree 07 Nov 2020
In reply to thread:

Updated plots from today's dashboard update.

Cases

  • As reporting lag resolves, Nov 1st sets a record high; the last two data points in particular could rise more with the next few days of data releases.
  • It could be rising cases, or it could be better direction of testing closing the gap on the (much higher) actual infections.
  • The jitter in the values (weekend effect etc) is getting much worse.  The middle plot (residuals) shows the difference from the smoothed trend line.  The bottom plot shows the residuals normalised to a nominal statistical (random) noise one might expect for random population processes.  However, the value of the normalised residuals keeps growing which says to me the process is getting more disorganised.  
  • As ever, I regard the cases data quite suspiciously as it has a lot of complications going on. 

Hospital Admissions

  • Genuinely seems to be levelling of, as indicated by the doubling time growing large.  
  • This is very welcome news, I hope the trend continues.  This would concur with the ONS and other random survey data showing infections levelling off 1-2 weeks ago in the highest prevalence areas.  Remember, about 10 to 14 days of hospitalisation was "locked in" by the time lockdown was announced. 
  • Don't forget that the lower prevalence areas are growing fast and could take over driving admissions, although the lockdown will hopefully prevent that.
  • This data seems close to "actuals" and not subject to reporting lag, so the existing points on the right side of this plot is hopefully aren't going to change.  
  • I would expect hospital occupancy to continue rising whilst hospitalisation rate remains constant over N days, where N is the average duration of a hospital stay from a covid admission. I don't have a figure for "N" to hand.

Hospital Occupancy

  • With the levelling off of hospital admissions, occupancy is pulling back from the brink.
  • The most recent extrapolations to lockdown + 10 days are falling just below the nominal capacity, and are well below the "surge" capacity.
    • Phew
    • A lot can change in a week, let's hope it keeps changing in the right direction
    • Things aren't looking quite as dire as they did when lockdown was announced, but it would in my view would have been irresponsibly bonkers to delay the announcement and longer to wait and see what happened.  I'm sure the usual quarters will push back on this however.
  • Influenza is going to grow and eat in to the same capacity.  Are social distancing and improved hygiene measures and increased vaccination enough to keep this minimal?  
    • Dammit Jim, I'm a scientist not a fortune teller.
  • Don't forget the hospitalisations "locked in" in the low prevalence, high growth rate regions not subject to T2/T3 before lockdown.

Deaths

  • I still see no support in the OPs claim embodied in their question "can someone explain why the death rate is rolling off over the last week?".
  • Deaths resolved by date of death (not reporting date) continue to rise.  The doubling time has been gently increasingly recently suggesting that the exponential rate of increase is slacking off, but it's nowhere near the death rate ceasing to grow (how I interpret "rolling off").
  • I always omit the most 5 recent days of data from "Deaths within 28 days of positive test by date of death" as these have significant reporting lag.  
    • The dashboard data "Deaths within 28 days of positive test by date reported" has no roll-off which would be a precursor to a genuine roll-off in deaths by date of death.  (The former leading the later).
    • This plot will have a drop over the next 2 days, that's a reporting lag over the weekend...
  • Because each person to be hospitalised takes a different amount of time to die, a "sharp" feature in the hospitalisations curve will be blurred out in the deaths curve but the distribution of that variation.  But if we get a long plateau in hospitalisations that should translate visibly in to deaths levelling off starting a week from now or so.

Summary

  • With the random sampling surveys showing a levelling off of cases, with hospitalisations levelling off with the appropriate lag, and with increased control measures coming in (lockdown) I am hopeful that we will see all measures falling over the next month, winding things back from the brink of healthcare overload and hopefully back to a lower level where test/trace/isolate can become a useful tool.
  • I think there needs to be a long, hard look at how control was lost before this lockdown from the giddy times of late summer when things were good.  If we want to release lockdown and not just have another surge > lockdown that - for a 3rd time - risks healthcare overload, we have to do something very different after this lockdown is released.  My thoughts on what we do differently:
    • I was a proponent for closing pubs when schools were opened.  Not just on simple additive basis but on the grounds that each creates different links between bubbles so they combine super-linearly in terms of enabling spread.
    • A lot of evidence suggests drunk people can't socially distance once they leave the pub, and that the police are unable to enforce this late at night.  Being outdoors is no protection against close social contact.
    • Household visits in violation of T2/T3 seem to be driving a lot of the problems.  Building a new trust between government and households is the only way to fix this.  It needs honesty, evidence based explanations, extreme clarity, a zero tolerance approach to public figures violating the rules and well funded, trusted public health teams pounding the ground.
    • If universities re-convene students next term, the inevitable wave of infection must not be allowed to push test and trace off a cliff as it appears to have done this time around.  
Post edited at 17:08

1
 Neil Williams 07 Nov 2020
In reply to wintertree:

Cheers for that.

One thing that concerns me is the possibility that cases will start going back up after half term, i.e. that schools are the problem, and shutting basically everything else isn't enough to offset it.  We will see over the next few days, but if I was the Government I'd be seriously looking at how they can be changed so that distancing is maintained, e.g. part time attendance.

 wintertree 07 Nov 2020
In reply to Neil Williams:

You could be right.  Given that very view people infected in schools will go in to hospital themselves, but will need to go on to infect someone else, the timing is quite tight for half-term to have contributed to this.  So, on the bright side - perhaps the benefit of half-term is yet to be felt...  

But yes, the statistics suggest that having everyone back in secondary school and especially six form is a pretty big problem.   Anecdotally, social distancing is not being enforced at all on the upstairs of double decker school busses either...

There have been suggestions of splitting classes for those aged 14+ so classes study from home one week (with remote teacher support) and from school the other.  By splitting half the classes to study-from-home in one week and half in the other, further improvements are gained by reducing the number of possible transmission events within schools and halving the occupancy on public transport. 

If older years at schools were to go mixed-mode either after lockdown or in the new year, there would need to be a decision made very soon to give time to prepare for it... 

Post edited at 18:13
1
 RobAJones 07 Nov 2020
In reply to Neil Williams:

In the week before half term attendance in Knowsley and Liverpool was 60% (secondary). After the testing in Liverpool, it will be interesting to see what attendance is like next week. Something like Yr 7 and 8 in full time,  9,10,11 3 days and sixth form 2 days might be sustainable. No way exams will go ahead as normal in the summer. Who expected students to "catch up" this year? Shifting exams to the winter has problems (but as a hay fever sufferer revising is November has it's attractions)

 Yanis Nayu 07 Nov 2020
In reply to Neil Williams:

I do contact tracing for an LA. I thought half term would be a nice fire break but it’s gone mad in the schools since they’ve been back. Lots of positives on Thursday. 

 Neil Williams 07 Nov 2020
In reply to Yanis Nayu:

> I do contact tracing for an LA. I thought half term would be a nice fire break but it’s gone mad in the schools since they’ve been back. Lots of positives on Thursday.

As expected

 Si dH 07 Nov 2020
In reply to wintertree:

Thanks for the analysis. Encouraging for hospitals especially.

I don't think looking at the national trend in cases (or indeed the wide regional trends) has been very insightful in recent weeks as different areas that had regulations imposed at different times have responded in different ways. However this might change with lockdown because everyone has seen a significant change in regulations and behaviour at the same time. I'd expect trends at a national level to become a bit more useful again.

I do think there is useful evidence at city region levels now that restrictions have had a positive effect and tier 3 has either flattened or reduced rates. Liverpool city region and now also now the Lancashire council area are on clear downward trends (edit: except Blackburn which has been peaky but basically around about the same mean since mid October, although even there it looks like there is the start of a drop showing). The effect is less clear elsewhere but we are still only looking at official data for cases averaged over the week beginning 3 days after Tier 3 started in GM and finishing 10 days after, so you probably wouldn't expect to see it strongly*. Other areas became Tier 3 later, as was half term. Despite all that there is still definitely a reduced average rate of increase in those areas, compared to many areas in Tier 1 and 2 which have shot up and even overtaken quite a few Tier 3 areas in the last couple of weeks. (Eg Bristol, Hull both up around the 400/100k weekly average, Derby now has more than Nottingham, etc.)

As a result of all this I think there can be good confidence that keeping Tier 3 in place would be enough to prevent a resurgence of the lockdown gets infection prevalence down significantly in the 4 week period. If in some areas that proved not to be the case, we can at least be sure any resurgence would be significantly slower and easier to get on top of.

I'm not sure about Tier 2, I suspect that it would be enough to hold rates down in most areas but not those that are particularly susceptible for other reasons (deprivation, type of local economy, etc). But there is no way of knowing for sure. Tier 1 has been shown to be inadequate everywhere.

So if I had a vote, it would be to exit lockdown into Tier 3 everywhere until Christmas. In an ideal world I'd like to then relax to Tier 2 on a city/county regional basis if infection rates have stayed low and flat for several weeks, then monitor carefully and go back to Tier 3 at the first obvious sign of increase. In practice I suspect if rates are low Ian we will get a few days of Tier 1 over Christmas and then be back to Tier 2/3 after that.

Hospitality would need a lot of support through all this but that's more affordable than furloughing everyone. In my eyes, the evidence we have is enough to have confidence this would get us through to next spring without any improvement in test and trace.

Everywhere will have to be kept a ratchet higher if:

1) prevalence doesn't come down a lot following the lockdown or

2) virus transmission increases significantly further due to cold temperatures in December and January

*It's also worth noting that Wales started it's firebreak 2 weeks ago on the 23rd, same day as GM entered Tier 3. According to current dashboard data on a weekly average basis most Welsh rates are still rising. That shows you how long it takes from a change being implemented to it feeding through to show a reduction in rates on a weekly average basis when the 5 most recent days are ignored. Answer - at least 2 weeks. I'm not aware of anywhere presenting city/council region or LA case rates on a daily basis rather than weekly average (in an easily digestible way.)

Post edited at 21:01
 wintertree 07 Nov 2020
In reply to Si dH:

> I don't think looking at the national trend in cases (or indeed the wide regional trends) has been very insightful in recent weeks as different areas that had regulations imposed at different times have responded in different way

I agree - and I'd suggest even the regional level and below is confusing at the moment as the various university outbreaks transiently dominated their host county's  data, and at least in the North East they confuse the regional level data I think - some examples in the plots I did up-thread [1].  That being said, I have some code to make nice plots of the top level data, and in terms of hospitalisations and deaths its a useful barometer.  I'm just very cautious of the cases data as the different data sources out there aren't in great agreement.   

> As a result of all this I think there can be good confidence that keeping Tier 3 in place would be enough to prevent a resurgence of the lockdown gets infection prevalence down significantly in the 4 week period. 

On the proviso that the shifting season doesn't make it worse I think you're right. The ONS regional level data in particular is pretty compelling that the required effect of R=1 was being achieved in T2/T3 regions - even with a high prevalence (give or take my concerns about the "university effect" masking the background rise, but that still seems to be slowing excluding the university age population if not yet hitting R=1) .  If the lockdown hammers prevalence down, test/trace should then work more effectively.  This suggests to me that T3 should be sufficient nation-wide as soon as the spectre of healthcare overload is passed and we exit lockdown.  (I'm assuming that's what gov will do, rather than go for close-to-elimination...)  

> Tier 1 has been shown to be inadequate everywhere.

Agreed - I'm still drawn to jkarran's observation that the rise in R travelled with the seasonal weather change spreading south; in the end everywhere ended up on an exponential track.

> Everywhere will have to be kept a ratchet higher if:

I would add option 3 to that - if the hospitalisations from influenza become high, as this reduces the safety margin for accommodating covid patients without system-wide overload; the situation ran right up to the red-line on healthcare overload this time round; there's only so many times that luck can be pushed.  

> So if I had a vote, it would be to exit lockdown into Tier 3 everywhere until Christmas.

If you had that vote, I would urge you to give it another week to be damned sure...  I'm minded to hold the lockdown as prevalence is just far too high, and if thinks take a turn for the worse we're only days away from locking in enough trouble to really mess up healthcare.  With the uncertainty over the flu season as well, I think getting cases down at least 40% before exiting to Tier 3 is important - especially as whatever relaxation there is over Christmas will drive a spike of hospital admissions, and it's hard right now to see a way of sending the undergraduate freshers home without some associated wave of parental hospitalisations.  A lot here hinges on how sensitive the lateral flow tests are that are being piloted for rapid, large scale screening . All the work I've found calibrates them against PCR positive individuals, which proxies the problem of the PCR false negative rate, for which there are many competing values...  

>  I'm not aware of anywhere presenting city/council region or LA case rates on a daily basis rather than weekly average (in an easily digestible way.)

"Daily COVID-19 cases by age and specimen date in unstacked CSV format" from the government dashboard has this at county level on a daily basis, but with a 7-day rolling average applied (go figure).  I suppose it can be differentiated but I don't know with enough detail what the averaging process is....

[1] https://www.ukhillwalking.com/forums/off_belay/more_covid_bants-727477?v=1#x93...

 Si dH 07 Nov 2020
In reply to wintertree:...

> > Tier 1 has been shown to be inadequate everywhere.

> Agreed - I'm still drawn to jkarran's observation that the rise in R travelled with the seasonal weather change spreading south; in the end everywhere ended up on an exponential track.

I hadn't seen that, there might be something in it. Although, I'm probably more inclined to a view that the renewed hotspots occurred first in areas that had never been driven quite as low through summer, were unlucky to have a super spreading event somewhere locally and had high deprivation that encouraged spread (primarily due to small terraced houses with high occupancy). This was happening from August, so it doesn't feel intuitive to me that temperatures would have had a big effect.

> > So if I had a vote, it would be to exit lockdown into Tier 3 everywhere until Christmas.

> If you had that vote, I would urge you to give it another week to be damned sure...  I'm minded to hold the lockdown as prevalence is just far too high, and if thinks take a turn for the worse we're only days away from locking in enough trouble to really mess up healthcare.  With the uncertainty over the flu season as well, I think getting cases down at least 40% before exiting to Tier 3 is important - especially as whatever relaxation there is over Christmas will drive a spike of hospital admissions, and it's hard right now to see a way of sending the undergraduate freshers home without some associated wave of parental hospitalisations.  A lot here hinges on how sensitive the lateral flow tests are that are being piloted for rapid, large scale screening . All the work I've found calibrates them against PCR positive individuals, which proxies the problem of the PCR false negative rate, for which there are many competing values...  

I might have been unclear... I didn't mean to say exit lockdown immediately, I was working on the assumption that lockdown continues until 2nd December and thinking about what we should do then.

 Dr.S at work 07 Nov 2020
In reply to Si dH:

Sensibly we would contine lockdown until schools break up for Christmas - its only another couple of weeks although I appreciate that would really screw over a lot of bricks and mortar retailers.

 wintertree 07 Nov 2020
In reply to Si dH:

> I might have been unclear... 

I should have realised I misunderstood it as I’m reasonably convinced that you’re not insane...

In reply to Dr.S at work:

> Sensibly we would contine lockdown until schools break up for Christmas - its only another couple of weeks although I appreciate that would really screw over a lot of bricks and mortar retailers.

If doing so concentrates footfall in to brick and mortar venues into a smaller amount of time, it may not have such a big effect?

 Si dH 07 Nov 2020
In reply to Dr.S at work;

I think that might be necessary if there are areas of high prevalence that aren't showing a strong downward trend in the data by 02/12.  Hopefully that won't come to pass, but who knows.

If possible I think it'd be best to have a few weeks of data from some state less than lockdown before Christmas that can be used to inform decisions in early January through to spring. Christmas period itself will be bound to cause additional transmission that will confuse things so I think dropping straight out of lockdown into a few days of lax Christmas behaviour could leave us not being as sure what should come next.

 RobAJones 07 Nov 2020
In reply to Si dH:

Have super spreader events happened outside? (genuine question) 

It's getting more difficult to keep windows, fire doors etc. open in schools.

There seems to be a school of thought that having heating on and no ventilation is making spreading within the home more likely.

I was going to say sorry to be pessimistic, but on second thoughts it would had help if the people in charge weren't so optimistic.

 Yanis Nayu 07 Nov 2020
In reply to Removed User:

Anyone hearing rumours of GPs being instructed to be ready to administer vaccines by December?

 Neil Williams 07 Nov 2020
In reply to Dr.S at work:

> Sensibly we would contine lockdown until schools break up for Christmas - its only another couple of weeks although I appreciate that would really screw over a lot of bricks and mortar retailers.

You could reopen shops but not pubs, I suppose?

 Neil Williams 07 Nov 2020
In reply to RobAJones:

> Have super spreader events happened outside? (genuine question) 

None that I've heard of.

 wintertree 07 Nov 2020
In reply to RobAJones:

> Have super spreader events happened outside? (genuine question) 

Trump’s recent rose garden event at the White House.  Lots of hugging, cheek kissing, hand shaking and so on.

https://www.google.co.uk/amp/s/www.bbc.com/news/amp/election-us-2020-544871...

 Dr.S at work 07 Nov 2020
In reply to Neil Williams:

Yeah maybe - I’d prefer to go as hard as we can for as long as we can. Two weeks of gleeful abandon at Christmas then back on it.

 RobAJones 07 Nov 2020
In reply to wintertree:

I actually thought about that just after posting, but was the whole event outside? At least half the people infected will have been in close contact with each other on a regular basis.

 Si dH 07 Nov 2020
In reply to RobAJones:

This article is quite interesting, no hard conclusions but some good discussion of this topic about 2/3 of the way down.

 Si dH 08 Nov 2020
In reply to Si dH:

> This article is quite interesting, no hard conclusions but some good discussion of this topic about 2/3 of the way down.

Argh.

https://www.vox.com/21296067/coronavirus-covid-white-house-superspreaders-s...

 RobAJones 08 Nov 2020
In reply to Si dH:

Thanks I'll read it this evening. I'll try and find a Chinese report I read in May. It tracked a few hundreds outbreaks. Of the thousands of points of transmissions only one was outside, as you say it isn't proof of anything but it did affect my behaviour in the summer.

 wintertree 08 Nov 2020
In reply to RobAJones:

> I actually thought about that just after posting, but was the whole event outside? At least half the people infected will have been in close contact with each other on a regular basis.

I've not seen a definitive study; but the transmission was so widespread the outdoor grouping seems implicated.  I imagine we'll get another one from his election night party.

Outdoors is one of many counter-measures to transmission.  No measure alone is sufficient; so even outdoors should be approached with some common sense on physical contact and distancing it.  I wouldn't want to spend an extended period of time up close and shouting face-to-face with an infected person outdoors, for example.

 wintertree 08 Nov 2020
In reply to Removed User:

Re: your YouTube video link.  This is it's description:

Graham Hutchinson, ex-Senior Chief Biomedical Scientist, Public Health UK, fully endorses the views of Dr Mike Yeadon and has an important message re: masks and the influenza vaccination

  • Who is "Graham Hutchinson"?  Googling for him returns this video, some Facebook profiles and some football links.   
  • What is "Public Health UK"?  There is no such Executive Agency or company in the UK and there never has been.
  • Searching for both terms together only returns this video and re-postings of it through various things like twitter.

This appears to be a random man in a study somewhere being blatantly misrepresented as an authority figure from an agency that doesn't exist. 

  • Shady as hell - it would be interesting to find out who the person in the video is and what their background is.

There is a "Royal Society for Public Health UK"

  • This is a charity - https://www.rsph.org.uk
  • Their biography (https://www.rsph.org.uk/about-us/who-we-are.html) includes executive positions, directors, members of a Council and a treasurer.  It's doesn't include titles relating to scientists or biomedical roles.  
  • I can't for one minute imagine anyone pushing this would have left out "Royal Society" from their PR push.
  • I will contact the charity and ask about any association, and update the thread if/when I hear back.

OP - perhaps the next time that you get sent a YouTube video, you might take a moment to research its author and check any claimed credentials, so as not to amplify the spread of what appears to be deliberate misinformation.    

1
Removed User 08 Nov 2020
In reply to wintertree:

His introduction covers this. He was a 'senior' scientist (whatever that means) at the Public Health England laboratory in Colindale. No Google scholar profile, I'll check Scopus later. Not necessarily unusual but removes weight from his polemic in my opinion if he has no published research track record.

You probably need to watch the video if you are going to comment further on this guy.

Post edited at 13:06
 wintertree 08 Nov 2020
In reply to Removed User:

So not "Public Health UK" as given in the description.  Good accuracy of the bat here...  

I don't watch YouTube videos, it does my nut in and rarely adds anything.  I found a transcript that I will not link as it's on an anti-vax website (which does rather add to the picture this is painting...) but people can search it if they want.

In the transcript he says

A little bit about myself, I really don’t like talking about myself, but who does? I was one of the Senior Chief Biomedical Scientists at the Centre of Public Health laboratory at Colindale. I worked in the Division of Hospital and Respiratory Infection and Anti-microbial Reference Unit. 

This is a genuine lab, and the kind of role he claims to have held is genuine so he could be authentic.  It's not what was presented in the title of the video...

The transcript is full of the usual same-old same-old bullshit.

We’re seeing a lot more cases and a lot less deaths all over the world

No evidence presented for this.  No case made.  Just an un-evidenced claim. 

and in my view COVID-19 is not the same organism as it was when it was first isolated and has now very low pathogenicity. 

Any half arsed examination of IFR estimations will quickly show that there is no evidence for the virus having become less lethal which is the picture he's painting here.  Demographics have shifted in good times and are shifting back, CFRs have changed wildly as a result of improved testing, but IFRs - I'm not aware of any credible evidence that there's been a substantial change excepting the effect of improved healthcare practices and treatments, and when compared to regions that had had healthcare overload.

I would recommend everybody to go to and look at a website called https://nextstrain.org and just look at the numbers of genomes and what’s going on is there are thousands and thousands now of different strains of COVID-19. 

The classic YouTube Gish Gallop - look over there!  Science!  Stuff!  He's presenting something that looks impressive without interpreting it at all, and without any scientific context.  Like noting that almost all them mutations apparently being in non-coding DNA and not affecting the virus' functioning and therefore potency. 

I could go on but the whole transcript is full of nothing more than the usual stuff that's been going around for 9 months backed up by a call to authority but no evidence what so ever.

Talking heads putting forwards words without presenting and interpreting any evidence are just noise.

>  No Google scholar profile, I'll check Scopus later. Not necessarily unusual but removes weight from his polemic in my opinion if he has no published research track record.

I couldn't find anything.  The role he says he held appears to be a high grade lab manager type position going off a past recruitment advert I found.  

Post edited at 13:31
In reply to wintertree:

"Senior Chief".

Really...?

In more hopeful news, I've just been looking at the figures for Belfast; cases are dropping as fast as they rose, after a plateau. Provided it's not some reporting anomaly, it shows that it is possible to bring the spread under control relatively rapidly. Looking at their restrictions, they don't seem particularly unusual. NI is the only region with R under 1. Whatever it is they're doing (better compliance, perhaps, after having had the shit put up them?), It looks like it's working.

 wintertree 08 Nov 2020
In reply to captain paranoia:

Better than being a Deputy Senior Chief Biomedical Scientist I should imagine.  I interpreted the "Senior" as a personal addition to indicate that they considered themselves one of the more important of the multiple people in the "Chief" role.  But who knows, I didn't give it more than a passing thought...

> I've just been looking at the figures for Belfast; cases are dropping as fast as they rose, after a plateau. 

One thing Neil Williams pointed out yesterday is that if you put the pessimist's hat on, the effect of the school half-term holidays will be feeding through to the data that we see right now, giving a double-whammy with places that had gone in to higher control measures before then; some of that progress may be transient with the return to schools.

It is good news that cases are dropping at that sort of speed - do you have a link so I can estimate the half-life?  It'd be interesting to compare to the first time around.   I think it'll be similar from what you've said.

 elsewhere 08 Nov 2020
In reply to wintertree:

> > Have super spreader events happened outside? (genuine question) 

> Trump’s recent rose garden event at the White House.  Lots of hugging, cheek kissing, hand shaking and so on.

I wonder about that one. TV coverage was outdoors but I did VVIP event really have no reception room/marquee?

In reply to wintertree:

> do you have a link so I can estimate the half-life? 

Just looking at the BBC covid stats page, not the raw numbers. Using their postcode search function, with BT1.

https://www.bbc.co.uk/news/uk-51768274

[edit: looks like the numbers have been updated since earlier: the drop was ~900]

Post edited at 15:08
 Neil Williams 08 Nov 2020
In reply to wintertree:

And if you look at today's 7 day average (which smooths out the messy graph) it's heading up again with a noticeable dip that appears to last for...you guessed it...half term.

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

Why won't the Government grasp the nettle and do something about schools?  It is clear that we must get distancing set up there, even if that means spending a fortune.  It might even be cheaper than closing everything else.

Post edited at 16:38
 wintertree 08 Nov 2020
In reply to thread:

Updated deaths plot with another day of data in it below.

I still see no support in the data for the OPs claim, embodied in their question "can someone explain why the death rate is rolling off over the last week?".

In the attached plot, an exponential is fit to the last 14 days with reasonably solid figures (small reporting lag effects).  This is 3 more days data than when I first addressed the OPs question.  The "roll off" they referred to in the government's plots is now gone, replaced with rising cases.  There is a new roll-off (in the dashboard plots) on days not yet reported when they posted.  Both are an effect of the reporting lag, nothing more.

If the data was "rolling off", there would be multiple data points to the right hand side of the plot that all fall below the exponential fit - this is, residuals to the right hand side would not be random (as expected from statistical noise) but would all be negative, indicating that the exponential model is a bad fit.

The "deaths within 28 days of positive test by date of death" plot on the dashboard always shows a rolling off, due to the reporting lag of recent cases.  I consider it poor form that the plot does not either truncate at -5 days or have an annotation indicating the provisional nature of the data.  The "deaths within 28 days of positive test by date reported UK total" plot on the dashboard shows a drop in line with last weekend's drop - due to the reporting lag.

Post edited at 16:49

 wintertree 08 Nov 2020
In reply to Neil Williams:

> And if you look at today's 7 day average (which smooths out the messy graph) it's heading up again with a noticeable dip that appears to last for...you guessed it...half term.

Yup; there's a similar up-tick in my version of the plot below.  

What bothers me about this data is the way the upper and lower envelopes of the data appear to be following two different trajectories with the "weekend" data levelling off and the "week day" data continuing to rise.  It makes me think that perhaps something is saturating testing at the weekend.

>  It is clear that we must get distancing set up there, even if that means spending a fortune.  It might even be cheaper than closing everything else.

I'm reasonably convinced.  I'm going to pour through the most recent week's "outbreak" data in the PHE reports when I have time.


 Neil Williams 08 Nov 2020
In reply to wintertree:

Another thought I had is whether it might, slightly paradoxically, make sense to vaccinate children first, if that vaccine might take them from "mostly asymptomatic" to "not even capable of spreading it".  That's another trial to do, though.

The thing about kids is that the amount of grandparent childcare is massive, and you're not going to stop that, so you need to consider that a younger child with it pretty much guarantees a grandparent getting it.

Post edited at 16:57
 RobAJones 08 Nov 2020
In reply to Si dH:

Thanks for that. I have read that the value of k had led to other countries (Korea, Japan) approaching track and trace slightly differently. Rather than just getting your contacts to isolate, they were interested in trying to work backwards to identify super spreaders.

 RobAJones 08 Nov 2020
In reply to wintertree:

I wouldn't want to spend an extended period of time up close and shouting face-to-face with an infected person outdoors, for example.

I don't think anyone would, but how often does that happen? Even before covid I'd guess that those type of exchanges would be between people who live together. Most shouting isn't directed in someones face, so it is much better done outside BLM protests vs singing inside.

even outdoors should be approached with some common sense on physical contact and distancing it

Again nothing to disagree with, but I'm not sure how much common sense is out there, even amongst "intelligent" people. Eight maths teachers in a small office with all the windows shut??  It's my view that "normal" behaviour outside mitigates this lack of common sense, to some extent.

For example how many people (even those with no common sense) shout/talk loudly in a beer garden compared to inside a busy pub?

 RobAJones 08 Nov 2020
In reply to Neil Williams:

It's only anecdotal but I've seen behaviour (in regard to covid) decline over the last five weeks. Mostly staff not students.

 RobAJones 08 Nov 2020
In reply to Neil Williams:

Another problem with proposing to vaccinate children will be the anti vax brigade. No long term studies and what is the benefit for my child?

 Neil Williams 08 Nov 2020
In reply to RobAJones:

Yeah, the UK one is just a "phone a friend" service that you can just do yourself, so a complete waste of time.  And because it isn't investigative, while it's illegal to give a false declaration nobody would know or have any way to know if you did.

Removed User 08 Nov 2020
In reply to wintertree:

Why are you only fitting 14 days. I'm not clear on your methodology here?

 wintertree 08 Nov 2020
In reply to Removed User:

> Why are you only fitting 14 days. I'm not clear on your methodology here?

I am testing your hypothesis from your OP: “Finally can someone explain why the death rate is rolling off over the last week?

By doing an exponential fit over two weeks and looking the residuals I can see if the most recent week (the one you refer to) is different from the previous week (what went before).  The random nature of the residuals tells me there is no significant difference in how well the two weeks are described by the same exponential.  So there is no support for a roll-off.  

My 16:48 Saturday post does the exponential fit over a sliding 14-day window and shows the doubling time of the exponential  vs date, so covers a larger period of time.  This shows the death rate has been slowly slacking off to longer doubling times.

If you’ve got a more specific methodology question let me know.   I tend to use windows an integer multiple of 7 days for consistency with cases where there is a clear 7-day periodicity and so an integer multiple of 7 days is required to minimise aliasing of the periodicity.

 RobAJones 08 Nov 2020
In reply to Neil Williams:

I think out track and trace was doomed when Matt Hancock said we would have 100000 test by the end of the month (or something similar) After that the focus was on increasing alleged capacity by whatever means, rather than developing  system that effectively prevents as many outbreaks as possible.

Removed User 08 Nov 2020
In reply to wintertree:

Got it. Ta.

 Dr.S at work 08 Nov 2020
In reply to captain paranoia:

> "Senior Chief".

> Really...?

Yup, Seal Team Six apparently!

 wintertree 08 Nov 2020
In reply to Neil Williams:

> And if you look at today's 7 day average (which smooths out the messy graph) it's heading up again with a noticeable dip that appears to last for...you guessed it...half term.

This is my estimated doubling time plot with the two common half-term periods (19-23 Oct and 26-30 Oct) blocked out.

The right hand side of the curves - cases and deaths in particular - are very "twitchy" and could change a fair bit over the next week with retrospective data releases due to reporting lag.


 wintertree 08 Nov 2020
In reply to Neil Williams:

> Another thought I had is whether it might, slightly paradoxically, make sense to vaccinate children first, if that vaccine might take them from "mostly asymptomatic" to "not even capable of spreading it".  That's another trial to do, though.

The problem is that the benefit of the vaccine to the child's direct health is basically 0.  It carries a risk - the risk of the unknown, as there's no longitudinal data over a reasonable period yet (nor can there be).  So whilst there's an argument to vaccinate children first, it's not on medical grounds and I don't think the situation is dire enough to over-rule medical best practice. 

It seems a lot of stuff is pointing to super-spreading being a key component of this pandemic, with figures like 10% of carriers infecting 80% of people being bandied about.  I've not dived in to the details mind.  I'm not sure children are well placed to be these super-spreaders, so I'm wondering how much of any "half-term" effect is actually through the indirect confinement of the parent(s) to child-care away from the usual work/socialising/etc.

 Toerag 09 Nov 2020
In reply to Removed User:

For what it's worth, my rough&ready 'live cases' calcs give a steady reduction in the rise in live cases as a percentage of the total i.e. the rate of rise since 24th October (down from 2.77% to 2.04%). This would imply things are improving, however, in terms of the actual numbers of live cases there is no drop, it's a steady 22-23k cases per day since the 27th. Thus I doubt things are going to improve much in the short term as far as hospital admissions and thus deaths are concerned.

In reply to Removed User:

I have a question, or more accurately, a few.

I have, shamefully, found myself getting wrapped up in more a more arguments lately, with anti-mask, COVID denying idiots online. Typically on FB, and more usually on BBC news feeds and Govt announcements. There are many single posters but many are regular contributors on many sites. Ive become so wrapped up in all this I have lost sight of what's going on. 

Who are all these people?  Why are they doing this? What do they have to gain in denying COVID and its affects. Why do they think it's a hoax, not deadly etc? Are they left, right, neither? Do they really believe what they are saying or has the world turned into mischievous trolls?

When I call them out you get called a sheeple,  a paid tool of a brainwashing government, despite really despising this government. 

I really need to stop reading the crap.

2
 Neil Williams 09 Nov 2020
In reply to TheDrunkenBakers:

I think it's generally human nature, if something bad is happening, to try to pretend it isn't.  Probably just that in most cases.

 Blunderbuss 09 Nov 2020
In reply to TheDrunkenBakers:

Engaging them is the very definition of futile experience......even if you clearly demonstrate they are lying they will ignore the point you made or claim your data/evidence is made up.

Basically anything 'official' is a lie so from the very outset a rationale discussion is off the table.

Best ignored.

 Offwidth 09 Nov 2020
In reply to Removed User:

Sorry I missed your reply. The physics is really all you need. A mask greatly reduces droplet spread and disrupts aerosol spread. It is absolutely clearly a benefit to others around the wearer. I agree any effect of mask wearing on reducing social distancing could be important but tell me where you see people indoors who are less than one metre for many minutes where a mask is required.

Too many people seem to be muddling the reasons for public mask use with the opposite science (does mask wearing protect the wearer) which is important for medical staff and although not proven yet for covid has strong evidence from other infectious disease. Nearly alll the stuff about touching masks is with reference to protection for the wearer.

 wintertree 09 Nov 2020
In reply to TheDrunkenBakers:

Never read the comments.  The only place I read the comments is on "Arstechnica", where the audience is smart, and where there's a downvoting system that hides the worst of the comments before they derail the informed discussion.

> Who are all these people?

A handful will be agents spreading misinformation (keep an eye out for claims of empty hospitals, as told by "a doctor friend", for example).  The rest are just people who have got drawn in, don't like reality, and think reading a view blog posts and watching some YouTube videos is "research".  You're best of moving on and not getting emotionally invested.  

If anyone was wondering why I'm so rabidly against people shitposting obvious crap about Covid on here, it's because I consider that you have seen the future that happens if we let the thin end of the wedge in.

> When I call them out you get called a sheeple,  a paid tool of a brainwashing government, despite really despising this government. 

I've had a lot of flack on here over Covid - apparently "all [I] want to do is lock everyone down", I've got this agenda, I've got that agenda, I'm for a tory(*) conspiracy, I'm anti-tory, I'm a batsh*t crazy conspiracy theorist, I'm the embodiment of Dunning Kruger, I'm an armchair expert.  (I'm not including comments from the obvious troll and enemy accounts...)  It seems many people have their own entrenched views and will project their usual bogeyman on to anyone who disagrees with them.  Only a small fraction of people engage with the intent of understanding, learning and developing.

(*) I wish there were some tories so I could be pro- or anti- tory.  They're just the hollowed out wreck of a party left by the ideologically purges.

3
 Offwidth 09 Nov 2020
In reply to wintertree:

Yeah well. In contrast I think you are the UKC poster of the year. A lot of the angry nuts also make the site a less pleasant place than it should be. The forums used to be more of a community and less of a debating chamber. Your contributions are directly community spirited.

 wintertree 09 Nov 2020
In reply to Offwidth:

Everyone needs a hobby...  I was pretty annoyed at the approach to Brexit on here over the last few years but almost nothing was clear cut.  Since March there have been objectively clear misrepresentations or misunderstandings of data about the Covid situation. 

 Stichtplate 09 Nov 2020
In reply to Neil Williams:

> I think it's generally human nature, if something bad is happening, to try to pretend it isn't.  Probably just that in most cases.

The hardcore nutters aren't pretending something bad isn't happening. What they are pretending is that covid is the prologue to something far worse is happening.

Removed User 10 Nov 2020
In reply to TheDrunkenBakers:

I've been debating hard with my Dad on Whatsapp this last week. I discovered to by disdain/digust that he has a twitter account and has been retweeting all kinds of conspiracy bollocks.

I have being trying to convince him why the hypothesis that false positives are driving this second wave - he is convinced this is the case and we have already reached herd immunity as per Yeadon's arguments.

He has previously been a rational person and is mathematically literate but seems to be having trouble accepting that the true false positives rate was shown to be low over summer and also that it is not possible for a constant (false postive rate) to change a linear growth in test carried out to an exponential function describing the increase in postive tests.

My advice to you is look hard on the stuff posted on the twittersphere - it just doesn't stack up

I'm still battling away...

Removed User 10 Nov 2020
In reply to wintertree:

Any idea why testing rates seem to be levelling off - even though capacity appears to be rising? Does testing data show a similar 5 day lag to death data?

Why is that correlating with a levelling off of case data - coincidence?

 wintertree 10 Nov 2020
In reply to Removed User:

> Any idea why testing rates seem to be levelling off - even though capacity appears to be rising? Does testing data show a similar 5 day lag to death data?

I have't been doing the daily downloads needed to look at the status of reporting lag in the reports for the number of tests actually taken; I'd naively imagine they're actuals but I don't know.  The number of tests actually processed reflects demand - through some sort of buffer representing under-sampling at weekends - so if cases are levelling off, tests processed will as well.  

The detected case rates reported by specimen date always has a reporting lag.  It varies quite a lot, but generally the most recent 3-4 days of data are substantially incomplete.  You should be able to tell by the annotation on the government plots, but there isn't any which makes it open to easy misinterpretation.

Re: arguing false positives with a parent - my extensive experience of arguing anything with Dad was that it was futile....  Less flippantly, the key thing to demolish the claim from Yeadon that there are false positive cases, admissions and deaths going on are the combination of the many credible news reports on ICUs and ward filling up in high covid areas combined with the very low prevalence of influenza as reported by PHE in their weekly combined covid and influence surveillance reports.  So, what is putting people in to hospital and ICU that is allegedly being misattributed to covid by false positives?  It's not the flu season...  Without a credible answer to that, the whole tower of cards falls down.

Post edited at 11:55
Removed User 10 Nov 2020
In reply to wintertree:

I've been supplied with this to critique http://inproportion2.talkigy.com/dashboard/

I have demonstrated to him that the deaths were rising exponentially (he denied this) by plotting a fit to the data myself. I also note on this chart that the roll off in deaths (I think this is now emerging clearly BTW) tallies with a lag from mid October when Tier 3 restrictions are in place. He is far to focussed on magnitudes than the underlying trends.

What I do not understand is why the linked charts plot deaths and cases as % changes?

 wintertree 10 Nov 2020
In reply to Removed User:

>  (I think this is now emerging clearly BTW) 

It may be.  Things always look better after the Sunday and Monday data releases as the reporting lag from the weekend takes a while to clear...  But it is starting to look real compared to on Friday.  It is happening at about the right time after hospitalisations started to slack off.

> What I do not understand is why the linked charts plot deaths and cases as % changes?

Who knows. Everyone has their own preferred way of plotting and interpreting things and this suits the way of thinking of whoever made it I expect.  

Day on day % change is an approach to the problem of visualising the underlying exponential mechanic and the exponential rate.  That rate is a far more useful measure in terms of evaluating how effective legislation and messaging are or aren't on the spreading of the virus than absolute numbers.  Absolute numbers are a key measure of how screwed healthcare is; horses for courses.  As with the doubling time plots I've made it, the % change curves going down just means "things are getting worse less quickly" rather than "things are getting better" so can be subject to misinterpretation.   Day-on-day % is more bounded than doubling time (which is some constant divided by that change) which tends to either infinity; I prefer the doubling time as it can be more readily linked to real world consequences but the % is just as good and doesn't need a log-y axes when rates are changing a lot.

Post edited at 12:25
Removed User 10 Nov 2020
In reply to wintertree:

I think that is key. To the layman a flat line on the percentage change plot look constant but actually represents exponential growth, and as you say a down tick is merely a reduction growth rate.

 wintertree 10 Nov 2020
In reply to Removed User:

Yup - it's very easy for someone who has worked with plots enough to develop an intuitive reading of them, to completely misunderstand this class of plot.  It's not an easy set of data to communicate effectively without banging on until people loose interest...

 wintertree 10 Nov 2020
In reply to Removed User:

> I also note on this chart that the roll off in deaths (I think this is now emerging clearly BTW)

Today's death tally is the highest since the first "wave".  This is the "by reporting date" figure and includes some of the delayed reporting from the weekend.  Tomorrow's is likely to be similarly high I think.

Generally, record high tallys by reporting date don't happen when actuals are falling.

Right now, plotting the data resolved by specimen date (below) with a 14-day exponential fit to the reasonably firm period (not the most recent 4-5 days) now looks like it's flattening a bit - seen by the sign of the residuals having some correlation meaning they're not random, and the shape of them suggesting more of a quadratic topping out.  But the right most point(s) could move up with tomorrow's data release.  

I still think a levelling off is coming to this to lag the hospitalisations (themselves levelling off ~ 1 weeks ago), and compared to Friday the data now has week support for this but - especially coming after a record high daily tally for this "wave" - I'm going to sit on the fence until tomorrow's data is out.


 Neil Williams 10 Nov 2020
In reply to wintertree:

Deaths are lagged about 3 weeks, though (they can't be lagged more than 4 weeks due to the 28 day cut off for reporting), so if cases level off now (as they seem to be doing) then deaths will continue to increase for the next 3 weeks, which is sobering as that means we will likely get over 1000 a day again before they come down.

 wintertree 10 Nov 2020
In reply to Neil Williams:

I think it's more optimistic than that.  I've put my latest plot of doubling times below.

I'll pick a doubling time of 100 days as an indication that a measure has levelled off - or as good as.

  • Cases hit this around Oct 23rd - although cases data is the lowest quality
  • Hospitalisations hit this doubling time around 11 days later on Sep 1st.  This is a more relevant and trustworthy measure than detected cases.
  • Deaths have their doubling time increasing as well, lagging hospitalisations.  The right side of the deaths curve in particular is twitchy and may get pushed down (bad) by tomorrow's update, but it looks on course to level off around November 12th
    • This is very soon - but we won't know for sure until Nov 17th due to reporting lag.
    • I'd estimate the level being around 350 deaths/day.

But - have cases really levelled off?  The doubling time is apparently dropping, and a plot of cases by specimen date has a corresponding uptick on the right hand side.  Last Monday was a record high following what looks like a saturated level of sample taking over the weekend.  This Monday's data won't be fully resolved until Thursday or Friday.  What happens next really depends on how much of the slacking off was due to T2/T3 and how much was due to school half terms.  If it was the former, the lockdown should rapidly take over moderating cases, if it was the later, we'll see cases rising again I think.  

ONS suggested a levelling off in the high prevalence areas last week (or two weeks ago by now) and ZOE  currently suggests infections are actually decreasing which isn't reflected yet in the testing case numbers.   The latest REACT update doesn't fit very well with these.  The lower prevalence regions were on track to take over driving cases before lockdown was announced.  The next MRC nowcast should be out soon.  I think there's a massive amount of uncertainty around cases right now - not surprising as things are fast changing.  

Post edited at 20:01

 Si dH 10 Nov 2020
In reply to Removed User:

I've just been doing a bit more qualitative analysis at the latest Covid case data by area, if anyone is interested. The data is now up to date for specimen dates up to 05/11, ie the day we entered lockdown.

All the interpretation below is based on seven-day average figures, which are necessary to remove the noise and be able to draw any conclusions without doing lots of stats, but do delay things a bit. The end of the latest 7 day period available being 05/11.

The UK wide figures almost flattened during the week leading up to lockdown but this gives a misleading picture of the cases in any one place.

The infection free rate in Northern Ireland has been dropping fast since mid October.

The infection rate in Scotland dropped for a week at the end of October but then started rising again through the first week of November.   I can't remember the exact dates of various Scottish interventions so won't speculate why that is. It's not countrywide - eg the uptick was strong in Glasgow but the fall continued in Edinburgh.

The weekly average infection rate started dropping in Wales from 02/11 and the rate of fall increased every day from there up until 05/11. So it now looks like the lockdown is clearly having a significant effect. Only two local authorities in Wales don't show a fall in the week and one of those, Ceredigion, had some sort of major spike on a single day which must have been some sort of significant workplace outbreak.

The infection rate in England shows a continued small increase but this masks significant variation.

In the North West the weekly average infection rate has been decreasing since mid-late October and the rate of fall increased slightly from ~02/11. This is nice because it suggests we are seeing the effect of Tier 3 in GM (which began on the same date as the Welsh firebreak, see paragraph above.) If you look at individual LAs in the north west, all those in Liverpool and most of Lancashire continue to fall (excl Preston and Burnley.) All of GM is either up a bit or down a bit, the biggest week on week change is about 11%.

In London there was a bit of a dip at the end of October and then a rise again but the rate is basically the same as it was two weeks earlier.

In all other regions of England the rate was continuing to increase but this was mostly masked by the decrease in the north west.

Even in the North East, after a small dip and flattening through the second half of October the rate took off again in the first week of November.

All of other areas in England were on continued upward trends, some very steep as of 05/11. Average rates in the East Midlands, West Midlands and the North East were all on track to pass the North West within another week (ie, by now or in a couple of days time, if lockdown hadn't happened. The data will probably show this in a few more days before the lockdown effect kicks in.) Yorkshire and Humber was already out in front but Hull and Grimsby had overtaken everywhere in Yorkshire as the worst hit parts.

Lots of cities in previously lesser hit areas had overtaken those under the tighter restrictions already by 05/11. Derby and Leicester both had higher rates than Nottingham. Bristol and areas of Birmingham/WM higher than Liverpool, most of Lancashire and even the southern end of GM. Hull and Grimsby worse than everywhere in the country except Blackburn and Oldham. Rates are also increasing markedly in rural areas across the country but obviously not to the peaks in the cities.

What all of this shows is that the regional restrictions have worked in some areas but that the overall centrally-managed regional approach failed once rates started going up generally because it wasn't clear where things were going to take off next. In practice to catch the recent rate increases in Hull, Derby, Bristol or parts of the WM you would have had to put them under Tier 3 restrictions 2-3 weeks ago when their declared infection rates were still quite low and not obviously differentiated from a number of other areas.

I think there is good news in the Welsh and North West figures and lockdown should work. It will be interesting to see which areas actually have the highest rates come the end of lockdown though and what that means for how it is released (my money is on Hull or Grimsby as their rates will have further to rise in the data over the next few days.)

I am also more convinced than ever that using regional or national data is generally useless for determining how the pandemic has been progressing in a given city (apart from in Wales and NI), which unfortunately includes all the data we have on hospitals and deaths. But hopefully the lockdown will put everyone on a more similar path again.

Post edited at 20:35
 wintertree 10 Nov 2020
In reply to Si dH:

Thanks for the detailed breakdown.  Not as optimistic as I was hoping when I started reading.

I think even regional level data is very misleading where there has been a significant university outbreak.

To go with your post I've made a plot of the 7-day rolling sum from the UTLA demographic breakdown from the UK dashboard, valid up November 5th.   I've pulled out demographic plots for the worst 6 regions (by the most recent count) and 6 that I identified by eye that had past spikes.  Some of the "past spike" regions are driven by university ages with the spikes clearly having given a false impression of stable or falling cases for Durham, Newcastle and Nottingham.

Post edited at 21:39

 Blunderbuss 10 Nov 2020
In reply to wintertree:

Do you know what the mean time from infection > hospitalisation and hospitalisation > death are? 

 Si dH 10 Nov 2020
In reply to wintertree:

I agree they gave a false impression for a while but at the same time they were very important in spreading infection to new areas that had previously been at low prevalence. I think that effect is probably now all washed through though?

Isee Govt are planning mass testing to avoid any major impact when students go home for Christmas but it strikes me they'll need another mass testing programme before they all go back again in January!

Post edited at 21:56
 wintertree 10 Nov 2020
In reply to Si dH:

We’ve been told in no uncertain terms that there’s no evidence of our outbreak spreading to the community...  I think for some of the ones I plotted it’s only just ceasing to distort the county and regional level figures - the north east was unlucky to have 3 major university outbreaks.

Mass testing for students - what’s the false negative rate?  The only data on lateral flow tests I’ve seen is validated against RT-PCR confirmed positives.  So there’s no validation against RT-PCR false negatives, such as newly infected students who will then go on to become highly infectious after getting home.  Screening them is better than doing nothing, but they’ll want some clarification about what a negative is, or is not, guaranteed to mean so it doesn’t sow false confidence.  

I really feel for the students, they’re having a shit time of study and are locked down to tiny rooms and are getting a lot of stick from locals over what was almost unavoidable.  Shared bathrooms for christsakes.

 wintertree 10 Nov 2020
In reply to Blunderbuss:

> Do you know what the mean time from infection > hospitalisation and hospitalisation > death are? 

Not with any great certainty.  It’s very demographically dependant and the demographics have been constantly shifting - hence the lack of correspondence between hospitalisations and deaths.  Any attempt to deconvolve the curves to get a distribution fails pretty badly.  I assume this data is gathered on a per-case basis centrally but isn’t openly published...?  

12 days for each step seems reasonably compatible with the plots but that’s not really science, more navel gazing.

 Neil Williams 11 Nov 2020
In reply to Blunderbuss:

> Do you know what the mean time from infection > hospitalisation and hospitalisation > death are?

Isn't it roughly:

Incubation period 1 week

Hospitalisation (if it's going to happen) 1 week later than incubation

Death (if it's going to happen) 2-3 weeks later than hospitalisation

 RobAJones 11 Nov 2020
In reply to wintertree:

Thanks for the graphs, they are now going to feature in a lesson plan. I also worry about the false positives, are they going to be isolated over xmas along the lines of the tourists in Italy?

 Neil Williams 11 Nov 2020
In reply to RobAJones:

If they're doing the tests on the 3rd they will be "released" on the 13th.

 wintertree 11 Nov 2020
In reply to RobAJones:

If you talk about the graphs perhaps ask them what they think of the colour map - it was carefully chosen as the colours change more rapidly at low values, so it gives visual dynamic range to small changes in small values and large changes in large values.  I like engaging students to think about the choice of colour map.    I've put a copy in below in a grey colormap - now the foothills of August/September are not at all clear.  It shows how a small tweak to the presentation of data can make a massive difference to people's ability to interpret it.  This can be used to help get all the details across or to mislead...

Particularly on the lower prevalence regions in the non-linear colormap plot you can see how earlier on there were bursts of infection that were wrapped up with the help of the then-useful test/trace/isolate system, but that's not noticeable on the grey maps.

> I also worry about the false positives, are they going to be isolated over xmas along the lines of the tourists in Italy?

The university students?  Given the way the LFT works as a self-contained, self-administered test, I think the risk of the cross-contamination or reagent problems that generates false positives in the existing RT-(q)PCR tests will not be a factor so I'd expect the false positives to be very low.  

As I understand it, they're going to encourage students to go down from term well early at the start of December - this leaves time to isolate and then re-test someone who gets a positive result, false or true, in time to send them home for Christmas. 

The bigger risk I think is from false negatives - in particular from students infected just before they test and get released to go home.  

If I was 15 years older and hence had a kid returning from a university halls of residences (*) and was at much higher risk, the offer would be that they could come home and isolate in the garden shed for 2 weeks - with a full meal service provided, and it is insulated and has a flushing loo and shares the fibre broadband...    Significantly larger and better equipped than the halls they've been locked down in, and an emergency kitchen can be assembled in a couple of hours...

Lateral flow test or PCR test, the false negative rate would put the wind up me at the age I'd then be. 

(*) I would have strongly encouraged them not to go, mind, and this return deal would have been explained before they choose to go...


 RobAJones 11 Nov 2020
In reply to wintertree:

As a student, I would have been grateful for the excuse not to go home at Christmas, but appreciate that makes me odd. I am now regretting, explaining to my mum, why at the time I thought it was on balance good thing (with regard to a family xmas), that my niece has tested positive in her second week at Warwick (mild symptoms). 

Removed User 11 Nov 2020
In reply to wintertree:

Isn't the false negative rate estimated at around 30%?

 wintertree 11 Nov 2020
In reply to Removed User:

> Isn't the false negative rate estimated at around 30%?

That's the impression I got the last time I trawled the literature for RT-PCR - starting at 100% false negative just after the moment of infection, dropping down to ~30% around the time of the onset of symptoms, and then rising back up to over time.

The lateral flow tests (LTFs) being prepared for university evacuation have I think been measured against RT-PCR confirmed positives, so we don't know what their accuracy is for RT-PCR false negatives.

To know the true false negative rate of LFTs would need a longitudinal study of the people tested by them to determine which of the LFTs negatives actually had covid by recording and studying their symptom history and looking at seroconversion by measuring seroprevalance before and after the testing etc.  I don't think there has been time between the LFTs being rolled out for trial at select universities and today's announcement for the seroconversion to finish...   So I don't think that study is there.  

I think it's a reasonable assumption that the LFTs are less sensitive than the RT-PCR tests, so the number of people being falsely cleared could be quite significant.  They've announced plans to test twice over a 5-day period which helps to hit the "optimum" window for testing each person, but this only really works if isolation is very robust during the extended testing period.

Post edited at 11:05
 jkarran 11 Nov 2020
In reply to oldie:

> Might not the fewer deaths be due to a lower initial viral load in many cases, due possibly to increased restrictions/precautions?

I've never quite been convinced by this viral load argument, I can't help but feel it should be quite easily proven or its significance quantified by studying infections in quarantined households. For a big enough data set we can assume the first person getting ill likely gets a small dose in the wider environment, certainly a random (average) dose. The person living with them in (generally close quarters) isolation is continually exposed to a symptomatic sick person, likely still sharing everything with little ability to effectively isolate from each other. It seems to me, though I could be missing something, that isolation acquired infections should be notably worse than those of unknown origin if there is a dose response. I haven't heard of that being proven.

jk

Post edited at 11:29
Removed User 11 Nov 2020
In reply to wintertree:

Study here - uses the RT-PCR test as a gold standard and employs some stats that is well beyond me.

Seems like RT-PCR selectivity may remain constant over 3 weeks from infection at ~70%.

LFT sensitivity seems to go from 30% i.e. 70% false negatives  (!) in week one but outperforms RT-PCR in weeks 2 (75% selectivity) and 3 (93% selectivity).

https://www.researchsquare.com/article/rs-33243/v1

Post edited at 11:55
 wintertree 11 Nov 2020
In reply to Removed User:

Thanks for the link.  It uses a methodology beyond my comfort zone and definitely beyond my detailed understanding...  (I don't mean this in a negative way about the work, but about my failure to ever really engage with Bayesian statistics...).  

"we used Bayesian latent class models (BLCMs), which do not require a gold standard for the evaluation of diagnostics"

It doesn't look like the best start for "clearing" a whole cohort of potentially asymptomatic carriers to go home to the more elderly parents.   

 freeflyer 11 Nov 2020
In reply to jkarran:

> I've never quite been convinced by this viral load argument ...

Interesting post - I've always had the opposite feeling, that there was a lot in it, and have been mystified by the many reports of one partner / household member who tests positive and the others don't. Perhaps the degree of viral shedding may vary from person to person, in line with their symptomatic response?

I'm going to try and see if there are any systematic studies along these lines.

 jkarran 11 Nov 2020
In reply to freeflyer:

> Perhaps the degree of viral shedding may vary from person to person, in line with their symptomatic response?

That seems reasonable.

> I'm going to try and see if there are any systematic studies along these lines.

It seems to me like quite a neat simple study since you tend to get couples of comparable ages, matched socioeconomic conditions etc and it should be pretty random whether the first infection is in the male or female (for male-female) households. Who got it first, who got it worse?

jk

 wintertree 11 Nov 2020
In reply to freeflyer:

> Perhaps the degree of viral shedding may vary from person to person, in line with their symptomatic response?

That fits with the super spreader mechanic - most people aren't very infectious, and a few are very infectious.  I don't know how much of that is behaviour and how much is biology but I hope you find some studies to share...

 Neil Williams 11 Nov 2020
In reply to wintertree:

It would be interesting were we to find that (almost) all of the spreading is coming from known-infected people who are refusing to isolate.

If that is the case we need to fix the reasons, which are mostly financial - i.e. you should get full pay, no cap, if you are asked to isolate and cannot WFH, and you should also get a guaranteed food delivery, and someone arranged for you to e.g. walk your dog.

Post edited at 16:35
 HansStuttgart 11 Nov 2020
In reply to wintertree:

> The bigger risk I think is from false negatives - in particular from students infected just before they test and get released to go home.  

Isn't this pretty minor compared to the number of people (not only) students, who don't quarantaine after confirmed or suspected infection?

A basic point is that if society wants a group of people to isolate for the greater good, society has to pay them.

 wintertree 11 Nov 2020
In reply to HansStuttgart:

> Isn't this pretty minor compared to the number of people (not only) students, who don't quarantaine after confirmed or suspected infection?

I think they’re similar?  Hard to get trustworthy evidence for the later as people aren’t likely to admit to it.

> A basic point is that if society wants a group of people to isolate for the greater good, society has to pay them.

Yes; when I look at the money spent on test and trace and how little it’s achieved in terms of infection control, I’d use most of it to pay for isolation if I had a time machine/reset button and I was boss.

 wintertree 11 Nov 2020
In reply to Removed User:

> I also note on this chart that the roll off in deaths (I think this is now emerging clearly BTW)

Now the weekend reporting lag is cleared as of this evening I think a levelling off it is definitely  shown in the deaths data by specimen date.  It's clearly dropped below the exponential and it looks like it's becoming level.  The doubling times look quite consistent with the doubling times for cases rising, then hospitalisations then deaths.  Interestingly, hospitalisations appear to have totally flattened where-as cases haven't - but the cases data is subject to more reporting lag than hospitalisations...

So, all quite positive in the top level data.  Hopefully the lockdown restrictions will moderate the lower prevalence areas that were rising fast and would otherwise take over driving cases.

If this holds for the next few weeks and the mass screening programs with the lateral flow tests help further reduce numbers, I could start to feel more positive about this than I have in the last couple of months.  Still - case numbers and hospitalisation numbers are uncomfortably close to some red lines - we're one loss of control away from very bad times.

Edit - the death rate may rise a bit more; the most recent four days are omitted from my plots as they’re normally subject to a lot of missing data from reporting lag, 3 of them are already at the level of the plateau and may go up over the next 3 days.

Post edited at 18:25

In reply to wintertree:

> If you talk about the graphs perhaps ask them what they think of the colour map

I was looking at the BBC map of infection. Their colour scheme goes up to 200+, one of six colours, with a log2 scaling.

But that upper limit misses the scale at the upper end; some areas have infections >700 cases/100k

I think some further colour scheme would be useful to help see how bad some areas really are. At least one more colour would be needed (400-800), and, I suspect, two at the peak (800-1600).

Post edited at 21:11
Removed User 11 Nov 2020
In reply to wintertree:

Is there a different reporting lag with the second chart down on the dashboard (deaths by reported date)? That is showing continued exponential growth.

 wintertree 11 Nov 2020
In reply to Removed User:

> Is there a different reporting lag with the second chart down on the dashboard (deaths by reported date)? That is showing continued exponential growth.

I'm struggling to get a precise form of words for this.  There is no "lag" in the reported date plot exactly - it's just the number that is released every day.  But the data it reports is from the last week or so.  

The plot below shows the number of deaths, by specimen date, after today's data release.

  • The red overlay shows where the 595 cases reported today (595 being the height of the final bar in the "by reported date plot") contributed to the count over the last couple of weeks.  Squint and you'll see a couple of deaths landing in mid-October but most are from the last week or so.
  • The orange overlay shows where the 480 or so cases reported on November 4th fell.  This was the part of the reporting spike from two weekends ago.   
  • The “by reporting date” is both more current - as it’s the most recent data to be released and stale - because it contains data from a week ago, so changes in it reflect quite a broad time period.

I think that's the worst job I've done of trying to explain anything yet.  I'm still trying to wrap my head around exactly how changes in actuals feed through to the by-reporting-date plot; it's difficult as the "smearing function" of when specimens are reporting changes depending on proximity to a weekend etc so I can’t think of it as a simple convolution with an asymmetric kernel.  

But I think deaths from the 6th onwards will rise over the next few days, so this plateau isn't as real as it looks, more a slacking off of the exponential climb.  I'd still expect the deaths to plateau fully over the next week, maybe peaking at ~380/day.

Post edited at 22:03

 The Lemming 11 Nov 2020
In reply to wintertree:

This thread is too long to read, and Covid Deniers wind me up, for obvious reasons.

Here is a cheery bit of news, my local hospital, right now, has zero free beds.

Zero.

I'm waiting to hand in over a positive confirmed Covid patient and there is no bed.

A cardiac arrest patient is inbound. There is no bed for that patient.

Stay safe everybody. And stay away from Covid Denying nut jobs.

I called one an idiot on youtube land.

Post edited at 22:40
Removed User 11 Nov 2020
In reply to wintertree:

I'm struggling to see the benefit of the 'by reporting date' data. It is supposed to be more current yet at the same time it is not! Furthermore the lack of methodology (which else where is quite good) is not clear.

Removed User 11 Nov 2020
In reply to The Lemming:

Chin up and keep up the good work man.

 wintertree 11 Nov 2020
In reply to The Lemming:

Stay safe.  

Timely story on the situation in America - https://arstechnica.com/science/2020/11/us-healthcare-on-brink-as-covid-19-...

 wintertree 11 Nov 2020
In reply to Removed User:

> I'm struggling to see the benefit of the 'by reporting date' data.

The weekly variations provide a very fertile ground for genuine misunderstanding and deliberate misinformation.  The radio news is back to reading out the numbers like a weather forecast; not very helpful or productive.  What matters really is if the situation is getting better or worse.  

> It is supposed to be more current yet at the same time it is not!

I made some more plots.  

  • The first one shows both measures for each date.  The data by reporting date always lags the specimen date data - it's shifted to the right by convolution with the lag kernel - so for a given day it's less current.
  •  But this is in retrospect when all the lag is resolved.  Tomorrow, when we get the reporting date date it's the most current data for that day at that point in time as the specimen date data isn't resolved yet.
  •  So in theory it's a useful barometer for change, in practice the 7-day periodicity is massive and makes it rather difficult to interpret.
  •  The second plot shows one vs the other for every day, coloured according to the gradient.  (Blue - falling, Red - rising).  It helps to understand how to interpret the reporting date data - when cases are falling, the data by reporting date is higher (right shifting a falling curve raises points at a fixed time) but when cases are falling, it's lower.  Right now the gradient of deaths is lowering (red fading to white) and the data markers are heading towards the 1:1 line - another sign of approaching a stable death rate.  (Both plots omit the most recent 5 days of highly incomplete data).

The main use of "by reporting date" appears to be to give a context free number for the radio and TV news to read out every evening.

Post edited at 23:27

 Stichtplate 11 Nov 2020
In reply to The Lemming:

> This thread is too long to read, and Covid Deniers wind me up, for obvious reasons.

It quite cheers me up when a denier pops up on a mate's Facebook. It's very therapeutic tearing them a new one, though it's not much of a challenge when it's invariably the sort of mentally challenged twonk who thinks 'meta analysis' is just a badly spelt pornhub search.

> Here is a cheery bit of news, my local hospital, right now, has zero free beds.

Nearly as bad here. Queued 2 hours for a handover last night and it seems to be a bit of a thing right across Europe currently. 

https://www.jems.com/2020/11/10/ambulances-park-outside-european-hospitals-...

Post edited at 23:29
 oldie 11 Nov 2020
In reply to jkarran:

> I've never quite been convinced by this viral load argument, I can't help but feel it should be quite easily proven or its significance quantified by studying infections in quarantined households. For a big enough data set we can assume the first person getting ill likely gets a small dose in the wider environment, certainly a random (average) dose. The person living with them in (generally close quarters) isolation is continually exposed to a symptomatic sick person, likely still sharing everything with little ability to effectively isolate from each other. It seems to me, though I could be missing something, that isolation acquired infections should be notably worse than those of unknown origin if there is a dose response. I haven't heard of that being proven. <

Might it also be the case that as soon as one household member is thought to have the virus they are isolated as much as possible, but when proximity is necessary for care etc the potential infecting dose of virus is less, eg patient and carer both wear masks, which was not the case at the start of the first lockdown (I think wintertree suggested this) Again if many people are not following rules and are visiting other households, but are taking greater care with separation, masks, and hygiene, then any infecting dose will be less.

However I'm making postulations with no evidence. Your argument is at least as likely.

Removed User 12 Nov 2020
In reply to wintertree:

WRT to first plot, is any of that lag associated with your filter?

 The Lemming 12 Nov 2020
In reply to Stichtplate:

> Nearly as bad here. Queued 2 hours for a handover last night and it seems to be a bit of a thing right across Europe currently. 

Last week we relieved the Day Crew who were waiting 3 hours outside. We waited a further 2 hours before the patient could be taken off the truck.

Today is just as bad

 Stichtplate 12 Nov 2020
In reply to The Lemming:

> Last week we relieved the Day Crew who were waiting 3 hours outside. We waited a further 2 hours before the patient could be taken off the truck.

> Today is just as bad

I wonder how long the 'no doubling up on patients' rule lasts as crews finish later and later and start having to come in later and later for the next shift? 

 wintertree 12 Nov 2020
In reply to Removed User:

> WRT to first plot, is any of that lag associated with your filter?

No, the smoothing filter is symmetrical.  It's good to check these things however so I did a test where I reversed the x-axis order of the data, then filtered it, then reversed it back, and compared the results - see below - the only difference is the rounding error from the finite precision maths.  (yP is the y-axis data by publication date).

The x-axis shift comes from the kernel for the reporting lag - all the values are off to one side of 0 days shift as it can't see in to the future.  

yPF_forward = scipy.signal.savgol_filter(yP, window_length=15, polyorder=2)
yPF_reverse = scipy.signal.savgol_filter(yP[::-1], window_length=15, polyorder=2)[::-1]
print (abs(yPF_forward - yPF_reverse).sum()) # Prints: 3.467448550509289e-12

Post edited at 09:56
 Toerag 12 Nov 2020
In reply to wintertree:

>  Now the weekend reporting lag is cleared as of this evening I think a levelling off it is definitely  shown in the deaths data by specimen date.  It's clearly dropped below the exponential and it looks like it's becoming level.  The doubling times look quite consistent with the doubling times for cases rising, then hospitalisations then deaths.  Interestingly, hospitalisations appear to have totally flattened where-as cases haven't - but the cases data is subject to more reporting lag than hospitalisations...

> So, all quite positive in the top level data.  Hopefully the lockdown restrictions will moderate the lower prevalence areas that were rising fast and would otherwise take over driving cases.

My crude active caseload supports this. It's been between 288&293k since the start of the month. New cases are also flat at 22-23k per day over the same period. No sign of a drop yet though. I guess we could be seeing 'at work' infections being replaced by 'within household' ones, and rises in the south being masked by decline in the tier3 zones?  Given our pet ambulance drivers are seeing problems things aren't looking very clever for a while if this isn't the case.

 wintertree 12 Nov 2020
In reply to wintertree:

> Edit - the death rate may rise a bit more; the most recent four days are omitted from my plots as they’re normally subject to a lot of missing data from reporting lag, 3 of them are already at the level of the plateau and may go up over the next 3 days.

Yup.  

Today's data release has seen deaths rise above the apparent plateau with the number for Nov 9th having risen up to 400 or so.  This looks like a massive jump but looking at the residuals to an exponential fit or filtered trend-line, normalised to the expected statistical noise, it's not actually out of the ordinary - something of a trick of the eye because it's falling over a shallow part of the curve and not a steep part, so there's more empty space around it on the screen.  

So for now I still think death's are plateauing over towards ~ 380 / day. 


 wintertree 12 Nov 2020
In reply to Toerag:

>  No sign of a drop yet though

Highest reported number of cases today by quite a margin (33k vs 26k previously) but the cases resolved by specimen date still look like they're barely rising - although the week/weekend fluctuations about the trendline are getting pretty extreme which makes things much harder to interpret on a day-to-day basis (compared to a week-to-week basis).

It looks to me a bit like the specimen taking / testing at weekends has saturated and so demand builds up and gets serviced on a Monday with the biggest case number so far being on Monday the 2nd.  Monday the 9th is already pretty high and could rise over the next few days - it's not shown on my plot until tomorrow as it's still in the "provisional" zone.   If it stabilises at or below the level of the Monday 2nd spike I'll start to believe that cases are levelling off.  

> Given our pet ambulance drivers are seeing problems things aren't looking very clever for a while if this isn't the case.

The hospital admissions rate is highest in the North East where I am (latest PHE surveillance report out today).  Not a comforting thought.  


 Si dH 12 Nov 2020
In reply to wintertree:

I agree this doesn't look like anything new when interrogated by specimen date. I also had a look to see if it changed anything I said about trends in different areas yesterday (eg if the additional cases were concentrated in any one region) but it doesn't, other than perhaps that London looks to be rising a bit rather than flat.

 wintertree 12 Nov 2020
In reply to Si dH:

> I also had a look to see if it changed anything I said about trends in different areas yesterday

I've been looking for a better way to get an overview visualisation of this.  Here's a plot from the latest data release (up to Nov 9th) that scatter plots the current rolling 7-day average against the rate of change of that over the last 7 days by UTLA.  Blue labels are falling cases, black are rising.  There's a lot more black than blue, and the rising rates are generally faster.

It's a quick and dirty plot - I haven't got got to shuffle the labels round so as not to overlap in high density areas - but it shows clearly places with high absolute numbers and high rates of change.

Ideally I'd normalise both to the UTLA population to give a better picture.


 RobAJones 13 Nov 2020
In reply to wintertree:

In general, although I have a few issues around half term in particular, I found the SAGE report into children and schools interesting 

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

but could they have avoided giving  those with an agenda, the headline

There is no direct evidence that transmission within schools plays a significant contributory role in driving increased rates of infection among children

In reply to captain paranoia:

> I was looking at the BBC map of infection. Their colour scheme goes up to 200+, one of six colours, with a log2 scaling.

Hah. Just looked at the BBC page, and they have finally added another colour, taking them up to 800 (well, 400+)...

https://www.bbc.co.uk/news/amp/uk-51768274

Post edited at 16:44
 wintertree 20 Nov 2020
In reply to captain paranoia:

I missed your previous post.  The relevance of the Red Dwarf "Shall I change the lightbulb" sketch never grows stale.

In reply to wintertree:

But at least I know to really avoid those areas on 'pomegranate alert'...

Not that I've been anywhere other than shops, work and the odd bike ride since March...

[edit] actually, having a colour scheme that better represents the magnitude of the problem does allow you to take in the situation pretty much at a glance. You might otherwise be fooled into thinking that maybe things in Rochdale or Oldham were only half as bad as they actually were...

Post edited at 17:21

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