UKC

A shorter update this week, as now all measures are so low, things don't change much on a weekly basis.  Nevertheless, there are some worry signals in the demographic and regional data, so I'll focus on those today.

There've been some wobbly moments over the last week as decay slowed down (exponentially speaking), but all nations except for Northern Ireland have remained in decay.  Northern Ireland looks to be in growth for cases, "by eye" and by the exponential rate constant plot.  Hospitalisations are currently still falling there.

Link to previous thread: https://www.ukclimbing.com/forums/off_belay/friday_night_covid_plotting_21-...

The decay in English cases previously stalled around 03-06 to 03-20;  the highly publicised use of asymptomatic LFD testing associated with the return to schools was clearly behind much of the stall in decay, and data on that was discussed and presented on previous threads.  This was clear in plot 9 - whilst the rate constant for cases became much less negative (a big blue hump in the pink zone between 03-06 and 03-20), the rate constants for hospitalisations and deaths remained quite negative meaning they continued to decay.  For admissions we'd expect changes ~ 2 weeks later in admissions than cases - one grey horizontal grid line.  So, we understand all that and there's nothing to worry about there.  Good.

There has been as second period of stalled decay over the last couple of weeks - perhaps not coincidentally during a late spell of winter weather - and the signals in plot 9 are behaving very differently this time.  The rate constant for admissions also became much less negative meaning that the decay of admissions has near stalled; this is quite clear from plot 7.1e becoming nearly horizontal away from the line y=0.  This appears to start happening before the decay in cases starts to fail, but I think that is because cases are confused by the switching off of LFDs giving some "artificial" decay.  Another possibility is that there's been a step change in lethality of circulating infections; that does not seem borne out by the sequencing data although that's quite hard for us to really interpret as discussed on other threads.  These worrying signals all start developing towards the far right of plot 9, and as always the far right of plot 9 is highly provisional until more data comes in.  To see the rate constant for admissions heading for growth (positive values) is worrying.  Let's hope that changes promptly.  It's a real shame the hospitalisation data is not broken down in the same demographic bins as other measures.  I'll look a bit more at demographic data below.

So, cases are still falling - just about - in England but we've seen a near stall in the decay of  hospitalisations.  What's going on?

It's hard to be anything like certain from the publicly released data as each dataset on the dashboard is a different reduction of the full data, and we're left starting at this different views trying to infer the most likely structure in the full data that would cause them.

Plot D1.c has a lot going on in it, so I've added an annotated copy.  Using a desktop browser you can use arrow keys to flip-book between images which helps.

• There's two islands of growth (orange) then stronger decay (dark blue) in rates for ages 5-20 associated with the return to schools.  This is near the bottom right.  Much (not all) of this will be the addition of asymptomatic LFD testing results giving a "false" growth in cases if we interpret them as representative of pillar 1 and 2 testing results, and then giving a "false" decay when the school term ended.
• There's a growth in cases of those aged 65-85 emerging in the top right - this is a series of rows turning orange.  Looking at the raw data, I think perhaps the rapid exponential decay has stalled, and that some noise in the data is being measured as growth by this plot; if so, the orange (growth) will perhaps fade to white (cases level) with the addition of a few more days data, so I'll revisit this in a few days.
  • A stall of decay would be concerning, growth would be Bad News IMO.   
  • Given the demographic dependance of hospitalisations, this is likely why the decay of hospitalisations has near stalled - and it could well be why it happened whilst cases were still falling; as the left plot shows the majority of cases are in younger ages (the red part is lower down on the far right), and so what's happening in cases there dominates the national level measures of decay.
• There's an orange growth signal emerging for young adults on the far right of this plot.  This is in the highly provisional zone, but it wouldn't surprise me if it develops to be concrete as more data comes in; the latest round of unlocking has pretty much thrown a switch that says "pandemic over" for a lot of people.

I've included plot 18 in this post as it has orange (growth) in the hospitalisations plots in the former "red" regions and the South West, as well as some orange in cases for the red regions.  So, it seems reasonably to tentatively concede that there's a demographically and geographically confined problem with a growth of cases and hospitalisations in older adults.   The absolute numbers are very small, but the appearance of the same behaviour in adjacent demographic and geographic bins strongly suggests the effect is not that of noise.  One to keep an eye on. 

Plot D1.d shows deaths continuing to be in decay for most age bins where there's enough data to justify a measurement of rate constants.  There's a bit of orange in the aged 90 bin; I'd say that is likely just noise in the low numbers data, but given the other signals, it's clearly also one to watch.

Plot 22 shows hospital occupancy continuing to decrease by both measures.

I've added a red line to the vaccine plot which, for a given day, estimates the lag from first to second dose for people receiving their second dose on that day.  It's derived by matching up the cumulative numbers for first and second dose which is not the correct way to do this, but it's all that is supported by the data and is indicative.  It looks like we're running to an average period of ~ 11 weeks between doses.

The variants plot is the same one from a few days ago [1].  It seems odd to me that there's been a flurry of news stories about the large number of SA variant cases recently discovered in London (and now beyond).  It seems odd, because I'd been assuming most SA variant cases were being detected at our borders, and so a large number of results from community transmission caught by surge testing should cause a step change in the number - and so the red data on this plot.  That doesn't seem to be the case, so perhaps more of this data was coming from community transmission that one would have hoped, or perhaps the new counts have yet to land.

The SA variant could perhaps be a contender for what we're seeing in the cases and hospitalisations data, and one of the news reports covered an outbreak in a care home.

I'm very much hoping that the various hints of growth tuck themselves away by next week, but I'm not convinced.  The good news is that more and more of the most vulnerable - and most likely behind the more concerning growth in the demographic plots - are now getting their second doses.

This to me still feels very much like a race against time for the vaccination program.  A Zeno's paradox style race for those of us in our early 40s waiting for our first dose, and concerned that if a variant gets loose, it's potentially a more lethal one at that.    

[1] https://www.ukclimbing.com/forums/off_belay/friday_night_covid_plotting_21-...

re 11 week gap - seems to be what is being targetted in England - got my invite yesterday for a first jab and it was stated when booking that the second woul dbe in 11 weeks.

(44)

It's sort of interesting that by Sunday evening, I think for the first time ever, there's been only one reply to this thread. Doesn't mean it's not appreciated, Wintertree - I looked out for it eagerly as ever. But I think there is a sense of watching and waiting...

I think we're all watching and waiting. With little movement in the numbers, and inadequate information to make sense of the numbers wrt the changes in behaviour, we're all a bit stuck. I hope those in charge have more information to enable them to unravel the changes. but, given that so many things have changed in the last few weeks, I suspect that meaningful analysis will be hard.

There's a band of persistence from Manchester to Hull, but elsewhere, it seems mostly random clusters. (I do like the BBC map that shows Salford when zoomed out, rather than Manchester; I guess Salford is now important to the BBC...).

I don't think the SA variant is supposed to be more lethal, just better at evading immunity. Having said that, there doesn't appear to be any decent scientific info coming out, the original study was a small sample size and didn't include old people.

That and vaccination.  Many people seem to think that vaccination is their personal shield of invulnerability, forgetting that it doesn't work for 1 in 10 people and doesn't prevent them from passing it onto someone else.  Yes, it is better than nothing, but if it encourages risky behaviour then it has the potential to more than nullify the effects of good behaviour unless there's enough of a level of herd immunity.

I wonder if the apparent growth in cases among the older groups is grandparent visiting/childcare related around Easter. With mass LFT of schoolkids temporarily switched off and generally asymptomatic/mild infections in the very young we'd not be seeing many (access filtered, symptomatic only) PCR detected cases there while they're not in school but those hidden infections may be being reflected up to the top of the chart where despite vaccination they do produce some infections, symptoms and some hospital admissions in the grandparents. Maybe previous exposure and more recently vaccination is now keeping child/family spread cases in the parents' age bins from showing up in symptomatic testing. It's all getting very hard to understand with home LFT kits so freely available and likely being (ab)used quite randomly.

It could also be older folk mixing a bit more freely now they're vaccinated but still developing symptoms if and when they get infected.

It's looking to me like this is going to rumble on quite significantly into 2022 at least.

jk

Some interesting comments; thanks all.  I'll reply when I get a chance - been enjoying getting out and about today.

Meantime, does anyone know how to convert the daily temperature values in the hadCET dataset in to something meaningful?  Perhaps I'm reading deficient but I couldn't see it anywhere...  I assume there is a scale factor, and potentially an offset, to be applied... 

https://www.metoffice.gov.uk/hadobs/hadcet/
https://www.metoffice.gov.uk/hadobs/hadcet/cet_mean_2021
 

Looks like https://www.metoffice.gov.uk/hadobs/hadcet/data/daily_format.html describes the format.

Thanks - reading deficiency on my behalf then.

I was asking as I was going to post my updated week-on-week decay plot.  This plots, for each day, (cases on that day) / (cases 7 days previously).  This gives a way of measuring the exponential decay that adapts naturally to the massive day-of-week bias in the sampling.  The data is all over the place, and there was a clear agreement between the first failure afraid decay and the cold period (blue markers), and I've had an impression of temperature dependance ever since.  On plotting #21, Si dH had suggested the most recent failure of decay was the Easter weekend sampling effects; they turn out to be behind the biggest two outliers of fast and slow decay.  My plot showing this didn't get finished as I went down a rabbit hole on temperature.

The second plot below uses the Hadley Centre Central England Temperature dataset and the English PCR cases data from the government dashboard.

Plotting temperature and exponential decay rate (in the form of a fractional decrease over 7 days) was hardly compelling but when one was falling, so was the other, and likewise rising although the baselines did not remain well aligned.

So, below, I plot the rate of change of (week on week decay fraction) vs time in black, and the rate of change of (central England Temperature) in red.  To really make the point, I've flipped the second y-axis for temperature so falling temperatures are up and rising temperatures are down.

When it's getting colder, the decay fraction is getting more positive (tending towards growth), and when the temperature is rising, the decay fraction is getting more negative (heading towards zero meaning faster decay).

Both rates of change were measured from the input data with a 9-point, 1st order Savitzky Golay filter.    

I've added a blue bar to annotate when schools were open.  Outside of this period there is an amazing correspondence between how the temperature is changing and how the exponential rate constant (plotted here as week on week decay fraction) is changing.

I need to think a bit more about what this means, but the most obvious candidates are:

1. Human behavioural responses to changing temperatures - getting out more when it's a bit nicer, closing windows more when its a bit colder.  
2. The effect is really about absolute temperatures, but that there is other stuff shifting the baseline over time, and that the differentiation acts as a high-pass filter rejecting slower baseline drift.

[1] https://www.metoffice.gov.uk/hadobs/hadcet/

Blue "schools open" line missing in second plot?

Otherwise, wow.

Indeed - although we can also go out now, and it's nice outside.  The last few days we've been making hay whilst the sun shines, getting in walks and even a trip to the pub...  The forum is back to multiple arguments about retro-bolting and bolt chopping instead of Covid.  Long may that continue.

In reply to captain paranoia:

Hopefully I've made a bit more sense of the way trends in decay have been all over the place recently.  That clarity is probably going to fade a bit more now with the new relaxations.

In reply to jkarran:

I think that's quite likely part of it.  Deaths had been halving faster than hospitalisations and hospitalisations faster than cases for quite a long time, which suggests to me that the vaccine (mainly the first dose thereof) is more effective at preventing death than illness, and more effective at preventing illness than infection, so there's plenty of scope for some growth in the less serious measures.

Indeed.  

Good point.  I hard coded it at y=0; doesn’t work on the absolute values plot...

Yes, I spent some time reading through the very simple code to see if I’d made a howling error.

The relationship with temperature seems really clear, that's impressive. Do you have the data to replicate the analysis over any other time period and see if the relationship holds, eg during last autumn? Obviously needing to be careful about when restrictions changed. If it holds up this seems like quite a significant finding that is worth trying to share further. Albeit might not be so important while weather is warmer over the summer.

Yes, I can adapt this to run over 2020; it’s a different field in the API for the cases data (pre LFD/PCR schism) so I’ll need to glue the two together, and I’ll need to do a better job of importing the hadCET data to automate it more lest I make a mistake.  If there’s an equivalent temperature dataset for the other home nations, I can also run those.

I thought of a third possibility for what the differentiation is doing; as the temperature data is quasi periodic (it’s always going up and down), differentiation introduces a phase shift; given typical time scales it could by total coincidence be time-shifting the temperature data to null the lag from infection to detection as a case.

I think it’s reason enough to sneak in to the local bus depot and weld all the ventilator windows open.  Doing another informal survey last Friday, about 70% of the ventilators I saw were closed.  

> Obviously needing to be careful about when restrictions changed.

If they manifest as a change in the baseline decay rate, which is then modulated by temperature effects, the differentiation goes some way to stripping that.  I can add some annotations as with the various plot 9s.

I'll just add another wow!, plus thanks for all the work you're putting in to this.

Very interesting stuff!

Given the temperature should hopefully be warm enough for people to be outside, open windows etc, from now until autumn, I’d put a small wager that rainfall starts to correlate better than temperature. 

BTW, where did you get your cubic Scotch eggs from?

Good point on the rain, although I'm wondering if we're going in to serious a drought in England....  Good news for the Covid recovery, bad news for my fruit trees.

Deep fried Brie with Rosemary.  I was rather hoping to make a competition out of recognising the pub from the photo!  The County Durham gang on here is surprisingly numerous...

Drought is a strong possibility. Bone dry everywhere. I’m still walking round in my Muckboots coz I can’t be bothered going shopping for new boots. I look like a fool and completely overdressed. 

Thanks.  The work paid off big style with this last plot - I'm happier watching the decay rate jump all over the place now I feel I have half an idea why it's doing so, and much happier that the weather is looking up.

In reply to willgriggsonfire:

Outside of the western mountains, March was exceptionally dry and I think April is going to be so too.  Currently there's no rain on the forecasts into May on the BBC, and the GFS ensembles are looking very dry.

Sterling work! +1 to Si dH's suggestion that this is worth sharing further, and to the suggestion that plotting last autumn might yield even more interesting results. The shift from last August into the first week of September and beyond would be really useful. If, when we get to this autumn, another wave does take place, your work combined with the new PHE/H&SE guidance on ventilation could help to persuade employers that there are definite mitigations that could and should be taken.

Predictably not much to talk about this week is there? Don't take silence as lack of appreciation though.

Just to muddy the waters a bit here's my plot what I do for my own curiosity sometimes. Sharing was encouraged last week so I'll throw it in again. Doesn't really add any value.
http://ukwalls.epizy.com/wtr/cases_18-4.png

That temp v cases correlation looks scarily good - close enough to make causation a possibility (*), certainly close enough to warrant further investigation.

(*) I think we can rule out the possibility of causation being the other way; i.e. changes in num cases causing changes in temperature 😁

Ok so what is our best guess on what’s happening in India? There was a massive drop in cases which was sustained and then recently.. boom. 
 

clearly their testing and data is sketchy in places, but the recent sudden increase is clearly real. New variants exist, some vaccination has been done. Should the rest of the world be very concerned or is this an India specific tragedy? Terrible either way..

I think India's vaccination rate is very low as a proportion of the population. So not to say there isn't a problem with this new variant, but what's happening in India isn't necessarily an indication of vaccine escape.

I shall try and make the time to throw an intelligible pre-print together with the observations and chuck it at the arXiv.  Perhaps Longsufferingropeholder can introduce it to the RAMP lists for me.  I'll see what I can do with 2020s data as well.  

It does seem like the answer to much of our Covid woes is literally blowing in the wind, we just have to let it in.  It's notable that when chest freezers sold out at the start of the pandemic, portable HEPA units didn't.  When I go in to Wintertree HQ I have an open window on one side and a HEPA unit on the other - particularly for still days where there's no through draft.

Here's a slightly more quantified look.

Fig 1 - I've converted the units of case decay to the exponential rate constant.  I should also note that for day 'x', it's calculated by (cases on day x+7)/(cases on day x), so it represents the average rate over a 7 day period starting on the day it's plotted.  I should perhaps shift the black data points to the right by 3.5 days to the middle of the period they correspond to, but as it happens this is about the lag between change in the weather data and change in the case data, so I left it as is.

I hope this approach of using the week-on-week ratios to get a good view of the time series of decay rates without needing de-weekending or polynomial filtering and exponential fitting addresses some concerns minimike raised long ago about the pitfalls of LOESS as translated to my approach.

Fig 2 - the time derivative of the plots in Fig 1 as measured with a 9-point, first order polynomial filter.

Fig 3 - a scatter plot of the time derivative of temperature (x) against that of the rate constant (y).   I've coloured days in black before schools returned, and blue after.  A linear least squares regression is carried out on the black points and yields a correlation coefficient (R-value) of -0.89 which is pretty strong.

Fig 4 - this is the correlation coefficient for different lags between the two data sets; add 3.5 days to the x-axis values to translate from the start of the exponential rate constant measurement period to the middle of the period.   The correlation is only strong for ~4 days from rising temperatures to faster exponential decay (and falling temperatures and slower exponential decay).  Although.... There is quite a strong periodicity in the temperature signal so if I expand the window on plot 4 there's weaker correlations further out.  Figure 5 shows this expanded window.  Weather is really interesting stuff; once I get the 2020 hadCET data imported I'm going to look at it's autocorrelation...

I think it shows what happens when somewhere lets their control measures down before immunity is wide spread enough.  Plenty of reason for the rest of the world to be concerned, as the variant rate is going to go up a lot as things continue to get worse there, and potentially the genetic distance of some of those variants from the strain embodied in the vaccines is going to increase significantly.  

Depends very much on if that switch to rising in older case rates keeps building or collapses as little more than a fleck of noise.  More recognition of the likelihood of a coming round of rising cases in the media today.  Interesting developments around the J&J vaccine which might put in to doubt the plans to offer that to younger adults in the UK; it’s not attracting the same heated discussions here as the AZ one despite the clear parallels. I like your plot now it’s just got the two bounding lines.  It’s very easy to interpret and very clear.

You and/or CT are cutting it fine on starting an SN15 thread, static fire perhaps today and NOTAMs out for Friday & Saturday.  This could be the one.  

Happy to throw stuff at the RAMP forums, but I really would encourage you to sign up if you still haven't. It's not onerous. I hardly ever post there; mostly just lurk and read the papers.

Can't see SN15 flying until late next week. Crew 2 Friday though... But there's loads of entertainment from the launch site anyway. That new crane is a beast and the tower is looking spectacular.

Also Venus seems to be doing that trippy disco scintillation thing at about bedtime again. I was half expecting a thread on that.

Oddly all 5 figures from my 20:03 post seem to have disappeared; some weirdness with the structural changes to the forums?  I don't have them at work with me but I'll re-post them tonight...

I've signed up but I can ill afford getting drawn in to a rabbit hole...  If I start reading things I'm going to get drawn in...

The crane is insane.  I went done a rabbit hole on those cranes -  youtube.com/watch?v=gYpMz63WAjM&

That rate of change plot 'BOOM' is rather convincing... The schools situation seemed to knacker it a bit.

Wow, ok, not even the top of the range model then...
I do wonder if they're still planning to use the BFC. Would (will?) be impressive to see that mounted on a 400ft tower.