Covid-19 Gambles

[echoes]

Your bias is showing

The data will be the arbiter of how successful I am!
And you're quite correct, I do have a position on this, and it's one which has been increasingly sceptical since last summer - when I really didn't know what to think until I started noticing increasing inconsistency, lies and blatant fear-mongering.

 

[echoes]

I realise there there are some flat-earthers, Holocaust deniers etc. and even 'Covid deniers' around but I do struggle to see how anyone can seriously believe that the falls in numbers ('cases', hospitalisations and deaths) following implementation of lockdowns/whatever in countries all over the world have not been related to those lockdowns.

I do think you should withdraw that. The comparison of holocaust denial with a scepticism of lockdown efficacy is rather surprising and unbecoming. But anyway...

There is an important nuance which we haven't discussed yet: that if we assume that you do have a valid point (and I'm willing to admit that you might), lockdown does reduce the spread of the virus to some extent, then we must consider that the burden on the vulnerable increases, since it takes longer for the healthy population to attain herd immunity. Once 'unlocked', the virus starts circulating again and puts the vulnerable at risk once more. The perverse, eventual outcome of this artificial stalling of infections is a higher number of deaths.

I did come across a study which had modelled this, and had also modelled the window in which an effective vaccine could mitigate this effect. Alas, the vaccine rollout in Europe came too late. I'll try and find a link to that paper.

This may be why deaths do not correlate to lockdown severity in the long run. (I'll support this assertion later).

 

[JohnW2]

I do think you should withdraw that. The comparison of holocaust denial with a scepticism of lockdown efficacy is rather surprising and unbecoming. But anyway...

Apologies if I offended you by mentioning the holocaust but (although I realise that you disagree) I was trying to underline my personal view that it is as ridiculous to deny a link between NPIs and the reduction in numbers which has virtually always followed the implementation of those NPIs in a vast number of countries, often on more than one occasion, as it is ridiculous to deny something as well documented as the holocaust (or the non-flatness of our planet!). If that also offends you, then I apologise again, but it is what I believe!

There is an important nuance which we haven't discussed yet: that if we assume that you do have a valid point (and I'm willing to admit that you might), lockdown does reduce the spread of the virus to some extent, then we must consider that the burden on the vulnerable increases, since it takes longer for the healthy population to attain herd immunity.

That's obviously true - the less one does to curtail spread of the virus, the quicker will all the infections (and a large number of deaths) happen and the quicker will herd immunity in the survivors be achieved. However, I thought that most people were agreed that a policy of seeking herd immunity by natural infection (rather than vaccination) would have been unacceptable, because of the vast number of deaths (and possibly vast number of 'after-effects' of Covid infection) that would have to be suffered?

As for whether I have "a valid point", it's not just that the number of cases starts falling at the expected interval after implementation of NPIs, but one can actually see (during a 'rising phase') the rate of rise increasing each time there is a significant relaxation in those NPIs (so I imagine that we'll probably be seeing another increase in the slope of the curve towards the end of next week!). As I illustrated in an early graph, there was even a short-lived, but marked, reduction in cases as a result of last November's brief lockdown, rapidly reversed once that lockdown ended - you surely don't believe that was due to 'seasonality'; or some sort of 'co-incidence', do you?

Kind Regards, John

 

[JohnW2]

... There is an important nuance which we haven't discussed yet: that if we assume that you do have a valid point (and I'm willing to admit that you might), lockdown does reduce the spread of the virus to some extent ...

Since you were effectively asking for an apology, I wanted to write some sort of reply last night (as above), but now it's no longer 'the middle of the night', I think I can probably respond a little more clearly/usefully ....

... I'm getting rather lost and confused. I accept that, in some situations, a part of the reductions in Covid cases/hospitalisations/deaths we see (in all countries) may be due to seasonal factors, or other factors unrelated to deliberate 'control measures' (NPIs).

However, you surely do not doubt that measures which reduce inter-personal contact/proximity will reduce transmission of a (any) virus, and that the degree of that reduction will increase as those measures become more strict/complete in reducing that inter-personal contact, do you? It's that (to me) 'obvious truth' which led me to cite the analogy which offended you.

Let's face it, at the extreme, if, hypothetically it were possible to literally 'lock up' every member (at least, 'every household') of a population for, say, 2-3 weeks, with individuals/households having no contact (with people or objects) from outside of their household, then the virus would presumably be almost eliminated from the population - all that would be remain would be a very small number of 'chronically infectious' people ('carriers') and whatever virus managed to survive outside of human bodies in the environment. That would obviously be impractical, but they may have got reasonably close to it during the initial outbreak in China.

... , then we must consider that the burden on the vulnerable increases, since it takes longer for the healthy population to attain herd immunity. Once 'unlocked', the virus starts circulating again and puts the vulnerable at risk once more. The perverse, eventual outcome of this artificial stalling of infections is a higher number of deaths.

You seem to have shifted the discussion from tactics to strategy.

One could, from the outset, go for a 'natural herd immunity' strategy, by making no attempt to restrict transmission and waiting until such a high proportion of the population had been infected to result in 'herd immunity'. That approach results in a large number of (direct Covid) deaths but, if one is going down that road, it may be best to do absolutely nothing to restrict transmission, allowing the NHS to become 'overwhelmed (such that many people with Covid infection might have to be denied NHS care), since the more rapidly it was allowed to proceed (hence the more rapidly 'herd immunity' would be achieved) the less would be all the 'collateral damage' (e.g. non-Covid morbidity and mortality).

Some (per the 'Barrington Declaration') proposed a half-way house, in which the 'vulnerable' (older age groups and underlying diseases) would be 'locked up', whilst the remainder of the population would be restriction-free. Most people did not believe that would be workable. For a start, there are very many multi-generational households, which would create big problems. More importantly, the 'vulnerable' population (60/70+, plus others) is sufficiently large that it's very probable that immunity in a high proportion of the (unrestricted) remainder would not be enough to reach the herd immunity threshold in the population as a whole (including the 'vulnerable') - in which case the vulnerable would 'drop like flies' the moment they were 'let out' of their isolation.

The above all assumes a non-vaccine strategy and does not attempt to reduce or limit the number of 'cases' - on the contrary, attempts to allow rapid development of effective (natural) 'herd immunity' are best served by allowing case numbers to become extremely high.

However, we (and most/all of the world) have based our strategy on the belief/hope (which has proved, so far, to be justified) that an effective vaccine would come along, so our measures have sought merely to keep cases (and hospitalisations/deaths) under a reasonable/'acceptable' degree of control until herd immunity primarily by vaccination can be achieved. That is obviously a much more protracted business than 'letting the virus do its thing' (due to the time taken to develop, test and deploy the vaccines) and therefore does unfortunately result in considerably more 'collateral (non-Covid) consequences' - but it is the route that we (and most of the world) has chosen to follow.

Kind Regards, John

 

[mikeey84]

Let's face it, at the extreme, if, hypothetically it were possible to literally 'lock up' every member (at least, 'every household') of a population for, say, 2-3 weeks, with individuals/households having no contact (with people or objects) from outside of their household, then the virus would presumably be almost eliminated from the population - all that would be remain would be a very small number of 'chronically infectious' people ('carriers') and whatever virus managed to survive outside of human bodies in the environment

One of the issues with that approach would be the growing evidence of the zoonotic potential of Covid with domestic and wild animals.

It doesn't seem as worrying as the flu, but there is significant evidence that MERS came from camels, and that the UK has seen both mink and cats infected

 

[JohnW2]

However, you surely do not doubt that measures which reduce inter-personal contact/proximity will reduce transmission of a (any) virus, and that the degree of that reduction will increase as those measures become more strict/complete in reducing that inter-personal contact, do you?

... if we assume that you do have a valid point (and I'm willing to admit that you might), lockdown does reduce the spread of the virus to some extent ...

As I illustrated in an early graph, there was even a short-lived, but marked, reduction in cases as a result of last November's brief lockdown, rapidly reversed once that lockdown ended ...

For clarity, herewith a magnified version of that 'early graph', concentrating on the period around the brief November lockdown. Following the start of that lockdown, cases, hospital admissions, ICU occupancy and deaths all start falling (appreciably) fell after roughly the expected time lags, and then everything started rising again almost immediately again when the lockdown ended. I would suggest that it's beyond belief that a 'seasonal effect' could be responsible for those falls (in November/December in UK) and cannot think of anything other than the lockdown which could have been responsible, can you?

Kind Regards, John

 

[JohnW2]

One of the issues with that approach would be the growing evidence of the zoonotic potential of Covid with domestic and wild animals. .... It doesn't seem as worrying as the flu, but there is significant evidence that MERS came from camels, and that the UK has seen both mink and cats infected

It obviously was not an approach that I was suggesting would be practical (or advisable!), but that's an interesting point. For those who would advocate that extreme approach, there's not a lot they could do about wild animals (other than try to shoot them all!) but I suppose that the domestic ones could be 'locked up' along with their owners!

Kind Regards, John

 

[echoes]

For clarity, herewith a magnified version of that 'early graph', concentrating on the period around the brief November lockdown. Following the start of that lockdown, cases, hospital admissions, ICU occupancy and deaths all start falling (appreciably) fell after roughly the expected time lags, and then everything started rising again almost immediately again when the lockdown ended. I would suggest that it's beyond belief that a 'seasonal effect' could be responsible for those falls (in November/December in UK) and cannot think of anything other than the lockdown which could have been responsible, can you?

View attachment 239103

I think (and we can all be guilty of this) that you may be seeing what you want to see. It's clear from your graph that cases had already reached a peak a week before before lockdown, then rose during lockdown had been imposed before falling away again.
Technical questions just to avoid doubt:
1) are your moving averages calculated by taking the mean of the current day + 3 previous + 3 next days?
2) And are those cases by reported date or specimen date?

 

[JohnW2]

I think (and we can all be guilty of this) that you may be seeing what you want to see.

I'm not at all sure that I am the one guilty (or most guilty) of that.

It's clear from your graph that cases had already reached a peak a week before before lockdown, then rose during lockdown had been imposed before falling away again.

They had seemingly reached somewhat of a plateau prior to lockdown but, particularly in view of my answer to both of your questions below, one would expect a lag of well over a week before the impact of the lockdown would be reflected in a fall in case numbers (per report date and with a retrospective average) - which is precisely what one sees.

Technical questions just to avoid doubt: 1) are your moving averages calculated by taking the mean of the current day + 3 previous + 3 next days?

No, the mean of the current day and the six preceding days. For the purpose in question, that is the more appropriate MA and, in any event, I am required to produce daily updates in real-time and don't have a crystal ball!

2) And are those cases by reported date or specimen date?

Reported date. I've been involved in a lot of discussion about the relative merits and the consensus is that this is generally probably preferable to specimen date, particularly when (as I/we are) one's interest is in real-time monitoring - and I am presenting here what I normally do, not creating graphs just for this forum. Using publish date obviously introduces a small lag (which averages out to a pretty constant lag) but if one uses specimen date one has to ignore the most recent several days' figures (since they will be artificially low) - which is no use when one wants real-time up-to-date figures.

In any event, if I produced the equivalent graph using specimen dates (which I could, retrospectively, if I could be bothered) it would presumably at least partially negate you initial comments/criticisms above, the since fall in cases would then start earlier after the start of the lockdown. Furthermore, that specimen/publish date issue only arises in relation to the 'cases' data - hospital admissions, ICU occupancy and deaths all relate to the most recent 24-hour period.

However, I think you are somewhat quibbling ('clutching at straws'?). Regardless of the above discussion about the onset in the fall in cases, what are you suggesting, if not the lockdown, resulted in the very substantial fall in cases (followed, in due course, by falls in everything else) during the second half of the brief lockdown, and the rapid reversal of those falls immediately the lockdown ended?

Kind Regards, John

 

[echoes]

They had seemingly reached somewhat of a plateau prior to lockdown but, particularly in view of my answer to both of your questions below, one would expect a lag of well over a week before the impact of the lockdown would be reflected in a fall in case numbers (per report date and with a retrospective average) - which is precisely what one sees.

There is no 'seemingly' about it. That's precisely what your data show - cases peaked before lockdown.

No, the mean of the current day and the six preceding days.

Right - in that case, your data is skewed and we can deduce that cases peaked 3 days earlier than your graph shows.

Reported date.

This adds a further skew to the data, also meaning that cases peaked another few days more before lockdown since test reports always lag specimen days by varying amounts (about 3 days, give or take, depends on weekends), and is sometimes quite lumpy.
I also mistrust the small secondary peak in the case data - the data is real enough, but it's not reflected in the death count which is most reliable proxy for cases. We would expect them to track, and this is an anomaly, probably related to your use of reported date.
You must also not use reported date for deaths, since these have an even bigger skew, sometimes up to 2 weeks.

This is not pedantry. I have also plotted the data using both specimen/actual date and reported date. The distortion is real and significant, especially during periods when there are interesting changes. If you're doing real-time monitoring, then I accept that using reported date is quicker (although suffers from built-in lag) but at least the data don't have to be constantly updated retrospectively as more become available. However, in doing a retrospective analysis to argue cause and effect, specimen date is a must.

You're trying to draw a correlation between cases, which are obviously skewed in the time-domain and lockdown which is not, to an accuracy of better than 7 days, and then introducing another assumption that it would take 'well over a week' for an effect to be shown, you may the one clutching at straws, and you are one trying to demonstrate not only correlation but causation. It may be there, but you'll need to do some more accurate analysis.

 

[mikeey84]

There is no 'seemingly' about it. That's precisely what your data show - cases peaked before lockdown.

If this is the case, why does the case line continue to rise for 2 weeks after lockdown begins?

 

[echoes]

There is a clear plateau before lockdown. That's the 'peak' I'm referring to (perhaps in a sloppy way - apologies). I.e. The line becomes quite flat. Then a clear hump to reach a maximum after lockdown. Note that this distinct hump is not reflected in the deaths line, so there is an anomaly. I've stated why this might be, and would like to see if it's still there if specimen dates are used.

But I'm not the one claiming correlation and causation, rather I'm scrutinizing that claim to see how well it holds up, and what assumptions are being made.
Likewise, I don't know why cases would plateau before lockdown either.

 

[echoes]

I did come across a study which had modelled this, and had also modelled the window in which an effective vaccine could mitigate this effect. Alas, the vaccine rollout in Europe came too late. I'll try and find a link to that paper.

Here it is.

 

[echoes]

Here's a scatter of lockdown stringency vs outcome for a range of countries.
Source: clicky
The authors found some factors, age, health etc produced some correlation with morbidity.

 

[JohnW2]

There is no 'seemingly' about it. That's precisely what your data show - cases peaked before lockdown.

As I said, and as you have conceded in response to Mikeey's comment, it was a plateau, not a 'peak'

Right - in that case, your data is skewed and we can deduce that cases peaked 3 days earlier than your graph shows.

It's not 'skewed', it is what it is intended to be, namely "the mean of the most recent 7 days" - which, as I suggested is probably the most appropriate sort of average for a purpose like this.

Any form of averaging over days obviously 'distorts' the data to some extent, and it doesn't really matter which one is used, since one cannot 'please all of the people all of the time', and anyone who knows what they are looking at can 'adjust' the dates by 3-4 days if they would prefer the other type of MA - but some sort of smoothing of the day-to-day variation is really needed.

However, I personally think that the way I've done it is probably the less potentially confusing ... imagine a (different) situation in which was expecting some fairly large change to show itself on the day after some intervention. If one useda 7-day average centred on the day of the intervention, then one's data would show 'the effects of the intervention' as starting to be seen 3 days before the intervention - which would be a little odd.

This adds a further skew to the data, also meaning that cases peaked another few days more before lockdown since test reports always lag specimen days by varying amounts (about 3 days, give or take, depends on weekends), and is sometimes quite lumpy.

I've explained why I routinely use 'reported date' for cases but, as you say, if one uses specimen dates (which one can only do retrospectively), it will move the curves about 3 days 'to the left'.

You must also not use reported date for deaths, since these have an even bigger skew, sometimes up to 2 weeks.

My death figures are PHE (or equivalent for other UK countries) "28day Deaths", not ONS ones. Whilst obviously imperfect, there is no significant lag, the figures attempting to relate to the actual number of qualifying deaths in the most recent 24-hour period. We know that these figures are pitifully low at weekends and bank holidays, but they 'catch up' on the day after the weekend/holiday, so the 7-day averaging takes care of that. Similarly, the hospital admissions figures are allegedly 'real-time' (most recent 24 hours) ones.

This is not pedantry.

I leave it to you to decide whether or not it is pedantry. I merely presented a graph I already had to illustrate what I've been saying about the November lockdown.

However, pedantic or not, as I suggested previously, you do seem to have shot your argument fairly well in the foot ... if one does as you would like (use specimen date for 'cases' and used a 7-day average centred on the displayed date, then what one gets is:

This shows 'cases' rising until about a week after the onset of the lockdown, whereupon the fall starts, which is precisely what one would expect if the fall was the result of the lockdown. You comment that:

... and then introducing another assumption that it would take 'well over a week' for an effect to be shown

... and maybe I should have said "at least about a week", but that IS what we expect. The average interval from exposure to onset of symptoms is around 5-6 days, and it usually takes a day or two after onset of symptoms or a test to be arranged, undertaken and a result obtained - so, if a lockdown does have an effect on new case numbers, one would not expect to see it until a week or so after the start of that lockdown. Furthermore, although you wrote:

I also mistrust the small secondary peak in the case data - the data is real enough, but it's not reflected in the death count which is most reliable proxy for cases. We would expect them to track, and this is an anomaly ...

... both the peak and the subsequent trough of the deaths curve above occur 3-4 weeks after the corresponding peak and trough of the 'cases' curve, which is what one would expect, isn't it?

... and you are one trying to demonstrate not only correlation but causation. It may be there, but you'll need to do some more accurate analysis.

As I'm sure you are aware, it's never possible to conclude causation from correlation by any sort of analysis of observed data, no matter how 'accurate' that analysis, and that it not what I am 'trying to do'.

I'm rapidly coming to regret that I did, but I presented the graph of the November lockdown simply to illustrate graphically what I (and I would imagine the vast majority of others) already 'knew' (on the basis of common sense and knowledge of how viruses are transmitted) - and I don't really understand why you don't also 'know' it (i.e. why you are sceptical).

We surely both know that the virus is transmitted by contact or proximity between people (or, less commonly by contact with an object that has been exposed to an infected person)? If we know that, it surely is inevitable that if we introduce measures which reduce that inter-personal contact/proximity, then that will have the effect of reducing the amount of transmission, hence the number of new cases (and all that follows therefrom). One surely doesn't need any mathematics, or even data, to understand that?

You recently started talking, by implication, about 'strategies', and not all strategies require control/reduction of the number of new (or prevalent) 'cases' (quite the converse for some strategies). Some of these strategies therefore do not require NPIs (whether 'full lockdowns' or whatever), but if one does introduce NPIs then (without the need for any maths to support the fact), one will inevitably get a reduction in new cases.

This all seems so obvious to me that I struggle to know how to sensibly quiz you further about your viewpoint! If you are truly unconvinced that lockdowns reduce cases, what mechanism/mode of transmission do you envisage would cause the number of new cases arising during a lockdown to be the same as had been the number prior to the lockdown?

Kind Regards, John