LondonIndicators - Revision history

Alex at 15:23, 21 December 2021

2021-12-21T15:23:16Z

← Older revision		Revision as of 09:23, 21 December 2021
Line 1:		Line 1:
			<div style="font-size:200%">'''(Old - from December 2020)'''</div>

	[http://sonorouschocolate.com/zoe/combined.London.png <img src="http://sonorouschocolate.com/zoe/combined.London.small.png">]		[http://sonorouschocolate.com/zoe/combined.London.png <img src="http://sonorouschocolate.com/zoe/combined.London.small.png">]

Alex: /* Description */

2021-01-15T09:33:57Z

Description

← Older revision		Revision as of 03:33, 15 January 2021
Line 4:		Line 4:

	= Description =		= Description =

			''Note/update: my description below of the green line, which was meant to be based purely on Zoe symptom data, is not correct. (Thanks to Alan Jenkins for pointing this out.) Effectively what it calculates is Zoe's incidence measure (though a slightly leading version of it), which is based on both testing and symptom data and is still useful. Over the next few days I will try to either change this curve to what I originally wanted it to be, or to rewrite its description to be more accurate.''

	The purpose of this page is to try to help see how containable the new variant (B.1.1.7 lineage) is. The idea is to take London, a large population which has been affected early and for which the new variant is now dominant, and look at two indicators derived from the [https://covid.joinzoe.com/ Covid Symptom Study run by Zoe, and accompanying swab/PCR tests] (known here as "Zoe" for short) in order to get an early idea as possible as to whether the stringent containment measures imposed on 20 December 2020 are enough to prevent growth in infections. This question has a bearing on how bad things are likely to get nationally and worldwide.		The purpose of this page is to try to help see how containable the new variant (B.1.1.7 lineage) is. The idea is to take London, a large population which has been affected early and for which the new variant is now dominant, and look at two indicators derived from the [https://covid.joinzoe.com/ Covid Symptom Study run by Zoe, and accompanying swab/PCR tests] (known here as "Zoe" for short) in order to get an early idea as possible as to whether the stringent containment measures imposed on 20 December 2020 are enough to prevent growth in infections. This question has a bearing on how bad things are likely to get nationally and worldwide.

Alex: /* Description */

2021-01-12T07:46:02Z

Description

← Older revision		Revision as of 01:46, 12 January 2021
Line 9:		Line 9:
	The most reliable indicators of virus prevalence are random sample swab/PCR test surveys, and these are carried out [https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/24december2020 for the ONS] and for Imperial College London as part of the [https://www.imperial.ac.uk/medicine/research-and-impact/groups/react-study/ REACT surveys]. Unfortunately there is a long lead time on results from these and I'm impatient to know as soon as possible whether we're going to be overrun by this variant.		The most reliable indicators of virus prevalence are random sample swab/PCR test surveys, and these are carried out [https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/24december2020 for the ONS] and for Imperial College London as part of the [https://www.imperial.ac.uk/medicine/research-and-impact/groups/react-study/ REACT surveys]. Unfortunately there is a long lead time on results from these and I'm impatient to know as soon as possible whether we're going to be overrun by this variant.

	By contrast, Zoe provides two streams of information which, although subject to biases in a way that a random sample survey isn't, should be faster to respond. The two streams are: (i) an indication, from people reporting their symptoms on an app, of how many people at any one time have symptomatic Covid-19 ("prevalence"), and (ii) a rolling swab/PCR-test based survey based on inviting selected app users to take a test, reporting a 14-day average of test results from 18 days before the present day to 4 days before the present day. (The symptom-based prevalence estimate is also adjusted to match known prevalence through random sampling, but I assume this is a fairly fixed or ~~flow~~-varying factor, and so won't change the multiplicative trends too much.)		By contrast, Zoe provides two streams of information which, although subject to biases in a way that a random sample survey isn't, should be faster to respond. The two streams are: (i) an indication, from people reporting their symptoms on an app, of how many people at any one time have symptomatic Covid-19 ("prevalence"), and (ii) a rolling swab/PCR-test based survey based on inviting selected app users to take a test, reporting a 14-day average of test results from 18 days before the present day to 4 days before the present day. (The symptom-based prevalence estimate is also adjusted to match known prevalence through random sampling, but I assume this is a fairly fixed or slow-varying factor, and so won't change the multiplicative trends too much.)

	Both of these indicators, taken at face value, are influenced by new Covid-19 cases stretching back three weeks or so: they are averages over the recent past. You can attempt to undo these averages to get a more up-to-date result - an estimate of new Covid-19 cases a few days ago - and that is what the purple and green lines in the above graph represent. The blue line is the number of daily confirmed cases in London, as obtained from the [https://coronavirus.data.gov.uk/ official data stream] and is included for reference.		Both of these indicators, taken at face value, are influenced by new Covid-19 cases stretching back three weeks or so: they are averages over the recent past. You can attempt to undo these averages to get a more up-to-date result - an estimate of new Covid-19 cases a few days ago - and that is what the purple and green lines in the above graph represent. The blue line is the number of daily confirmed cases in London, as obtained from the [https://coronavirus.data.gov.uk/ official data stream] and is included for reference.

Alex: /* Description */

2021-01-12T02:14:41Z

Description

← Older revision		Revision as of 20:14, 11 January 2021
Line 9:		Line 9:
	The most reliable indicators of virus prevalence are random sample swab/PCR test surveys, and these are carried out [https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/24december2020 for the ONS] and for Imperial College London as part of the [https://www.imperial.ac.uk/medicine/research-and-impact/groups/react-study/ REACT surveys]. Unfortunately there is a long lead time on results from these and I'm impatient to know as soon as possible whether we're going to be overrun by this variant.		The most reliable indicators of virus prevalence are random sample swab/PCR test surveys, and these are carried out [https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/24december2020 for the ONS] and for Imperial College London as part of the [https://www.imperial.ac.uk/medicine/research-and-impact/groups/react-study/ REACT surveys]. Unfortunately there is a long lead time on results from these and I'm impatient to know as soon as possible whether we're going to be overrun by this variant.

	By contrast, Zoe provides two streams of information which, although subject to biases in a way that a random sample survey isn't, should be faster to respond. The two streams are: (i) an indication, from people reporting their symptoms on an app, of how many people at any one time have symptomatic Covid-19 ("prevalence"), and (ii) a rolling swab/PCR-test based survey based on inviting selected app users to take a test, reporting a 14-day average of test results from 18 days before the present day to 4 days before the present day.		By contrast, Zoe provides two streams of information which, although subject to biases in a way that a random sample survey isn't, should be faster to respond. The two streams are: (i) an indication, from people reporting their symptoms on an app, of how many people at any one time have symptomatic Covid-19 ("prevalence"), and (ii) a rolling swab/PCR-test based survey based on inviting selected app users to take a test, reporting a 14-day average of test results from 18 days before the present day to 4 days before the present day. (The symptom-based prevalence estimate is also adjusted to match known prevalence through random sampling, but I assume this is a fairly fixed or flow-varying factor, and so won't change the multiplicative trends too much.)

	Both of these indicators, taken at face value, are influenced by new Covid-19 cases stretching back three weeks or so: they are averages over the recent past. You can attempt to undo these averages to get a more up-to-date result - an estimate of new Covid-19 cases a few days ago - and that is what the purple and green lines in the above graph represent. The blue line is the number of daily confirmed cases in London, as obtained from the [https://coronavirus.data.gov.uk/ official data stream] and is included for reference.		Both of these indicators, taken at face value, are influenced by new Covid-19 cases stretching back three weeks or so: they are averages over the recent past. You can attempt to undo these averages to get a more up-to-date result - an estimate of new Covid-19 cases a few days ago - and that is what the purple and green lines in the above graph represent. The blue line is the number of daily confirmed cases in London, as obtained from the [https://coronavirus.data.gov.uk/ official data stream] and is included for reference.

Alex: /* Conclusion at 1 January 2021 */

2021-01-12T02:12:30Z

Conclusion at 1 January 2021

← Older revision		Revision as of 20:12, 11 January 2021
Line 24:		Line 24:

	This simple indicator, [[LondonTraffic]], suggests that traffic on the road has been considerably down from usual since 20 December 2020, maybe comparable to the latter part of May 2020, but higher than in April 2020 where the country was at its most locked down, and [https://covid19.apple.com/mobility Apple mobility data] says something similar. So there is perhaps some scope for tightening up, but maybe not a great deal.		This simple indicator, [[LondonTraffic]], suggests that traffic on the road has been considerably down from usual since 20 December 2020, maybe comparable to the latter part of May 2020, but higher than in April 2020 where the country was at its most locked down, and [https://covid19.apple.com/mobility Apple mobility data] says something similar. So there is perhaps some scope for tightening up, but maybe not a great deal.

			= Conclusion at 11 January 2021 =

			[[File:Combined.London.2021-01-11.png\|200px\|2021-01-11 snapshot]]

			The green (symptom-based) and purple (test-based) lines have been on a decline since around 2 January. The blue line (confirmed cases) is roughly stable or possibly declining - a little too early to say for sure. I don't really believe the purple line can drop this fast (would mean R~=0.5), and it really should not be going below confirmed cases, itself probably a significant underestimate. I don't know if there is something wrong with my treatment or something wrong with the underlying test-based data.

			Not that this is a very scientific assessment or anything, but the green line looks fairly plausible by eye - it stays above confirmed cases which is good and doesn't jump too fast. If it were to anticipate a fall in confirmed cases over the next week, then that would be a little evidence that it has some predictive value. On the assumption that the green line is reliable, what would it mean? Its slope over the last week would correspond to an R of roughly 0.8. Unfortunately since writing the previous conclusion, I've learnt that 20-25% is a rather low-end estimate for the current attack rate (number of people who have been infected) in London, and other analysts estimate [https://www.thetimes.co.uk/edition/news/neil-ferguson-puts-his-faith-in-the-herd-dnvqjgg70 25-30%] or even 40%. If it is 30%, then that means that roughly 0.7 of the cutdown on R could be due to immunity, which means the current level of suppression would result in R~=0.8/0.7~=1.14 in a completely susceptible community. It's true there is more we could do to suppress the virus (schools are closed, but in fact open to quite a lot of people, places of worship still open, more jobs classified as essential compared to April, support bubbles etc), but it would obviously be quite painful to do so. So if the preceding analysis/guesswork is correct, it looks like B.1.1.7 is going to be very tough to suppress if it arrives in big numbers around the world in a couple of months, as it looks like it might.

	= Appendix =		= Appendix =

Alex at 19:53, 4 January 2021

2021-01-04T19:53:38Z

← Older revision		Revision as of 13:53, 4 January 2021
Line 1:		Line 1:
	[http://sonorouschocolate.com/zoe/combined.London.png <img src="http://sonorouschocolate.com/zoe/combined.London.small.png">]		[http://sonorouschocolate.com/zoe/combined.London.png <img src="http://sonorouschocolate.com/zoe/combined.London.small.png">]

	(This graph should update three times daily, once for each curve.)		(This graph should update three times daily, once for each curve. The last few values may be subject to some retrospective change as a consequence of the estimation and smoothing algorithms that use future values.)

	= Description =		= Description =

Alex: /* Description */

2021-01-04T02:47:40Z

Description

← Older revision		Revision as of 20:47, 3 January 2021
Line 7:		Line 7:
	The purpose of this page is to try to help see how containable the new variant (B.1.1.7 lineage) is. The idea is to take London, a large population which has been affected early and for which the new variant is now dominant, and look at two indicators derived from the [https://covid.joinzoe.com/ Covid Symptom Study run by Zoe, and accompanying swab/PCR tests] (known here as "Zoe" for short) in order to get an early idea as possible as to whether the stringent containment measures imposed on 20 December 2020 are enough to prevent growth in infections. This question has a bearing on how bad things are likely to get nationally and worldwide.		The purpose of this page is to try to help see how containable the new variant (B.1.1.7 lineage) is. The idea is to take London, a large population which has been affected early and for which the new variant is now dominant, and look at two indicators derived from the [https://covid.joinzoe.com/ Covid Symptom Study run by Zoe, and accompanying swab/PCR tests] (known here as "Zoe" for short) in order to get an early idea as possible as to whether the stringent containment measures imposed on 20 December 2020 are enough to prevent growth in infections. This question has a bearing on how bad things are likely to get nationally and worldwide.

	The most reliable indicators of virus prevalence are random sample swab/PCR test surveys, and these are carried out [https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/24december2020 for the ONS] and for Imperial College London as part of the [https://www.imperial.ac.uk/medicine/research-and-impact/groups/react-study/ REACT surveys]. Unfortunately there is a long lead time on results from these and I'm impatient to know whether we're going to be overrun by this variant ~~as soon as possible~~.		The most reliable indicators of virus prevalence are random sample swab/PCR test surveys, and these are carried out [https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/conditionsanddiseases/bulletins/coronaviruscovid19infectionsurveypilot/24december2020 for the ONS] and for Imperial College London as part of the [https://www.imperial.ac.uk/medicine/research-and-impact/groups/react-study/ REACT surveys]. Unfortunately there is a long lead time on results from these and I'm impatient to know as soon as possible whether we're going to be overrun by this variant.

	By contrast, Zoe provides two streams of information which, although subject to biases in a way that a random sample survey isn't, should be faster to respond. The two streams are: (i) an indication, from people reporting their symptoms on an app, of how many people at any one time have symptomatic Covid-19 ("prevalence"), and (ii) a rolling swab/PCR-test based survey based on inviting selected app users to take a test, reporting a 14-day average of test results from 18 days before the present day to 4 days before the present day.		By contrast, Zoe provides two streams of information which, although subject to biases in a way that a random sample survey isn't, should be faster to respond. The two streams are: (i) an indication, from people reporting their symptoms on an app, of how many people at any one time have symptomatic Covid-19 ("prevalence"), and (ii) a rolling swab/PCR-test based survey based on inviting selected app users to take a test, reporting a 14-day average of test results from 18 days before the present day to 4 days before the present day.

Alex: /* Conclusion at 1 January 2021 */

2021-01-03T18:43:22Z

Conclusion at 1 January 2021

← Older revision		Revision as of 12:43, 3 January 2021
Line 19:		Line 19:
	[[File:Combined.London.2021-01-01.png\|200px\|2021-01-01 snapshot]]		[[File:Combined.London.2021-01-01.png\|200px\|2021-01-01 snapshot]]

	At the time of writing, IF these green and purple curves are an accurate, and IF they are up-to-date enough for the last values to be mostly based on the last 12 days, then we could be in trouble because just flatlining isn't good enough. Bear in mind that ~~London has~~ [https://www.mrc-bsu.cam.ac.uk/now-casting/ roughly 20% ~~immunity~~] ~~through prior infection~~, mostly ~~during~~ the first wave, so we might expect the growth elsewhere in the country and world will be even more difficult to contain.		At the time of writing, IF these green and purple curves are an accurate, and IF they are up-to-date enough for the last values to be mostly based on the last 12 days, then we could be in trouble because just flatlining isn't good enough. Bear in mind that [https://www.mrc-bsu.cam.ac.uk/now-casting/ roughly 20%] of London has already had Covid-19, mostly from the first wave which we expect should give some degree of immunity, so we might expect the growth elsewhere in the country and world will be even more difficult to contain.

	Of course the effect of Christmas muddies the waters somewhat. It's possible that the Christmas period causes more (and varied) contacts than usual, but it's also possible that it's the other way around, with schools shut, many off work, and only a minimal Christmas exemption against restrictions on personal meetings.		Of course the effect of Christmas muddies the waters somewhat. It's possible that the Christmas period causes more (and varied) contacts than usual, but it's also possible that it's the other way around, with schools shut, many off work, and only a minimal Christmas exemption against restrictions on personal meetings.

Alex: /* Appendix */

2021-01-02T13:16:29Z

Appendix

← Older revision		Revision as of 07:16, 2 January 2021
Line 35:		Line 35:
	\[\sum_{t=0}^d \left(P_t-\sum_{i=0}^{k-1}\mathbb P(T>i+\tfrac12)I_{t-i}\right)^2 + \lambda^2\sum_{t=-k+1}^{d-1}(I_t-I_{t+1})^2\]		\[\sum_{t=0}^d \left(P_t-\sum_{i=0}^{k-1}\mathbb P(T>i+\tfrac12)I_{t-i}\right)^2 + \lambda^2\sum_{t=-k+1}^{d-1}(I_t-I_{t+1})^2\]

	This is fairly crude stuff but hopefully not too far removed from reality. $$\lambda$$ represents the ratio of the variances of the assumed prior normal error distributions of $$P_t$$ and $$I_t$$ from their respective predictors. I would like to say that $$\lambda$$ was chosen in a principled way that maximised the predictive ability of the output, or from knowing from basic biology how much $$I_t$$ can vary from day to day, or something, but I confess I just fiddled around with it until it looked like it was producing something reasonable (neither implausibly flat nor implausibly wiggly to use technical terms) which ended up being at $$\lambda=1$$. (To be fair, varying it a bit didn't make a huge difference.)		This is fairly crude stuff but hopefully not too far removed from reality. $$\lambda^2$$ represents the ratio of the variances of the assumed prior normal error distributions of $$P_t$$ and $$I_t$$ from their respective predictors. I would like to say that $$\lambda$$ was chosen in a principled way that maximised the predictive ability of the output, or from knowing from basic biology how much $$I_t$$ can vary from day to day, or something, but I confess I just fiddled around with it until it looked like it was producing something reasonable (neither implausibly flat nor implausibly wiggly to use technical terms) which ended up being at $$\lambda=1$$. (To be fair, varying it a bit didn't make a huge difference.)

Alex: /* Appendix */ Delete joke because it's not accurate

2021-01-02T02:11:46Z

Appendix: Delete joke because it's not accurate

← Older revision		Revision as of 20:11, 1 January 2021
Line 35:		Line 35:
	\[\sum_{t=0}^d \left(P_t-\sum_{i=0}^{k-1}\mathbb P(T>i+\tfrac12)I_{t-i}\right)^2 + \lambda^2\sum_{t=-k+1}^{d-1}(I_t-I_{t+1})^2\]		\[\sum_{t=0}^d \left(P_t-\sum_{i=0}^{k-1}\mathbb P(T>i+\tfrac12)I_{t-i}\right)^2 + \lambda^2\sum_{t=-k+1}^{d-1}(I_t-I_{t+1})^2\]

	This is fairly crude stuff but hopefully not too far removed from reality. $$\lambda$$ represents the ratio of the variances of the assumed prior normal error distributions of $$P_t$$ and $$I_t$$ from their respective predictors~~, or more concisely a fiddle factor~~. I would like to say that $$\lambda$$ was chosen in a principled way that maximised the predictive ability of the output, or from knowing from basic biology how much $$I_t$$ can vary from day to day, or something, but I confess I just fiddled around with it until it looked like it was producing something reasonable (neither implausibly flat nor implausibly wiggly to use technical terms) which ended up being at $$\lambda=1$$. (To be fair, varying it a bit didn't make a huge difference.)		This is fairly crude stuff but hopefully not too far removed from reality. $$\lambda$$ represents the ratio of the variances of the assumed prior normal error distributions of $$P_t$$ and $$I_t$$ from their respective predictors. I would like to say that $$\lambda$$ was chosen in a principled way that maximised the predictive ability of the output, or from knowing from basic biology how much $$I_t$$ can vary from day to day, or something, but I confess I just fiddled around with it until it looked like it was producing something reasonable (neither implausibly flat nor implausibly wiggly to use technical terms) which ended up being at $$\lambda=1$$. (To be fair, varying it a bit didn't make a huge difference.)