Meet the Other Phone. Protection built in.

Meet the Other Phone.
Protection built in.

Buy now

Please or to access all these features

Covid

Mumsnet doesn't verify the qualifications of users. If you have medical concerns, please consult a healthcare professional.

Daily numbers, graphs, analysis thread 15

999 replies

BigChocFrenzy · 13/08/2020 21:37

Welcome to thread 15 of the daily updates

Resource links:

Uk dashboard deaths, cases, hospitals, tests - 4 nations, LAs, English regions
Slides & data UK govt pressers
UK added daily by PHE & DHSC
PHE Surveillance report infections & watchlists every Thursday with sep. infographic
ONS England infection surveillance report
ONS UK death stats each Tuesday
ECDC rolling 14-day incidence EEA & UK
Daily ECDC country detail UK
Worldometer UK page
Plot FT graphs compare countries deaths, cases, raw / million pop
Covidly.com world summary & graphs
Plot COVID Graphs Our World in Data test positivity etc

We welcome factual, data driven, and civil discussions from all contributors 📈 📉 📊 👍

OP posts:
Thread gallery
104
Derbygerbil · 16/08/2020 22:45

Its not it’s!

AnyFucker · 16/08/2020 23:25

.

whatsthecomingoverthehill · 17/08/2020 09:18

Does anyone know whether there is any good info on the false positive rate for testing? Just commented on another thread, but I read this article which seemed to have some pretty glaring issues (I'm not convinced by Heneghan's analyses in general at the moment).

NeurotrashWarrior · 17/08/2020 09:45

T cell test is v interesting.

Could change how people are screened for many things thinking about it.

pinkbalconyrailing · 17/08/2020 10:09

www.zeit.de/wissen/gesundheit/2020-08/corona-impfstoff-russland-hans-georg-eichler-ema/komplettansicht

interview (in german) with the chief medical officer of the European Medicines Agency about the regulatory aspect of the covid-19 vaccine authorisation.

Cloudburstagain · 17/08/2020 10:58

With the change in reporting deaths, and not currently on dashboard, is there an up to date rolling 7 day average?
Cases are rising, so I wondered how that affected the death rate?

walksen · 17/08/2020 11:08

Death rate hasn't really changed/ increased has it? The theory being that most infections are now under 45's so there are fewer hospital admissions yet.

DearHeidiHansen · 17/08/2020 11:29

www.jackiecassell.com/coronavirus-testing-how-to-be-less-confused/
This deals with false positives and why they are more significant as infection rates fall.

whatsthecomingoverthehill · 17/08/2020 12:12

@DearHeidiHansen

www.jackiecassell.com/coronavirus-testing-how-to-be-less-confused/ This deals with false positives and why they are more significant as infection rates fall.
Thanks. I was more wondering whether they have any idea what the false positive rate is. Or is that something else that is confused by the different laboratories etc?

My thinking at the moment is that as the ONS prevalence survey is around 1 in 2000, that it is not that significant to the daily numbers - even if all of the ONS results were false positives that would be about 80-90 daily false positives, so less than 10% (and nothing like the 50%+ that Heneghan claims might be possible).

AnyFucker · 17/08/2020 13:06

What a fucking shit show

AnyFucker · 17/08/2020 13:06

Wrong thread, sorry Blush

BigChocFrenzy · 17/08/2020 13:11

I have read that some tests in the USA are v poor quality and give a significant number of false positives

However, in the UK, false positives seem to be about 1% or less,
whereas false negatives - according to NHS advice for staff testing - are about 30%

Heneghan has been saying for months that he thinks that COVID is far less serious than public health experts in the WHO and most countries do
and he seems to oppose local lockdowns, thinks pubs etc are low risk

Of course, false positives affect the statistics far more when rl cases are v low, whereas false negatives do when cases are higher.
However, in the latter case, false negatives could delay the realisation that exponential growth is taking off

So long as deaths and hospitalisations remain low, I wouldn't expect e.g. whole schools rather than classes to close down

Local lockdowns have a high enough threshhold not to be caused by false positives

This was an example of what the false negatives and positives could mean, in an article by statistician David Spiegelhalter a few weeks ago:

Daily numbers, graphs, analysis thread 15
OP posts:
BigChocFrenzy · 17/08/2020 13:13

@AnyFucker

Wrong thread, sorry Blush
... You're probably right for any thread on COVID atm !
OP posts:
BigChocFrenzy · 17/08/2020 13:16

Spiegelhalter's example above with false positives and negatives would give results reasonably close to the rl situation, within the usual error margin

OP posts:
AnyFucker · 17/08/2020 13:19
Smile
AugustBreeze · 17/08/2020 13:21

@whatsthecomingoverthehill there's this summary of a research paper, which gives anything between 67 and 20% false negatives, depending on day of testing:

www.sciencedaily.com/releases/2020/06/200610094112.htm

I also read another article recently about the meaning of a negative test result which I'm sure said about 30%, but I can't find it at present, sorry.

whatsthecomingoverthehill · 17/08/2020 13:44

@BigChocFrenzy

Spiegelhalter's example above with false positives and negatives would give results reasonably close to the rl situation, within the usual error margin
Thanks BCF, but that's what I can't rationalise. He uses 1% as the false positive rate, but it surely can't be anything like that when the ONS survey is only giving a positive test rate of ~0.05% at the moment?
BigChocFrenzy · 17/08/2020 13:49

Yes, as an example of consequences, the flowchart is fine
However, the assumptions of how many flase positives may indeed be too pessimistic

OP posts:
wintertravel1980 · 17/08/2020 14:01

Heneghan’s calculations are actually identical to Spiegelhalter’s (it is pure maths so this is entirely expected). He also (1) uses more realistic assumptions for prevalence (0.1% rather than 6% in the other example) and (2) assumes COVID tests are 99.9% specific.

If we apply these assumptions to the population of 10,000 people where 0.1% (i.e. 10 people) are infected:

  • 7 individuals will be correctly tested positive
  • 3 infected individuals will be missed and tested negative
- 10 individuals will be erroneously tested positive
  • 9,980 individuals will be correctly tested negative

Heneghan’s point (which was also raised separately at one of the latest SAGE meetings) is that even when we use highly specific tests, there will always be a “floor” of false positives and we should take care when interpreting results of the ONS survey. While it is a useful tool to monitor trends and identify early spikes, it may produce inaccurate results at an individual level when the prevalence is very low.

whatsthecomingoverthehill · 17/08/2020 14:02

I'd have thought it was a pretty crucial thing to be understanding, as it could be having a massive impact on the ONS survey figures, which I'd previously thought were a better indication of the general prevalence than the daily test results.

whatsthecomingoverthehill · 17/08/2020 14:06

Thanks winter, I understand the maths behind it. What I don't understand is how Heneghan claims 99.9% is overly optimistic when that would imply a floor level in the ONS survey of 1 in 1000, when their results have been about 1 in 2000 prevalence.

wintertravel1980 · 17/08/2020 14:12

The outputs of population level surveys might actually be adjusted for known test deficiencies (sensitivity/specificity). I have not had a chance to read the latest weekly ONS report but I have noticed that they stopped giving the actual number of people tested positive over the last week.

BigChocFrenzy · 17/08/2020 14:13

"While it is a useful tool to monitor trends and identify early spikes, it may produce inaccurate results at an individual level when the prevalence is very low."

As the exams fiasco has again demonstrated, no maths tool applied over a large population is accurate when applied to individuals

OP posts:
wintertravel1980 · 17/08/2020 14:16

In other words, 1 in 2000 may be a modelled number. It might not necessarily mean 1 in 2000 actually tested positive.

Most antibody testing estimates are adjusted for sensitivity/specificity gaps. I am not sure if it is also the case with the estimates for community prevalence.

BigChocFrenzy · 17/08/2020 14:17

We have seen the NHS estimating false negatives at 30% for their own staff

Has there been a reliable study in the UK to estimate false positives ?
I've read 1% and also 0.1% for them

The worked examples by Spiegelhalter and Heneghan use different input data and assumptions as they were addressing different situtations,
but of course the maths algorithm is very basic and the same for both

OP posts:
Swipe left for the next trending thread