Can we give @DHSCgovuk and @educationgovuk a lesson in probablilty? This is A level stuff - if there are Year 12 and 13s in isolation please try it (you can do it with Venn diagrams or a Tree diagram) and send your full workings to @GavinWilliamson@MattHancock@SMHopkins
(a) In 1,000,000 children of which 0.1% have Covid-19, if a test gives +ve results with probability 0.5 in those with Covid-19 and -ve results with probability 0.999 in those without, how many +ve results would you get, and how many of these would actually have Covid-19?
For Further Maths geeks:
(b) If those +ve on the first test and tested with a second test which gave -ve results in those without disease with probability 0.990, how many will not have Covid-19 infection AND have a +ve result on the 1st test AND have a +ve result on the 2nd test?
Answers
(a)1,499 positive results, 500 with Covid
(b)10.
β’ β’ β’
Missing some Tweet in this thread? You can try to
force a refresh
Great that @ab4scambs shows some MPs understand the laws of probability (beginning to doubt that there was one) but this justification why we don't need to PCR kids who have positive LFT sadly starts with a fatal flaw
This is the second time I have tweeted this as one decimal place went for a wander in my first set of tweets. Nothing else changes. Thanks to @d_spiegel for spotting it.
2/8
It presumes that the prevalence to use for Covid infection in these calculations is that in the general population - 0.5% or 1 in 200. And then shows that 30% of those who are LFT+ and then PCR - will still have Covid infection. Can you spot the error?
3/8
Great that @ab4scambs shows some MPs understand the laws of probability (I was beginning to doubt that there was one) but this justification why we don't need to PCR kids starts with a fatal error.
It presumes that the prevalence to use is that of Covid in the general population - 0.5% - then shows that 30% of those who are positive on LFT and negative on PCR will have Covid-19. Can you spot the error?
2/7
Its that only asymptomatic kids get LFT - if you have symptoms kids will not be at school. And as Matt Hancock tells us most days, one third of cases are asymptomatic (not sure how he knows that but he seems very sure).
3/7
So are Lateral Flow Tests are becoming more useful?
Eh β No
Last week 2,764,845 tests done (new record high) and 4,353 were positive. Thatβs 0.16% (new record low).
So probably lots are false positives. Lots of money spentβ little found
In secondary school 663,332 were tested, only 328 positive. Testing has move than doubled on previous week, numbers detected have not.
We are now down to 0.05% being positive β that is only 1 in every 2000 tested β another record low.
Given what we know about the infection rates from ONS and REACT β these data raise serious concerns that the LFTs are not doing the job that we expect. Many more cases should have been detected.
@UnHerd asked me to write about school testing β and why PCR matters so much.
From @DHSCgov data I estimate 70%-95% of school positive cases could be false. Hope Iβm wrong, but unless PCR confirmation is done we will never know.
The mistake being made is to think that all that matters is that false positives are rare. However, if true positives are even rarer then we are in trouble. Test positivity data from Test-and-Trace (1 in 1500 positive) raises serious concerns that is the case.
2/8
If 1 in 100 pupils had Covid-19, we would be fine. Five out of every six test positives would be true positives. But it looks like we are somewhere between 1 in 1,000 and 1 in 10,000 (we donβt know).
This is a real lesson in LFTs not working well as "tests to enable". Lessons from two outbreaks in sporting teams. Clear evidence that the undetectable infectious period can be longer than hoped, and -ve tests increase risk by disinhibition.
Sorry @sbfnk I don't see how this analysis takes us forwards. We need to know the +ve predictive value, not specificity, and we cannot get that without verifying LFT+ves with PCR. Your analysis only tells us that the specificity must be =< observed total +ves, not by how much.
Last week total +ves in staff and students was 0.07%, so we know specificity was >=99.93% (we don't need to do any maths to conclude that). So we know it is between 99.93% and 100% but have absolutely no clear where. Without that knowledge we cannot compute the PPV.
If it was 99.93% that means none of the +ves were real cases. If it was 100% then all +ves would be real cases. So we can't tell whether the PPV was 0% or 100% or where it is inbetween. No maths will solve it, only PCR verification.