Read on Twitter
BAYES' THEOREM: The basic reason we get so many false positives to COVID19. The disease is so rare that the number of false positives greatly outnumbers the people who truly have the disease: THE MATHS:
https://math.hmc.edu/funfacts/medical-tests-and-bayes-theorem/
https://math.hmc.edu/funfacts/medical-tests-and-bayes-theorem/
"Suppose that you are worried that you might have a rare disease. You decide to get tested, and suppose that the testing methods for this disease are correct 99 percent of the time"
"Suppose this disease is actually quite rare, occurring randomly in the general population in only one of every 10,000 people. If your test results come back POSITIVE, what are your chances that you actually have the disease? LESS THAN 1% chance that you have the disease!"
"The basic reason we get such a surprising result is because the disease is so rare that the number of false positives greatly outnumbers the people who truly have the disease"
Say mass testing of the contagious virus was done to 1 million people. In that million, 100 will really have the disease, 99 will be correctly diagnosed as having it. 999,900 of the million will not have the disease, but of those about 9,999 will be false positives! #BayesTheorem
1 in 2,000 INCLUDING FALSE POSITIVES classifies Covid19 as a rare disease in winter 20-21 https://twitter.com/robinmonotti/status/1336270787393753091?s=19
Stefano Scoglio, Nobel prize candidate 2018, has calculated a real false positive rate of 95% from official Italian Health Service numbers. This is in line with #BayersTheorem. Calculation in links in thread:
https://twitter.com/robinmonotti/status/1331500903564775426?s=19
https://twitter.com/robinmonotti/status/1331500903564775426?s=19
If we apply the 95% false positive (Scoglio) back to England positive test % incl. false positive: 0.05% (ONS), we get a real % of positives in England of 0.0025% of the population, or 1 in 40,000 people. This would confirm Covid19 as a rare disease as per #BayersTheorem.
Try this calculator for up to 100 tests. https://twitter.com/robinmonotti/status/1336675905796771841?s=19
Severe Covid19 is a rare disease in England, if tests are 100% accurate, acc. to hospitalization numbers, it's 0.02% or 1 in 5,000 people.
https://www.england.nhs.uk/statistics/statistical-work-areas/covid-19-hospital-activity/
https://www.england.nhs.uk/statistics/statistical-work-areas/covid-19-hospital-activity/
#BayesTheorem applied to LF tests: https://twitter.com/robinmonotti/status/1328818961715142656?s=19
A simple example of #BayesTheorem with a prevalence of 0.1% (much higher than Covid19) an error range of 1% (RT-PCR Charité range est. 0.8-4%) and only 1,000 people tested: 91% false positives.
#BayesTheorem in simple terms: when medical mass testing includes asymptomatics & the disease affects a minority of the population, a very small margin of error in the testing process will mathematically result in the false positives being many times more than the real positives.
"Covid19"[
] mass testing graph from The Economist. Y axis being % of test results either true or false. As share of population with active infection (X axis) is well under 1%, most positive tests are false, & most negative results are true. This is called #BAYESTHEOREM.
!["Covid19"[] mass testing graph from The Economist. Y axis being % of test results either true or false. As share of population with active infection (X axis) is well under 1%, most positive tests are false, & most negative results are true. This is called #BAYESTHEOREM.](https://pbs.twimg.com/media/Eo4gg23XYAMVkPA.jpg "\"Covid19\"[] mass testing graph from The Economist. Y axis being % of test results either true or false. As share of population with active infection (X axis) is well under 1%, most positive tests are false, & most negative results are true. This is called #BAYESTHEOREM.")
] mass testing graph from The Economist. Y axis being % of test results either true or false. As share of population with active infection (X axis) is well under 1%, most positive tests are false, & most negative results are true. This is called #BAYESTHEOREM.
Latest England estimates: https://twitter.com/robinmonotti/status/1337464319274004480?s=19
How to increase the prevalence of a rare disease from 0.01% to 1%? Test the asymptomatics. What prevalence do we estimate Covid19 at including asymptomatic tested? Less than 1%. What could the true prevalence be if we exclude asymptomatic testing? 0.01%. It's called #BayesTheorem
Proof that Matt Hancock is aware of #BayesTheorem, he mentions Bayesian mathematics: https://twitter.com/EdmundFordham/status/1332750239158120450?s=19
As Matt Hancock is clearly aware of #BayesTheorem, if he wanted to avoid the false positives being many more than the true positives, he would not say ONS is applying rigorous Bayesian mathematics, he would instead not implement testing of any asymptomatics not linked to a case.
Matt Hancock is pre-empting the #BayesTheorem false positive trap by mentioning Bayesian mathematics himself in reply. A Freudian slip, a lapsus which reveals what he is really thinking: how do I increase false positives to make Covid19 prevalence appear worse than it really is?
So how does he do it? He implements mass testing of asymptomatics in Universities, then schools. He uses the hierarchical power structures in these institutions to convince healthy students they need to be tested. The schools get closed on false positives, false fear is created.
#BAYESTHEOREM @ Cambridge University. 0.4% of 262 students came back as positive after the first "test". All came back as negative after the second. Government only tests once. ONS would say there is 0.4% prevalence instead it's 0%.
#BAYESTHEOREM @ Cambridge University. 0.5% of 1,937 students came back as positive after the first "test". All came back as negative after the second. Government only tests once. ONS would say there is 0.5% prevalence instead it's 0%.
Sorry this should say 0.4% of 263 students https://twitter.com/robinmonotti/status/1338043356677623810?s=19
See what happens in #BayesTheorem? Number of asymptomatic testing increases & the estimated prevalence of the disease increases!! This can be addressed by requiring confirmatory tests of those who test positive when numbers are small, otherwise DON'T test asymptomatics.
Cambridge Pooled Testing Report #BayesTheorem
https://www.cam.ac.uk/sites/www.cam.ac.uk/files/documents/pooled_testing_report_30nov-6dec.pdf
https://www.cam.ac.uk/sites/www.cam.ac.uk/files/documents/pooled_testing_report_30nov-6dec.pdf
To have the same number of false negatives as false positives you need a disease that is present in 30% of the population. Covid19 affects less than 1%. This means the false positives VASTLY outnumber both the real positives & the false negatives. It's called #BayesTheorem.
Matt Hancock claim: "the ONS report..address directly the question how the ONS adjusts for potential false positives, due to the high but not perfect specificity of the PCR test. I am very happy for one of my academics to take him through the rigorous Bayesian mathematics"
"I am very happy for one of my academics to take him through the rigorous Bayesian mathematics, which I am sure will help to elucidate the debate on this matter still further." @MattHancock to @DesmondSwayne https://hansard.parliament.uk/commons/2020-09-17/debates/92B2DEA8-5D74-42FA-923B-3A3C478494B7/Covid-19Update
I am waiting for one of @MattHancock's academics to take us through this as I have seen no evidence of @ONS adjusting for false positives according to #BayesTheorem https://twitter.com/robinmonotti/status/1338788051015913472?s=19
Professor Emeritus in Public Health, University of Arizona: https://twitter.com/prof_shahar/status/1339217272338935809?s=19
It seems that ONS is deliberately using a figure for false positive calculations which does NOT correspond to reality: https://twitter.com/grazza611/status/1340018889531543552?s=19
@threadreaderapp perfavore unroll
BAYES TRUE/FALSE + CALCULATOR:
The Bayes Lines Tool is a set of software tools for back solving disease prevalence, test performance and confusion matrices for diagnostic tests.
@waukema @BorgerPieter @goddeketal @Bobby_Network @Kevin_McKernan
https://bayeslines.org/
The Bayes Lines Tool is a set of software tools for back solving disease prevalence, test performance and confusion matrices for diagnostic tests.
@waukema @BorgerPieter @goddeketal @Bobby_Network @Kevin_McKernan
https://bayeslines.org/
In which the WHO confirms everything I wrote is correct: https://twitter.com/dimgrr/status/1353705418145595392?s=19
