Tweet

Jon Deeks

15 Jan, 9 tweets, 2 min read

https://twitter.com/ScienceShared/status/1350141171994144768

Great summary and we really need to take the toxicity out of this. Agree with >95% of what is here. Thx for taking time to do this. The big challenge is how we make quick progress without being able to get a ref std for infectiousness. Cluster RCTs great but hard to do.

https://twitter.com/ScienceShared/status/1350141171994144768

Part of my 5% disagreement arises in whether we are making a mistake in trying to force “infectiousness” into the test accuracy paradigm, or whether we could do better considering it in a different way. It is a probabilistic rather than binary concept which causes the problem.

We can classify people as more or less likely to infect others, but I don’t think we will ever successfully put people into two groups of those who do or do not infect somebody else from a test. That is unlike our ability to say that people do or not have detectable virus.

Any Ct threshold chosen to define “infectious” or “not infectious” is really a division between different risks of being infectious. Infectiousness is a continuum, and our methods need to deal with that.

We need good and validated predictive models which convert things we know about a person into probability estimates of infecting others. Good research is developing these, but they need validating in different settings.

Then we can either “integrate” probabilities across test positives and test negatives (similar to what has been done here) to get a measure of the risks of infection in the two groups (we would have to think carefully about what those numbers mean).

Or we define a minimally accepted risk of infectiousness and quote percentages above that using the model. The acceptable risk of infectiousness of course will be situation dependent. When no risk is accepted, then we end up back with the “being infected” situation.

None of this is going to be as scientically certain as information about "infected or not", but it could give us some quantities which allow us to more honestly summarise what we know about infectiouness and how certain we are about it.

Putting this out there for thoughts and reactions?

• • •

Missing some Tweet in this thread? You can try to force a refresh

This Thread may be Removed Anytime!

Twitter may remove this content at anytime! Save it as PDF for later use!

More from @deeksj

Jon Deeks

@deeksj

16 Jan

@educationgovuk

Testing in schools saga - again

What is the explanation behind this from @educationgovuk? Why has the @MHRAgovuk said DfE does not need regulatory approval for daily “assisted testing” given it does not approve daily “self testing” of contacts of infected cases in schools?

Daily self-testing of contacts was not approved because MHRA considered that the test misses too many cases, allowing infected and infectious individuals to remain in class. The MHRA would have come to this decision based on a forensic and balanced assessment of the evidence.

Hopefully somebody can help answer why this does not also apply to “assisted testing”. I can think of three different explanations. There may be more.

Read 7 tweets

Jon Deeks

@deeksj

15 Jan

@DHSCgovuk

The testing in schools story is still unfolding.

@DHSCgovuk told journalists today that yesterday's Guardian story was "not true".

But then the @MHRAgovuk have come and said the opposite.

schoolsweek.co.uk/rapid-covid-te…

The MHRA has now confirmed to Schools Week it has “not issued an Exceptional Use Authorisation for that self-test device for ‘serial testing’ for school pupils who have been exposed to a confirmed positive COVID case that would enable them to attend school as normal”.

MHRA: “continues to advise that close contacts of positive cases identified using the self test device continue to self-isolate in line with current guidelines. Discussions with Test and Trace regarding any future exceptional use cases are ongoing.”

So - contradicting DfE policy

Read 6 tweets

Jon Deeks

@deeksj

7 Jan

I'm rather concerned by the inability of Test-and-Trace to explain the diagnostic accuracy of the lateral flow test. This was the statement about the performance of Innova in the School Handbook sent out by Test-and-Trace before Xmas.

"As accurate in identifying as PCR" is simply not true. Specificity tells you about false positives, not true positives. It tells you the proportion of those without COVID who correctly get negative results. It doesn't tell you how could the test is at identifying cases.

And interesting that the high figure for specificity is stated, but no numbers for sensitivity are given (which are much lower - 40% in Liverpool). We don't want selective reporting from our Health Department.

Read 7 tweets

Jon Deeks

@deeksj

2 Jan

MHRA exceptional authorisation of Innova for asymptomatic individuals was issued on 22nd Dec.

This is a LIMITED approval, and the details of how and why it is LIMITED are important. It does not cover all current uses of INNOVA by Government

1/8

gov.uk/government/new…

@MHRAgovuk

Details here have been provided by the @MHRAgovuk by email. If any fact is incorrect here @DHSCgovuk @MHRAgovuk please let me know and I will change – it is a bit complicated and rather unusual.

2/8

First, approved test is called the “NHS Test and Trace COVID-19 Self-Test kit” - the LFT test in it is the INNOVA test.

Second, the test manufacturer is stated as “Department of Health and Social Care”, and not Innova.

This determines who has the duty of care and safety.

3/8

Read 9 tweets

Jon Deeks

@deeksj

21 Dec 20

What was the sensitivity and how many false positives were there from Mass Testing of University students?

Results from University of Birmingham and Universities in Scotland don’t make good reading.

SENSITIVITY 3% (not a typo)
42% of Innova positives were FALSE POSITIVES

1/15

@unibirmingham

Testing at @unibirmingham was done in our Great Hall – impressively now a testing centre. We retested a random sample of 710 Innova test negs using PCR. We haven’t heard of anybody else doing this. Preprint soon, but here are key results.

2/15

.

@alanmcn1

Our results were posted on Twitter by @alanmcn1 who organised the testing in real time, and were sent to @DHSCgovuk at the same time.

3/15

Read 19 tweets

Jon Deeks

@deeksj

19 Dec 20

https://twitter.com/GenerallyUnwell/status/1340308187451371521

This is a really good question. If a test has an NPV of 99% usually we would think that is great as it means if you have a negative result you have a 1% chance of having Covid. However, a 1% chance of having Covid is actually very high.

https://twitter.com/GenerallyUnwell/status/1340308187451371521

Case rates where I am at the moment are 200 per 100,000, so 0.2%. So a totally "random test" would give an NPV of 99.8%. How low should the rate be to "rule out"? Maybe 20 per 100,000? That would be an NPV of 99.98%. Clearly we are getting into decimal place madness with NPVs

Thats why I don't use them, and because they are going to vary between places and times. My preference is to talk about how much a negative result reduces the chances of Covid. Given the event is rare, the specificity is nearly 1, this is simple to get from the sensitivity.

Read 6 tweets

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Share this page!

Jon Deeks

Try unrolling a thread yourself!

More from @deeksj

Jon Deeks

Jon Deeks

Jon Deeks

Jon Deeks

Jon Deeks

Jon Deeks

Did Thread Reader help you today?

Like this author's thread?