Thread by @ewanbirney

A primer (journalists, part written for you) on false positives and why you should (a) know about them but (b) feel confident the system does the right thing for them in a SARS CoV2 world.

False positives is when someone who does not have the thing of interest (in this case, "SARS CoV2 infection") is reported positive by a test. When one does things at the scale of 100,000s, *everything* has a false positive rate, the question is do you understand it+manage it

Just to make this confusing there are different types of false positive for any test, this included. There are samples swaps/tracking errors (very rare but not 0) - this flips a sample in the system, and one of the flips is positive. There is lab contamination (again super rare)>

<< then there is technical false positives (the actual test reported incorrectly) and final a sort of "scope" or "interpretation" false positive where everything is right but you'd actually, in an ideal world, wanted a slightly different test.

Before anyone freaks out; all of these are either *rare*, *very rare* or *insanely bonkers rare*; the professionals in the labs understand them, are super-paranoid about them (hence their rarity) and finally the overall system understands them (eg, interpretation).

It's not so useful to dive into the deep weeds of the different types of false positive and their mitigation - be reassured that clever people think v. hard about it and people stay up all night worried about their labs etc. The rates are in the 10,000s to millions to 1.

Of course, with 100,000s of tests, *some* of them at these rates should be false positives, but as the ONS drily points out, even if *all* the tests over the summer were false positives then the false positive rate would still be super super low.

This means change in positive rates on the same volume (magnitude) of testing implies a change in underlying infection rates. This goes to the fact that some things we're interested in (eg, infection levels) we can either ignore or model out false positive rates

One "false positive" type that bares some exploration is the interpretation or "Ct value" or "Hot" vs "Cold" positives in vernacular. This is a true complication, but again is less important than the sometimes shrill view by people on twitter in my view,

Stepping back - what is going on? - when the virus infects someone it goes from a small number of viral particles (possibly just one) to many via the process of hijacking cells in our airways to make more virus (very clever of the virus). Our body responds in a variety of ways

As such, when you swab (or spit) the virus is not present in detectable levels in fact for some time (roughly 7 days on average) and then rapidly expands; part of the basic default coronavirus strategy is to get in+get out quickly before the mammalian immune system knows whats up

(The virus is super-sneaky molecularly - not just in how it gets into our cells but also its clever, "under the radar" replication strategy - our cells bristle with all sorts of "expect the unexpected" strategies - many of them leading to the "autodestruct" button every cell has>

So the virus has to get in, not activate any of the "autodestruct if you have been infected by a virus" trip wires and then when it goes for it, it has to go hard because then it can expect a full on response from the host).

Post peak there is actually quite a time when the virus is at a low level and/or bits of virus are being left in cells and sloughing off. Think of this as a battleground with debris. The virus is there but quite possibly the person's immune system has "won".

Many tests are super sensitive - they can detect really shocking low numbers of molecules, and a number of tests has a read out of how many molecules there are. For "classic" RT-PCR tests, this is called the Ct value (cycle threshold)

(Just to confuse everyone, high Ct numbers means *low* molecular detection, low Ct numbers means lots of molecules. It makes sense as it is about the number of cycles needed to detect but it catches everyone out first round).

Now... ideally we'd like to distinguish "virus is there, hot and infectious" vs "bits of virus are hanging around the complex battle ground of this person", and, indeed, on aggregate there is clearly a difference between a "hot" positive (low Ct) vs cold positive (high Ct).

This is particularly true if you do culture experiments or well controlled animal experiments.

*However*, in the real world, there is annoyingly all sorts of other reasons why you get a high Ct value; the swab was not great. Someone slipped up in the lab. The robot piercing on the foil on step 12 was not quite as deep as possible.

Basically, in dry statistical terms, there is considerably additional variance to Ct values in the real world than ideal. One might be able to get into this in more subtle ways (there is a whole business about which bits of the virus is measured) but ... it is unlikely to work

So - it can be absolutely valid to say "an effective test only has to work with infectious individuals, usually with low Ct values" but this does not mean "it is sensible to truncate off high Ct values and declare them 'false positives'"

(there's a bit of semantics here - this definition of a false positive is absolutely about moving goalposts - not about test behaviour).

I get most people are already bamboozled somewhere up this thread. The main thing to realise is (a) the rise in case numbers is *not due to false positives*. It really isn't. You have to be a bit loopy to cling onto this and (b) finessing Ct values is not going to change things.

In the long shopping list of "things to do better in our collective COVID response" false positive management is really really down the list. Think about isolation support/compliance, contact tracing waaay before this. And trust the lab pros - they worry on our behalf about this!