Yesterday Irish supreme court upheld "absolute confidence" bar on negative results from #CervicalCheck - by chance, preprint by @donalb5, @CiaranORiain, & several others on false positives / negatives just dropped. So let me explain why most scientists & docs unhappy (1/n)
First off, the idea behind screening is that you cast a wide net, and over a population, you catch some warning signs before they become a problem. The net is inherently imperfect - but on AVERAGE it saves lives. So what defines a test's reliability? These things (2/n) 
So what do you want in an ideal screen? Some test with high sensitivity (correctly identifies the thing) and high specificity (doesn't accidentally say the thing is there when it's not). But prevalence matters too - take LBC, the standard modality. (3/n)
In 1000 women, about 20 (2%) have CIN2+. We want to catch these, as even though most harmless, some will be bad. LBC is 75% sensitive, 90.3% specific. That means:
-> Catches 15 CIN2+, misses 5.
-> Of 980 without, says 95 have it
So 110+ tests, only 15 "true +" - 13.6%
(4/n)
-> Catches 15 CIN2+, misses 5.
-> Of 980 without, says 95 have it
So 110+ tests, only 15 "true +" - 13.6%
(4/n)
These women get invasive colposcopy, & while most overtreated, it saves lives. But conversely, we miss 5. So of our 890 negs, 5 (0.56%) are false negs. This is what we wish to avoid, but is a limitation of the test - the net isn't perfect, but it saves lives. (5/n)
We really want to reduce this, and there are other modalities too - but there's always a trade-off between false + (over-treatment) and false - (missing stuff). We simulate some of this in paper (6/n)
But imagine we're told this is not good enough - we need to get the false- to zero, to be absolutely confidence in negative results. Well, we could keep retesting.. we'd catch more. But (a) could never catch all (b) seriously detrimental for over-treatment. Gets worse too: (7/n)
..if you decide you're going to test more aggressively or more often, the net effect is you massively overtreat with all the inherent harm that brings. So brute force doesn't work - more testing isn't better for outcome - it's worse (8/n)
Also it's more complicated, as there are many modalities we can use - but worth noting that cervical cancer is typically slow to develop. This in practice means you don't need to have perfection - you just gotta catch it within window of development. (9/n)
Even then, it's not perfect - some things get missed even with best strategies. Absolute confidence is an impossible bar - a cytologist can't know whether a slide is perfect, nor if it caught all cells, or any of that - you can't fix by over-testing either. It won't work. (10/n)
This is precisely why both the high court & supreme court ruling have made scientists and medics uneasy - it's asking for the impossible. Samples are, by definition, samples - we don't have a magic ball to know if they're perfect, & cytology esp. prone to ambiguity (11/n)
We'll see what happens with the programme. I hope it goes on. But the fact that CervicalCheck saves lives has been lost in the furore, and it's hard to see how it can be continued when held to an impossible bar by people who don't seem to get that it's impossible. Fin. (12/12)
