Calibrating COVID (part 1)

Using a neural network to assess diagnostic reliability of the PCR test for in-hospital death

I am going to assume that subscribers to this newsletter are pretty much aware of the extraordinarily bizarre situation we find ourselves in when it comes to diagnostic methods adopted for COVID-19. A person with no symptoms other than joy and happiness can poke themselves up the nose at home and claim SARS-COV-2 infection because a plastic device says so. When we come to question the reliability of that device and ask for performance data revealing specificity and sensitivity we are branded as anti-vaxxers. Hospitals are no better places for a poke either - I’ve had several nurses confide that they test unvaccinated admissions relentlessly until they get a positive test result. The word is BIAS, though I am tempted to use stronger.

So here I am, sitting on anonymised electronic patient records for 11,156 in-hospital deaths over the period 1st Feb - 7th Dec 2020 for an undisclosed NHS trust wondering if there’s a way of assessing the reliability of COVID designation using information c…