Experts warn yearly checkups carry risks and do not reduce mortality

Re:Experts warn yearly checkups carry risks and do not reduc... by amluto on Monday April 03, 2023 @12:33PM (#35426803)

Looking at this from the other direction is interesting. Suppose hospitals had a way to monitor fetal heart rate with no possible side effects (e.g. no possible placebo or psychological effects from wearing the monitor). And they discover that intervening based on the results leads to poor outcomes, and they do better if they completely ignore the data except for one sample taken every hour roughly on the hour.

One solution would be to say “great! Let’s sample once an hour!” But surely one could come up with much better filters and heuristics.

Of course, doing this in an ethical way might be complicated.

[ Parent | on HN ]

Re:Experts warn yearly checkups carry risks and do not reduc... by travisjungroth on Monday April 03, 2023 @08:02PM (#35433194)

I work in software A/B testing. We have similar problems.

The better filter and heuristics is using an algorithm that’s fit for anytime valid testing. The issue here is a high false positive rate. If you come up with all your treatment guidelines in an environment where you check every few hours and change it to every second, you’ve increased your samples by 3,600. Something like heart rate isn’t independent, but that’s still a lot.

You need to use a different algorithm (or algo at all) and/or relax the thresholds. Just random numbers, but if 45bpm was an alert before maybe it needs to be 50bpm.

[ Parent | on HN ]