our hospital rolled out an ai alert system a little while ago that's supposed to catch things like early sepsis or patients who might be getting worse. i've mostly seen it on night shifts. around 2 or 3 a.m. the unit is quieter, but somehow every monitor still seems to be beeping at once. some of the alerts make sense. some don't. after enough false alarms, people stop reacting right away, which is probably human nature more than anything. then every now and then one pops up that actually points to something important, and that makes it harder to know how to feel about the whole thing.
i don't think this is really an ai problem by itself. if the alerts don't fit into how people actually work, even a good model is going to struggle.
for anyone building these systems or using them every day, where do you think the biggest disconnect is? is it the models, the way they're presented, or just the reality of working in a busy hospital?
[link] [comments]