[jandrese]

False positives are definitely a problem. When you read industrial accident reports one extremely common theme is some sensor that was notifying the controller of the problem, but that sensor had a history of false positives so it was disregarded. Companies that don't take false positives seriously are inherently dangerous.

[krisoft]

> When you read industrial accident reports one extremely common theme is some sensor that was notifying the controller of the problem, but …

I remember an accident report. It was about a container ship which had a bad flooding incident in their engineering spaces. One thing the report pointed out that the engineers had ways to fight the flooding, but they were not doing them because they were playing whack-a-mole with all the alarms caused by the flood. If i recall correctly the engineers kept ignoring the waist deep and rising water and prioritised silencing the alarms. (And not because they were stupid, but just because the many independent blaring alarms task-saturated them.)

[Johnny555]

they were playing whack-a-mole with all the alarms caused by the flood

That's common in computer monitoring systems, at my last job when we had a serious outage, we'd get dozens of pager alerts, it was hard to figure out the root cause because so many alerts fired that were caused by the root cause. I.e. like if the root cause was a root volume was out of disk space, the "unable to log in" alert was superfluous and not helpful. Eventually we moved to a better system that had a betrer sense of hierarchy for alerts as well as a way to easily silence them.

[teeray]

Many of those companies fall into the trap of “well, we’d rather a noisy alarm that catches the problem than a silent one that doesn’t.” Both are problems. The former just makes management feel like a problem would be caught be the on-call.

[throwaway173738]

The ventilator company I worked for tried very hard to avoid false positives because we were very concerned about alarm fatigue. We also tried to ride the line on false negatives. It’s really hard.

Sometimes the alarm limits are set incorrectly by the RT or aren’t forgiving enough to allow some motion. When you see an entire ward of nurses totally ignoring alarms it’s a management failure. Either there aren’t enough nurses available to manage the issue or there aren’t enough technicians to properly configure the equipment for each patient. If someone dies because of that then it’s ultimately the hospital’s fault.

[LorenPechtel]

The day I encountered it I have no idea of what sensitivity controls might have existed but the problem was unquestionably the system failing to recognize that what had just transpired was a beat. The trace on the screen looked like a beat to me, but not always to it.

I will not say it was a management failure because I don't know if management could have done anything about it. Given the total indifference of the nurses I strongly suspect they couldn't do anything.

[makeitdouble]

Management could be the most relevant part. A silent alarm is management's fault, a wrongly ignored noisy alarm can be pushed as staff's fault.

[heavenlyblue]

Pretty certain management have 0 control over which alarms can be disabled on the equipment. And I would bet that the equipment from other brands have the same issue.

[kmoser]

It's hard to solve the problem of false positives when the decision to sound an alarm is reliant on a single sensor that may start to become detached (e.g. glue/tape failure). If you think the solution is multiple sensors, well, what happens when one sensor indicates an alarm condition and the other doesn't? Now you have another potential false positive. Not to mention it's untenable to connect twice as many leads to a patient.

[bee_rider]

If they’d use three sensors, they could vote. If one sensor often votes differently from the other two, it could be marked as defective and replaced or re-seated.

Three times as many leads would be pretty annoying, though.

[lostlogin]

You’ve hit the nail on the head. It’s often tedious getting one to work.

[LorenPechtel]

That could partially be addressed by making the sensor include the concept of not working. Run a small electric current across the sensor, if that current fails the sensor knows that it's not monitoring and can report it as a loose sensor rather than as a failure of whatever it's supposed to be sensing.

[kelnos]

> Companies that don't take false positives seriously are inherently dangerous.

Alarms with incessant false positives are inherently dangerous. Sure, there's some threshold of false positives, under which we should still expect people to investigate all alarms. But above that threshold, how can we continue to blame the people involved? The hardware is at fault.

[nradov]

Please propose a design for better hardware then. You'll make a fortune and do a lot of good in the process.

Seriously, what would motivate you to make a comment like that? Do you think medical device engineers and clinicians are unaware of the false alarm issue and haven't already tried a variety of improvements? There is an inherent trade-off between false alarms and missing a real problem. And devices need to be not only accurate but also affordable, durable, and cost effective. It's not easy to get this right.

[pjerem]

I think GP was talking about the people who don’t try to reduce false positives (by actively searching for solutions to reduce them), not the ones ignoring them because they are used to.