What night shifts on a cardiology ward taught me about alert fatigue

Before I became an engineer I was a nurse for six years, the last three on a cardiology ward. I want to describe one thing from that time, because I see the same pattern now in software: in dashboards, in on-call rotations, and in the apps we build for clinicians.

On a night shift you hear alarms all the time. The monitor alarms when a patient turns over and an electrode slips. It alarms when the saturation probe loses contact for a few seconds. Infusion pumps alarm when a bag is nearly empty, which is useful, and then again with the same sound for things that are not urgent at all. After some weeks your brain does something you do not decide: it starts to filter. You hear the alarm, you know which bed, you know it is "probably the probe again", and you finish what you are doing first.

Most of the time this is correct, and exactly that is the danger. The one alarm that matters sounds the same as the forty that did not. The literature calls it alarm fatigue, and studies on monitored wards keep finding that most alarms are not clinically actionable. I did not need a study. I could feel it in my shoulders.

What helped on the ward was never "try harder to pay attention". It was changing the system:

• thresholds adjusted per patient instead of the default for everyone
• short delays for conditions that resolve by themselves within seconds
• changing electrodes every day, so the false alarms went down
• different sounds and escalation for the things that really cannot wait

When I now see a Slack channel with three hundred automated messages per day, or a clinical app with a red warning banner on every second screen, I recognise it immediately. If everything alarms, nothing alarms. People are not lazy when they ignore your alerts. They are adapting, the same way I did, and the adaptation is rational.

So my rule, for clinical software and for monitoring: every alert must be actionable, must go to someone who can act on it, and must be rare enough that it still surprises. If an alert fires and the right response is "ignore it", delete the alert or fix the cause. There is no third option that works for long.