The Alert Nobody Reads Is Not a Safety Feature

When I started in informatics, my job was mostly additive. A near-miss would happen, a root-cause review would recommend a new alert, and I'd build it. Over several years, that produced a monitoring system with an alarm rate high enough that the nursing staff had, without any of us formally deciding it, started treating most alerts as background noise.

This is a well-documented phenomenon — alarm fatigue — but living through it as the person who'd built the alerts made it land differently. Every individual alert had a defensible justification. The system as a whole had become something closer to a fire drill that never stopped, which meant that when a genuinely urgent one fired, it competed for attention against dozens of routine ones that looked identical on screen.

The reframing that changed our approach was treating attention as a budget rather than a checklist item. Every alert we added was a withdrawal from a shared, finite resource — a nurse's attention during a shift — and we'd been spending it as if it were free. Once we started asking 'is this worth the withdrawal' instead of 'could this theoretically matter,' the alert count dropped by more than half.

The safety metrics improved, not despite fewer alerts, but because of them. A nurse who trusts that an alert means something responds differently than one who's learned, correctly, that nine out of ten don't. We hadn't been building a safer system by adding more alarms. We'd been quietly training our own staff to stop hearing them.

I don't think this is unique to nursing, or to alerts specifically. Any system that asks for a clinician's attention is spending something scarce, and the systems that respect that scarcity — that earn the interruption instead of assuming it's owed — are the ones that actually get listened to when it counts.