Alert fatigue among triage nurses

1. System failure identification

   Nurses dismiss the majority of clinical alerts within seconds. The dismissal is rational, since most alerts are not actionable, but it applies equally to the minority that are. The behaviour is an adaptation to the system, not a lapse in care.

2. Data pipeline diagnostics

   Alert volume is monitored and alert precision is not. No pipeline computes the proportion of alerts that changed a decision, so a firing rule and a useful rule look identical in the logs. Volume is easy to count and precision requires linking an alert to an outcome.

3. Predictive computational model

   Instrument each alert rule with an action rate: the share of firings followed by a relevant clinical action within a defined window. Rules below a threshold become candidates for retirement. Action rate is recoverable retrospectively from orders placed after each firing.

4. Workflow integration

   Review action rates on a standing schedule with the authority to switch rules off. Without a retirement mechanism the measurement changes nothing, and volume only ever grows. Measurement without the power to act on it changes nothing.

Back to the case study database