Batch 03 · EHR alerts
Duplicate alerts from parallel systems
-
System failure identification
The same deterioration generates alerts from the bedside monitor, the records system, and a separate scoring tool. Staff receive three notifications for one event and reconcile them by hand. Volume triples while information stays constant, which is precisely how fatigue is manufactured.
-
Data pipeline diagnostics
Each system was procured independently with its own alerting. No layer deduplicates across sources, because no component owns the concept of a clinical event as distinct from a message. Each vendor is behaving correctly within its own boundary.
-
Predictive computational model
Introduce an event identity keyed on patient, finding, and time window, and collapse messages that resolve to the same event. Report duplication rate as a direct measure of noise. Deduplication needs only patient, finding, and time, all present in every message.
-
Workflow integration
Deduplicate at a notification layer placed in front of the existing systems. Replacing any of the three is a procurement cycle, and interposing a layer is not. A thin layer is reversible in a way a replacement programme is not.