Left-without-being-seen as a lagging indicator

1. System failure identification

   Patients who leave before assessment are counted after they have gone. By the time the daily rate is reviewed, the shift that produced it has ended and the conditions that caused it have changed. The measure describes a population that has already left the building.

2. Data pipeline diagnostics

   The measure is a terminal event recorded at the point of loss. Nothing models the interval before it: how long a given patient has waited relative to those who previously left under similar load. Risk accumulates continuously while the record holds only a final binary outcome.

3. Predictive computational model

   Model departure hazard from current wait, queue depth, and time of day. The output is a live count of waiting patients whose profile matches those who historically left before assessment. Historical departures supply the labels, so the model can be fitted from existing data.

4. Workflow integration

   Route the at-risk list to the waiting room nurse as a rounding task, not a numeric alarm. The intervention is a conversation with a specific patient, so the output has to name patients. An aggregate number gives a nurse nothing specific to do.

Back to the case study database