Emergency department overcrowding is one of the most studied problems in Australian healthcare, and one of the least instrumented in real time. Triage systems record who has presented. Bed management systems record who has been admitted. But the waiting room itself, the physical space where crowding actually happens, is typically measured by a clerk's line of sight and a patient-record timestamp. This post is about what an anonymous occupancy layer adds to hospital patient flow, and why the privacy architecture is the difference between a tool a hospital can deploy and one that dies in ethics review.
01 · THE PROBLEM IS FLOW, AND IT IS MEASURED TOO LATEThe problem is flow, and it is measured too late
Australian EDs work against the National Emergency Access Target: 90 per cent of patients departing the ED within four hours. The chronic obstacle is access block, admitted patients waiting in the ED because no inpatient bed is free, and research programs like the Emergency Medicine Foundation's patient flow scheme have spent a decade quantifying its consequences: longer waits, worse outcomes, higher mortality.
The operational signals that precede a bad afternoon are physical and visible: the waiting room filling faster than it clears, dwell times stretching, the corridor outside triage becoming a queue. Today those signals live in staff intuition. The data that would confirm them arrives hours later, extracted from patient records after the shift is over.
02 · WHAT AN ANONYMOUS OCCUPANCY LAYER MEASURESWhat an anonymous occupancy layer measures
An edge sensor over a waiting room emits four numbers, continuously and anonymously: how many people are in the zone, how dense the seating area is, how long people have been dwelling, and the rate of change in each. Nothing else. For a patient flow coordinator those map directly onto decisions:
- Count and trend say whether the room is filling faster than triage is clearing it, an hour before the four-hour clock makes it official.
- Dwell distribution separates a busy-but-moving room from a stuck one. A rising median dwell is the earliest quantitative signature of access block reaching the front of house.
- Density per zone covers the safety and dignity questions: the corridor crowd, the families standing because seating is full, the pressure building at reception.
- Surge slope feeds escalation. A threshold alert that does not cry wolf can page the flow coordinator when count and dwell rise together, the combination that predicts a bad evening.
None of this replaces clinical triage, and none of it needs to know anything about any individual patient. It is the waiting room as a physical system: load, throughput, backlog.
Triage knows who is waiting. Nobody currently measures the waiting itself, live, as a physical system.
03 · WHY THE PRIVACY ARCHITECTURE DECIDES DEPLOYABILITYWhy the privacy architecture decides deployability
A hospital waiting room is close to the most privacy-sensitive space in civic life. People arrive frightened, unwell, sometimes in crisis. A proposal to point recording cameras at them, stream the footage to a server and run analytics on it deserves to fail governance review, and generally does.
This is where the architecture matters more than the feature list. In the system we build, inference runs on the device: raw frames are processed and discarded on the sensor, no video is stored, no video is transmitted, and no faces are recognised. What leaves the device is a count. There is no footage of a patient because there is no footage of anyone, which changes the governance conversation from "how will recordings be protected" to "no recordings exist". We wrote up the same architectural argument for aged care, an equally sensitive environment, in our piece on anonymous corridor monitoring in aged care, and the general case lives on our privacy by design page.
04 · WHAT A PILOT WOULD LOOK LIKEWhat a pilot would look like
We are not running hospital deployments today; health is a vertical we approach deliberately, pilot first, governance first. The pilot we would scope looks like this:
- Two zones, four weeks. The waiting room and the triage corridor, one sensor each, anonymous counts per second, no integration with clinical systems.
- Baseline against the record system. Compare the occupancy curves with the ED's own presentation and departure timestamps. The correlation validates the sensor; the divergence is the discovery, the crowding the paperwork never saw.
- One escalation rule. A single, conservative alert (count plus dwell rising together for fifteen minutes) routed to the flow coordinator, tuned to be quiet. As with our counting accuracy work, the goal of a first deployment is calibrated trust, not a dashboard full of numbers nobody has verified.
The output of four weeks is an evidence pack: how load actually moved through the front of house, where the dwell built, and whether the early-warning rule would have fired before the bad afternoons. That pack is the business case for going further, or the honest reason not to.
05 · THE WIDER PATTERNThe wider pattern
Waiting rooms are not unique to emergency departments. Outpatient clinics, imaging departments, pharmacy pickup, pathology collection: every one of them has the same shape, arrivals, a queue, a constrained service point, and no live measurement of the queue as a physical thing. The same anonymous architecture covers all of them, which is why we treat the ED as the hardest case worth designing for: if the governance and the measurement hold up in the most sensitive waiting room in the building, they hold up everywhere else in it. The vertical overview lives on our health and aged care page, and if you run a flow program and want the four-week evidence pack, that conversation starts here.