Canadian ER wait times & closures, in one place

How accurate is the ERstat number?

A hospital posts the wait right now. That is not the wait you will face, because you still have to get there. ERstat publishes a forecast of the wait at the time you would actually arrive.

Both numbers can be checked the same way. Wait until that time arrives, then compare each one against what the hospital itself posted at that moment. The hospital's own reading is the yardstick, not ours. Everything below is scored that way, and it updates every night.

Scored against what hospitals later posted

Based on 444,681 forecasts across 202 hospitals that publish a live wait, made between 2026-08-28 and 2026-09-23.

What “off by” means here

It is the typical miss: half the forecasts land closer than that figure and half land further. We use the middle miss rather than the average so that a handful of chaotic nights cannot flatter the result.

The average miss is always the larger number, because a small share of forecasts are badly wrong. We publish it too, since quoting only the typical miss would understate how far the worst ones run.

One hour ahead is the honest exception. Our typical miss beats the posted number there, but our average miss is worse, which means our occasional bad forecast is worse than theirs. The further ahead you look, the more the posted number goes stale and the further ahead ERstat gets. That is the whole point of forecasting rather than repeating.

By province, two hours ahead

Only provinces with enough live-publishing ERs to score appear here. Where we lose, we say so. New Brunswick publishes total visit length rather than time to a clinician, so its posted numbers describe a different quantity from the one our model was built to forecast, and we do not yet beat it there.

The longer test

The table above covers the forecasts we have actually published, which is a short window so far. To check the method over a longer stretch, the model is also retrained each night on older data and tested on months it has never seen.

Held-out test over 2026-06-12 to 2026-09-23 (103 days), 136,658 scored predictions across 231 emergency departments. The model never saw these hours during training. Each horizon is stated separately because an accuracy figure without a horizon cannot be checked: forecasting five minutes ahead and three hours ahead are different problems.

What this does not prove

• It measures our forecast of the hospital's own posted number. Provinces define that number differently, and none of them measure the thing a patient experiences end to end. Beating it is not the same as being right about your evening.
• It covers only the roughly 202 ERs that publish a live wait. Hospitals that publish nothing cannot be scored this way at all, because there is no later number to check against. For those, the held-out test above is the only evidence, and it is evidence about the method rather than about that specific hospital.
• The live window is still short. It lengthens every night, and this page always shows the current figures rather than a frozen best result.

Ready to quote: “Forecasting the wait two hours ahead, ERstat was typically off by 33 min against what hospitals themselves later posted, compared with 41 min for the hospitals' own current posted figure, and was the closer of the two 58% of the time across 444,681 forecasts (ERstat).”

What the open data covers

• ER wait times: live posted waits for 800+ hospitals, normalized into one national schema, with monthly snapshots archived for trend analysis.
• Closures & disruptions: every ER currently closed or on reduced service, with status message and expected reopening where published, and a flag for scheduled vs unplanned.
• Coverage by province: how many ERs each province has, and how many publish a live wait time.
• Monthly snapshot archive: point-in-time CSVs of the above, preserved each month.

Free for non-commercial use under CC BY-NC 4.0 with attribution to erstat.ca. Full event-level history, wait-time time series and bulk export are available under a commercial licence.

How to access it

OpenAPI spec · API index (JSON) · Live feed

How to cite this data

In news copy, a visible link is all we ask: “according to data from ERstat”. For academic work, each dataset has its own DOI:

APA

ERstat. (2026). Canadian emergency-room wait times [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21940685
ERstat. (2026). Canadian ER closures and service disruptions [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21853002

MLA

"Canadian Emergency-Room Wait Times." ERstat, 2026, erstat.ca/data. Accessed 2026-09-23.

BibTeX

@misc{erstat_waits,
  author = {{ERstat}},
  title  = {Canadian Emergency-Room Wait Times},
  year   = {2026},
  doi    = {10.5281/zenodo.21940685},
  url    = {https://erstat.ca/data},
  note   = {Accessed 2026-09-23}
}

Each DOI resolves to the latest archived version on Zenodo; every monthly snapshot is preserved as its own citable version.

Methodology in brief

Wait times and statuses come from official health-authority feeds and hospital announcements, normalized into one national schema and refreshed continuously. Live wait figures reflect what each ER is publicly posting at the moment of the snapshot; they describe posted waits, not a modelled patient-experienced wait. Closures are marked scheduled (routine, posted operating hours) or unplanned. Coverage began in February 2026. For per-province detail on which health systems publish live waits, and which do not, see the transparency page.