Winter 2026/27 A&E forecast: what to plan on
One-page memo for planners. Prepared by Ellen Marsh, Fedelm Advisory, 31 October 2026.
The forecast. Between December 2026 and March 2027, emergency admissions via A&E across England are expected to peak at TBD: central estimate a month, with a 90% range of TBD: low to TBD: high. Ranges for each of the 36 integrated care boards and the seven regions are in the attached spreadsheet. TBD: populate from the 31 October run
What to plan on. The range is the planning object, and the top of it is the number to hold capacity against. On the sealed test described below the ranges held more often than they claimed, so planning on the top of the range is conservative. A trust-level table of breach probabilities is attached; every row is labelled illustrative, because the occupancy reconstruction behind it missed its registered accuracy test (median error 8.7 percentage points against a tolerance of 5). Use it to compare trusts, not to set a number.
What this forecast is. Monthly forecasts of attendances and emergency admissions, built from NHS England’s published A&E statistics with a classical exponential-smoothing model and its own simulated ranges. That model was fixed by a rule written before the results were in. A machine-learning model with a calibration layer was tested alongside it and produced tighter, better-calibrated ranges; it is the candidate for next winter, once a second sealed test can decide it, and it did not change this year’s forecast because doing so after seeing the results would contaminate the record.
How much to trust it. On a test window sealed before evaluation, nineteen forecast rounds from January 2024 to September 2025, the published model’s 90% ranges contained 97–98% of ICB-level outturns: wider than they need to be, erring towards over-warning rather than under-warning. In the development years outside the pandemic its ICB ranges held 89–93%; in every month including 2020–21 they held 78–81%, and the misses fell above the range more often than below it. From 12 November each month’s forecast is scored against the outturn as NHS England publishes it, so by January you will be able to see whether this winter is behaving like the backtest.
What it cannot do. It forecasts demand, not beds; the bed table depends on an occupancy and length-of-stay assumption that has not passed validation. It will not anticipate a regime change of the 2020 kind; it recovers after one. It is independent work on public data, with no NHS endorsement.
Assumptions. Data as published on 8 October 2026; forecasts at ICB, region and England level; winter months December to March; ranges are 90% intervals from the model’s own simulation. Every figure here traces to a file in the public repository at github.com/ellenmarsh1612/nhs-ae-forecast; the technical account is at DOMAIN-TBD/projects/nhs-ae-forecasting.html.
Questions: email address TBD.