NHS bed occupancy simulator

Does the simulator match reality?

The simulator is tuned to reproduce a handful of published figures for 2025/26. That proves little on its own: with enough settings, any model can hit its targets. The real test is everything it was not tuned to: other years, other months, individual trusts, and measures it never saw. This page runs those tests and reports the failures as plainly as the successes. The pass marks were set before the tests were run (validation plan).

In brief.

1. How full hospitals are: the model gets this mostly right

Occupancy by year: published and simulated

Each year run with its own beds and emergency admissions, first with stays held at 2025/26 values, then with each year's measured overnight stays and same-day share.

Occupancy each month: published and simulated

G&A beds occupied at 08:00, April 2022 to March 2026, each year with its own beds, admissions, stays and flu wave.

Weekly pattern, winter 2025/26

Average 08:00 occupancy by day of the week.

2. Long waits: right today, but only with a change after 2020

Patients spending over 12 hours in A&E, counted two ways

Type 1 A&E departments, England. From arrival: all attendances, admitted or not (HES A&E to 2019/20, ECDS from 2022/23). From the decision to admit: admitted patients only (A&E monthly situation reports). Only years in NHS England Digital's comparison tables are shown.

Show as table

12-hour waits by year: published and simulated

Share of emergency admissions waiting over 12 hours from the decision to admit. The dashed grey line converts each year's waits counted from arrival into the same units (scaled to 2024/25), which is what the pre-pandemic setting is fitted to. The simulated years hold stays at 2025/26 values, so 2022/23, when stays were longer, comes out emptier than it was and with too few waits.

12-hour waits each month: published and simulated

Share of emergency admissions waiting over 12 hours from the decision to admit, today's flexibility.

Corridor care

Since May 2026 NHS England has published two counts of corridor care, defined as care for 45 minutes or more in a clinically inappropriate place (source; experimental statistics). Ward corridor care counts patients on a G&A ward but outside a bed space at 08:00. That matches the model's corridor spaces, used only when every proper bed is full, and their number is fitted to the average count. Corridor care in emergency departments counts patients over each day, including people never admitted, and has no exact equivalent in the model.

Ward corridor care by month

Average patients in corridor care on G&A wards at 08:00.

Ward corridor care by day of the week

Average patients at 08:00, Monday to Sunday.

3. What changed after 2020? Testing the flexibility explanation

The model can only reproduce today's waits if beds behave as if split into small pools that cannot lend to each other, about beds each, and 2018/19's waits (counted from arrival) only with pools of about . "Bed flexibility" is a name for whatever makes a full hospital produce longer waits. Four questions test whether that explanation holds up.

Could anything else in the model close the gap?

Each pre-pandemic year is run with its own beds, admissions and occupancy, but today's flexibility, while everything else is varied over a wide range: discharge delays, corridor spaces, steadier stays, 7-day discharging and no flu season (scripts/flexibility_check.mjs).

Show the runs

Is the change real, or an artefact of how waits are recorded?

A pool size is one number, but it makes several predictions it was never fitted to: how 4-hour and 12-hour waits relate, how waits rise with occupancy, and what happens on the busiest days. These tests check them, and check how much of the change survives on measures that do not depend on when a decision to admit is recorded (pool_tests.mjs, deadline_test.mjs, pool_slopes.py, daily_tests.py, infection_tests.py, site_tests.py, clock_test.py).

Show the eight tests (A to H)

Which trusts saw the biggest rise in long waits?

A panel of acute trusts for three years before the pandemic (2017/18 to 2019/20) and three after (2022/23, 2023/24, 2025/26) asks whether, at the same occupancy, long waits rose more in trusts with a given characteristic (Poisson regression with trust and year fixed effects, 95% confidence intervals clustered by trust; scripts/trust_flexibility.py).

Show the trust panel

Do long-stay patients explain the rise?

Four follow-up tests: trusts compared with themselves month by month (A), long stays measured before the pandemic (B), competing explanations (C), and a separate simulation of a hospital split into 20 specialty wards, testing whether long-stay patients blocking particular beds can produce the pattern (D) (scripts/longstay_tests.py, scripts/longstay_mechanism.mjs).

Show the four tests

4. Trust by trust: one national model is not enough

Acute trusts: 12-hour waits, simulated and published

Each trust run with its own beds, admissions and occupancy and the national flexibility, against its published share of emergency admissions waiting over 12 hours from the decision to admit (2025/26).

Does each trust have its own flexibility?

Here every acute trust gets its own pool size in each post-pandemic year (2022/23, 2023/24, 2025/26), with the model scaled to the trust's beds and emergency admissions, its stays set to match the trust's occupancy, and the pool fitted to its 12-hour waits. Are those pools stable, do they predict later waits, and what do they go with (scripts/trust_pools.mjs, scripts/trust_pools_analysis.py)?

Show the trust-pool tables

5. How far to trust the answers

Show the detailed results

6. Can area data measure the effect of public-health policies?

This project also tried to estimate the effects of smoking, obesity and flu vaccination on emergency admissions by comparing ~150 local authorities, both between areas (adjusted for deprivation) and within areas over time (scripts/effects_analysis.py). A placebo test settles it: flu vaccination should not change falls or hip-fracture admissions, but in these data it appears to. The designs pick up confounding, so the simulator uses OHID's attributable-admission estimates and published studies instead (see Data & sources).

Show the area estimates

What this means for using the simulator

Scorecard: every pre-set check

All results are produced by scripts/validate.mjs and scripts/validate_extra.mjs. "Reported" means a figure is shown without a pass mark, usually because it was used in calibration or is not the same measure.

Show the core checks (V, C and L)
Show the further tests (E1 to E6)