NHS bed occupancy simulator

Analysis ·

Why England's hospitals are full, and why the waits exploded after 2020

England's hospitals run at about 92% full, with fewer general and acute beds per person than ever. Waits in A&E were already lengthening before the pandemic. Since 2022 the same fullness has produced several times as many 12-hour waits in A&E. We put three decades of published NHS data together with a simulation of every emergency and planned admission to find out why, who is to blame, and what would help.

In brief.

1. A system with no slack

NHS England counts general & acute (G&A) beds open overnight in its KH03 return. On a consistent definition, England had such beds in 1987/88 and has today. Over the same period the number of people aged 75 and over, who account for nearly half of emergency bed days, grew from to .

G&A beds open overnight, England

Dashed: years before 2010/11, converted to today's definition (× 0.895).

Show as table

Beds per 1,000 people aged 75+

People aged 75+ account for of emergency bed days (HES 2023/24).

Show as table

Most of the fall in beds came before 2010. Since then the number of beds has barely changed while the population kept growing, so beds per person have kept falling. Occupancy is beds in use divided by beds, so its rise splits exactly into two parts:

G&A beds per 1,000 people: open and in use

KH03 beds open overnight and occupied overnight (average for the year), ONS mid-year population, England. The gap between the lines is the empty beds.

Show as table

Length of stay per admission

Mean nights per admission among patients who stayed at least one night (day cases and same-day admissions left out), as defined by the Health Foundation. HES, England, general and acute care, all ages, financial years. HES's published tables do not split bed days by admission method, so emergency stays are measured on diagnoses where at least 85% of admissions are emergencies (about half of all emergency admissions), and planned stays on diagnoses where at least 85% are planned. Diamonds: the Health Foundation's own figures from patient-level HES (adults, calendar years 2019 and 2022). Same-day stays were first published in 2013/14.

Show as table

Emergency admissions who go home the same day

Share of emergency admissions with no night in hospital ("zero bed day cases", HES, general and acute care), e.g. same-day emergency care. Not published before 2013/14.

Show as table
More trends: admissions, age, bed days per person and long stays

Emergency admissions per 1,000 people

Finished admission episodes (one per admission) with an emergency admission method (HES), ONS population. 2020/21 is the first pandemic year.

Show as table

Emergency admissions aged 75 and over

Share of emergency admissions, general and acute care (HES). HES gives age for all episodes of a diagnosis, not by admission method, so each diagnosis's emergency admissions are split by its age profile, as in the simulator.

Show as table

Emergency bed days per 1,000 people

Admissions times nights per admission: nights spent in hospital after an emergency admission each year, general and acute care (HES), per 1,000 people (ONS). Up to 2019/20, HES bed days per person moved with beds in use (KH03), both down about 12% from 2010/11. Since 2022/23 HES bed days have risen faster than beds in use, which were flat, so part of the recent rise may be in how stays are recorded rather than in beds.

Show as table

Beds held by patients in hospital for a long time

Share of occupied G&A beds, England, mean of December to February (winter daily sitreps). Not collected in the winters of 2020/21 and 2021/22.

Show as table

These follow the Health Foundation's definitions ("Longer hospital stays and fewer admissions", 2023), which count stays per admission, not per consultant episode as HES's own "mean length of stay" does (about 4.5 nights in 2019/20). From published tables, emergency stays come out at (2019/20) and (2022/23), against their 7.9 and 9.1 days; all overnight admissions at and , against 7.3 and 8.3. Planned stays come out longer than theirs ( against 5.2), so that series is less reliable.

By 2019 hospitals were running above 90% full with little room to absorb a bad week. That made them fragile. It does not explain what happened next.

2. After 2020: the same fullness, far longer waits

A full hospital makes patients wait. When no bed is free, the patient stays in A&E, sometimes on a trolley in a corridor, until one is. NHS England counts these waits in two ways. Its monthly figures start the clock when a doctor decides to admit: on that measure 12-hour waits happened to , and to in , the year the simulator is set up for. Before 2020 they were almost unheard of. But that measure exaggerates the change, because before the pandemic the decision to admit was often recorded late. Hospital records also time each patient's whole stay in A&E from arrival, and on that clock long waits were common before the pandemic too.

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

Type 1 A&E departments, England, each financial year with published data. 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

Counted from arrival, patients spent over 12 hours in A&E in 2018/19, against counted from the decision to admit. From arrival, 12-hour waits have risen about -fold since then, not -fold. Other measures that do not depend on when a decision is recorded tell the same story at the same occupancy: 4-hour waits counted from arrival roughly doubled, and ambulance handover delays rose about fourfold. The real change is a several-fold rise in long waits at the same level of fullness.

The arrival-based 12-hour count is published only once a year. The arrival-based measure published every month is the share of patients in major (type 1) A&E departments who spend over 4 hours there, and it shows the same break against occupancy. Each dot below is one quarter.

Occupancy and 4-hour waits in A&E, each quarter since 2010

x: G&A occupancy (KH03). y: share of type 1 A&E attendances spending over 4 hours in A&E, counted from arrival (A&E attendances and emergency admissions time series). Includes patients who are not admitted.

Show as table

Why the same fullness can mean far longer waits

Queueing theory explains how. Picture 600 beds running 92% full. On an average night about 48 are empty, and even on a busy night a few are free somewhere, so a patient coming up from A&E rarely waits long. Now split the same 600 beds into ten wards of 60 that cannot take each other's patients, each also 92% full. Each ward has about five empty beds on an average night, but arrivals and discharges bunch together, so on any night some wards are full while others have beds they cannot lend. Same beds, same patients, same average occupancy, far more waiting. The smaller and less interchangeable the groups of beds, the more often a patient arrives to find their group full.

Simulated 12-hour waits by occupancy, with beds shared in pools of different sizes

Today's patients run with between 98,000 and 122,000 beds; everything else as in today's simulator. A pool is a group of beds whose patients can use any bed in the group. Share of emergency admissions waiting over 12 hours for a bed.

Show as table

That is how the simulator represents the break. It reproduces today's waits only if England's beds behave as if split into pools of about beds, and 2018/19's waits (counted from arrival) only with pools of about . On the chart, the system jumped from one curve to a steeper one while occupancy barely moved. "Pool size" is not something anyone counts: real beds are split by hospital, ward, specialty and single-sex bay, and how easily a patient can be placed elsewhere varies from day to day. In the simulator it stands for whatever now makes a full hospital produce longer waits. The tests point to patients being harder to place on other wards, not to infection control or hospital layout, and bigger trusts did only slightly better, so it is not simply a matter of hospital size (tests). We show both versions throughout, as "today's hospitals" and "pre-pandemic flexibility".

Nothing else in the model closes the gap. Give it 2018/19's own beds, admissions and occupancy with today's flexibility, and it predicts of emergency admissions waiting over 12 hours, against the that 2018/19's arrival-based waits imply. At the same occupancy, removing discharge delays makes almost no difference (delays matter by making hospitals fuller), and nor does any other single change. Only every favourable assumption at once (steadier stays, 7-day discharging, no flu season and double today's corridor spaces) comes close, at .

So something about how hospitals work changed. The next section tests the four usual suspects.

3. What broke? Beds, social care, GPs, or how hospitals run

Four explanations dominate the debate. There are simply too few beds. Social care cannot take patients who are ready to leave, so they block beds. GPs are too hard to see, so patients end up in A&E and then in hospital. Or hospitals themselves manage the flow of patients badly. We tested each against published data, then asked the simulator how much each could matter. No single test proves cause, but together they point one way.

What changed between and today

Start with what actually changed. The simulator reruns and switches its measured conditions to today's one at a time: how many patients arrive, how many beds there are, how long overnight patients stay (including waiting to be discharged), how many go home the same day, and how flexibly beds are used. Averaging over every order of switching splits the change exactly between the five.

What drove the rise in 12-hour waits, to

Simulated change in the share of emergency admissions waiting over 12 hours for a bed, split between five measured changes (Shapley values; they add up to the total). Positive bars pushed waits up. Waits here are the simulator's, counted from the decision to admit; its pool size is fitted to that year's waits counted from arrival.

Show as table

Suspect 1: too few beds

Suspect 2: social care

Why patients who are ready to leave are still in hospital

Patients in hospital 7 days or more who no longer meet the criteria to reside, by the reason they are still there, grouped by who has to act. Average of weekly snapshots, , England (acute discharge situation report). "Service not available": the care home place, home care or rehab is not there. "Arrangements under way": it is being organised.

Show every reason

Do delays track long waits? Trusts compared with themselves, month by month

Change in 12-hour waits for each extra patient per 100 G&A beds delayed 7+ days for each reason, at the same occupancy. . Poisson regression with trust and month fixed effects; bars show 95% confidence intervals. Within-trust changes remove fixed differences between hospitals, such as size and case mix.

Show as table

Suspect 3: general practice

A&E attendances and emergency admissions per person

Indexed to = 100. All A&E attendances and major (type 1) A&E attendances, and all emergency admissions, per head of population (A&E monthly situation reports, ONS). 2020/21 is the first pandemic year.

Show as table

Admissions for conditions GPs can often manage

Chronic ambulatory care sensitive conditions (angina, asthma, COPD, diabetes, heart failure, epilepsy, dementia, atrial fibrillation, hypertension, anaemia; NHS Outcomes Framework 2.3i, three-character codes): share of emergency admissions and of emergency bed days, general and acute care (HES).

Show as table

Suspect 4: how hospitals run

Some hospitals lost far more flexibility than others

England's average hides a wide spread. Across acute trusts in , trusts at the same occupancy had very different long waits.

Acute trusts: occupancy and long waits,

Each dot is a trust. x: average 08:00 G&A occupancy. y: share of emergency admissions waiting over 12 hours.

Show as table

Are those differences new? One clue looked strong. Group acute trusts into fifths by the share of their beds held by patients in hospital three weeks or more. Since the pandemic, the fifth with the fewest long-stay patients has had of emergency admissions waiting over 12 hours for a bed, and the fifth with the most , at about the same occupancy (). On that measure the gap looks new, because before the pandemic every group had almost no 12-hour waits. But that is the recording problem from section 2 again: the decision-to-admit measure was near zero everywhere before 2020. On the arrival clock, waits rose about equally in every group, from of patients over 4 hours in A&E before the pandemic to after, whatever the long-stay share. Comparing each trust with itself, each standard deviation of long-stay share predicts a bigger rise in 12-hour waits from the decision to admit () but not in 4-hour waits from arrival ().

4-hour waits in A&E before and after the pandemic, by trusts' share of long-stay patients

Acute trusts grouped into fifths by the share of G&A beds held by patients in hospital 21 days or more (2025/26). Before: 2017/18 to 2019/20. After: 2022/23, 2023/24 and 2025/26. Share of type 1 A&E attendances spending over 4 hours in A&E, counted from arrival. The table also shows 12-hour waits from the decision to admit.

Show as table

So long-stay patients go with longer waits for a bed today, but there is no sign that they explain why waits broke. The 4-hour measure mostly covers patients who are never admitted, so it is a blunt test of anything that blocks beds. But the other long-stay evidence points the same way: long stays grew only a little, they move with long waits but do not lead them, and in a simulation of a hospital split into specialty wards, blocked beds are too small an effect to matter.

The four long-stay tests in detail
  • Long stays grew only a little. The winter sitreps of 2017/18 to 2019/20 show of beds held by patients in hospital three weeks or more, against since, and the same trusts had high shares in both periods. Before the pandemic, long stays did not predict long waits on either measure (12 hours from the decision to admit: ; 4 hours from arrival: ). After it, both a trust's earlier level () and its rise since () predict the increase in 12-hour waits from the decision to admit, but neither predicts the increase in 4-hour waits from arrival (; ).
  • The total number of patients fit to leave does not predict which trusts got worse (); the delay tests above show which kinds of delay do track waits. On the decision-to-admit measure, the long-stay result holds when each region has its own change ().
  • Long stays and long waits move together, but neither leads. Months with more long-stay patients have more long waits ( for each extra point of long-stay share), but last month's long stays do not predict this month's waits (), and earlier waits do not predict later long stays.
  • Blocked beds are too small an effect. In a separate simulation of a hospital split into 20 specialty wards, waits stay near zero when beds are fully shared () and reach today's level when patients can rarely be placed outside their own ward. At today's level of sharing, raising long-stay patients from , whether they are spread out or concentrated on a few wards.

The verdict

How much each fix would cut 12-hour waits today

Simulated change in emergency patients a year waiting over 12 hours, , today's system. Colour shows which explanation each fix tests. Delay fixes remove the bed days lost to that group of reasons (acute discharge sitrep). Waits in the simulator respond more steeply than hospitals' do, so these are upper bounds; compare the bars with each other.

Show as table

If the break is mostly about flow, the fixes are about flow too. But hospitals that run this full need room as well.

4. What would help

There are two kinds of fix: more room, and better use of the room there is. The simulator runs a full year of patients through England's G&A beds, with numbers set from published data for , so both can be compared on the same footing.

More beds

Simulated occupancy by number of beds

demand. Vertical line: beds today.

Show as table

Simulated waits over 12 hours by number of beds

Thousands of emergency patients a year.

The simulator can also give today's patients the number of beds that would match an earlier year's beds per person, with everything else as it is today.

Simulated waits over 12 hours if beds per person matched an earlier year

demand. Thousands of emergency patients a year. Each pair of bars is one bed count; the label shows the year whose beds per person it matches.

Show as table

Needing fewer beds

Fewer people needing admission, shorter stays and faster discharges all free up space. We ran each policy at a size that has been proposed or achieved somewhere. The effect of each on admissions and length of stay comes from measured data, not assumption: OHID attributable-admission estimates for smoking, a large cohort study for obesity, a test-negative study for flu vaccine effectiveness, NHS England's discharge data for delays, and HES readmission counts for readmissions (sources). Bars show the change from today. The biggest flow fix, restoring pre-pandemic flexibility, is in the verdict above and in every chart as the "more flexible" system.

Change in occupied beds

Average beds in use. Negative means space freed. Capacity levers add beds rather than freeing them, so they are not shown.

Change in waits over 12 hours

Thousands of emergency patients a year, today's system.

Show all policy results as a table
Show all policy results as a table

Beds occupied by patients well enough to leave

Patients no longer meeting the criteria to reside, England, 7-day average.

Show as table

Readmissions: patients who come straight back

Emergency readmissions within 30 days of discharge, England

Share of eligible discharges followed by an emergency readmission within 30 days. Excludes day cases, maternity, cancer and deaths. The 2020/21 peak reflects the pandemic, with fewer and sicker admissions. OHID Fingertips 41101, from HES.

Show as table

Flu: the winter wild card

Flu fills thousands of beds for a few weeks each winter, and no two seasons are alike. In 2024/25 a sharp early-January peak put flu patients in G&A beds on a single day. In 2025/26 the peak came earlier and lower, at .

G&A beds occupied by patients with flu, daily

UEC daily situation reports, winters 2024/25 and 2025/26, aligned by date.

Show as table

5. The next ten years

What this analysis can and cannot tell you

Try your own scenario in the simulator. Every input and its source is listed on Data & sources, and all code is on GitHub.