Hospital bed capacity and usage across secondary healthcare providers in England during the first wave of the COVID-19 pandemic: a descriptive analysis

Objective In this study, we describe the pattern of bed occupancy across England during the peak of the first wave of the COVID-19 pandemic. Design Descriptive survey. Setting All non-specialist secondary care providers in England from 27 March27to 5 June 2020. Participants Acute (non-specialist) trusts with a type 1 (ie, 24 hours/day, consultant-led) accident and emergency department (n=125), Nightingale (field) hospitals (n=7) and independent sector secondary care providers (n=195). Main outcome measures Two thresholds for ‘safe occupancy’ were used: 85% as per the Royal College of Emergency Medicine and 92% as per NHS Improvement. Results At peak availability, there were 2711 additional beds compatible with mechanical ventilation across England, reflecting a 53% increase in capacity, and occupancy never exceeded 62%. A consequence of the repurposing of beds meant that at the trough there were 8.7% (8508) fewer general and acute beds across England, but occupancy never exceeded 72%. The closest to full occupancy of general and acute bed (surge) capacity that any trust in England reached was 99.8%. For beds compatible with mechanical ventilation there were 326 trust-days (3.7%) spent above 85% of surge capacity and 154 trust-days (1.8%) spent above 92%. 23 trusts spent a cumulative 81 days at 100% saturation of their surge ventilator bed capacity (median number of days per trust=1, range: 1–17). However, only three sustainability and transformation partnerships (aggregates of geographically co-located trusts) reached 100% saturation of their mechanical ventilation beds. Conclusions Throughout the first wave of the pandemic, an adequate supply of all bed types existed at a national level. However, due to an unequal distribution of bed utilisation, many trusts spent a significant period operating above ‘safe-occupancy’ thresholds despite substantial capacity in geographically co-located trusts, a key operational issue to address in preparing for future waves.

codes where appropriate. In some cases it was necessary to resort to using site names where no codes were present on the sheets containing hospital-level information regarding general and acute and critical care bed availability and occupancy. It was immediately apparent that extracting comprehensive site-level data from these records was non-trivial and for reasons discussed later, we maintain two datasets moving forward: one at site-level and one at trust-level that is used to aggregate to STP, Regional and Total figures as well.
Bed availability and total occupancy was recorded directly for G&A and critical care beds, alongside percentages of covid-confirmed occupants allowing for the calculation of a covid / non-covid / unoccupied breakdown for the G&A beds only (due to discrepancies in the definition of HDU / ITU and critical care beds, the percentage occupancy for critical care beds often resulted in impossible values of over 100%; it was decided to forego calculating a covid-breakdown for these beds due to how prolific these inconsistencies and issues were). For all of the other bed types, data was recorded in a different way. The number of covid positive patients (and in some cases covid-suspected patients), non-covid patients and the remaining unoccupied beds were recorded, allowing for total occupancy and availability to be calculated through simple transformation of these columns.
There are two key dates and several more minor milestones in the period we have data for (26th March to 5th June) where significant, non-trivial changes occurred in the site report structure and content. Prior to April 1st there was no information on bed availability beyond G&A and Critical Care beds; only the number of covidpositive patients were recorded for each type of bed. After the 1st of April, more granular bed availability was provided along with the means to work out the covid/non-covid breakdown of occupancy for Ventilated beds. From the 27th April onwards similar breakdowns and availability were recorded for HDU / ITU, IDU and most other types of bed at a site-level.
After loading in the data and accounting for the above described changes to its composition, the trust-level data used for the majority of our analysis had: -8.7% of Ventilated bed non-covid and unoccupied numbers missing across all records (no missing records for covid occupancy) - No missing records for G&A, Critical care bed availability and occupancy -No missing records for HDU / ITU after April 26th, otherwise 45.0% -All other columns containing information regarding the hospital, trust, etc. were complete Both datasets were filtered to remove children's hospitals, mental health hospitals and other sites / trusts that were not relevant to the analysis. STP linkage data was acquired via NHS Digital's library of public datasets source: (https://digital.nhs.uk/services/organisation-data-service/data-downloads/other-nhs-organisations) and augmented to include populations within STPs to facilitate our "beds-per-capita" figures (values were scrapped manually from the NHS England website, source: https://www.england.nhs.uk/integratedcare/stps/view-stps/). It was found that 7 trust codes were duplicated across 2 STPs; it was inappropriate to double count them so they were arbitrarily assigned to one of the STPs. The following table contains the STP and Trust code pairs that were chosen / removed from the linkage data to ensure a one-to-one mapping: Additionally, due to some trust-level mergers that took place and missing data in the source, 4 updated STP-Trust pairs were manually added to the linkage data to facilitate their inclusion in the analysis (source: Finally, it was found that two STPs spanned two regions. It was decided that QHM should fall under the North West region (all but one of its trusts are in that region) and QF7 should fall under the Midlands region (all but one of its trusts are in that region). The region definitions are inferred from the regions assigned to each trust in the site reports making up our primary dataset.
Despite our best efforts there were some missing values that persisted in critical columns outside of the key milestones mentioned in the section above. Moreover, in preparing the data it was noted that on several occasions there were substantial and improbable changes in the number of available beds that lasted 24 hours (even after allowing for the weekly trend of cyclical fluctuations in beds availability), prior to reversion to a value that fit the overall trend. These outliers follow from the reasonable assumption of the presence of data entry errors; it was decided that a cleaning rule should be applied to the data to avoid these seemingly impossible daily fluctuations and outliers.
First, a rolling median centred on each record was calculated using the 5 applicable days surrounding the record (smaller windows used at extremities of the data with correction not being possible at its absolute extremes). Missing values as well as values deemed to be outliers (a change greater than the 95th percentile of all differences between each record and the centred median spanning five days around it) were replaced with the aforementioned rolling median values. Highly improbable fluctuations were filtered out and missing values could be imputed in a robust way. This imputation and outlier detection process was applied to every applicable bed column spanning every type contained in the data. Only after this cleaning took place were other columns created through transformation, e.g. the number of available ventilated beds etc. The effect of cleaning the data is shown below in a before and after comparison, 4 trusts were chosen for their high initial volatility in G&A bed occupancy (See SFigure 1 & SFigure 2).

Statistical Analysis Notes
Temporalized values, i.e. hospital-days, were calculated by multiplying the absolute number of each functional unit for which data was available, and the number of days for which data is available for each.
After cleaning the data, two more key issues had to be dealt with in the trust-level and site-level datasets respectively: 1. Due to the aforementioned trust-level mergers, the composition of the data changed slightly throughout its duration. In an effort to achieve consistency, we merged and coalesced records prior to each mergers' appearance in the data to match their state post-merger. I.e. any rows corresponding to trusts that were eventually merged into some other trust were merged consistently throughout the dataset, even before this change actually took place. This was applied to records for the trusts RQ8, RDD and RAJ which were merged to fall under the single code RAJ on April 1st. This was also applied to RC9 and RC1 merged into RC9, and RA7 and RA3 merged into RA7: mergers that also occurred on April 1st and were reflected in the data shortly after. 2. It was observed that in one of the sheets relied upon for ventilated bed numbers, separate rows were included for both the sites and the corresponding trusts (given a "catch-all" label as their organisation type rather than "site"). In cases where only one hospital was associated with a trust, the numbers for that hospital were sometimes -inconsistently -recorded in the catch-all row rather than the site row as was done fairly consistently across all other situations. To achieve consistency without losing significant portions of the site-level data, we coalesced those rows where only one site was present and the catch all row contained numbers whilst the site row had zeroes or missing values. In order to achieve this, the organisation types of the two rows were swapped so that the catch-all row would be used in place of the site row, such that the site code and name was consistent throughout the entirety of the data.

Data Limitations
One persistent concern was the formulas by which bed occupancy proportions were generated. For example, the COVID-19 specific G&A bed percentage-occupancy was initially calculated as the sum of COVID-19 patients in IDU (infectious disease unit) beds and COVID-19 patients in "any other beds" divided by the total number of available G&A beds. This eventually changed to being the sum of the number of mechanical ventilated beds, non-invasive ventilated beds, oxygen-supporting beds and "any other beds" occupied by COVID patients minus the number of HDU / ITU beds occupied by confirmed COVID patients, all divided by the total number of available G&A beds. Whilst this is not in-and-of-itself problematic, the nature of the "any other beds" item was deemed concerning by the authors.
To understand the aforementioned concern, we first need to explain the data specification in more detail. It was noted that columns of the form "Number of Covid-19 confirmed patients in … beds at 0800" did not seem to contain values consistent with "Number of … beds available, as at 08:00 (COVID)", which we expected to have mirrored values. Importantly, the latter set of columns did not contain an "any other bed" column. As such, the formula used by NHS-E in the above calculation of G&A bed proportions drew the "any other beds" value from the first set of data, whereas all of the other information was drawn from the latter columns as they were internally consistent. We acknowledge that the use of this formula could have introduced an error of unknown magnitude or direction (as the two versions of data reporting were not consistent). Similar issues were seen with the independent sector data as well.     ). See SFigure 4 for a visual summary of these results. See SFigure 5 for the aggregate occupancy, stratified by COVID-19 status at the regional level.
In the context of baseline capacity, at the trust-level, 2620 trust days (22.1%; median number of days per trust = 27 [range: 1 to 69]) were at or above 100% capacity, which corresponds to 92 trusts spending at least 1 day at, or above, their-pre-pandemic baseline. 230 trusts days (median number of days per trust = 9 [range: 1 to 49]) were at or above 200% capacity, which corresponds to 19 trusts spending at least 1 day more than 100% above their-pre-pandemic baseline. At the STP-level, 620 STP days (median number of days per STP = 24 [range: 1 to 63]) were at an occupancy-level above 100% of baseline availability, which corresponds to 27 STPs spending at least 1 day at, or above, their-pre-pandemic baseline. 44 STP days (median number of days per STP = 14 [range: 10 to 20]) were at an occupancy-level above 200% of baseline availability which corresponds to 3 STPs spending at least 1 day more than 100% above their-pre-pandemic baseline. See SFigure 8 for a visual summary.

HDU/ITU Beds
The following results should be interpreted in the context of the date range available, i.e. data is only present after the 27th of April. Thus, the results are likely a significant underestimation of peak occupancy as, in retrospect, the peak number of cases and fatalities in the UK were near the beginning of April.
In the context of surge capacity, at the site-level, 315 hospital days (2.7%; median number of days per hospital = 2 [range: 1 to 39]) were at or above 85% of capacity, which corresponds to 59 hospitals spending at least 1 day at, or above, the aforementioned threshold. 216

Legend: The proportion of all trusts, and sustainability and transformation partnerships (STPs), at varying general and acute (G&A) bed occupancy thresholds relative to their baseline (mean availability January-March 2020) capacity, across England, from April 1st to June 5th. The superimposed colours represent how long the trusts spent at each specific threshold.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)