Article Text

Role of hospital strain in determining outcomes for people hospitalised with COVID-19 in England
Free
  1. William K Gray1,
  2. Annakan V Navaratnam1,
  3. Jamie Day1,
  4. Johannes Heyl1,2,
  5. Flavien Hardy1,
  6. Andrew Wheeler1,
  7. Sue Eve-Jones1,
  8. Tim W R Briggs1,3
  1. 1 Getting It Right First Time programme, NHS England, London, UK
  2. 2 Department of Physics and Astronomy, University College London, London, UK
  3. 3 Department of Surgery, Royal National Orthopaedic Hospital NHS Trust, London, UK
  1. Correspondence to Annakan V Navaratnam, Getting It Right First Time programme, NHS England, Wellington House, 133-155 Waterloo Road, London, SE1 8UG, UK; annakan.navaratnam{at}nhs.net

Abstract

Background In England, reported COVID-19 mortality rates increased during winter 2020/21 relative to earlier summer and autumn months. This study aimed to examine the association between COVID-19-related hospital bed-strain during this time and patient outcomes.

Methods This was a retrospective observational study using Hospital Episode Statistics data for England. All unique patients aged ≥18 years in England with a diagnosis of COVID-19 who had a completed (discharged alive or died in hospital) hospital stay with an admission date between 1 July 2020 and 28 February 2021 were included. Bed-strain was calculated as the number of beds occupied by patients with COVID-19 divided by the maximum COVID-19 bed occupancy during the study period. Bed-strain was categorised into quartiles for modelling. In-hospital mortality was the primary outcome of interest and length of stay a secondary outcome.

Results There were 253 768 unique hospitalised patients with a diagnosis of COVID-19 during a hospital stay. Patient admissions peaked in January 2021 (n=89 047), although the crude mortality rate peaked slightly earlier in December 2020 (26.4%). After adjustment for covariates, the mortality rate in the lowest and highest quartile of bed-strain was 23.6% and 25.3%, respectively (OR 1.13, 95% CI 1.09 to 1.17). For the lowest and the highest quartile of bed-strain, adjusted mean length of stay was 13.2 days and 11.6 days, respectively in survivors and was 16.5 days and 12.6 days, respectively in patients who died in hospital.

Conclusions High levels of bed-strain were associated with higher in-hospital mortality rates, although the effect was relatively modest and may not fully explain increased mortality rates during winter 2020/21 compared with earlier months. Shorter hospital stay during periods of greater strain may partly reflect changes in patient management over time.

  • COVID-19
  • hospitalisations

Data availability statement

Data may be obtained from a third party and are not publicly available. This report does not contain patient identifiable data. Consent from individuals involved in this study was not required. Requests for any underlying data cannot be granted by the authors because the data were acquired under licence/data sharing agreement from NHS Digital, for which conditions of use (and further use) apply. However, individuals and organisations can access HES data by direct request to NHS Digital.

This article is made freely available for personal use in accordance with BMJ’s website terms and conditions for the duration of the covid-19 pandemic or until otherwise determined by BMJ. You may use, download and print the article for any lawful, non-commercial purpose (including text and data mining) provided that all copyright notices and trade marks are retained.

https://bmj.com/coronavirus/usage

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

WHAT IS ALREADY KNOWN ON THIS TOPIC

  • In England, in-hospital COVID-19-related mortality rates rose during winter 2020/21 relative to earlier summer and autumn periods.

  • It has been suggested that this was due to high levels of strain on services due to large numbers of people hospitalised with COVID-19.

  • There have been very few studies globally investigating the link between patient numbers and in-hospital mortality rates.

WHAT THIS STUDY ADDS

  • We identified an association between higher daily COVID-19 hospital activity (new admissions and current inpatients) and higher in-hospital mortality rates.

  • The absolute difference in mortality rates between the highest and lowest quartiles of activity did not fully explain the increase in mortality rates during winter 2020/21.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

  • Clinicians and service managers should be aware of how strain impacted on patient outcomes during the COVID-19 pandemic to inform planning for future pandemics.

  • The reasons why hospital stay decreased during periods of greatest strain should be investigated more fully.

Introduction

The COVID-19 pandemic placed an unprecedented burden on health services globally. In England, the number of patients hospitalised during the second wave of winter 2020/21 was far higher than during the first wave of spring 2020.1 Outcomes for hospital patients with COVID-19 improved dramatically during spring 2020 and these improvements appear to have been maintained into the summer and early autumn.2 3 However, in late autumn/early winter 2020/21 mortality rates started to increase as patient number increased.4 Possible explanations for this increase in mortality rates include: (1) increased disease severity in those admitted due to changing admission policies or viral mutations (eg, Alpha or Delta variant)5 or late presentation, (2) long-established seasonal trends towards poorer outcomes during hospital stay in winter months,6 (3) strain on healthcare services.4 7

The aim of this study was to examine factors influencing mortality and length of stay for patients with a diagnosis of COVID-19 during periods of greatest activity.

Methods

Ethics

The analysis and presentation of data follows current NHS Digital guidance for the use of Hospital Episode Statistics (HES) data for research purposes. Reported data are anonymised to the level required by ISB1523 Anonymisation Standard for Publishing Health and Social Care Data.8

Study design and data collection

This was a retrospective exploratory analysis of HES data. HES data are mandatory and collected by NHS Digital for all NHS-funded patients admitted to hospitals in England. Hospital trusts run all NHS hospitals in England and a typical trust provides secondary and/or tertiary care for a geographically defined catchment population.

Timing, case ascertainment, inclusion and exclusion criteria

Data were extracted from the HES dataset by WKG, who has worked extensively with the HES data. HES data are entered by trained clinical coders at each trust and collated centrally by NHS Digital. The data undergo quality and validation checks (including duplicate removal, data field consistence checks, mandatory field completion checks) prior to being made available for onward processing.9 With regard to COVID-19 data in HES, various sensitivity analyses have been conducted in previous research to assess for potential biases introduced during data processing.2–4

We reviewed HES data for all completed episodes of hospital care in England with an admission date from 1 July 2020 to 28 February 2021 that involved a diagnosis of COVID-19. The start date was chosen as a period where much of the rapid learning during the early phase of the pandemic in England had been done and outcomes were relatively stable.2 We excluded patients aged <18 years and patients admitted to specialist trusts (eg, non-acute, mental health, community and single-specialty trusts), defined as those not contributing to the NHS England’s daily COVID-19 situation reports.10 Cases of COVID-19 were identified using the International Statistical Classification of Disease and Related Health Problems 10th edition (ICD-10) codes U07.1 (presence of COVID-19 has been confirmed by laboratory testing) and U07.2 (clinical or epidemiological diagnosis of COVID-19 where laboratory confirmation is inconclusive or not available). The data extraction process is summarised in online supplemental figure 1.

Supplemental material

Where a patient had multiple admissions during the study period, only the chronologically last admission was retained. This ensured that all admissions were independent at a patient level and avoided biasing the data by including cases where survival was predefined by virtue of a subsequent admission.

Outcomes

The primary outcome was in-hospital mortality as recorded by the Office for National Statistics, linked to HES at a patient level. An in-hospital death was assumed if the date of death was the same as or within 1 day of the hospital discharge date recorded in HES. Length of hospital stay in survivors and patients who died in hospital was a secondary outcome.

Exposure

The exposure was the level of COVID-19-related strain the hospital was under at the time of the admission of patients with COVID-19. We calculated two measures of strain, admission-strain and bed-strain. Admission-strain was defined as the total number of people with COVID-19 admitted to the hospital trust on the same day as the index patient divided by the maximum number of patients with COVID-19 admitted to that trust on any given day during the study period. Bed-strain was defined as the number of beds occupied by patients with COVID-19 in a trust on the day of the index admission divided by the maximum number of beds occupied by patients with COVID-19 on any given day. Strain data were categorised into quartiles for analysis. These definitions ensured that each trust had its own ‘maximum’ capacity which reflected actual activity and avoided including elective activity in the measures of strain.

Covariates

Age: categorised as 18–39 years, 40–49 years, 50–59 years, 60–69 years, 70–79 years and≥80 years for exploratory analysis and treated as continuous in the final multivariable model.

Sex: male or female.

Ethnicity: coded in categories used by NHS Digital (white, Bangladeshi, Indian, Pakistani, other Asian, black African, black Caribbean, other black, mixed, other, not stated).

Deprivation: recorded using the Index of Multiple Deprivation (IMD) for the lower super output area (LSOA) of the patients’ home address, with scores categorised into quintiles based on national averages. The IMD categorises all households in England into percentiles of relative deprivation based on their LSOA of residence. It includes items measuring income, employment, health and disability, education and skills training, crime, barriers to housing and services and living environment.

Comorbidities: these were the 14 comorbidities used in the Charlson Comorbidity Index (peripheral vascular disease, congestive heart failure, acute myocardial infarction, cerebrovascular disease, dementia, chronic pulmonary disease, connective tissue disease/rheumatic disease, peptic ulcer, liver disease (mild and moderate/severe), diabetes (with and without chronic complications), paraplegia/hemiplegia, renal disease, cancer (primary and metastatic), HIV/AIDS).11 The comorbidity was deemed present if it was recorded in HES as a secondary diagnosis in the index admission or as a primary or secondary diagnosis in any admission during the previous year, in accordance with the recommendations of Quan et al.12

Obesity: recorded as present if the ICD-10 code E66 was used as a diagnostic code during the admission.

Severity of COVID-19: in the absence of markers of disease severity on presentation (eg, blood biomarkers, clinical signs), severity of COVID-19 was assessed based on ICD-10 codes for a number of key complications/sequelae of COVID-19 recorded during the index admission (pneumonia, renal disease, blood clotting, cardiology/circulation, neurology, digestive system and sepsis). The ICD-10 codes used to identify these categories are listed in online supplemental table 1.

Admission date: categorised as weekly or monthly for descriptive purposes and modelled as day of admission.

Elective surgical activity

Elective inpatient, day-case and outpatient surgery data for patients aged ≥18 years during financial years 2019/20 and 2020/21 were extracted from HES as activity which met the intermediate definition of surgical activity set out by Abbott et al (requires use of an operating theatre and/or a regional or general anaesthetic).13

Data management and statistical analyses

Data were extracted onto a secure encrypted server controlled by NHS England and analysed using Microsoft Excel (Microsoft, Redmond, Washington, USA), Stata (StataCorp, College Station, Texas, USA) and Alteryx (Alteryx, Irvine, California, USA). For descriptive analysis, length of stay and age were summarised using the median and IQR. All other data are described by frequency and percentage.

In-hospital mortality was modelled using multivariable logistic regression including the covariates listed above. Length of stay was modelled separately for those who died in hospital and those who survived to discharge using negative binomial regression. A flexible parametric survival model was also constructed, which combined the effects of strain on in-hospital mortality and length of stay. In-hospital mortality was the event of interest and time was modelled as the time from admission to discharge using a spline with three equally spaced knots (25th, 50th and 75th percentile). A flexible model was preferred over a standard Cox model due to violation of the proportional hazards assumption based on analysis of Schoenfeld residuals.14

In all models, age was treated as a continuous variables and modelled using restricted cubic splines due to non-linearity. The optimal number of knot (three) and knot position (10th, 50th and 90th percentile) were identified using the recommendations of Harrell15 and with reference to Akaike Information Criterion. IMD score was modelled as a continuous variable. There was no evidence of non-linearity, thus IMD score was modelled as a linear term. All other variables were modelled as categorical terms. Strain was modelled as quartiles to aid interpretability. Given the potential of confounding of the strain covariate with admission date, sensitivity analyses were conducted with the covariate admission date excluded.

Fixed-effects models were preferred over mixed models due to prior evidence of limited intratrust clustering for the primary outcome.16 The model outputs are presented as adjusted estimates (‘margins’ command in Stata) and as ORs for the logistic model, as incident rate ratios for the negative binomial models and as HRs for the survival model. All models are presented with 95% CIs. Non-overlapping CIs were taken as an indication of statistical significance at the 5% level. For length of stay, the adjusted estimates from modelling are mean values; the median value cannot be obtained through modelling.

Other than for ethnicity, missing data were relatively rare (table 1), and no attempt was made to impute missing values. For ethnicity, a number of patients declined to state their ethnicity. In these cases, HES was searched for a prior hospital admission where ethnicity of that patient had been recorded and this value was used. Where more than one ethnicity was recorded for different admissions, the most recent record was used. Where data were missing, the numbers involved are stated.

Table 1

Patients’ demographic profile, timing of admission, illness severity and outcomes

Results

Patient demographics and outcomes

The data of 253 768 unique patients who had a diagnosis of COVID-19 either on admission or during their stay were extracted (online supplemental figure 1). Of these, 241 717 (95.3%) had test confirmed COVID-19; the remaining patients were diagnosed clinically.

The demographic profile of patients and their outcomes are summarised in table 1. There were 59 931 (23.6%) in-hospital deaths, among whom 48 168 (80.4%) had COVID-19 listed as the primary cause of death on their death certificate. Greater age was associated with higher mortality and longer hospital stay. Crude outcomes were generally poorer in the least deprived socioeconomic groups and in patients of white ethnicity (compared with all other ethnicities), although these data are confounded with age. Compared with those who survived to discharge, length of stay was longer for those who died in hospital. Severe COVID-19 was associated with a fivefold increased mortality rate and a doubling of the median hospital stay in survivors.

Trends in admission over time

The number of COVID-19 admissions rose rapidly from October 2020 and peaked in mid-January 2021 (table 1 and figure 1). The proportion of patients with severe COVID-19 followed the same timing (figure 1). The highest in-hospital mortality rate was in December 2020, prior to the peak in admissions. Length of stay decreased as patient numbers increased, and this was more noticeable for those who died in hospital than in those who survived to discharge. Median patient age peaked in November 2020 (73 years, IQR 57–83) compared with 69 years (IQR 53–82) in July 2020 and 67 years (IQR 52–80) in February 2021.

Figure 1

Measures of COVID-19 severity and patient admissions per week.

Table 2

Bed-strain and admission-strain quartile for each patient outcome

Elective surgical activity

Data on elective surgical activity for financial years 2019/20 and 2020/21 are presented in online supplemental figure 2 and show a rapid fall in activity in March–April 2020, followed by a steady recovering in activity levels from May 2020 onwards. There is a small fall in activity from November 2020 to January 2021, corresponding to the peak of patient numbers during the second wave of COVID-19 in England. However, by March 2021 the level of activity seen in October 2020 had already been surpassed.

Hospital strain

Crude in-hospital mortality increased with each increasing strain quartile for both measures, although most obviously for bed-strain (table 2). Length of stay in those who died decreased as both measures of strain increased; there was no strong association between strain and length of stay in those that survived (table 2). Both measures of strain showed a very similar trend over time and followed the same timing as the number of weekly admissions (figure 2).

Figure 2

Measures of admission-strain and bed-strain and patient admissions per week.

Since bed-strain showed a clearer unadjusted relationship with in-hospital mortality than admission-strain (table 2), it was chosen as the measure of strain during multivariable modelling. The model-adjusted relationship between each of the outcomes and bed-strain quartile is shown in table 3. The adjusted in-hospital mortality rate increased 1.7% from the lowest quartile (Q1) of bed-strain to the highest (Q4). Length of stay decreased with increasing bed-strain both for survivors (Q1–Q4 1.6 days fewer) and those who died in hospital (Q1–Q4 3.9 days fewer). The survival model showed a similar and consistent trend across quartiles of bed-strain, with an HR of 1.33 in Q4 compared with an HR of 1 in Q1. For all outcomes and models, the Q1–Q4 difference was larger when admission date was excluded as a covariate (online supplemental table 2).

Table 3

Adjusted outcomes and model parameters per bed-strain quartile

The adjusted mortality rates per month are presented in table 4 and suggest that our measure of bed-strain did not fully account for the increased mortality rate during winter 2020/21. As for the crude data, the highest adjusted in-hospital mortality rate was during December 2020, prior to the peak in admissions.

Table 4

Adjusted in-hospital mortality rates per month

Discussion

This study found that increased hospital strain was associated with higher in-hospital mortality rates. The absolute effect was relatively modest and did not fully explain the increased mortality rate during periods of highest patient numbers. Greater hospital strain was associated with shorter hospital stay. Compared with in-hospital mortality, length of stay was more clearly influenced by hospital strain.

Strain-outcome associations

A previous study of 144 116 patient admitted between March and August 2020 in 558 hospitals in the USA concluded that almost one-in-four deaths could be attributed to case surges.17 Using a reference category of below median case surges, the 95th–99th and >99th percentile of a surge index developed by the authors had increased odds of mortality of 1.59 (95% CI 1.41 to 1.80) and 2.00 (95% CI 1.69 to 2.38), respectively. These results are supported by other US-based studies.18 The influence of strain on mortality was more modest in our study. The reasons for this are likely to be multifactorial, but include the measure of strain used, the setting and the study timing.

In our study, there are a number of possible explanations for the observed higher mortality rates at times of greatest strain and a direct causal relationship should not be assumed. COVID-19 severity was a major contributor to in-hospital mortality, and patients admitted in winter 2020/21 had greater disease severity than those admitted in summer and autumn 2020. Greater severity of illness may reflect changing admission criteria, with only those deemed in urgent need of hospital care admitted at times of highest case numbers. Those with less severe presentations could be discharged with advice and monitoring (eg, home pulse oximetry and virtual wards).19 20 If this is the case, increasing disease severity of admitted patients may indirectly lead to increased service strain. Increased disease severity in winter could also reflect a general propensity for respiratory infections, such as COVID-19 and influenza, to be more severe in winter.21 Given that the greatest strain occurred during winter 2020/21, it is hard to fully disentangle the effects of strain and wider patterns related to seasonality.22

Length of hospital stay in survivors and those who died in hospital declined with increasing strain and over time. As for mortality, improved understanding of optimum patient management strategies and likely prognosis and initiatives to support early discharge in survivors will have contributed to shorter hospital stay.23 24 Implementation of such changes will have become more pressing during winter 2020/21 as the need to create bed space became a priority.

Measures of strain

We developed two measures of strain: bed-strain and admission-strain. Although neither measure will fully capture the strain on hospital services due to COVID-19 admissions, we feel they are pragmatic and, given their simplicity, have the potential to be applied to other settings and diseases. The measures do not require data on bed availability, staffing levels and skill-mix which are difficult to capture consistently, particularly during times of greater strain.10 Similarly, prepandemic data are often of limited value due to staff redeployment, staff illness and ward repurposing. Even if reliable, contemporary data were available, the relative importance of each bed availabilty, staffing levels and skill-mix in any model is likely to vary substantially from one setting to another and over time.

Although elective activity was largely suspended in England from 1 April 2020,25 it had partially restarted by July 2020. Our data on elective activity reveal a dip in activity during winter 2021/21, suggesting that during periods of strain, elective activity was reduced, and staff and beds redeployed to prioritise COVID-19 care. Furthermore, during the study period elective activity levels were often constrained by heightened infection control measures and new working practices. As such, any association between elective activity levels and strain is likely to be weak and including elective activity as part of a strain metric is unlikely to improve performance.

Our strain metrics did not include critical care admissions, as other studies have.17 18 In the UK, differential rates of access to critical care, particularly for frailer patients, make this an unsuitable marker for strain. Although the maximum number of patients with COVID-19 in a trust on any given day may not represent true capacity for all hospitals, the vast majority of hospitals in England experienced significant strain at some point during the study period.

Strengths and limitations

Our study period avoided the peak of the first wave of the pandemic in England, when much of the treatment was experimental and where hospitals had to adapt rapidly. By July 2020, hospital case numbers were relatively low and much of the rapid learning on patient management had been completed. Although many clinical trials were ongoing, or yet to report, significant changes in treatment options during the study period are unlikely to explain our findings.26–28 Our dataset covers all people admitted to hospital in England and thus collider bias should be minimal when considering hospital populations.29 Nevertheless, care should be taken when comparing our findings with other countries and healthcare settings. As with most administrative datasets, clinical data are lacking.30 We included data for test-confirmed and clinically diagnosed COVID-19 to ensure as complete coverage of hospital activity as possible. However, we recognise that some clinically diagnosed cases may not have had COVID-19. Likewise, we included deaths for any cause, including those unrelated to COVID-19 infection. Previous sensitivity analyses by our team suggest that this is unlikely to have introduced significant bias.2

Summary

Increased strain on hospitals caused by surges in patient numbers was associated with higher mortality rates and shorter hospital stay. However, the measures of strain used do not fully explain the increased COVID-19-related mortality observed in winter 2020/21.

Data availability statement

Data may be obtained from a third party and are not publicly available. This report does not contain patient identifiable data. Consent from individuals involved in this study was not required. Requests for any underlying data cannot be granted by the authors because the data were acquired under licence/data sharing agreement from NHS Digital, for which conditions of use (and further use) apply. However, individuals and organisations can access HES data by direct request to NHS Digital.

Ethics statements

Patient consent for publication

Ethics approval

Ethical approval was not sought for the present study because it did not directly involve human participants. This study was completed in accordance with the Helsinki Declaration as revised in 2013. Consent from individuals involved in this study was not required.

Acknowledgments

We acknowledge NHS Digital for permission to use their data in this report. The GIRFT programme is providing a framework for examining contemporary clinical practice in unprecedented detail and breadth. We also thank all staff within individual NHS trusts who collected and entered the data used in this study and GIRFT Clinical leads for advice: Michael Jones, Philip Dyer, Chris Moulton, Anna Batchelor, Michael Swart, Christopher Snowden, Martin Allen, Adrian Hopper, Partha Kar, Gerry Rayman and GIRFT Clinical fellows: Ini Adelaja and Pratusha Babu.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Footnotes

  • Handling editor Mary Dawood

  • Contributors This study was designed and organised by AVN, WKG, JD and TWRB. Data cleaning, analysis was by WKG, supported by JD, JH and FH. Writing of the first draft was by WKG. All authors critically reviewed the manuscript and agreed to submission of the final draft. WKG is guarantor for this study and accepts full responsibility for the work and the conduct of the study, had access to the data, and controlled the decision to publish.

  • Funding JH and FH received a fellowship from Distributed Research Utilising Advanced Computing (DiRAC), which paid their salaries.

  • Competing interests None declared.

  • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

  • Provenance and peer review Not commissioned; internally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.