Background
In 2020 Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) caused a worldwide pandemic of COVID-19 disease resulting in substantial excess mortality and global disruption to healthcare and social care. The first major peak in COVID-19 cases in England occurred between March and May in 2020 and prompted a national lockdown. Lockdown measures were adjusted during the pandemic and widespread vaccination in 2021 triggered relaxation of the restrictions. During the pandemic, healthcare was rapidly restructured in anticipation of predicted needs [
1]. This included redeployment of people and resources, especially to acute general medicine, emergency medicine and critical care, reductions in non-COVID-19 research activity, reductions in elective procedures, and increased use of remote telephone or video consultations [
2,
3].
Reports of reductions in hospitalization for non-COVID-19 acute illnesses have raised concerns that patients may not have attended hospital for an acute illness and might subsequently experience increased morbidity or mortality as a result [
4‐
9]. Factors influencing hospital attendance during the pandemic may have included fear of acquiring COVID-19 infection, a desire to reduce the pressure on hospitals or a higher threshold among referring and receiving clinicians for hospital review or admission. Conversely, others have suggested patients avoiding ED had more minor illness and this had a beneficial effect by reducing crowding in ED [
10].
In the UK, acute hospital care is provided by direct presentation of patients to the emergency department (ED) or referral of patients by their primary care general practitioner or paramedics to the hospital [
11]. We sought to understand both the COVID-19 and non-COVID-19 activity in the ED and the acute medicine department and how this changed across the course of the pandemic. It is important to understand both the dramatic changes that occurred during the first wave of the pandemic and the subsequent patterns of acute care usage after this early phase.
The aim of our study was to determine the clinical characteristics of emergency department attendances and medical admissions during the COVID-19 pandemic and whether particular groups were under-represented during the pandemic peaks and the time frame of any changes.
We identify and characterise major changes in ED attendances and in medical admissions during the pandemic and highlight changes in physiological severity, patterns of diagnoses, and outcomes including mortality. These findings demonstrate the profound impact of a pandemic on urgent care, even for non-pandemic illness, and will form a foundation for planning to minimise the impact in the future.
Methods
We extracted data from the hospital electronic health record (EHR) for all ED patients aged 18 or over from the Oxford University Hospitals NHS Foundation Trust (John Radcliffe and Horton sites) between 17th March 2019 and 12th July 2021 (n = 243,667). We extracted data for medical admissions from the EHR between 1st January 2019 and 18th August 2021 and pruned the analysis to 17th March to 18th July 2021 (n = 82,899). The longer time period for admissions compared to ED attendances allows proper analysis of medical patients who were inpatients during the period of interest but admitted or discharged outside this time period. ED attendances and acute medicine patient data incorporate critical care and high dependency unt (HDU) admissions because these admissions pass through ED or for medical patients remain under the duty medical physician. Day-case attendances (consisting of procedural admissions for e.g., endoscopy or bronchoscopy) were removed from the data by filtering out admissions to a day-case unit location or where the admission method was ‘planned’, ‘booked’, or ‘elective’. For analysis of acute medical patients, the hospital attendances were sub-divided into ‘medical attendances’ where the patient was discharged directly from the ambulatory emergency care unit (AEC) and into ‘medical admissions’ where the patient was admitted directly to a bed-based pathway or admitted to a bed-based pathway from the AEC. At our institution medical admissions mostly occur through the emergency medical assessment unit (EAU) but can also occur directly from ED to medical wards or directly from the ambulance service to cardiology for suspected ST-elevation myocardial infarction.
Clinical ED data are recorded according to the UK emergency care dataset (ECDS) parameters [
12]. The ED diagnosis is recorded in real time by the clinician selecting from a curated list of SNOMED terms. We modified this list by adding diagnostic codes for COVID-19 and categorizing COVID-19 into the ‘respiratory’ group of diseases in group 2 of the ECDS diagnostic tree (Supplementary file
ECDS data – table of ECDS codes and groupings).
For acute medical admissions the diagnosis of COVID-19 was derived from the primary diagnosis data field using ICD-10 codes of U07.1 (COVID-19, virus identified) and U07.2 (COVID-19, virus not-identified). The medical diagnoses are recorded by professional medical coders after the admission is completed using aggregated data from the EHR. Inpatient COVID-19 diagnoses in our institution are generally made by a consultant using a combination of clinical data, PCR testing, lateral flow testing and chest X-ray or CT findings. For ED patients, PCR testing was not widely available during the first wave of the pandemic.
Mortality data were obtained from the EHR and by querying the NHS Digital Personal Demographics Service Database using the Demographics Batch Service [
13].
Our hospitals serve a population of around 650,000. The UK containment phase of the pandemic ended on 12th March 2020 and from 16th March onwards a ‘lockdown’ was officially advised and enforced from 23rd March 2020. We defined the first wave of the pandemic period as from 17th March 2020 to 31st May 2020 and this captures the major first peak of COVID-19 in the UK. Patients with suspected COVID-19 (fever, respiratory symptoms) or confirmed COVID-19 are assigned into side-rooms, grouped bays, or designated wards according to a traffic light system of Green (not COVID-19), Amber (suspected COVID-19) or Blue (confirmed COVID-19). Further details on COVID-19 infection control and pathways at our Institution are published elsewhere [
14]. We define the second wave of the pandemic period as from 26th November 2020 to 10th February 2021. For numerical comparisons we used a pre-pandemic period in 2019 that matched the first pandemic peak and a late-pandemic period matching the equivalent time period in 2021 one calendar year after the first peak period.
Analysis was undertaken using R [
15]. Rolling averages over time were calculated using the ‘rollmean’ function of the zoo (v1.8.9) package in R [
16]. A centred rolling window of 14 days was used for daily deaths and daily medical admissions and a window of 28 days for all other plots. ED patients were considered to be ‘Admitted’ if admitted from ED for more than 24 h or ‘Discharged’ if discharged from ED directly or within 24 h of attending hospital. For medical attendances diagnoses were stratified as COVID-19 if the primary diagnosis code was either U07.1 or U07.2 and as non-COVID-19 for all other primary diagnoses.
For age group analysis patients were stratified into 10 age groups of equal time width using the binning function in the R package dlookr [
17]. For categorical variables including gender and ethnicity, the difference in distribution between the pre-pandemic period and pandemic first wave peak period was compared with a chi-squared test and if significant, then a row-wise proportion test with multiple testing correction to evaluate the difference between categories using the prop test function in the rstatix package [
18]. To calculate distance from a patient’s domiciliary address to hospital we used the code for the UK census area of their address, termed the ‘lower layer super output area’ (LSOA) and obtained the latitude and longitude for the centroid position of the LSOA from the 2011 Office of National Statistics Census data [
19]. To obtain the latitude and longitude from the Northing and Easting positions in the census data we converted them using the web tool
https://gridreferencefinder.com/batchConvert/batchConvert.php. To calculate the straight-line distance from the hospital to the respective centroid coordinate we used a webtool distance calculator:
https://stevemorse.org/nearest/distancebatch.html. The significance comparison of distance was calculated with a Student’s t test. A deprivation decile was assigned by matching the LSOA code to the Index of Multiple Deprivation data from the English Indices of Deprivation 2015 [
20]. A value of 1 indicates the most deprived area and an overall chi squared test was performed between comparator groups and if significant then a row-wise proportion test.
For medical patients, co-morbidities were determined from the ICD10 codes for secondary diagnoses up to a depth of 50 co-morbidities. A combined co-morbidity score was calculated according to the mean weighted Elixhauser score system using the R package ‘comorbidity’ [
21,
22]. We amalgamated the comorbidity R package categories of ‘hypertension’ and ‘hypertension-uncomplicated’ as well as ‘diabetes’ and ‘diabetes-complicated’. Alcoholic liver disease was parsed separately using an ICD10 code of K70. To calculate a NEWS2 score (National Early Warning Score 2) we used the first set of observations including temperature, heart rate, systolic/diastolic blood pressure, peripheral oxygen saturation (without correction for chronic type 2 respiratory failure status), Glasgow Coma Score and the use of supplemental oxygen [
23]. We report mean scores compared using a Wilcoxon Rank sum test, and calculated incidence rate ratios for each of the 4 NEWS2 alert categories. NEWS2 score were binned into the NEWS2 clinical risk alert levels whereby a score of 0–4 = “low”, unless any individual category is 3 in which case = “low-medium”, 5–6 = “medium”, 7+ = “high”. The number of attendances/admissions without available observations is shown in supplementary tables referred to in the Results. To calculate an incidence rate ratio between the pandemic and pre-pandemic period we used a Poisson regression model of daily counts of each category implemented by the R mfx package ‘poisonirr’ [
7,
24]. The proportion of patients using oxygen was compared with a chi-squared test.
For the ED presenting complaint and primary diagnosis we selected the 10 commonest diagnoses across the whole dataset. The ED primary diagnosis data is displayed at the 2nd group level of the ECDS system (with COVID-19 grouped in ‘Respiratory’). For the primary diagnosis for medical admissions, we selected the 10 commonest primary diagnoses at the ICD-10 tier level of 3 alphanumeric characters. Statistical analysis was performed using incidence rate ratios as described above. For ED data we also show a table of manually selected diagnoses at the most granular level of diagnosis. Student’s t-test was used to analyse ED length of stay. For ED mortality we calculated a mortality rate using the number of patients who died during their ED attendance or before leaving hospital if they were admitted and applied a chi-squared test of the proportion. For medical patients we calculated the mortality rate during admission and compared the proportion of admissions with mortality using a chi-squared test.
The study formed part of an institutional service evaluation using retrospectively collected anonymized routine clinical data and was deemed not to require further ethical approval or informed consent from patients.
Discussion
Beyond the direct effect of COVID-19 illness, there have been concerns that the pandemic has deterred hospital attendances for acute non-COVID-19 illness, triggering increased downstream mortality or morbidity due to late presentation of disease [
7,
8,
26‐
30]. In the UK, despite an attenuation in the pandemic from the spring of 2021, hospitals have experienced unprecedented ED attendances over the usually quiet summer period. In this context, understanding how the pandemic influenced hospital attendance and admission and which groups were under or over-represented is important for future health and pandemic planning.
As anticipated, ED attendances and medical admissions fell rapidly with the first UK death and the lockdown. In the UK, medical attendances are mostly referred by general practitioners or ED doctors and the rapid fall with the first wave suggests reduced referrals and/or a reduced acceptance rate. ED attendances and medical admissions fell less sharply in the second wave, despite higher numbers of COVID-19 cases and higher daily death rates, indicating acclimatisation to the pandemic by clinicians and patients.
ED attendances have a high proportion of young patients, whereas medical admissions have a high proportion of older patients [
31,
32]. The fall in ED attendances by young patients during the pandemic reflects experience elsewhere and is likely multifactorial [
26,
33]. In some areas, closure of higher education institutions will have prompted some students to return to their family homes [
34,
35]. However, similar falls have been seen elsewhere and may be consistent with younger patients not attending ED with self-limiting conditions and possibly experiencing less minor trauma with reductions in some activities during the lockdown [
7,
10]. The greatest relative reduction in medical admissions were in the oldest age groups possibly due to concerns about COVID-19 acquisition in hospital or about the futility of admission [
36].
The increase in co-morbidities among medical admissions without COVID-19 suggests that admission for highly co-morbid patients was not easily avoidable, whilst those with lower co-morbidities presumably remained in the community. Medical admission with COVID-19 had a distinct comorbidity profile compared to non-COVID-19 admissions. Metastatic cancer was less common, perhaps because these patients shielded themselves from infections or because of reduced referral or admission rates; there is evidence of increased non-COVID-19 cancer mortality in the community [
37]. In contrast, the increased prevalence of dementia among COVID-19 admissions could reflect outbreaks in nursing homes or difficulty shielding effectively when reliant on external carers. Other factors with higher prevalence in COVID-19 admissions, such as obesity and diabetes, reflect known susceptibility factors for severe COVID-19 illness [
38,
39].
The NEWS2 severity scoring system has been adopted by the NHS to highlight patients at high risk of clinical deterioration [
40‐
43]. The fall in lower severity ED attendances that we observed during the first wave is consistent with several other studies identifying the greatest reductions in low acuity attendances though these studies also identify some reductions in higher severity cases [
44‐
46]. In a healthcare system in the Netherlands where there are generally fewer low acuity ED attendances, a 30% reduction in attendance was nevertheless still observed [
28]. Reduced attendances among lower acuity ED patients may not be immediately harmful if they represent non-severe illness. However, the reductions we observed in medical admissions across all severity categories suggests that some patients with severe illness did not attend hospital.
Several factors might have contributed to the steady rise in low severity ED attendances after the second peak, including diminished concern about acquiring COVID-19 in hospital, especially in the vaccinated, publicity about the dangers of avoiding acute care, increased exercise and traffic trauma, symptoms of long COVID-19 and a perceived reduction in access to primary care which has experienced increased demand in additional to providing vaccines. Delayed presentations of major illness might be expected to result in greater numbers of patients with higher severity scores subsequently, but this was not observed. Nor was there an increase in cardiac arrest presentations, although this may have been influenced by increased attention to resuscitation status during the pandemic [
47].
The marked reduction in ED attendances for trauma during the first wave likely reflects in part reduced physical activity, sport, and road traffic. UK road traffic dropped by 59% and a Spanish trauma unit described significant reduction in most trauma [
48,
49]. The common ED diagnoses of ‘no abnormality detected’ dropped substantially during each COVID-19 peak, which is unlikely to have led to clinical harm and may suggest scope for reducing ED attendances in the future. During the first wave, there were almost no ED attendances where the patient walked out and walk-outs remained low for some time thereafter. This might reflect a faster ED service, as length of ED stay was reduced, or resulted from non-attendance of people with low severity complaints. Walkouts increased in the summer of 2021 in association with increased ED attendances.
There were no changes in ED attendances during the pandemic for certain conditions that might be acutely disabling, such as subarachnoid haemorrhage or diabetic ketoacidosis. Other centres observed either a reduction or no change in sub-arachnoid haemorrhage [
50,
51]. In aggregate, there were reductions in cardiac diagnoses and at a granular level there were reductions in atrial fibrillation as well as myocardial infarction, with potential downstream consequences, for example if fewer people with atrial fibrillation received stroke preventing anticoagulants. We observed no change in cerebral infarction strokes during the first wave, although a study in Israel found a reduction of 29% for stroke and a UK-wide registry study found only a 12% reduction in stroke admissions at the peak of the first wave [
52,
53]. Our data suggest a possible trend towards increased strokes in the last months of our analysis which, if confirmed over time, might reflect reduced opportunities for stroke prevention during the pandemic. Medical admissions for myocardial infarction, with and without ST-elevation, were reduced during the pandemic and this is consistent with reports from the UK and elsewhere [
7,
9,
54‐
56]. The relatively constant inpatient mortality for non-COVID-19 medical admissions throughout the study might suggest that in our catchment area similar numbers of patients with the most severe life-threatening non-COVID-19 conditions were attending hospital throughout. However, nationally there is evidence of reductions in hospital mortality with concomitant increases in community mortality [
37].
There are necessarily limitations to any retrospective study, and we cannot determine the precise factors that influenced whether a patient attended hospital and then was or was not admitted. ED diagnoses are provided by the initial clinician, often a junior clinician, and therefore may not be as robust as the diagnoses for medical admissions, which are ratified retrospectively by professional coders. The NEWS2 score had not yet been fully adopted by our institution during the first wave of COVID-19, so scores were not explicitly available to clinicians and were calculated retrospectively.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.