Exploring overcrowding trends in an inner city emergence department in the UK before and during COVID-19 epidemic

The spread of COVID-19 during 2020 has put pressure and induced a number of changes to the National Health Service (NHS) in the UK. To prevent overwhelming the health services, including Emergence Departments (EDs) and to suppress the increasing number of COVID-19 cases in the UK earlier this year, the UK Government imposed strict social distancing measures (“lockdown”) from March 23, 2020. As the number of cases started to decline following the lockdown and over the following months, a phased relaxing of the lockdown measures started with primary schools opening on June 01, 2020 and wider relaxing of imposed measures from July 4, 2020. After remaining low during July and August, the number of COVID-19 cases and associated deaths, started to increase again in late August and throughout September 2020. After trialling a Tier system for more localised strict control measures, which was insufficient to curb the large recent resurgence in cases and COVID-19 associated deaths, a second national lockdown was imposed across the UK from November 5, 2020.

EDs provide immediate assessment and care to patients, who may be critically ill. During the first wave of COVID-19, the level of visit to EDs declined sharply; 48% reduction was reported nationally in April 2020 compared to April 2019 [1]. While awaiting roll-out of an effective vaccine, maintaining social distancing, in conjunction with effective testing, tracing and isolation of positive cases, are the main interventions for virus suppression. Maintaining social distancing is difficult to achieve in overcrowded EDs.

ED treatment in the pre COVID-19 era was often hindered by overcrowding. Contributing factors include seasonal increases in demand, onset of an epidemic or pandemic [2, 3], shortage of available inpatient beds, staff shortage [4, 5] and inefficient patient flow through the system [6, 7]. Previous studies have shown that overcrowding in ED is associated with increased mortality, increased length of stay, reduced quality of care, poor patient experience and increased number of serious incidents [8‐11]. However interventions have had varying levels of success [12], and despite ongoing efforts, there is understanding on how to improve ED patient flow [11].

The Royal College of Emergency Medicine suggests that interventions focusing on changes in the input, throughput and output parts of the ED system may be one way of relieving overcrowding [13‐15]. Input interventions involve targeting aspects responsible for managing the number of patients attending the ED e.g. relocation of primary care services within emergency care services [16]. However the evidence for this is weak, limited and outdated [17]. Throughput interventions comprise processes within the ED and can include aspects such as staffing, co-ordination with inpatient teams, local protocols and the physical layout of the department. For example training nurses to order an X-rays at triage [18] is a throughput intervention. Finally, output interventions focus on improving the exit of patients from the emergency department, through either admission, discharge or transfer to another service. For example, boarding patients, i.e. sending them to an inpatient ward while waiting for a bed is an output intervention [19, 20]. However, the evidence base of the effectiveness of these interventions is small, limited and heterogeneous [21, 22].

To improve emergency care provision and reduce prolonged inpatient length of stay, the UK Department of Health and Social Care in 2004 produced a white paper [23] that set a standard target for acute hospitals as part of the National Health Service (NHS) that at least 95% of patients attending EDs must be seen, treated, admitted or discharged in under 4 h. This still represents the UK’s key indicator of an ED performance [24].

Modelling can aid assessment of the impact of different interventions. To date a number of studies have used modelling to quantify the impact of ED system changes on ED overcrowding as outlined in recent reviews for the UK system [25] or the USA system [26]. Different modelling approaches such as statistical (regression) modelling, mathematical (i.e. queueing-theory based) modelling and discrete event simulations can be used for this purpose and have different advantages and disadvantages. Generally, statistical modelling is used to evaluate impact of implemented interventions, queueing theory can evaluate flow of people through the ED system and hence explore impact of potential new interventions, while the discrete event simulators allow tracking of individual patents through the system and hence identify potential “clogs” within the system. In existing studies, specific type of models are constructed for the posed question, but often the details of the modelling framework have not always been transparent and difficult to replicate.

In this paper we examine overcrowding in the ED within the University College London Hospitals NHS Foundation Trust, UCLH. Our study is using data from this ED over a long period before the COVID-19 epidemic (April 01, 2017- December 31, 2019) as well as during the first 10 months of the COVID-19 epidemic in 2020. We compare the impact on overcrowding of three trialled interventions at different times pre-COVID-19 to the impact of the presence of the epidemic and the imposed first national lockdown to suppress it and protect the NHS. The three interventions targeted the influx of patients at ED (interventions A), reduced the pressure on in-patients’ beds (interventions B) or improved internal ED processes to increase outflow of patents from ED (interventions C) (details in Methods).

All three sets of interventions A-C trialled in the pre-COVID-19 era led to small reductions of waiting time over the 2 years before the epidemic. These reductions were significant in the case of interventions A (that targeted the influx of patients at ED) and interventions C (that improved internal ED processes to increase outflow of patents from ED) but were not in the case of interventions B (that aimed to reduce the pressure on in-patients’ beds). In addition, all three trialled interventions were also associated with a small, but significant decrease in the number of patients leaving the ED within 4 h of arrival. Therefore with the trialled interventions pre-COVID-19 no improvement was evident in the four-hours target.

During 2020 and in presence of the COVID-19 pandemic, there was a drop in attendance at this ED and as a result, waiting times were reduced and the percentage of people leaving within 4 h or arrival increased. This result is unsurprising: the drop in attendances at UCLH ED compared to the same periods in 2019 reflects the national data [1, 40]. Crucially, however when the lockdown was lifted following the first wave of COVID-19, there was both an increase (6.5%) in the percentage of people leaving within 4 h , together with a larger (12.5%) decrease in waiting time. This occurred despite the increase of 49.6% in ED attendance after lockdown ended.

The mixed results found pre-COVID-19 (significant improvements in waiting time with some interventions but fewer people leaving UCLH ED within 4 h of arrival), may be due to indirect impacts. For example, clogging up one part of the ED system e.g. triage will have a knock on effect and increased pressure on the outflow of patients. This is simply due to the fact that large influx will produce a bottle neck effect within the system. In a situation where there is limited capacity in terms or inpatients’ beds or a complex system of discharging from hospitals, this will lead to lower outflow through the system i.e. elevated overcrowding. Therefore, in line with existing literature [41, 42], there is a need for the entire pathway of flow through the ED system to be examined and a system-wide improvement to be implemented. This needs to include separate parts (a) deciding the priority measure of overcrowding (b) identifying the stream of blockages within different parts that affect it and (c) taking an informed approach with engagement of staff to design and implement a modular strategy.

Our findings reflect those in the literature. For example, a systematic review [38] of the effectiveness of interventions on reducing ED crowding by older patients between 1990 and 2017 showed that none of the implemented interventions reduced all overcrowding measures. Our findings of the need for multifaceted interventions and a system-wide approach is also supported by the literature [11]. For example, results in [38] suggest that rapid assessment and streaming of care for older adults arriving at the ED lead to a statistically significant decrease of ED length of stay. This is analogous to how implementation of interventions A led to statistically significant reduction of the time spend in ED in our study. Similarly, existing work showed that the presence of an ED-based consultant geriatrician was associated with significant time reduction between patient admission and geriatric review compared to an in-reaching geriatrician [41]. This has parallels with our findings that implementation of interventions C leads to reduction to the time spend in ED.

Our results also reflect the existing literature on the impact of COVID-19 on ED overcrowding [1, 43, 44] and add to it. For example, our findings of the drop in visit to EDs during the first lockdown, reflect the nationally reported decline of 48% reduction in attendances in April 2020 compared to April 2019 [1, 40]. The drop in attendance during the first stages of the pandemic spread, compared to these being stable in the pre-COVID-19 period, are also in agreement with the findings from a study in the inner city ED of the University Hospital of Parma, Italy [43] and across 24 EDs from five American states [44]. The literature across these studies agrees that the COVID-19 pandemic lead to reduced attendance of EDs and as a consequence overcrowding was curbed.

A shift in public behaviour during the first wave of the pandemic can explain these results. For example, during the first lockdown period in the UK, when the public was instructed to ‘stay at home, save lives and protect the NHS’, anxiety over presenting in hospitals and higher use of NHS 111 and other advice lines occurred [1]. Nationally, this led to a decline in number of arrivals within EDs with minor injuries [1, 40]. This allowed ambulance services to more effectively respond to people with more severe needs. In addition, primary care services made a substantive shift to virtual primary care consultations [40], which gave immediate access to patients with less severe injuries. Hence, the reduction in overcrowding in 2020 compared to 2019, would have been due to changes in the ways in which people with less severe needs gained immediate access to services virtually and hence attended EDs less frequently, leading to more appropriate use of the ED services.

After the first COVID-19 wave, an increase was evident in attendances in this ED and notably from patients walking in (Fig. 2(a)). Despite this, waiting time at UCLH ED was reduced while the number of people leaving UCLH ED within 4 h of arrival was increased. We note two things. Firstly, the increase in UCLH ED attendance after the first lockdown has not yet reached baseline pre-COVID-19 level (Fig. 2(a)). Secondly, a large restructuring of EDs, including UCLH ED has occurred as part of the COVID-19 epidemic response. Hence, the combination of shifted public behaviour of more appropriate use of EDs and the system-wide restructuring changes during COVID-19 epidemic would have increased flow of patients in this ED and hence curb ED overcrowding during 2020. It is important to assess whether this remains the case in future, and while our study is the first to study thus, we are aware that a longer timeframe analysis is necessary to confirm this.

We note that our study has some limitations. Firstly, when comparing the first 10 months of 2019 and the first 10 months of 2020 we have taken a simplistic approach that doesn’t account for the fact that these period are inhomogeneous. Specifically, during the first 3 months of 2019 interventions C were not applied yet, while in the first 10 months of 2020, and in response to the COVID-19 epidemic a number of changes and responsive interventions were employed in this ED. Furthermore, when comparing the “lockdown” timeframe (March 23, June 01) in 2019 and 2020 we note that there is a theoretical difference between the two periods, since for example interventions C were applied from April 01 in 2019. Secondly, we have not untangled in our analysis the key patients cohorts (e.g. those with respiratory symptoms, or cardiovascular symptoms, or trauma patents) most likely to be attending this ED during the COVID-19 epidemic. While this is an interesting follow on study, it is beyond the scope of the current work. Thirdly, we note that interventions A and B overlap for some of the pre/post-intervention period, but we didn’t feel they are confounding factors to each other’s impact. This is because, although taking part at the same time over the overlapping periods, the relevant interventions were independent of each other. An alternative, would have been to use 3 months as the evaluation period across all interventions but since interventions take time to start, we felt that 6 months was in this study a more realistic evaluation period.

The choice of statistical modelling in this study also has some limitations. Our method, interrupted time series, can effectively compare changes in outcomes for successive groups of patients before and after ED interventions/lockdown started, and is therefore fit for the purpose of this study. But whilst such regression modelling is useful in drawing conclusion for the duration of the study where fitted curves mimic the data, the presence of turning points in the non-linear fits makes them unreliable for prediction beyond the period for which data are available. An alternative would be to develop and utilise queuing theory models e.g. [45], that model the flow of patients through a system.

Another possible future direction in research would be use of automated learning systems, rooted in higher-order statistical methods e.g. machine learning, that will allow real-time assessment of patients arriving at ED. A recent review [46] has collated the existing literature on the application of AI methods in emergency medicine, concluding that although some attempts have been made in the field, this an emerging field that will benefit from application of tailor-made machine learning algorithms for real-life assessment of, for example, arrival and triage assessment. Future work could combine the use of Electronic Health Records and robust machine learning algorithms.

In summary, our study found that trialled interventions pre-COVID-19 lead to mixed results: significant improvements in waiting time with interventions that targeted the influx of patients at ED (A) or improved internal ED processes to increase outflow of patents from ED (C), but fewer people leaving UCLH ED within 4 h of arrival. This is likely because of the indirect impacts of these interventions, where increasing pressure on one part of the ED system affected other parts. This underlines a need for multifaceted interventions and a system-wide approach to improve the entire pathway of flow through the ED system.

During 2020 and in presence of the COVID-19 epidemic, a shift in public behaviour with anxiety over presenting in hospitals and higher use of NHS 111 or virtual primary consultations, UCLH ED attendance dropped notably, and as a consequence waiting time was reduced while the percentage of people leaving the ED within 4 h was increased.

Importantly, once the lockdown was lifted, there was an increase in arrivals at UCLH ED, but not yet reaching the pre-COVID-19 attendance level. As a result, even with this increase the overcrowding metrics were reduced. Therefore, it may be possible that shift in public behaviour to more appropriately use EDs combined with the system-wide restructuring changes during COVID-19 epidemic would be able to curb overcrowding in future, but longer timeframe analysis is required to attest this.