Introduction

The daily average attendance at major hospital emergency departments (ED) in the 2022/23 financial year in England was 44,400 people [1]. Over the previous decade in England there was a consistently steady annual increase in attendance (13% increase between 2012/13 and 2022/23 [1]), with exceptions during the COVID-19 pandemic (March 2020 to December 2021) [2], and an increase in the average length of time that people spend in ED [3].

Waits of over four hours from ED arrival to departure have become increasingly common. In 2011/12, 5% of patients spent longer than four hours in EDs [2], increasing to 29% in 2022/23 [1]. Despite the number of attendances in 2022/23 being similar to pre-pandemic levels (2% increase in 2022/23 compared to 2019/20 [1]), waiting time performance considerably worsened over this period [3].

Investigations into the determinants of urgent and emergency healthcare utilisation have led to the targeting of interventions at groups with higher risks of ED attendance, with the aim of reducing the frequency of attendances. These interventions have included improving self-management of long-term health conditions [4, 5], improving primary care access for repeat attenders [6], and improving care coordination in older people [7]. Identifying groups with higher ED attendance rates also aids understanding of help-seeking behaviours that lead to ED attendance. For example, progress has been made for children with high attendance rates [8–10].

Predictors of ED attendance previously identified by research are broad, making it difficult to target interventions. Small-scale studies across England have attempted to understand the predictors for ED utilisation using survey data [11, 12], and aggregated general practice (GP) level data, including practice characteristics such as distance from ED, practice list size, and Index of Multiple Deprivation (IMD) based on GP location [13–15]. They found that higher deprivation, lower distance to EDs and higher morbidity (particularly mental health problems) were strongly associated with ED attendance rates [11–15]. Hull and colleagues used linked patient-level secondary and primary care data to investigate whether the association between deprivation and ED attendance in inner urban areas of East London could be explained by clinical variables in GP records in 2014/15. They found that multimorbidity was the strongest clinical predictor for ED attendance [16].

Using individual level data, with over one million linked records from primary and secondary care covering almost the entire population of Norfolk and Waveney, an area with rural, urban, and coastal communities and three acute National Health Service hospital trusts, this study aimed to investigate factors associated with ED attendance. In the whole study population we investigated associations between frequency of ED attendances and deprivation, age, ethnicity, and distance from home to EDs. In a subset of individuals with GP data on presence or absence of a range of long term conditions (LTCs,) and frequency of GP appointments, we also examined associations between these factors and ED attendances.

Methods

Results

Results for the total population

The dataset included a total population of 1,027,422 individuals with a residential address within N&W and a recorded sex of male or female. Table 1 shows the characteristics of the study population and the frequency of ED visits associated with each explanatory variable. The population included all age and deprivation bands, with roughly half males and half females. Recorded ethnicity was predominantly White British (59.9%), with 37.9% unknown. All explanatory variables in Table 1 were statistically significantly associated with the mean numbers of ED attendances in univariable models (P<0.001). Children under five years of age had the highest mean number of attendances compared to other age groups. Mean ED attendances were higher in individuals living in the most deprived areas compared to the least deprived.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of participants and association between participants’ characteristics and emergency department attendance.

https://doi.org/10.1371/journal.pone.0303270.t001

Results for the multivariable analysis that included the total population are reported in Table 2. All covariates were statistically significantly associated with the rate of yearly hospital attendances. Individuals under five years old had a 25% higher rates of hospital attendances compared to those aged 15 to 35 years (adjusted incidence rate ratio (IRR) 1.25, 95% confidence interval (CI) 1.23 to 1.28). Both the age groups of five to 14 years and over 70 years had approximately 8% lower ED attendance rates (IRR 0.92), while those aged 36 to 70 years had 28% lower attendance rates (IRR 0.72, 95% CI 0.71 to 0.73). Males had 2% higher rates of ED attendances compared to females (IRR 1.02, 95% CI 1.01 to 1.02). Individuals whose ethnicity was not known or missing had a much lower rate of ED attendances (IRR 0.11, 95% CI 0.11 to 0.12) than the white British category. For each unit increase in IMD decile, there was 5% (IRR 0.95, 95% CI 0.95 to 0.95) decrease in the rate of hospital attendances. For every 10km increase in distance to the nearest hospital, the rate of ED attendances decreased (IRR 0.86, 95%CI 0.85 to 0.87). However, the deterrent effect of distance on ED attendances decreased with increasing distance (IRR>1 for distance squared). With IMD decile modelled as a categorical variable, ED attendance rates decreased steadily with increasing IMD decile (that is, with decreasing deprivation). Individuals in the least deprived decile (IMD10) had rates of ED attendances 39% lower compared to those in the most deprived decile (IMD1) (IRR = 0.61, 95%CI 0.60 to 0.63).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Association between sociodemographic factors and number of emergency department attendances in the total population (n = 1,027,422): Poisson regression model.

https://doi.org/10.1371/journal.pone.0303270.t002

Results for individuals with long term condition and general practice appointment data

Of the total population, 589,564 (57.4%) individuals had LTC data and were included in analyses which used LTC data. Characteristics of individuals with and without LTC data are compared in Table 3. Those with missing LTC data had higher probabilities of being younger and male, more deprived and not identified as white British and had, on average, fewer ED visits and lived further from EDs.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Characteristics of individuals with and without long term condition data.

https://doi.org/10.1371/journal.pone.0303270.t003

Results of multivariable models 1–4, including only individuals who had LTC data, are shown in Table 4. In Model 1, (without LTC covariates) the IRRs associated with each covariate were similar to the equivalent Model 1 results for the whole study population, except that the IRR for age under 5 was greater (IRR 1.97 compared to 1.25 in Table 2). In Model 2 (including the number of LTCs as covariate) there was a 16% (IRR 1.16, 95% CI 1.15 to 1.16) increase in the rate of ED attendances for each additional condition. This adjustment for number of LTCs decreased the IRRs associated with age 36–70 (IRR 0.53 in Model 2 compared to 0.63 in Model 1) and >70 (IRR 0.47 compared to 0.76). In Model 3 (including specific LTCs) every condition was associated with increased rates of ED attendances, with atrial fibrillation, stroke, depression, and heart disease having the highest IRRs. In Model 4, ED attendances increased by 3% (IRR 1.03, 95% CI 1.03 to 1.03) for every extra primary care appointment per year. The IRRs for the other variables in Model 4 are similar to those in Model 3, except that in Model 4 the small sex difference changed direction, the IRRs for diabetes and pre-diabetes were smaller, and for hypertension the IRR was 1.00 (95% CI 0.99 to 1.02).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Association between explanatory variables and frequency of emergency department visit in individuals with data on long term conditions or general practice attendances: Poisson regression models.

https://doi.org/10.1371/journal.pone.0303270.t004

Models 1–4 implemented as mixed models with GP as random effect produced very similar results to the primary analysis (S1 and S2 Tables in S1 File).

Discussion

This study shows that the frequency with which individuals in N&W attended EDs was dependent on factors previously known to be associated with poor health, including socioeconomic deprivation. Sex and ethnic differences were relatively small, although missing ethnicity data may have impacted those results. The growing burden on emergency services has been partly attributed to the increasing proportions of the UK population that are older. In this study, individuals aged over 70 years had lower rates of ED attendances than any age categories younger than 36 years, but they made up 25.9% of all attendances. Children aged under five years had the highest rates of ED attendances, and were responsible for 7.3% of total attendances. However young children do not necessarily place a disproportionate burden on EDs. Previous studies have shown that children spend less time in EDs compared to the elderly [24], and are more likely to be classified as non-urgent compared to other age groups [25]. ED attendance rates increased steadily with deprivation. Greater distance from home to the nearest ED was associated with lower attendance rates, which presumably reflects less accessibility of care. However, this effect was attenuated at greater distances.

Analysis of data restricted to individuals with LTC data showed that LTCs were associated with more frequent ED attendances. This is to be expected as LTCs are known to be associated with increased risk of medical emergencies. The results suggest that the co-existence of additional LTCs increased the risk of ED attendance multiplicatively, whether LTCs were modelled as single conditions or as number of LTCs. Adjustment for long-term health conditions reduced the incidence ratios for those aged 36–70 and those over 70 (Models 2–4 compared to Model 1 in Table 4), suggesting that in those two age groups ED attendance rates are largely explained by their higher prevalence of long-term health conditions. Frequency of GP appointments was also associated with greater ED attendance, presumably because it is also an indicator of ill health.

While previous research has attributed the association between deprivation and ED attendance to higher levels of multimorbidity in more deprived populations [13–16], in our study adjustment for LTCs did not change the association between deprivation and ED attendance rates (models 2–4 compared to Model 1 in Table 4). This suggests that LTCs may not be the main reason that more deprived people attend EDs more often. However this inference should be cautious, given the limitations of our LTC data.

Our findings broadly agree with previous research in the UK about correlates with higher ED attendance rates. Hull and colleagues found that higher attendances were linked with higher GP consultation rates [14]. Other studies have found that ED attendances rose with increasing co-occurrence of long-term health conditions [12, 13, 16], and that attendance rates were higher in individuals living in the most deprived English areas [13–16]. In an analysis of data available in 2008 for the West Midlands in central England (population 5.4 million), Rudge and colleagues [15] found declining ED attendance rates with increasing travel distance, although this relationship was modified by deprivation. Hull and colleagues [14] and Giebel and colleagues [11] also reported declining attendance rates with increased distance to home residence. Previous research is less consistent about which age groups have the highest utilisation rates of EDs. In a retrospective cohort analysis of London residents, Hull and colleagues found that crude and adjusted ED attendance rates were higher for persons aged 70 years and over than for persons under age five years [14]. Conversely, in a cross-sectional study of all English GPs, Scantlebury and colleagues [13] found that the proportion of the population aged under five years in the same geographic area was positively related to ED attendances. Differences in the findings of these previous studies may partly be attributed to differences in the range of confounding variables they were able to account for.

A key strength of this study is the large dataset, linking primary and secondary care data together with sociodemographic and geographical data, enabling us to investigate the independent effects of many interlinked variables in a geographically defined population. The data included almost all of the population residing within the study area, and the large sample size provided high statistical power and precision. This is the first external use of the N&W ICBs developmental linked patient level dataset. Such analyses have the potential for improving patient care and has created opportunities for collaboration between local authorities, the National Health Service and academic researchers.

However, the dataset is still developmental and subject to data quality issues. Of the 106 GPs in N&W, 15 opted out of sharing any of their data. The dataset also had many missing values, particularly for ethnicity (37.9% missing or not known), diagnosis codes for LTCs (42.6% missing) and GP appointments (56.9% missing). This level of missing information is common across NHS healthcare datasets [26, 27], and has particularly been highlighted in the use of linked datasets where GPs may use different coding procedures [28]. To our knowledge this is the first use of individual level data to analyse the associations between ED attendance and specific long-term health conditions in a large geographically defined English population. However, the large proportion of participants with missing data on long-term health conditions and GP appointments, with more missing data at younger ages, raising doubts about the accuracy of these variables and the analyses that used them. The results of the analyses involving long-term health conditions are therefore not necessarily generalisable to the whole study population. For future epidemiological research linking sociodemographic, primary and secondary care data for geographically defined populations in Norfolk and Waveney, and the United Kingdom more generally, this study shows that it is a priority to enable more complete linkage of informative primary care data about individuals’ health states.

Another limitation of the study is its cross-sectional design. The dataset aggregated events over a single twelve-month period, which precludes identification of earlier causes and later effects. For instance, it is plausible that greater access to and use of primary care could have led to less ED use, but it is also plausible that some primary care visits might have followed from ED visits. Our finding that GP appointment rates were associated with higher ED attendance rates may also be due to higher rates of GP appointments for those in poor health.

Conclusion

This study demonstrates what can be learned from individual patient record data, its use for informing the evidence base about the challenges faced in recent years by urgent and emergency care services in England, and the caveats that apply. Average ED attendance rates during 2022/23 were greater in people aged under five years; living in more socioeconomically deprived areas; living closer to ED; with a long-term medical condition; or with more GP attendances during the same year. Our findings align with the current evidence and further add to the evidence base, particularly about how different long-term health conditions affect ED attendance frequency. The findings support the use of interventions and policy aimed at preventing and managing LTCs, for example the National Health Service Prevention Programme in England [29]. Future research could investigate characteristics of frequent attendees using similar methods and linked data.

Supporting information