Article Text
PDF
Abstract
Background EDs globally are under increasing pressure through rising demand. Frequent attenders are known to have complex health needs and use a disproportionate amount of resources. We hypothesised that heterogeneity of patients’ reason for attendance would be associated with multimorbidity and increasing age, and predict future attendance.
Method We analysed an anonymised dataset of all ED visits over the course of 2014 in Yorkshire, UK. We identified 15 986 patients who had five or more ED encounters at any ED in the calendar year. Presenting complaint was categorised into one of 14 categories based on the Emergency Care Data Set (ECDS). We calculated measures of heterogeneity (count of ECDs categories and entropy of categories) and examined their relationship to total number of ED visits and to patient characteristics. We examined the predictive value of these and other features on future attendance.
Results Most frequent attenders had more than one presenting complaint type. Heterogeneity increased with number of attendances, but heterogeneity adjusted for number of attendances did not vary substantially with age or sex. Heterogeneity was associated with the presence of one or more contacts for a mental health problem. For a given number of attendances, prior mental health contact but not heterogeneity was associated with further attendance.
Conclusions Heterogeneity of presenting complaint can be quantified and analysed for ED use: it is increased where there is a history of mental disorder but not with age. This suggests it reflects more than the number of medical conditions.
- crowding
- emergency care systems
- emergency departments
Statistics from Altmetric.com
Key messages
What is already known on this subject
Frequent attendance at the ED is a ubiquitous problem and frequent attenders are a diverse group, often with complex needs.
Few methods are available to predict which frequent attenders will continue to attend.
We hypothesised that heterogeneity of reasons for attending (either because of multimorbidity or for other factors) may be a predictor of future attendance.
What this study adds
We examined reason for ED attendance in over 15 900 frequent attenders from a population of 5 million throughout 2014.
Heterogeneity of reason for attending was more strongly related to mental health problems than to age suggesting that it is driven by more than multimorbidity.
Heterogeneity of reasons for attending did not predict further attendance.
Background
ED care is under increasing pressure both in the UK and internationally due to growing numbers of patients presenting to EDs.1 One aspect of this which has attracted attention is the issue of frequent attenders of EDs.2 There is no single definition of frequent attendance, and in different studies it varies between 3 and 12 or more attendances in a single year.3 However, regardless of the definition, frequent attenders make up a disproportionate amount of attendances. UK based studies show 3%–4% of ED patients are frequent attenders and account for 8%–12% of all attendances.4 5
The literature suggests that frequent attenders are a diverse group of individuals, often with complex health needs. Previous studies have found frequent attenders are more likely to be male,6 older7 and to suffer from mental health or substance misuse problems.4 8 9 Several studies have shown that frequent attenders are more likely to suffer from chronic health problems,10 11 more likely to be admitted and more likely to have a poorer outcome during a hospital stay.11 Several studies have indicated that frequent attendance is also associated with the presence of functional12 or ‘medically unexplained’ symptoms.13–15 Patients with functional symptoms commonly have multiple symptoms in multiple body systems12 for which they seek medical attention.16 and the presence of multiple reasons for attendance may be a pointer to this.
We hypothesised that heterogeneity of the reason for attendance17 may be informative in understanding frequent attenders. First, it may indicate the presence of multiple medical comorbidities; second, it may indicate a tendency to experience and consult for multiple different physical symptoms as is seen in many patients with functional disorders12 14 15; or third it may indicate a lower threshold for consulting due to anxiety or concurrent mental distress. As such it may have value both as a marker of risk for further attendance and as an indicator that attention should focus beyond the immediate presenting complaint and onto the patient’s wider condition and context.
In this study, we aimed to measure the heterogeneity of presenting complaints among frequent attenders of ED care, to describe its association with patient characteristics and to examine whether it was predictive of further ED usage.
Methods
Study design
Cross-sectional analysis using routinely collected ED data.
Patient and public involvement
Patients were not directly involved in the planning or execution of this research which involved routinely collected healthcare data. Three public members of the Sheffield Emergency Care Forum commented on the final manuscript.
Data collection
The study used data collected as part of the NIHR Collaboration for Leadership in Applied Research and Care Yorkshire and Humber (Y&H). This involved a retrospective cross-sectional database linking routine NHS health data from a number of UEC providers within a single geographical region in England (Y&H). The region has a population of 5.5 million people with a mixture of urban (large and small), suburban and rural settings and as such is representative of the UK. At the time of data collection, the region included 13 acute hospital trusts with 19 type 1 EDs (consultant-led, with multi-specialty 24-hour services and full resuscitation facilities). The Y&H region is served by a single ambulance service (Yorkshire Ambulance Service).
Hospital data relating to ED attendances was extracted by the acute trusts from their ED administration systems. Data were linked at the patient level in order to identify and code attendances for the same patient at different EDs. All data were pseudonymised before use. We used data for new attendances and unplanned reattendances for adults aged 18 years and over which occurred during the calendar year 2014.
Data management
Data items collected and used in this analysis included: age, sex, date of attendance, presenting complaint and diagnosis. Coding of presenting complaint varied between EDs: some reported patient complaints verbatim, while others collected them in categories. For this analysis we chose to use categories relating to the Emergency Care Dataset.18 This contains 14 categories of presenting complaint; cardiovascular, respiratory, gastrointestinal, neurological, skin, head and neck, eye, orthopaedics/trauma, genitourinary, obstetrics/gynaecology, environmental, mental health, substance misuse, general/minor. Due to the sometimes brief details given for presenting complaint, it was impossible to satisfactorily separate mental health and substance misuse encounters, so these were merged together as a single mental health/substance misuse category. These 13 categories formed the basis of our set of presenting complaint categories. One author (RH) created extensive lookup tables of words and phrases within the presenting complaints field and used them to recode each presenting complaint to one of the categories. Where there was uncertainty, this was resolved by discussion with CB. While we did not run independent checks of the mapping of entries to Emergency Care Data Set (ECDS) codes, the automation of this by lookup tables meant that mapping was consistently applied. We were able to recode 99% of the data into one of the 13 categories, the remaining 1% of encounters (a mix of blank, unclear or very infrequently occurring terms) were placed in an additional ‘other’ category.
Definition of frequent attendance
We used a cut-off number of attendances of five within the 12-month period for inclusion in this analysis. There is no standard definition for frequent attendance of EDs. While several recent studies have used a cut-off of four or more attendances,3 we chose five as this was the threshold used in a recent study of GP Out of Hours service attendance which used a similar approach to analysis.17 The data contained some instances where patients had multiple encounters on the same day. Where the reason for encounter was the same we used only the first encounter.17 19 Where the reasons for encounter were different we included all encounters.
Estimation of heterogeneity of presenting complaint
We used two measures of heterogeneity of reason for encounter, both of which are based on the presenting complaint category. The first was count of categories and the second was Shannon entropy of categories. Count of categories was defined as the number of different categories present in a sequence of consultations. Shannon entropy extends this by describing the unpredictability of the distribution of categories (if most consultations are in one category it is quite likely the next one will be in that category too; if consultations are spread across several categories, it is harder to predict what the next one will be). These measures which are widely used for instance in ecology have previously been used in a study of primary care out of hours services.17 Figure 1 illustrates the relationship between these, using the example of patients with eight ED attendances for up to four categories of presenting complaint (indicated by the letters A, B, C and D). Shannon entropy increases when the attendances are more evenly distributed between presenting complaints categories and decreases when one or two presenting complaints account for most of the attendances. Both the count and the entropy of presenting complaint categories are strongly related to the total number of attendances and must be interpreted in the light of that.
Schematic representation of the relationship between number of attendances, number of presenting complaint categories and entropy of presenting complaint categories. Letters in italics refer to hypothetical presenting complaint categories and are provided as illustration only.
Statistical analysis
Association of measures of heterogeneity with patient characteristics
We plotted the relationship between count of presenting complaint categories, the entropy of presenting complaint categories and the total number of contacts by age, sex and whether patients had any encounters for a mental health problem. We did not have data on other diagnoses so used age as a proxy for multimorbidity. We then examined the relationships between either the count of entropy of presenting complaint categories with age, sex and mental health contact using univariable linear regression, followed by multiple linear regression. Because of the large number of individuals in the data, we entered all variables into multiple regression models and model fit was estimated by Aikake information criterion.
Predictive value of variables for further contact
To assess whether measures of heterogeneity could be clinically useful, we examined whether calculation of the measure after a given number of contacts was predictive of future contact. We used a range of values for the number of contacts (N) between 5 and 15. For each value of N we identified all patients with at least N contacts and calculated the heterogeneity measures after those N contacts. We then used logistic regression to examine whether the heterogeneity measures predicted having N+1 contacts. We used count and entropy of presenting complaint categories (split at the median into high and low) and attendance with a mental health problem (present or absent) over the first N contacts as predictors. We calculated the OR adjusted for age and sex for each value of N.17
Sensitivity analysis
During the analysis we found marked variation between EDs in the extent to which presenting complaints were coded. To examine whether this influenced our results we conducted a sensitivity analysis in which we repeated the analysis using only those EDs with many (>200) unique presenting complaints.
Data processing and statistical analysis was carried out using R 3.6.0.
Results
The dataset initially contained 1 272 817 ED encounters by adults aged 18 or over. 47 900 (3.8%) encounters were excluded because there was no linked pseudonymised ID number, leaving 1 224 917 encounters by 805 180 adults. A further 193 204 encounters were excluded: 169 165 (13.8%) without any presenting complaint data; 17 340 (1.4%) duplicate encounter records and 6699 (0.5%) with incomplete recording of age or sex. This left 1 031 713 (81.1%) encounters by 691 571 individuals. 15 986 (2.3%) of these individuals made five or more attendances over the year, representing 118 501 (11.5%) of eligible encounters and were therefore defined as frequent attenders.
The median age of frequent attenders was 50 (IQR 31–75) and 7778 (48.7%) were men. 3652 frequent attenders (22.8%) had one or more contacts for a mental health problem. The distribution of contacts per patient was heavily skewed as has been found elsewhere.19 Frequent attenders had a median of 6 contacts (range 5–152, IQR 5–8) with a median of 3 (IQR 2–4) unique encounter categories. 1794 (11.2%) frequent attenders had more than 10 encounters and 2395 (15.0%) had 5 or more different presenting complaint categories. Table 1 shows the distribution of different presenting complaint categories per patient by total number of contacts.
Number of presenting complaint categories by total number of attendances per individual patient
Association of measures of heterogeneity with patient characteristics
Figure 2 shows the relationship between the number of unique encounter categories and both the total number of contacts (left hand plots) and entropy of encounter categories (right hand plots). These are split by median age (plots A and B), sex (plots C and D) and whether patients had any encounters for a mental health problem (plots E and F). These show little difference between groups until applied to patients with at least five different encounter categories.
Association of number of unique PC categories with total ED contacts and entropy of PC categories. PC, presenting complaint.
Table 2 reports the results of univariable and multiple variable linear regression with either number of presenting complaint categories or entropy of presenting complaint categories as the outcome variable. In light of the very large number of patients in the analysis, the finding of statistically significant differences in groups does not automatically render them clinically meaningful. While the strong association between number and entropy of categories is to be expected, more striking is the small effect of the other variables (mental health consultation history, age or sex) on these measures of heterogeneity. Individually these accounted for only between <0.1% (sex) and 4%–8% (mental health consultation history) of the variance of both number of categories and entropy of categories. The finding that less than 1% of the variance in count and entropy of presenting complaint categories is attributable to age suggests that comorbidity of long term conditions, which is strongly determined by age, is not a major contributor to heterogeneity of presenting complaint.
Results of linear regression (univariable and multiple variable) for two measures of heterogeneity of presenting complaint category (PCC)
Predictive value of variables for further contact
Table 3 considers whether heterogeneity after a given number of attendances is useful in predicting whether a patient is likely to attend again. It shows the probability of one or more additional contacts, after each given number of contacts, expressed as an OR comparing those with an above-median or below-median value of a variable. The three variables shown are count and entropy of presenting complaint categories (comparing above-median value to median or less), and whether one or more attendances had been for a mental health problem (compared with no mental health problem). After each number of attendances (between 5 and 10), patients with higher levels of both measures of heterogeneity were less likely to reattend. For example, after seven attendances the OR for an eighth attendance for a patient with a high rather than low entropy of presenting complaint category was 0.8 (95% CI 0.72 to 0.90). In contrast a patient with at least one mental health presenting complaint was more likely to reattend than a patient with no mental health presenting complaints: OR 1.57 (1.37 to 1.80)
Prediction of probability of further attendance for each number of attendances
Sensitivity analysis
The sensitivity analysis which was limited to EDs with >200 presenting complaints (8 EDs, 6605 frequent attenders, 49 017 attendances) found no material differences in the results from the full analysis (data not shown), suggesting that the absence of major effects of heterogeneity on future attendance were not simply due to insufficient granularity of coding.
Discussion
Summary of principal findings
We quantified and examined the heterogeneity of presenting complaints in frequent attenders at EDs. There was little increase in heterogeneity of presenting complaint with age suggesting that factors other than multimorbidity (for which age is a proxy) drive heterogeneity. In contrast heterogeneity was associated with one or more attendances for a mental health problem. Heterogeneity did not predict further ED attendance in frequent attenders.
Strengths and limitations
The main strength of this study is the use of a very large, population-wide dataset involving multiple different EDs. By examining associations across a range of thresholds for frequent attendance we were able to demonstrate that our initial assumptions have not affected our results. The main limitation was the inconsistency of recording presenting complaint between EDs. Some used a very wide range of statements—which we subsequently mapped to categories—while others used very few categories. Particularly in these low category number services we found examples of unhelpful coding (eg, weakness due to a stroke being coded as musculoskeletal). However, the sensitivity analysis which was restricted to departments with many different recorded presenting complaints found no material differences in the findings. Our choice of 13 categories mirrored that in a study of a GP out of hours service.17 The use of presenting complaint from ED data which was not linked to other medical records meant we were unable to include any measure of actual multimorbidity. This means we could not disentangle the effects on attendance of concern about multiple symptoms from the presence of multiple diseases. The data also did not include information about individual patients’ socioeconomic status.
Comparison with other studies
While this is the first study examining heterogeneity of presenting complaint in the ED, the findings are similar to those described in a study of users of a GP out of hours services17 which used the same approach. In both studies, there was little increase in heterogeneity with age. Neither study was able to triangulate heterogeneity of presenting complaint against documented multimorbidity.
Implications for practice and further research
This study demonstrates that heterogeneity of presenting complaint in frequent attenders of EDs is common and measurable. The weak association of heterogeneity with age suggests that while heterogeneity is likely in part driven by the number of medical conditions an individual has—which strongly increases with age—this is unlikely to be the whole explanation. An alternative explanation for high heterogeneity, is due to different illness perceptions,20 greater experience of multiple symptoms,14 15 or a lower threshold for attendance (whatever the presenting complaint),21 rather than to the experience of more illnesses of such a severity as to require emergency medical care. While this might indicate an unrecognised mental health problem such as anxiety or high emotional distress, greater heterogeneity of reasons for attendance was less predictive of future attendance than explicitly presenting with one or more mental health problems. As the ECDS coding system becomes widely used, it should be possible to automate the sequence generation, which was labour intensive for this study, and use artificial intelligence approaches to identify additional signals in attendance data which are predictive of particularly high future attendance. These could be used to flag patients as at particularly high risk of reattendance and trigger active case management22 particularly where mental health problems appear significant.23
Conclusion
Heterogeneity of presenting complaint can be quantified and analysed for ED use and is only weakly associated with age. This suggests it reflects more than the number of medical conditions and should alert clinicians to the possibility of underlying functional somatic disorders or mental disorders.
Ethics statements
Patient consent for publication
Ethics approval
Ethical approval for the dataset used in this research was gained from the NHS Research Ethics Committee (REC reference 14/YH/1139). The NHS Health Research Authority Confidentiality Advisory Group also approved the research (CAG REC reference 14/CAG/1015).
Footnotes
Handling editor Kirsty Challen
Twitter @ProfSueMason
Contributors CB had the original idea for the study and all authors were involved in design. TS, SMM and CO’K provided the data and RH and CB carried out the analysis. RH wrote the first draft of the paper which was edited and reviewed by all authors.
Funding The study used data collected as part of the NIHR Collaboration for Leadership in Applied Research and Care (CLAHRC) Yorkshire and Humber (Y&H). SMM, CO’K and TS are funded by the National Institute for Health Research Yorkshire and Humber ARC.
Disclaimer The views expressed in this publication are those of the author(s) and not necessarily those of the National Institute for Health Research or the Department of Health and Social Care.
Competing interests None declared.
Patient and public involvement Patients and/or the public were involved in the design, or conduct, or reporting, or dissemination plans of this research. Refer to the Methods section for further details.
Provenance and peer review Not commissioned; externally peer reviewed.