Article Text
Abstract
Aims Ambulance services in England receive around 8 million calls a year, and no known studies have explored characteristics of frequent callers. This study aimed to identify the characteristics of the most frequent callers to Yorkshire Ambulance Service (YAS) between April 2010 and March 2011.
Methods Top 100 frequent callers to YAS were analysed using population comparison, case control and multiple regression methods. 7808 calls were made by the frequent callers, and data were analysed to predict total number of calls made, and explore characteristics of frequent callers.
Results Six call codes were associated with a higher number of calls. Frequent callers were assigned slower response levels, or often no call code. Calls increased during the times of 4:00–9:00, 16:00–20:00 and 22:00–2:00, and in the months of December, January and February. Men and patients with all but the very highest conveyance rates had a higher number of different reasons for calling. Patients with a medical diagnosis were more likely to be conveyed, while patients with a psychiatric classification had a higher number of different reasons for calling, were older and were more likely to call for ‘assault/sexual assault’ or ‘haemorrhage/laceration’.
Conclusions Frequent callers to YAS were a heterogeneous group that differed from the overall population served, resulting in numerous implications for the delivery of services for this group of patients. Further research is required to determine if and how frequent callers differ from frequent attenders at emergency departments.
- emergency ambulance systems
- prehospital care
- statistics
Statistics from Altmetric.com
Background
Frequent users of emergency departments (ED) have received much attention within emergency medicine literature. They have been identified to be a heterogeneous group, whose characteristics differ from those of infrequent users.1 It is unclear how these groups are comprised, although it is believed they may be differentiated by chronic illness,2 ,3 psychiatric conditions,4 substance misuse,3 or be based upon gender, age or other demographics.1 Interventions aimed at reducing use of the ED by frequent users have been suggested to reduce costs while increasing patient outcomes.5 ,6
Despite this growing evidence base, there have been few studies exploring frequent users of ambulance services, who predominantly access the service via telephone. Ambulance services in England received 8.08 million calls during the period April 2010 to March 2011, of which 6.61 million resulted in an emergency response and 4.29 million were conveyed to accident and emergency (A&E) destinations.7 In the UK, Ambulance Quality Indicators (AQI) consist of both systems and clinical indicators.8 The systems indicator entitled ‘emergency calls from patients for whom a locally agreed frequent caller procedure is in place’ makes it a requirement for ambulance services to monitor the number of calls made by frequent callers.
Current studies that have explored frequent users of ambulance services9–11 only include patients who are conveyed by ambulance, therefore excluding potentially half of all patients who were treated but not conveyed, or called for other reasons such as advice. One study exploring ED interventions for frequent users6 found that as ED use dropped, so did ambulance service use. This was coupled with a relative decrease in ambulance costs, although the cost of each visit remained the same to the ambulance service. Studies on frequent use of the ED have identified that frequent users are more likely to attend via ambulance than non-frequent users,3 ,12 ,13 although this is the opposite in psychiatric emergency services.14
This study aimed to address this gap in the literature by exploring the characteristics of the top 100 frequent callers to an ambulance service in England. This was done using multiple methods of analyses depending upon available data, including within-group analysis of the frequent callers, a comparison with the total population and a case control analysis.
Methods
Ambulance service description
Yorkshire Ambulance Service NHS Trust (YAS) works closely with hospitals, health trusts and healthcare professionals, as well as the other emergency services. It covers the whole of Yorkshire, from isolated moors and dales to urban areas, coastline and inner cities, providing a 24 h emergency and healthcare care service to more than 5 million people. In particular, YAS operates
-
Two emergency operations centres where staff receive 999 calls and deploy the most appropriate response to meet patients’ needs,
-
An accident and emergency service in response to 999 calls,
-
A non-emergency patient transport service which takes eligible patients to and from their hospital appointments,
-
A general practitioner (GP) out-of-hours call handling service for some primary care trusts across Yorkshire and beyond.
Dataset description
The sample of frequent callers consisted of the top 100 frequent callers to the ambulance service over the study period of April 2010 to March 2011. Patients of all ages were included in the sample. Those that made regular hoax calls were automatically excluded. Variables included ‘date of call’, ‘time of call’, ‘conveyed’ and ‘despatch code’, the last consisting of the primary call reason and the assigned response level. Primary call reason is based on the Advanced Medical Priority Despatch System (AMPDS), and is not meant as a means to clinically diagnose patients via the telephone, but rather to assist in assigning responses. Response assignment consists of a non-linear grid encompassing responder capability (basic life support and advanced life support) and response time (cold is routine and hot is lights-and-siren). While Alpha to Delta calls are measured on this scale, Omega calls are referrals or alternate care, and Echo calls require the most urgent of responses. Further information on response levels is available.15
Demographic information (age, gender and primary diagnosis) was also obtained for the top 100 frequent callers. Primary diagnosis was split into three groups: (1) physical (respiratory, cardiac or medical conditions) (n=47); (2) mental health (n=20) or (3) a combination of physical and mental health (n=33). A separate variable based upon primary call reason was also developed and entitled psychiatric classification. Frequent callers with >10% of calls for ‘overdose/poisoning’ and ‘psychiatric/abnormal behaviour/suicide attempt’ were coded as psychiatric, while those not meeting this criteria were coded as non-psychiatric.
The population consisted of every caller, including frequent callers who had contacted the ambulance service. Data were available on the primary call reason, of which 553 439 (83.3%) calls had a valid call code, and the date and time of calls. Data were not available for response level. A case-control group was created by randomly selecting 100 callers from the total population in order to compare the response levels assigned to calls. The inclusion criteria for the control group were that they had to have a valid despatch code assigned to their call, and they should not have already been identified as a frequent caller.
Data analysis
Data were analysed in three categories: frequent caller characteristics, call patterns and predictors of the total number of calls made. All data were analysed using PASW Statistics V.18. The level of significance for all tests was set at α=0.95. For the top 100 frequent callers, the Shapiro–Wilk test indicated that the total calls (p<0.001), number of different reasons for calling (p=0.016) and number of times conveyed (p<0.001) were not normally distributed, but the age of the callers (p=0.357) was normally distributed.
Multiple regression analysis was conducted to determine which primary call reasons were able to predict the total number of calls made by the top 100 frequent callers. The assumptions of linearity, normality, reliability and homoscedasticity were not met,16 so the data was cleaned. A Cook's Distance (Di) value of >1 was used to remove patients (n=4, Di=1.281 to 2.841) with a large influence on the data. Once these outliers were removed, variance inflation factors (VIFs) of >10 were used to determine any collinearity (n=0). A plot of the studentised residuals was then used to remove any outliers (n=4), identified by being >3 SDs away from the mean. The Breusch–Pagan test was then used to determine homoscedasticity (x2=0.00015, df=1, p=0.990) and VIFs ranged 1.075–3.612. The final multiple regression analysis contained 92 participants.
This study was classed as an audit by YAS research and development (R&D) department as defined by the Research Governance Framework for Health and Social Care,17 and therefore did not require NHS research ethics committee approval, or R&D approval from YAS.
Results
Call information
The top 100 frequent callers to the ambulance service made 7808 calls over the study period, of which 5818 (74.5%) had a valid despatch code. Frequent callers made a greater proportion of their calls than the population during the hours of 4:00–9:00 (16.2% compared with 12.2%), 16:00–20:00 (22% compared with 20.2%), and 22:00–2:00 (17.6% compared with 15.5%) (p<0.001). Frequent callers also made a greater proportion of their calls (p<0.001) than the population during the winter months of December, January and February (30.1% compared with 26%), represented in figure 1.
Response level
There was a significant difference between the case and control groups (x2=24.520, p<0.001) in the response levels assigned to calls. Frequent users have 27.2% of their calls assigned to the low-level responses of Omega and Alpha in comparison with 12% of calls in the control group. This contrasts with Delta-level responses, which make up 28.5% of frequent users in comparison with 44% of the control group. The differences are represented in figure 2 in the same format as the original non-linear response level matrix.15
Reason for calling
Total number of calls made
Calls made for each call code were used to develop a multiple regression model of the total number of calls made by the top 100 frequent callers. Using the enter method of multiple regression, a significant model emerged (F23,91=79.993, p<0.001, adjusted R2=0.958). Significant predictors are shown in table 1. ‘breathing problems’ and ‘sick person’ contributed the most to the model, followed by ‘psychiatric/abnormal behaviour/suicide attempt’, ‘abdominal pain/problems’, ‘chest pain’ and ‘falls’. For most call reasons, for one call increase in the independent variable, there was approximately one overall call increase. The two main exceptions to this were ‘choking’ and ‘assault/sexual assault’, which increased by 4.5 (95% CI 1.216 to 7.825) and 3.9 (95% CI 1.297 to 6.581) calls respectively for each call made. Calls for codes ‘healthcare professional call’ (95% CI 0.402 to 2.885) and ‘sick person’ (95% CI 1.398 to 1.694) resulted in around 1.5 more calls overall.
Population comparison
There was a significant difference (x2=4539.966,i p<0.001) between the primary call reasons of the top 100 frequent callers and the population (figure 3). Key primary call reasons by frequent callers, where they are considerably higher (>1%) than the general population are ‘abdominal pain/problems’ (6.6% to 3.4%), ‘breathing problems’ (16.4% to 10.5%), ‘chest pain (non-traumatic)’ (13.6% to 10.0%), ‘headache’ (3.0% to 0.8%), ‘psychiatric/abnormal behaviour/suicide attempt’ (10.3% to 2.4%) and ‘sick person’ (19.1% to 8.1%); 24.5% of all calls by frequent callers were not assigned a call code in comparison with 16.7% of the population.
Frequent caller attributes
Demographics
Of the top 100 frequent callers, 46% were men and 54% were women. There was no significant difference in age (p=0.510, 95% CI −4.592 to 9.180) of men (=55.4, SD=15.94, range 19–86) and women (=57.7, SD=18.36, range 13–90). Frequent callers were conveyed 2729 (35.0%, range 0–100%) times from the 7808 calls. There was no significant difference in the number of times conveyed between men (=26.7, SD=24.91, range 0–133) and women (=27.8, 26.74, range 1–172) frequent callers (p=0.777). However a Pearson correlation revealed that frequent callers making more calls were more likely to be conveyed (r=0.308, p=0.002). Men (=76.02, SD=120, range 32–846) and women (=79.8, SD=98, range 18–705) did not differ significantly in the total number of calls made (U100=1094.5, z=−1.020, p=0.308). They did differ significantly in the number of different reasons for calling (U100=934.5, z=−2.139, p=0.032), with men calling for more reasons (=8.57, SD=3.81, range 1–20) than women (=7.02, SD=2.95, range 1–14).
Men (=2.83, SD=4.02, range 0–18) were significantly more likely than women (=1.94, SD=4.39, range 0–25) to call for ‘overdose/poisoning’ (U100=959, z=−2.124, p=0.034), and men (=8.46, SD=17.95, range 0–106) also made more calls than women (=3.89, SD=9.43, range 0–48) for ‘psychiatric/abnormal behaviour/suicide attempt’ (U100=944, z=−2.205, p=0.027). There were no other significant differences in primary call reason based on gender (table 2).
Primary diagnosis
There was no significant difference (x2 (2, n=100)=1.465, p=0.481) in the total number of calls made between patients with a physical diagnosis (=80.79, SD=121.98, range 18–846), mental health diagnosis (=72, SD=46.15, range 35–235) or a combination (=77.91, SD=116.06, range 33–705). There was also no significant difference (x2 (2, n=100)=0.585, p=0.746) in the number of different reasons for calling between patients with a physical diagnosis (=7.77, SD=3.82, range 2–20), mental health diagnosis (=8.15, SD=3.36, range 1–14) or a combination (=7.42, SD=2.97, range 1–14).
Psychiatric classification
There was a significant difference (p<0.001) in the age of the patients between the non-psychiatric group (=61.73, SD=17.76, range 13–90) and psychiatric group (=48.98, SD=13.33, range 19–77). The total number of calls made did not significantly differ (p=0.138) between the non-psychiatric group (=73.42, SD=92.95, range 18–705) and the psychiatric group (=85.08, SD=128.51, range 33–846). However, there was a significant difference (p=0.002) between the non-psychiatric group (=6.88, SD=3.289, range 1–20) and the psychiatric group (=9, SD=3.31, range 2–15) for the total number of different reasons for making a call. Those in the psychiatric group (=0.75, SD=1.3, range 0–5) made significantly more calls (p=0.001) for ‘assault/sexual assault’ than those in the non-psychiatric group (=0.12, SD=0.454, range 0–3), and those in the psychiatric group (=2.05, SD=3.16, range 0–13) also made significantly more calls (p=0.023) for ‘haemorrhage/lacerations’ than those in the non-psychiatric group (=1.12, SD=3.36, range 0–25⇓).
Discussion
Frequent callers displayed many different characteristics compared with the population, and there appeared to be different subgroups of frequent callers based broadly on their primary medical diagnosis or psychiatric classification. These subgroups appear to be similar to those identified in EDs,2–4 although further research is required to determine where similarities and differences may exist.
Frequent callers were more likely to be assigned ‘abdominal pain/problems’, ‘breathing problems’, ‘chest pain (non-traumatic)’, ‘headache’, ‘psychiatric/abnormal behaviour/suicide attempt’ and ‘sick person’ call codes. These were identified primarily via the population comparison, and with the exception of ‘headache’ were triangulated with the multiple regression analysis. The lack of the headache call code in the multiple regression analysis suggests that although headache is a more common call code for frequent callers, it had little impact upon the total number of calls made, and is assigned sporadically by the call taker, most likely as a result of headache calls being combined with a more serious complaint which takes priority.
Unassigned call codes were also more common for frequent callers than the rest of the population. It is possible that this is a result of frequent callers making duplicate calls within a short time of the initial call without completing the triage process, or hanging up on the call after enquiring if an ambulance has been dispatched. This may impact upon the ability for ambulance services to achieve their frequent caller AQI target8 by skewing the total number of calls made by frequent callers for clinical needs. Frequent users of the ED are often deemed to be misusing healthcare resources, although this is acknowledged to be a misassumption.18 While call takers could have responded differently to patients recognised as frequent callers, the process of triaging patients using the AMPDS reduces this possibility as the system relies on a linear triage process as opposed to clinical judgement. A further possible explanation is that patients’ healthcare needs are unmet in a timely manner, with further investigations required.
Comparison of months with the population suggests that frequent callers are significantly more likely to call during the winter months of December, January and February. It is possible that this was a result of the frequent callers being more vulnerable, and has operational implications for ambulance services due to increased demand for services during the busiest period of the year.
Frequent callers were also more likely than the population to call during the hours of 4:00–9:00, 16:00–20:00, and 22:00–2:00. By 9:00, frequent callers plateaued until 16:00, whereas the population peaked between 9:00 and 16:00. These times coincide with the availability and provision of other services that frequent callers may need, such as primary care or community care services, suggesting that when other services are available, the number of calls made by frequent callers reduce.
Frequent callers were also more likely to be assigned lower urgency call codes than non-frequent callers, demonstrating a potential lack of insight into the difference between the need for emergency treatment and a condition that could be managed via an alternative pathway in the community, such as mental health pathways, community nursing services and out-of-hours GPs. The speed at which ambulance services respond to calls could also be a contributing factor that impacts on the behaviour of frequent callers by providing instant access for patients who may require reassurance and attention.
There were apparent subcategories of frequent callers, in particular, those with either a medical or mental health condition. Patients with a psychiatric classification had a greater number of different reasons for calling, were older and were more likely to call for ‘assault/sexual assault’ and ‘haemorrhage/lacerations’, as well as their inclusion criteria of ‘overdose/poisoning’ and ‘psychiatric/abnormal behaviour/suicide attempt’ calls. Patients with a medical diagnosis were more likely to be conveyed more often than those with a mental health diagnosis. No similarities existed between the characteristics of patients with psychiatric classifications, suggesting that these two methods of identifying patients highlighted different subgroups. Additionally, it was identified that patients who were male, had a psychiatric classification or were not conveyed a very high amount of times were likely to have a greater number of reasons for calling. A summary of significant findings is presented in table 3.
Based on these findings, it may be possible to develop a tool which would identify patients who are likely to become frequent callers to ambulance services, which in turn could be used as part of a targeted intervention. Similar interventions in the ED have been suggested to be successful,6 ,19 reducing ED costs while improving social and clinical outcomes for patients19 by delivering appropriate care that incorporates the unmet needs of the patient before they become a frequent user.1 As ambulance services are often a first point of contact for patients, reducing their frequent users may reduce frequent use of other services, such as EDs.
Limitations
The overall population sample was inclusive of the top 100 frequent callers. Analyses including the overall population sample will have been either underestimated or overestimated, depending on the particular statistical test or comparison. As the frequent callers only constitute a small proportion of the overall population sample, the impact of this will have been minimal and was not anticipated to have had a large impact on significance levels.
It is possible that some patients were not identified as being frequent callers as the caller's name is not always collected by the call taker. Similarly, the use of addresses and postcodes to identify participants is limited when it is possible that people will have made a call or been the subject of a call to the ambulance service when away from home. The use of The International Academy Quality Assurance Guide V.12, AMPDS20 in the identification of the primary call reason was also problematic, as the call codes assigned may have been incorrect.
The subject to variable ratio (svr) for the multiple regression analysis of 10(s):2.6(v) was below the rule-of-thumb minimum of 10(s):1(v).21 The high adjusted R2 value may actually compensate for the low svr using a Monte Carlo simulation,22 though further research is required to pinpoint the most appropriate number of subjects for 26 predictor variables.
The >10% level used to distinguish between psychiatric and non-psychiatric classifications was chosen arbitrarily. This level was chosen by the research team based on clinical experience, and a requirement for a means of identifying patients with a psychiatric need which was based on routine data collected by the ambulance service. Although the actual medical history may be a more valid indicator, the 10% level still provided a useful distinction between patients’ medical conditions identified by their primary diagnosis, and the primary call reason assigned to their call.
Finally, due to the use of routine data used by the ambulance service, it was not possible to match the control group with age or gender, nor was it possible to provide demographic information for the control group. Ambulance services in the UK do not collect or hold demographic information on callers as it is often not collected during the triage process. In order to address this limitation, future studies should use a larger control group or collect demographic information for the control group from other sources such as GPs or hospital records. Similarly, the use of routine data meant that some call codes, such as ‘sick person’ or ‘unknown’ have limited clinical relevance.
Conclusion
This retrospective analysis of top 100 frequent callers to an ambulance service begins to address the dearth of research into frequent callers to ambulance services. The findings suggest that frequent callers differ in many ways to non-frequent callers, with the call reason, response level, month and time of day all differing. In addition to these differences, there also appear to be subgroups of frequent callers, although their exact composition requires further research. The variables identified in this study provide a foundation on which to develop a method of identifying potential frequent callers which could be used in the development of an intervention aimed at reducing use of ambulance services by frequent callers.
References
Footnotes
-
Contributors JS was involved in the conception, design, analysis, interpretation of data and drafting the article. APS and KW were involved in the conception, interpretation of data and critically revising the manuscript. PD was involved in the conception, design, interpretation and critically revising the manuscript. All authors provided final approval of the version to be published.
-
Funding Yorkshire Ambulance Service NHS Trust.
-
Competing interests None.
-
Provenance and peer review Not commissioned; externally peer reviewed.
-
↵i 4 cells (5.9%) had an expected count of <5. Minimum expected count was 1.47.