Drivers of 'clinically unnecessary' use of emergency and urgent care: the DEUCE mixed-methods study

The review question and initial theoretical framework

We focused on three services: emergency ambulance, EDs and general practice. The research question was ‘What are patients’ perspectives of why they make use of services providing emergency and urgent care that is judged subsequently by health professionals to be “clinically unnecessary”?’.

Realist synthesis allows the initial theoretical framework or rough programme theories to be identified in different ways. 38 Coster et al. ’s28 recent review of the demand for emergency and urgent care offered a set of potential rough programme theories, but it was based on qualitative and quantitative research of all users of a range of emergency and urgent care services. Because this did not focus solely on those patients who were judged to have made ‘clinically unnecessary’ use of services, we decided not to use these theories as the basis for developing our programme theories, but rather as a theoretical framework to guide our exploration.

Identifying and selecting primary studies for inclusion

Although we could have identified our literature through database searches, we decided not to do this as we were aware of a number of high-quality reviews of demand for emergency and urgent care that were likely to include the literature we were seeking. We therefore selected three recently published, complementary reviews that were likely to contain relevant articles about emergency ambulance service demand,17 ED demand29 and emergency and urgent care,28,30 together with one ongoing review by a member of our team (Joanne Turnbull). 39 These reviews focused on demand for emergency and urgent care generally, self-referred ED users or the use of ambulance services for ‘primary care sensitive conditions’. We identified the studies in these four reviews and requested copies of the articles from their lead authors. In addition, in February 2017 we updated these searches by using MEDLINE and Google Scholar^TM (Google, Inc., Mountain View, CA, USA) to identify material published during 2015 and 2016. Finally, because of a lack of articles identified that focused on ‘clinically unnecessary’ use of same-day GP appointments, in April 2017 we undertook specific searches in MEDLINE and Google Scholar for relevant general practice-focused studies from the start of each database to March 2017. All articles were in English as this was an inclusion criterion of the four reviews, and we applied this to the updated searches. All countries were included. Figure 2 summarises the search processes and the selection of studies, and Appendix 2 provides the search strategies.

FIGURE 2.

Summary of search processes and selection and extraction of articles.

Quality appraisal

The nature of realist reviews means that the formal quality assessment process that is associated with other review types is not used. 35 The study proposal outlined our intention to assess the relevance, context and methodological rigour of potential literature. As part of this process, the relevance of all literature identified from the reviews and additional searches was appraised and we included only literature related to ‘clinically unnecessary’ use of our three services. We identified studies as relevant if they focused on ‘appropriate use of services’, ‘non-urgent ED users’, ‘ED users who could be managed in alternative urgent care services’, ‘ambulance users or ED users with primary care conditions’ or ‘service use for minor disorders/ailments’, or when it was established in the paper that a group made high levels of contact with a service when they could have used alternative services (e.g. ‘use of out-of-hours services for febrile child’). For the qualitative evidence, we then undertook an assessment of how well the context of each included study was reported (see Appendix 3, Table 15) and assessed the study’s methodological rigour using the Critical Appraisal Skills Programme (see Appendix 4, Table 16). 40 We had initially stated that we would exclude studies based on the detail of their reporting of context but not on rigour, but we later decided not to exclude for either reason to ensure that all relevant literature was included. Poorly described context and low methodological rigour tended to coincide, and we took account of these issues in the analysis so that programme theories were not based solely on poorer-quality articles.

Data extraction

For the qualitative evidence, each article identified in the first search iteration was coded by context (year, country, service and type of user, e.g. not clinically necessary, common user, frequent user) and subgroup (e.g. young adults). We also identified key themes, which are reported in Appendix 5, Table 17. We then identified provisional context (C) and mechanism (M) chains for the outcome (O) of using a higher-acuity service than was considered clinically necessary. We defined the mechanism as the trigger for or driver of the decision, which arose from an ongoing situation that we defined as the context. As has been found in other realist reviews, it was often difficult to distinguish context from mechanism41,42 because of the multiple mechanisms operating that were often contexts for further mechanisms. In the early stages of the process, we undertook duplicate data extraction of some articles and discussed context–mechanism–outcome chains and potential programme theories among our team (JCon, JCos, JL and AOC). We supported the findings with direct quotations from participants in the included studies.

Developing and refining programme theories

During team discussions, we sought to identify chains of inference and connections across the data. Our initial proposal stated that we would identify between three and six rough programme theories to refine and test, but as the review progressed we decided to retain all of the programme theories we identified, as we considered them to be of equal importance and too inter-related for any to be appropriately excluded. We had multiple long team discussions (JCon, JL, JCos and AOC), as well as further discussion with the wider team and PPI members in the Project Management Group, to refine the context–mechanism–outcome chains until we had finalised 10 detailed programme theories. We further developed our understanding by focusing on what created an urgent need to contact a service within each programme theory, and developed six mechanisms across the 10 programme theories to account for this urgency. We presented the programme theories at a conference and to our Project Advisory Group for external validation.

Description of the qualitative evidence base

We included 32 qualitative articles in the review and rated their relevance to the research question. Eighteen were rated 1, directly relevant, and 14 were rated 2, partially relevant (see Appendix 3, Table 15, for details of the relevance ratings). Articles predominantly focused on EDs, either adult/mixed (n = 16) or specialist paediatric (n = 7). Only four studies focused on GP out-of-hours services and two focused on emergency ambulance services. No studies considered daytime general practice. Most articles were from the USA (n = 12) or the UK (n = 10), with others from continental Europe (n = 5), Australia, Canada and the Caribbean. Almost all were from high-income countries, although some explored the perspectives of communities with high levels of deprivation in those countries. The health-care context varied widely between countries, particularly in relation to how patients paid for services.

Mechanisms and programme theories

An overview of our six underlying mechanisms and 10 inter-related programme theories for seeking ‘clinically unnecessary’ emergency and urgent care, together with their potential relevance to particular subgroups, is shown in Table 1.

Testing the programme theories

There was considerable support for all of the programme theories from existing theories. Only some of the programme theories were supported by quantitative research, and this was largely from cross-sectional quantitative studies rather than comparative studies (Table 2).

Selecting subgroups

We planned to select the subgroups using the following criteria:

• those more likely to be ‘clinically unnecessary’ users based on the literature

• those more likely to be ‘clinically unnecessary’ users based on the views of our Project Advisory Group

• those on whom a lack of qualitative research has been undertaken.

To identify potential subgroups, Emma Knowles and Alicia O’Cathain read reviews and quantitative papers from the WP1 literature. Articles reported sociodemographic status (age, sex, ethnicity, relationship status), socioeconomic status (income, education level, employment status, housing tenure), presenting clinical complaint, health-care utilisation (registered with a GP, prior use of care, attendance by day/time, arrival mode) and distance to care. The findings from the literature were sometimes inconsistent, so we used them to identify a list of candidate subgroups to be discussed with our Project Advisory Group. Following consideration of the candidate subgroups, the group identified the following as of interest to them: people of Eastern European origin, individuals living in socially deprived areas, ambulatory ED patients, parents with young families, young adults and care home residents. We decided not to focus on Eastern European people, as an ongoing study was focusing on their help-seeking behaviour relating to urgent care services such as WICs,22 ambulatory ED patients, because we were interested in three services rather than EDs only, or care home residents, because the decision to seek care was likely to be taken by a third party rather than the patient. The Project Advisory Group members strongly supported focusing on people from socially deprived areas and parents of young children. They identified young adults as an interesting group to study because they viewed this group as accessing care for convenience. Some relevant qualitative research had been undertaken on these subgroups, particularly on parents of young children, but this had been undertaken largely outside the UK.

We decided that our qualitative research should focus on the following three subgroups.

1. Young adults: people aged 18–25 years (later expanded to 18–30 years). Reviews in WP1 identified that younger adults were more likely to be ‘clinically unnecessary’ users of EDs,13,14,86 and a study of medically unnecessary GP out-of-hours users showed that they were more likely to be aged 25–44 years. 88 This choice of subgroup was supported later in our study, when a UK-based research study was published showing that those aged 16–44 were more likely to use EDs for non-urgent reasons105 and a media study identified young people aged 18–29 as more likely than older working-age groups to use EDs and less likely to visit their GP. 106 None of the qualitative research identified in our WP1 review had focused on this subgroup.

2. Socially deprived: people residing in postcodes with an Index of Multiple Deprivation (IMD) score of ≤ 3. Reviews in WP1 identified that use of ambulance services for non-urgent conditions was higher in deprived areas17 and that people from deprived communities may prefer to access EDs rather than GPs. 28 A review of ‘clinically unnecessary’ use of EDs agreed with this,14 but two other reviews identified that affluent groups were more likely to be ‘clinically unnecessary’ users of EDs. 13,86 Although the evidence was conflicting, our Project Advisory Group members also identified those who are socially deprived as ‘clinically unnecessary’ users. A few of our qualitative articles in WP1 focused on this subgroup, but these were mainly US based.

3. Parents of young children: parents (aged ≥ 18 years) attending on behalf of a child aged 0–5 years (later expanded to 0–10 years). The odds of being a ‘clinically unnecessary’ ED user were found to be 1.5 among children aged 0–2 years. 107 One-third of children attending EDs in London received reassurance only rather than investigations, and researchers concluded that many could have been managed without attending an ED. 108 Our Project Advisory Group identified this as an important group to study. However, a number of the qualitative studies identified in our WP1 review focused on parents of young children attending paediatric EDs, EDs or GP out-of-hours services. We decided to focus on this subgroup because most of these qualitative studies were US based and so did not apply to the UK.

Emergency ambulance service

In the two ambulance service areas (Sheffield and Wolverhampton), ‘clinically unnecessary’ users were defined as patients who received only telephone advice or were referred to another service by clinical staff, that is cases where an ambulance was not sent. We briefed relevant staff about the study, and they asked patients before the end of a call if they would be interested in participating in the study and willing to have their contact details passed to the research team. It was not possible for staff to ascertain whether or not someone was living in an area of deprivation based on our criteria (i.e. IMD score of 1–3). As a result, staff were asked to invite all patients who were not excluded and to establish the age of young adults and of young children whose parents were calling on their behalf, and the postcodes of all callers who agreed to be contacted. This information, together with patients’ contact details (usually a telephone number), was telephoned to the research team so that we could check eligibility with regard to deprivation. Once ineligible patients were excluded, a member of the research team attempted to contact those remaining to explain more about the study and arrange to post or e-mail further information (for copies of the letter of invitation, participant information sheet and consent form, see the project web page: www.journalslibrary.nihr.ac.uk/programmes/hsdr/1513612/#/; accessed 5 February 2020). A follow-up call was made a few days later to find out if the person was still interested and, if so, to arrange a convenient time for the interview. We confirmed this by letter or e-mail.

We faced considerable challenges in recruiting from the two ambulance service areas. Whereas managers were supportive of the study, front-line staff assessing patients were initially unaware of the study or their role in it. Owing to the pressured environment of the control centre, which operates tightly structured procedures, it was also challenging to introduce a new element into staff contact with patients. As a result, the researchers spent considerably more time liaising with front-line staff to establish effective procedures for patient contact and passing on information, and to keep the study in their awareness. Despite these measures, recruitment remained difficult, particularly in Sheffield, where it did not prove possible to establish a successful system for ongoing recruitment. We encountered specific difficulties with two subgroups. As ambulances are usually dispatched when a call relates to a young child, very few parents of young children were identified to be contacted. Recruitment of young adults also proved extremely difficult. These difficulties led to the expansion of recruitment from the original Sheffield and Wolverhampton areas to include the whole of South Yorkshire and the entire West Midlands region.

Emergency departments

Recruitment strategies varied between the hospitals involved in the study depending on the way in which staff identified relevant patients. A researcher (JL or LBE) was based at each ED for a number of sessions and liaised with relevant staff to identify potential recruits. Recruitment was generally undertaken during the ‘in hours’ period, but some attempts were made to recruit over the weekend or during the early evening.

The definition of ‘clinically unnecessary’ was more complex than that used in the ambulance service, as staff with a triage role were asked to identify patients whom they considered could have used an alternative service or self-care to manage their presenting problem, or, if they had used an emergency ambulance to access the ED, could have arrived by another means.

General practitioners

The researchers established links with a particular general practice in each area (LBE in Sheffield and JL in Wolverhampton) that agreed to be recruited to the study. Both general practices were in areas of social deprivation to increase the likelihood of finding participants from this group. We asked GPs and practice nurses to identify patients whom they considered had made ‘clinically unnecessary’ use of same-day/urgent contact, either in a face-to-face consultation or by a telephone call, and whom they believed could have managed their problem through either self-care or a routine appointment. To reduce the impact on GPs’ time, they were not asked to speak to patients about the study but were asked to pass the patient’s details to a research nurse or the practice manager at the end of their session. The research nurse or practice manager then telephoned the patient to ask if they were interested in the study and willing to have their contact details passed to the research team. Details of potentially eligible patients were then telephoned through to the team, who screened the patient for eligibility and made contact, as described previously. Different general practices employed slightly different approaches to identifying ‘clinically unnecessary’ users, so the inclusion criteria varied to some extent between practices as well as between GPs.

As with the other services, recruitment was much slower than anticipated, with a number of difficulties encountered, especially in Wolverhampton. GPs, particularly those with less experience, found it challenging to categorise patients as ‘clinically unnecessary’, and engagement with the recruitment process varied among GPs in each practice. Practices also struggled to find time to make the follow-up calls to patients at certain points. The recruitment of young adults proved particularly challenging, which led to it being extended to a second practice in Sheffield that had a younger and more affluent demographic.

Sample characteristics

Sixteen parents were recruited, ranging in age from 22 to 42 years (mean 33 years). The sample predominantly consisted of interviewees with either high or low deprivation IMD scores, rather than medium scores. Seven had a score of between 1 and 3 (i.e. indicating that they lived in an area of social deprivation, according to our study criteria), whereas another seven had a score of 8–10 (indicating affluence). All but two of the interviewees were female and were accessing care for the child in their role as the child’s mother. Just over half of the interviewees described themselves as white British, with four identifying as black African or Caribbean, two as other white and one as Asian. Twelve interviewees indicated that they were married or living as married, and the other four indicated that they were single or not married. Parents in the sample had between one and five children, with six having one child. The children in the sample were predominantly at the young end of our age range, with nine aged ≤ 2 years and another four aged 3–5 years, and only two aged > 5 years. Of the 16 parents interviewed, one had been recruited following contact with the emergency ambulance service, nine had been recruited following contact with an ED and six had been recruited following contact with a general practice. All but one of the interviewees who were living in areas of deprivation had used emergency ambulance or EDs, whereas those with higher IMD scores had been recruited from their GP.

Overview of findings for parents of young children

Two key drivers were most evident in parents’ decision-making regarding seeking health care: (1) concerns about the seriousness of symptoms that the child was experiencing and (2) a complex interlinking of a sense of responsibility as a parent of a young child and a lower threshold of concern regarding children’s health. In many cases, these concerns prompted a need for some form of reassurance from a health-care provider. There was a strong sense of interaction between these drivers, with each potentially triggering the other. Although these drivers were most prevalent, a number of other factors were also present and contributed to the decision to seek care: in particular, the perceptions or experiences of services, the influence of others and the timing of the incident.

Sample characteristics

Sixteen young adults were recruited, ranging in age from 18 to 30 years (mean 25 years). Most interviewees described themselves as white British, with two being Asian (one Chinese and one Pakistani). Approximately two-thirds of the young adults were female, and all but one had lived in the city from which they were recruited for > 1 year, with most living there for > 5 years. The majority were occupied full-time as students and/or employed full- or part-time. Three described themselves as unemployed because of health problems. Five interviewees were recruited from a general practice in an affluent area and were working full- or part-time, whereas the interviewee recruited from an area of deprivation was unemployed. Half of the sample (n = 8) were living in areas of deprivation (IMD quintile 1 or 2), and, although their child was not the focus of their service contact, four were also parents. All four parents were from areas of deprivation.

Interviewees sought help for a wide range of reasons, and sometimes included more than one reason. Reasons included pain after injuries (n = 4); exacerbations of chronic or ongoing symptoms, including pain, bleeding, digestive problems, fatigue and anxiety (n = 8); breathing difficulties; suspected concussion; vomiting blood after drinking alcohol; suspected miscarriage; and seeking medication to manage anticipated rather than actual symptoms (n = 2). The interviews focused mostly on contact with an ED (n = 8) or a GP (n = 6), with only two relating to emergency ambulance use. The service contact in this sample seemed to be differentiated by sex and deprivation. All of the men had used the ED or emergency ambulance service, whereas just over half of the women had been recruited after contact with the GP. Almost all of the young adults living in areas of deprivation (7 out of 8) had contacted the ED or emergency ambulance service and only one had contacted the GP; both of the young adults who received care from the emergency ambulance service were living in areas of deprivation and were unemployed because of long-term health problems, particularly mental health. By contrast, five of the young adults living in more affluent areas had used a GP and only three had used the ED.

Overview of findings for young adults

Three drivers were most evident in young adults’ decisions to seek health care: concerns about the meaning/seriousness of the symptoms that they were experiencing; an unwillingness to further delay care-seeking; and perceptions or experiences of services. Although these drivers were the most significant among this sample, a number of other factors were also frequently present, notably the influence of others, particularly those in their social network, and the impact of poor mental health. Most interviewees identified between two and four drivers of their decision to seek care, which were usually interconnected.

Sample characteristics

Sixteen individuals residing in areas of social deprivation were recruited, ranging in age from 32 to 80 years (mean 57 years). The majority of interviewees (n = 9) resided in an area with an IMD score of 1 (the highest level of deprivation). All but one interviewee had lived in the city in which we recruited them for > 5 years; the other interviewee had lived in the city for > 1 year. The sample was evenly split between men and women. Interviewees predominantly identified their ethnicity as white (n = 11); four other interviewees were Asian/Asian British (Kashmiri, Indian, Pakistani and Afghan) and one was black African. Nine interviewees described themselves as married, three separated or divorced, three single and one widowed. Half of this sample were retired (n = 8), with one other describing being both retired and working part-time. Four other interviewees were working full-time or part-time, two were full-time homemakers and one was permanently unemployed.

Of the 16 interviewees, five were recruited following contact with the emergency ambulance service, four were recruited following contact with an ED and seven were recruited following contact with a general practice. More people in this subgroup (n = 4) had used an emergency ambulance than in the other subgroups (young adults, n = 2; parents, n = 1), and one had been taken to the ED by ambulance following a call to NHS 111. Concerns prompting contact varied widely across this sample, and included chest, abdominal or back pain (n = 4), sometimes as part of an ongoing problem; illness, injury or confusion of someone they had caring responsibility for (n = 3); ear, throat and eye problems (n = 5), sometimes as part of a longer-term period of ill-health and other times prompted by health anxiety; anxiety attack; and headache after an injury.

Overview of findings for people living in areas of social deprivation

There was no single dominant driver of interviewees’ decision-making about seeking health care. Instead, a number of key drivers were evident, including concerns about the severity of their symptoms, an unwillingness to further delay care-seeking, and their perceptions and experiences of services. The influences of others and of social isolation were also significant drivers for different participants in this subgroup, together with poor mental health and geographical proximity. Most interviewees in this sample had between two and five drivers, with only one interviewee having a single driver evident in their decision to seek care. With regard to making contact with a particular service, a lack of access (either perceived or real) or dissatisfaction with general practice were factors for all interviewees in their decision to contact either the emergency ambulance service or an ED. Conversely, positive experiences of general practice, or negative perceptions about urgent care services such as WICs, were the predominant factors in the decision to access general practice. The influence of others, either within a social network or health-care professionals, could also determine whether or not an interviewee sought help, or where help was sought from.

Links to programme theories

Evidence of all 10 programme theories was found in the interviewees’ accounts and, mirroring the findings discussed above, some were present more frequently and to varying degrees across the subgroups (see Appendix 6, Table 18). Reflecting the areas of commonality identified above, three programme theories featured strongly in all three groups: programme theory 1, uncertainty about symptoms causing anxiety, programme theory 8, following advice of trusted others, and programme theory 9, perceptions or experiences of services. Parents’ particular concerns regarding their responsibility for their child and lower threshold of concern for their child’s health are apparent in the much more frequent evidence in this subgroup of programme theory 3, fear of consequences when responsible for others, and programme theory 2, heightened awareness of risk as a result of experience or knowledge of traumatic health events leading to anxiety. By contrast, programme theory 6, waited long enough for things to improve, and programme theory 7, stressful lives and difficulty coping, are more apparent in the young adult and social deprivation subgroups, the latter perhaps reflecting the impact of mental health problems and social isolation. Young adults’ concerns about the impact of symptoms on their ability to function are evident in the more frequent occurrence in this subgroup of programme theory 4, inability to get on with daily life. The number and range of programme theories in many of interviewees’ accounts also highlights the complexity of individuals’ experiences. Only seven interviewees showed indications for fewer than four programme theories and 10 interviewees had evidence of six or seven of the theories.

Our findings add new insights to the existing programme theories, indicating areas where these need to be refined or extended to more accurately reflect the diversity of experiences in the data (Table 6). Interviewees’ accounts also identify further drivers of their decision-making that are not reflected in the existing programme theories, particularly in relation to the influence of the timing of the event and the proximity of services, which suggests that there is a need for additional programme theories.

Links to existing theories

A number of the existing theories and models explored in the WP1 review are evident in the interview findings. For all three subgroups, concern about the seriousness of the symptoms links clearly with Leventhal’s58 self-regulation model, which recognises the importance of both cognitive and emotional dimensions of the experience of illness. In relation to the cognitive dimension, all of Leventhal’s five domains of illness, namely identity (label/name), consequences (including impact on function), timeline (including duration), cause and controllability, can be seen to varying degrees in many aspects of interviewees’ accounts of their symptoms. Uncertainty about the meaning of symptoms, which was a frequent concern for interviewees, has been identified by Mishel and Braden59 as having a significant impact on help-seeking; they also identify the role of social support in coping with uncertainty. All three subgroups frequently identified the influence of others on their decision-making, and Pescosolido’s71 network episode model, recognising both the direct and the indirect impact of social networks, is evident in our findings. The role of recursivity95 and candidacy74 is also evident in this context.

For parents and others in a caring role, the sense of responsibility can be understood partly in relation to Beck’s70 work on the Risk Society, where blame is avoided by acting responsibly, including in relation to health. Dingwall and Murray75 also highlight how children are considered as a different category of patient, with an accepted lower threshold for help-seeking. For those experiencing a range of stresses, which is seen in both the young adults and the social deprivation groups, including poor mental health and social isolation, Antontovsky and Sagy’s89 work recognises how a lack of ‘generalised resistance resources’ can have an impact on people’s ability to cope with additional challenges, including illness. The overall complexity of drivers, including both individual and contextual factors, is recognised in Andersen’s33 model of health-care utilisation, with later iterations increasingly acknowledging many of the social and structural influences on decision-making evident in interviewees’ accounts.

Link to wider literature

The clear correspondences between our findings and the programme theories developed from the realist review inevitably mean that there is a significant overlap with the literature on which the programme theories are based. In particular, our findings support existing qualitative evidence that highlights the influence of uncertainty about symptoms, fear of consequences when responsible for others, need for immediate pain relief, the impact of stressful lives, the advice of others, and perceptions and experiences of services. More recently published work suggests further overlap. Although focusing on different population subgroups, Pope et al. ’s22 research found that perceived risk, anxiety and a need for reassurance were common motivations for accessing urgent care services. Similarly, Pope et al. 22 also found a heightened sense of risk associated with children and babies, corresponding to Kai’s117 work, published over 20 years ago, which describes parents’ experiences of managing young children’s acute illness. His findings identified parental worry about the nature of symptoms, the threat that these symptoms might pose to their child’s health and the responsibility that parents felt to keep their child safe, which mirror the accounts of parents in our study. Given that we found a number of qualitative articles focusing on children using GP out-of-hours services and paediatric EDs in our review, we reflected on our additional contribution to knowledge for this subgroup. We feel that we more clearly identified the difference between people’s behaviour with regard to their own health and their behaviour with regard to the health of their child, how problems with children’s ongoing health creates an anxiety and unwillingness to wait, how a parent’s own illness can have an impact on their coping capacity, and the influence of family on decision-making. In addition, in the literature there is a general perception or experience of not being able to access GP appointments or of a dissatisfaction with GPs, but our inclusion of people accessing same-day GP appointments shows the ease of access for some parents of young children.

The evidence from the realist review showed how a range of social and psychological stressors could have an impact on individuals’ help-seeking, reducing their coping capacity and increasing their use of emergency and urgent care. Although poor mental health was identified as one of these stressors, the findings from our interviews suggest that this specific issue was often more dominant and was frequently a significant driver of seeking care. Although interviewees’ overall physical and mental health was not explored routinely during the interviews, a number described mental health symptoms for which they were being treated or seeking treatment through their GP, and a very small number reported more severe mental health problems. It was clear that, even for those whose problems were less severe, their symptoms or associated reduced coping capacity were frequently a factor in their help-seeking, although the problem for which they were seeking help was frequently a physical one.

Recent research supports our findings, identifying that those with a mental health diagnosis (classified as mild, moderate or severe) are more likely to use EDs than those without a diagnosis, with more severe problems associated with more frequent visits. 118 Other studies from the USA and Australia identify increases in both the absolute numbers and the proportion of ED use that were linked to mental health problems,119–121 particularly in adolescents/young adults121–123 and those with lower incomes. 119 Although most of this use is for problems requiring urgent attention, Hsia and Niedzwiecki124 also identified that 6.8% of all ‘avoidable visits’ to US EDs were for mental health conditions, and that mood disorders resulted in the highest proportion of avoidable visits by diagnostic grouping. In relation to the emergency ambulance service, two recent studies in the UK125 and Australia126 identified that approximately 10% of calls are related to mental health problems. Both studies emphasise the need for alternative solutions, particularly more community-based provision, transportation level of < 50% for those with less severe problems126 and high levels of re-attendance. 125

Mental health problems have also been found to have an impact on levels of primary care use. Lockett et al. 127 found that people with anxiety, depression or bipolar disorder were more likely to use primary care in New Zealand but were also more likely to report having unmet needs and a less positive experience with their GP. Even among those without a diagnosed mental health problem, self-perceived stress has been found to increase use of primary care, with higher stress levels associated with greater numbers of people having visited their GP in the past year in a Danish population study. 128 Notably, stress was also associated with greater use of out-of-hours services and, among those with multimorbidity, with decreased use of chronic care services. This reflects the findings from our interviews, suggesting that poor mental health can have am impact on help-seeking behaviour, even when not at a clinically recognised level.

Convenience has been identified as a driver of navigating emergency and urgent care services. 22,28 For example, an ED may be perceived as offering more availability or accessibility. Although there was some evidence of this in our study, convenience did not appear to play a significant part in decision-making among our participants. It is possible that interviewees did not want to appear to be inappropriate users of health care, but the researchers also perceived that convenience did not generally seem to be an important consideration. A far stronger driver of accessing ED services in our sample was the perceived lack of access to primary care services. This is widely supported in the existing research literature, has been shared in a patient perspective editorial34 and was evident in the recent NHS GP Patient Survey,129 in which around one in five patients (21.6%) reported that they had tried to contact an NHS service in the previous 12 months when they had wanted to see a GP but could not because their general practice was closed.

Strengths and limitations

A key strength of this part of the study is the collection of in-depth qualitative data eliciting drivers of decision-making among three population subgroups that were identified as having a tendency for ‘clinically unnecessary’ use of services. The inclusion in the study of three service settings, in particular same-day GP appointments, also extends the existing literature in this area, which has until now focused largely on ED use, with limited material relating to use of emergency ambulance and GP services. The process of data analysis, particularly the use of the principles of IPA, helped us to identify the complexity of the decision-making process. The findings highlight how, rather than being prompted by one single driver, an individual’s decision both to seek care and to use a particular service to obtain it is influenced by the interaction of multiple drivers. Providing insights into this complexity is vital in understanding and addressing ‘clinically unnecessary’ use.

We encountered a number of difficulties with recruitment that had an impact on the composition of the final sample. Most significantly, despite extending recruitment by a number of months, we recruited considerably fewer participants from the ambulance service than originally planned. Study set-up meetings led us to feel confident that we could recruit our relatively small sample from the large pool of patients calling the ambulance service and receiving telephone advice only. Initial engagement by staff at the ambulance services was also encouraging, but lessened in one service over time. Staff reported that pressures of demand on the service (we attempted to recruit between the months of November 2017 and June 2018) were a barrier to recruitment, and the lack of a researcher present regularly on site is also likely to have affected engagement with the study. To address the lower levels of recruitment via this route, we also recruited a few patients at the ED who had been conveyed by ambulance but whose transportation had been considered unnecessary by ED staff. We also encountered difficulties in recruiting in both the ED and the GP setting in Wolverhampton and, because of time pressure, had to recruit more than half of the overall sample in Sheffield. As a result, it has not been possible to make comparisons between two operationally different emergency and urgent care systems.

The study recruitment processes meant that the majority of participants were recruited between 09:00 and 18:00, with recruitment from only the emergency ambulance service occurring 24/7 (i.e. 24 hours per day, 7 days per week). An attempt to recruit at the ED during the weekend was not successful, and we did not attempt to recruit from this setting overnight. Therefore, there may be some differences between our interviewees and those using the services at other times of the day. For example, the lack of availability of other services or a reduction in social support at night may have an impact on the levels of anxiety that drive help-seeking.

Participants in the qualitative research were not aware that they had been included in the study because they had been identified as ‘clinically unnecessary’ users of a health service; our interest was in the decision-making that led to this contact. This approach was recommended by PPI members and approved by the ethics committee. To have been explicit about this with participants at the recruitment phase may have had an impact on the study in two ways. First, it may have had a negative impact on recruitment if potential participants felt that they had been perceived as using health services inappropriately. Second, if recruited participants felt that they were being labelled as ‘clinically unnecessary’ users, this may have affected their openness during the interviews, and, subsequently, the integrity of the data. The presence of the interviewer and the topic being discussed may have affected interviewees’ responses, prompting them to offer more socially acceptable reasons for their decision-making to avoid being judged. This may have been particularly evident for parents wanting to present themselves as ‘good parents’ who were doing the ‘right thing’ for their child. Although this cannot be ruled out, the researchers did not feel that it was an issue during the interviews.

Interviewees were recruited for their membership of a particular subgroup, but a significant number also belonged to one or even both of the other subgroups. For instance, eight young adults were also living in areas of deprivation, four of whom were parents. This means that there is some overlap between the subgroups. However, analysis was conducted both on those recruited to a particular subgroup and on those who had the characteristic of that subgroup, and this was not found to significantly influence the findings.

Implications

The programme theories identified in WP1 were clearly apparent in the three subgroups, and we refined some programme theories based on the interviews reported in this chapter. The drivers were largely similar across the subgroups but there were differences between subgroups in the strength of the drivers. This implies that potential interventions may need to be different for population subgroups.

Changes in service provision

The group viewed improvements to GP provision as key, which was also echoed in the interview data. Better availability of GP appointments was a key priority, with interview participants identifying a number of specific suggestions, including more weekend, early morning and evening appointments, quicker access to appointments for non-urgent problems (i.e. within a few days rather than a number of weeks) and the ability to book appointments further ahead (e.g. 2–3 months). Focus group parents also considered that there should be prioritisation of urgent appointments for children, a point echoed by interview participants, who variously suggested specific appointments, telephone calls or drop-in sessions for children. Six-monthly GP appointments for children were also suggested, providing regular opportunities to discuss health concerns, gain knowledge and reassurance, and build a helpful relationship with an expert in child health.

In relation to accessing appointments, the group described diverse experiences, with some reporting that their GP ‘always sees my children on that day’ [focus group (FG) P3] and others describing the frustration of repeatedly trying to book an appointment:

They tell you to call at 8:30 for my GP. You call, by the time it’s 9 o’clock they’ll be like ‘Oh sorry, we’re out of . . . appointments’ and then, they’re all ‘Call in tomorrow’ you do the same thing for the whole week and then I’m like ‘Oh yeah . . . can I book it for next week?’ and they’ll be like ‘Oh no, there’s no appointments, call in on the day’. It’s just always ‘Call in on the day’ [agreement from group] and you never find one.

P4, parent, FG

Interview participants echoed these frustrations, which were heightened by the fact that the phone-in time was often when they were taking children to school. Suggestions for improvement included an online or app-based booking system for same-day GP appointments and/or call-back requests, and the ability to book an urgent appointment for the next day rather than have to telephone repeatedly each morning.

Recognising that parents’ sense of urgency, combined with the lack of access to GP appointments, may prompt them to use the ED, locating a GP service at the ED was recommended, allowing non-urgent problems to be dealt with via another route, which would also reduce demand on local GPs. The specialist skills available in the ED were also seen to be a significant driver of parents’ decision-making, particularly in the context of there being a children’s hospital in the city, with one parent observing that ‘if your kids are seen by a paediatrician, a specialist (. . .) I feel more you know, more safe’ (P1 FG). Therefore, the group recommended introducing specialist GPs for children, based either in a general practice or in a WIC, suggesting that parents’ greater confidence in the skills of these professionals could reduce their perceived need to use the ED. This suggestion was also made by an interview participant who did not take part in the focus group, and this seems to have been based on their experience in their country of origin. Interview participants also highlighted the need for better information about how to access urgent GP appointments out of hours or when no same-day GP appointments are available, together with increasing out-of-hours capacity to avoid attending an ED because of lack of GP appointments.

Focus group parents identified a number of other service changes that could potentially reduce the pressure that GPs were under. An increased role for pharmacists was suggested, including offering follow-up appointments. Some parent-specific changes to the WIC were also identified, many of which focused on minimising the length of the visit when bringing young children for treatment. These included having a priority system for children, having a separate WIC for children, reducing waiting times and introducing an appointment system to avoid the need to sit in a waiting room with an ill child for long periods. The location of the current WIC was seen as problematic, with the group suggesting that a more child-friendly location, and the introduction of local centres rather than one central service, could encourage greater use. Interview participants also recommended an increase in the number of WICs, enabling people to drop in and guarantee being seen at a convenient location.

More broadly, the group saw increased support, particularly for first-time parents, as essential, including more proactive engagement by health visitors and other community services, and offering parenting lessons in schools. The group also felt that there was a need to improve patient experience across the services and highlighted that a specific priority should be to offer language assistance at other services apart from the ED, as this could be a significant determinant of where to seek help:

. . . I don’t think they had anybody that would translate at certain GPs but in the A&E there’s always somebody (. . .) ‘cause there’s Somali doctors and there’s so many different doctors and nurses, so I think that’s why A&E’s more helpful.

P4, parent, FG

Education/awareness

Improving publicity about the alternatives to emergency and urgent care was seen as important, including promoting the GP out-of-hours service; encouraging the use of NHS 111 and emphasising its value in increasing parents’ health knowledge; and highlighting the knowledge, training and skills of pharmacists and their role in offering advice:

Explain what advice and what they [pharmacists] can actually offer advice on. Do they know about rashes? Do they know about different temperatures? Do they know about the actual symptoms of what my child is getting? Or do they – I don’t really know what they can fully offer, apart from just general advice on ‘Do I use this medicine or do I use this medicine?’

P3, parent, FG

Promoting the services offered by family centres was also seen as important in facilitating access to additional resources to help parents gain the skills and confidence to manage their children’s health.

Changes in service provision

The key service priority identified by this group was the need to significantly improve mental health provision, a concern that may have been driven by at least half of the group having experienced significant mental health problems, as well as the prevalence of such issues in their friendship groups. Improvements in the range, quality and speed of access to mental health services were seen as important aspects of this, with the current gaps in provision leaving people with no option but to use other services, particularly at times of crisis. Lack of appropriate mental health support was also seen to have an impact on decision-making about physical illness, making them likely either to inappropriately delay seeking help or to ‘panic and try to grab out for every service’ (FG P12). In this context, the group suggested that a specific mental health ED should be introduced to provide urgent specialist support and advice, significantly improving patients’ experience and reducing pressure on the main ED (which patients are currently advised to use in crisis):

. . . it doesn’t seem like maybe the best use of the time for the people in A&E because they can’t really do anything really that practical to help, but also the person who’s feeling that way to be sat in there for how many hours to just go home again at the end of it.

P10, young adult, FG

Better support for mental health distress was also highlighted by interview participants, who emphasised the need to improve the quality and speed of access to mental health services, and how the current lack of provision contributed to increased use of the ED. The need for a more sympathetic response from clinical and reception general practice staff over the telephone was also suggested, together with clearer information on how to access immediate support.

Reflecting the concerns raised by parents, easier access to GP appointments was highlighted as important in encouraging people to reduce their use of the ED, with more availability of appointments outside working hours (i.e. early mornings, evenings and weekends) being a particular issue:

I think my GP does one day, where it’s like a late opening I think. But that’s no good if they’re all booked up, so then it’s like you’re going back into the week cycle aren’t you, waiting for the next appointment.

P13, young adult, FG

Interview participants also focused on the need for improvements to GP provision, including having extended opening hours, more same-day appointments and shorter waiting times for non-urgent appointments, introducing drop-in sessions and using a triage system to prioritise urgent cases. A text-reminder system for medication reviews was seen as valuable in preventing urgent appointments.

A number of uses of online technology were proposed by this focus group. One GP-related initiative was the establishment of an online database of GP appointments available across the city, which would allow people to travel to another practice to access a convenient appointment:

(. . .) particularly if younger people are more able to travel around that if they could somehow make, sharing of notes a little bit easier, and maybe people could, opt to travel to get an appointment, rather than going to their same GP and also it prevents, empty appointments going wasted [agreement].

P12, young adult, FG

Other proposed online interventions included improving the accessibility of GP services by using Skype^TM (Microsoft Corporation, Redmond, WA, USA) or online GP consultations, although not all of the interviewees favoured this option. Views were also mixed on other internet-based consultation services, particularly in terms of potential health and privacy/confidentiality risks. However, extending NHS 111, a service known to be credible, to include an online chat facility was seen as having the potential to increase use of this service, particularly among younger people, although the detailed questioning process was identified as a barrier by some.

Young adults taking part in interviews also made suggestions relating to other services, although these were a small proportion of the total. As in the parents group, increasing the number of local WICs was seen as important in improving access to this service. Additionally, co-locating GP and ED provision was recommended, which would enable people to be readily transferred to the service most appropriate to meet their needs.

Education/awareness

The importance of education and awareness raising was a key focus in this group, with significantly more recommendations relating to this rather than to changes in service provision. There was a strong feeling that many people, including young adults, were unaware of the services available and what these can offer, thus making appropriate decision-making difficult. In particular, the group considered that raising awareness of the range and scope of alternative services, including NHS 111, MIUs, WICs and pharmacies, was a key priority:

I think a lot of the problem is people don’t know these things exist, or don’t know what they’re for, and so they know A&E exists, they know something’s wrong, they just go to A&E because it’s probably fine. Some people I think are under the impression that you can’t walk in to A&E, that you have to be ambulanced in so the only way to get to A&E is by ambulance. And I just feel like there needs to be more education and publicity and, ‘These things exist, here’s how you can access them [pauses] there you go’.

P10, young adult, FG

Promoting NHS 111 on social media, in television adverts and by other routes was recommended, including gaining celebrity endorsement. Interview participants also echoed the need for greater publicity of NHS 111. Those in the focus group emphasised the importance of placing publicity in non-clinical locations, noting that ‘most posters about medical services are in the hospital when you are already waiting for one service’ (P15 FG), at which point it is too late to inform people. Specifically, they recommended promoting MIUs as an alternative to the ED, focusing on locations where people are more likely to need this service such as gyms or sports facilities. Increasing awareness of the skills and knowledge of pharmacists, and promoting their role in providing advice, was also highlighted as important in potentially reducing GP use in particular. Specifically, the group recommended that GPs and other practice staff advise patients whose needs could have been addressed by pharmacists about this option for future consultations.

Considering the issue on a wider level, the group felt that there was a need for education around what constitutes an emergency or urgent health problem, as this lack of understanding obviously has an impact on people’s decision-making. The importance of beginning the process of education in schools was emphasised, in terms of informing children about what should be considered an emergency and about the appropriate use of different health services:

I think like the more we’ve talked about it the more like educating in school seem the most plausible [agreement] because we keep going, ‘Oh, more publicity, more publicity’ there’s only so much money in the NHS anyway and it’s, you can’t just throw it out for every single service all the time can you?

It’s just starting people young isn’t it? [agreement] getting them educated.

An app or another online resource to help decision-making was also seen as a key tool to support change in service use, helping people decide on when and how to use different services:

(. . .) mainly directing to the right kind of place. Like if there could be some sort of red flag system of like, ‘Oh you’ve got that symptoms, that’s serious’. You know for people who’ve got a cold and like automatically worry that they’ve maybe got meningitis or something, like some sort of, ‘Have you got any of these three symptoms that would indicate meningitis?’ [agreement] OK, you do? Then yes, you do need a GP appointment. Yes, you do need to go to A&E.

P10, young adult, FG

Addressing the problem of inappropriate use, the group suggested raising awareness of the cost of service use and missed appointments, and explored the issue of penalties for misuse, although they acknowledged the difficulties that prevented this being a viable option. Finally, encouraging people to take more responsibility for their own health was emphasised, as this would ultimately lead to less service use overall. Returning to the concern around mental health, the group felt that improving mental health literacy was a priority, enabling people to identify and address their symptoms before they need urgent or emergency help.

Service provision

In common with the other subgroups, accessibility and availability of GP appointments was highlighted as a key priority in changing decision-making. In relation to access, a number of participants reported difficulties in either getting through to their practice to make an appointment or being able to access a timely appointment once they did get through to their practice, leaving them feeling that they had no option but to use another service:

You either go there or you wait to phone your GP then you don’t get through so in the end you go to A&E. Then you’re getting worse so you might as well just go.

P9, social deprivation, FG

Reflecting the concerns raised by young adults, both the focus group and the interview participants emphasised the need for more appointments outside normal office hours and greater availability of appointments within a few days of making contact, rather than the current dichotomy of same-day appointments or waiting several weeks. Summing up a problem experienced by many, one interview participant highlighted how this dichotomy prompted her to leave things until they became urgent rather than book an appointment for some weeks ahead, by which time the problem may have resolved. Interview participants focused almost exclusively on issues relating to general practice and made a wide range of suggestions for improvement, including having drop-in appointments for children; reviewing the current same-day appointment booking system, including offering the option to book a next-day urgent appointment rather than having to telephone each day; being able to book appointments in person rather than over the telephone, which was important for those not confident about speaking English; having greater continuity of care from an individual GP; being able to speak directly to a doctor about a problem rather than communication being mediated by reception staff; having priority appointments for elderly people; and removing the triage system so that appointments could be allocated on request rather than being based on potentially inaccurate information.

One recommendation made in the focus group to reduce the use of GPs was increasing the prescribing role of pharmacists, which could particularly have an impact on those who obtain their prescriptions free of charge:

. . . it sounds really mercenary but some of time, because I get my prescriptions free, I’m over 60, I think I’m going to pharmacist but then, whatever they prescribe you, you’ve got to pay for. And you can go to your doctor and you can get on prescription and I know that sounds really awful (. . .) but that is one of things that goes through me mind (. . .) I’m sure it would be for a lot of these people who are, if they’re on benefits they get free prescriptions anyway don’t they but I certainly think that is a stumbling block for some people, ‘Why should I pay when I can get it free?’

P8, social deprivation, FG

Improvements to the consultation space in pharmacies were also seen as essential in this context, with privacy being a particular concern.

Although NHS 111 was valued, and promoting it more widely was seen as a priority, changes to improve the NHS 111 service were considered a prerequisite. In particular, echoing concerns raised by young adults, the group felt that there was a need to streamline the assessment process, with staff able to access patients’ clinical records so that they could avoid the lengthy questioning that currently discouraged people from using the service:

I think where it’s fallen down is, there’s a boring, protracted rigmarole whereby you’ve got to speak to somebody for 10 minutes and give your details and your favourite colour and everything before you can actually get to the point of the call. If they, you know, streamlined that service and made it more efficient then I think that would be a viable option.

P7, social deprivation, FG

Similarly, although increasing the number of WICs was recommended, there was a perception that the skill and specialism of practitioners based in them needed to be improved if people were to have sufficient confidence in this service for it to become an alternative to the ED. Combining the accessibility of WICs with the quality of GP services was also identified by interview participants.

Perhaps reflecting poorer availability of their own transport, access was a key concern in this group, potentially driving the recommendation to co-locate a wide range of services within a number of local centres, such as the ED, WICs, MIUs, general practices and baby clinics.

. . . instead of just having one central minor injuries, whatever, A&E, you know, Sheffield’s a big city, people have to travel from really far places, you know, so we could have something as (name) said east, west, north and south (. . .) So it would be really good to have something like that around the city where you have so many services at one building.

P9, social deprivation, FG

However, the group also highlighted the widespread confusion about the range and scope of different services available, and the consequent difficulty of making an informed decision. Rather than promoting this diversity, one suggestion was to significantly simplify provision to offer only the familiar, more clearly delineated services of the GP and ED and thereby reduce the complexity of decision-making.

Education and awareness

As noted in the previous section, the group considered it a priority to address the widespread confusion about when and how different services should be used, in particular by raising awareness of what each health service can offer, including the type and severity of conditions it can deal with, and when and how it can be accessed. In particular, the need to raise awareness of the NHS 111 service was emphasised to enable it to be promoted as an alternative source of advice when a GP appointment could not be accessed, with one participant observing:

. . . there’s only four of us and we’ve all got individual ideas. Like to me it’s just an out-of-hours service. I wouldn’t dream of calling it [NHS 111] in the day, if your GP’s open (. . .) We’ve all got such different, it’s just amazing for such a small group (. . .) So if four of us have got our own opinions, you can imagine what 40 would have. So it certainly needs something to get people a lot more aware of.

P8, social deprivation, FG

Promoting the role of pharmacies was also mentioned. Echoing comments made by the young adults’ group, the need to raise awareness about service use in non-clinical locations was highlighted, rather than this being done at a point when individuals had already made a decision about where to seek care:

. . . it seems like closing the stable door after the horse has bolted because you’re already there (. . .) but also a lot of people are worried when they’re in waiting rooms and I don’t think it’s necessarily the best time to be reading things. I mean you’re not at your most perceptive, when you’re about to go into an appointment that you’re concerned about, so I think that more needs to be done to get people before they’re in that situation.

P7, social deprivation, FG

Concern and frustration were also expressed about the misuse of services, prompting the group to recommend greater education to encourage people to take responsibility for managing their health, and to also consider applying penalties to ‘repeat offenders’, although the highly problematic nature of this was also acknowledged.

Sociodemographic characteristics

NatCen collect a range of characteristics of respondents within the British Social Attitudes Survey. We considered the effects of:

• sociodemographics – age, sex, ethnicity, region, household with children aged < 5 years, deprivation as measured using the IMD, social class and rurality

• resources – car ownership and access to internet

• health – long-term limiting illness and general health

Deprivation and rurality were not asked about in the questionnaire but were added to our data set by NatCen from background information held for sampling purposes.

Vignettes

Vignettes used in emergency and urgent care population surveys have varied in terms of whether they focus on a third party or ask an individual to say what they would do, the amount of detail in the vignettes, and the potential seriousness of the problems addressed within them. 6,134,135 For example, some focus on chest pain and finding a lump,135 whereas others focus on less serious issues of an ankle sprain/bruise on leg after a fall and a child with a cold. 6

In the proposal we said that we would construct three vignettes: one in which care is needed immediately from an urgent care service (e.g. NHS 111, GP in hours same day, GP out of hours, WIC), one in which care is needed from a GP within a few days for a non-urgent problem and one in which self-care could be undertaken. In practice we constructed three pairs of vignettes, with a half of the sample receiving the same three vignettes. This allowed us to assess the effect of different symptoms and different times at which events occurred. We constructed these vignettes with clinical input from Steve Goodacre, who is an academic ED consultant, and Jon Dickson, who is an academic GP. Our PPI members and attendees at our large PPI event commented on the evolving versions of these vignettes.

We focused the vignettes on minor health problems. We identified a pair of illness vignettes with an expected action of self-care or pharmacy. We identified use of an emergency ambulance, ED or GP as ‘clinically unnecessary’:

• Illness 1: imagine you have had a cough and sore throat for 3 days.

• Illness 2: imagine you have had diarrhoea and vomiting for 2 days.

We identified a pair of injury vignettes where we expected respondents to contact a GP, contact an urgent care service such as an urgent care centre or a MIU, self-care or use a pharmacy. We identified use of an emergency ambulance or ED as ‘clinically unnecessary’:

• Injury 1: imagine you have fallen and have a very painful rib. It is 8.30 in the evening.

• Injury 2: imagine you have had back pain for 2 weeks and have not been able to sleep.

We identified a pair of child vignettes where we expected respondents to contact a GP in or out of hours, contact an urgent care service such as a WIC, self-care or use a pharmacy. We identified use of an emergency ambulance or ED as ‘clinically unnecessary’:

• Child 1: imagine your young child or a young child in your care has a high temperature and cried throughout yesterday and last night. Today, which is a Saturday, you do not think the child has improved.

• Child 2: imagine your young child or a young child in your care has a high temperature and cried throughout yesterday and last night. Today, which is a Wednesday, you do not think the child has improved.

We explored these symptoms on the NHS Choices website and this validated our conclusions about the clinically recommended actions to take. In the case of all of these symptoms, NHS Choices confirmed they are unlikely to be serious and that self-care is usually adequate. We show summaries of the advice offered by NHS Choices in Appendix 8. The response options offered for each vignette are shown in the questionnaire available on the project web page (www.journalslibrary.nihr.ac.uk/programmes/hsdr/1513612/#/; accessed 5 February 2020). In the proposal we said that if a respondent selected an emergency service we would ask why. However, attendees at our wider PPI event and NatCen expressed concern about this because it might have alerted respondents to what was considered to be a right or wrong answer.

Ten programme theories from work package 1 realist review

We included questions to measure aspects of the 10 programme theories identified in WP1. We struggled to construct questions that measured some of the programme theories (see Results, Description of programme theories).

Other influences

Some issues in the literature did not become programme theories but we felt that they were important enough to be included in the questionnaire:

• Awareness of services because it appeared in articles in the review.

• Recursivity, in that patients learn to attend services because services have sanctioned previous decisions,68 because it is a relatively new issue identified in emergency care literature.

• Health literacy because it is a relatively new issue identified in the emergency care literature. In the proposal we highlighted that researchers in the USA were studying the effect of low health literacy on ED use and developing interventions to address this. 136 Although health literacy is associated with social class and educational attainment, 29% of people attending EDs for non-urgent issues have a college education and low health literacy. 137 We explored the use of two validated instruments to assess health literacy: Newest Vital Sign (NVS) and HLQ. We had originally intended to use the NVS in the first pilot because Duell et al. 138 undertook a systematic review of measuring health literacy in a clinical environment and found that using the NVS was the most acceptable approach. We identified an alternative because of concerns expressed by NatCen about NVS. The HLQ was developed and validated in Australia139 and consists of nine domains of health literacy. We selected two five-item domains most relevant to our study: domain 6, ‘ability to actively engage with health-care providers’ (we labelled this ABILITY), and domain 9, ‘understand health information well enough to know what to do’ (we labelled this UNDERSTAND). Its strengths were that it addressed two aspects of health literacy and was preferred over the NVS at the PPI event, and the attendees at the wider PPI event found it very easy to complete.

We also wanted to take account of the following:

• Recent use of health care because people recently using a service may be more likely to base their answers on experience. We used the question from the General Practitioner Patient Survey 2017140 and adapted it to ask about emergency ambulance and ED use as well as general practice.

• Beliefs about overuse of services because we were interested in whether or not the population shared views about ‘clinically unnecessary’ use of services.

Weighting

NatCen uses a robust sampling and weighting method, with weights applied for unequal selection probabilities and non-response and then calibration weights. Therefore, it enables analysts to make generalisations to the population at large. NatCen produced separate weights for interviewer-administered questions and self-completed questions because of differential response rates:

• We used both unweighted and weighted data when presenting descriptions of variables and simple bivariate analyses.

• We used unweighted data in the logistic regressions because we could not undertake this type of regression in SPSS (IBM Corporation, Armonk, NY, USA) or Stata (Stata Press, College Station, TX, USA) using weights.

Description of attitudes and tendency

We summarised the responses to questions to display the prevalence of population attitudes. This included a summary of responses to each pair of vignettes, comparing responses to show how different characteristics, such as symptoms, age and day of the week, affected tendency.

Differences by subgroups

We compared items related to the 10 programme theories by different sociodemographic characteristics using chi-squared tests. We limited this testing to variables related to the three subgroups from WP2 (young adults, parents of young children, deprivation levels) and a subgroup identified commonly in our programme theories in WP1 (presence of chronic conditions).

Explaining the tendency to make ‘clinically unnecessary’ use of services

We considered fitting a multinomial model comparing ambulance, ED and GP use with pharmacy/self-care, but were concerned about the difficulty of interpreting the results of multinomial analyses. Instead, we undertook a logistic regression on a series of binary comparisons. We considered the tendency to make ‘clinically unnecessary’ use of six health services: emergency ambulance, ED adults, ED children, ED all, GP and any service. We divided ED into adult and child because of the radically different answers to the child and adult vignettes. We wanted to do this for the ambulance analysis also, but the numbers were too small. We created binary variables based on whether or not the respondents chose ‘clinically unnecessary’ options in the vignettes. In the proposal, we planned to measure the proportion of people selecting a higher or lower level of care recommended for each scenario. In practice we selected only a higher level of care because of our focus on ‘clinically unnecessary’ demand.

We undertook a logistic regression comparing those ticking ‘clinically unnecessary’ options for a vignette with those who did not. We used SPSS version 25. First, we conducted a univariate analysis on each of the binary tendency outcomes, testing 54 independent variables. We then tested only the significant independent variables in a complete-case multivariable logistic regression using backwards elimination with a cut-off point of 0.05 for selection. We calculated the number of events per variable: the number of events divided by the number of degrees of freedom required to represent all of the variables in the model. The events per variable need to be at least 10 and preferably 20 for reliable parameter estimation. If fewer than this, there is an increased risk of overfitting of the models. Backwards elimination has advantages over forward selection when variables are correlated. 141 We collapsed some categories of some variables for the regression where numbers were small. We present odds ratios and 95% confidence intervals (CIs).

Once we had our final multivariable regressions we tested for multicolinearity by calculating generalised variance inflation factors using the statistical package R (The R Foundation for Statistical Computing, Vienna, Austria). We could not use SPSS because this software calculates variance inflation factors for linear regressions only. The variance inflation factors for each of the six regressions ranged from 1 to 1.7, well within the limit of < 5, indicating that multicolinearity was not at play.

We took the final multivariate regressions and tested whether or not the results depended on which vignettes were completed. We tested interactions between factors in each final regression and the two samples that completed different vignettes (cough/rib/Saturday and vomiting/back/Wednesday).

Missing data

As the number of missing data was small (apart from the missing values for self completion questions), missing data were treated as missing and no methods of imputation were used. However, in some cases, missing data were recoded as a category ‘missing’ to increase the sample size in the analysis.

Tendency to call the emergency ambulance service (adult or child)

A total of 11% (328/2906) of respondents ticked ‘call an ambulance’ for any of the six vignettes. In a univariate analysis respondents had a higher tendency to call an ambulance if any of the following applied (see Appendix 9, Table 20):

• sociodemographic characteristics – male, BAME

• resources – social classes III manual/IV/V, most deprived communities, no personal access to the internet, no car

• health – in fair/poor health, with limiting long-term illness

• programme theories – worry if in pain (programme theory 1), if losing sleep or loss of daily activities (programme theory 4), if overwhelmed when faced with health problem (programme theory 7), do not have someone to care for them when they are ill (programme theory 7), find it difficult to travel to the ED (programme theory 7), prefer service with no appointments (programme theory 7), prefer the ED for getting tests done quickly (programme theory 9), think ED doctors know more than GPs (programme theory 9) and have no confidence in their GP (programme theory 9)

• awareness of services – not confident about finding out when NHS services are open, unlikely to look up on the internet what a problem is or what to do about it

• recursivity – feeling that if tests are undertaken they were right to contact a service

• health literacy – low levels of understanding information and ability to engage with health professionals

• recent service use – recent use of ambulance, frequent use of ambulance, frequent use of ED

• attitudes to overuse of services – disagree that too many people use EDs and GPs when not needed.

There were some unusual findings:

• People who did not have work or family responsibilities that made seeing their GP difficult had a higher tendency to call an ambulance.

• People who disagreed that people use an ED because they cannot get a GP appointment were more likely to call an ambulance.

Based on a complete-case analysis of 2203 out of 2906 (76%) of the data set, the full model containing all of the statistically significant variables identified above is given in Appendix 9, Table 21. The events per variable was 8.7 (235/27), showing that there was some risk of overfitting within this model. The following variables made an independent contribution in a multivariable analysis, with the following groups having a higher tendency to contact an emergency ambulance (see Appendix 9, Table 21):

• sociodemographic characteristics – male, BAME, social class III manual

• resources – households with no car, no personal access to the internet

• programme theories – programme theory 1, people who worry that a pain is sign of something serious, and programme theory 7, people who feel overwhelmed when they have an unexpected health problem. Findings for programme theory 8 and programme theory 10 were not straightforward. People who were fairly likely to check with family and friends had a lower tendency than those who were not likely to do this; this finding was not similar for those ticking very likely, as might be expected. Programme theory 10, work or looking after family make it difficult to see a GP, was the opposite of what might be expected, in that people who did not face this problem had a higher tendency to contact the ambulance service

• health literacy – people with low understanding of information

• recent service use – people who have used an emergency ambulance, particularly in the past year.

All six vignettes were included here, but the tendency to call an ambulance was dominated by the child vignettes in that around three-quarters of those with this tendency had ticked a child vignette. There were no statistically significant interactions between the samples completing the cough/rib/Saturday and the samples completing the vomiting/back pain/Wednesday vignettes.

In summary, the key findings were:

• The two child vignettes dominated the tendency to call an ambulance.

• A large number of variables related to people having low resources, including social class, low health literacy and feeling overwhelmed when faced with a health problem.

• Lack of a car featured here, indicating that an ambulance may act as transportation to an ED for people with low resources.

• People who had used the ambulance recently had a higher tendency to select it in the vignettes. This could be explained by someone having an ongoing health problem or by learnt behaviour in that having used it once, they feel happier to use it again.

Tendency to attend an emergency department for an adult

In total, 18% (512/2906) of respondents ticked ‘go to an ED’ for any of the four adult vignettes. In a univariate analysis respondents had a higher tendency to go to an ED for adults if any of the following applied (see Appendix 9, Table 22):

• sociodemographic characteristics – male, BAME, social classes III manual/IV/V, most deprived quintile, urban areas

• resources – none

• health – none

• programme theories – very likely to worry that pain was a sign of something serious (programme theory 1), lost functioning of daily living such as work or sleep (programme theory 4), overwhelmed when they have health problem (programme theory 7), very likely to check with family or friends if they are ill (programme theory 8), prefer EDs because can get tests done quickly (programme theory 9) and think ED doctors know more than GPs (programme theory 9)

• awareness of services – none

• recursivity – if tests are done they were right to go to a service

• health literacy – none

• recent service use – recent and frequent use of ED

• overuse of services – think that too many people use ambulance and ED.

Based on a complete-case analysis of 2309 out of 2906 (80%) of the data set, the full model containing all of the statistically significant variables identified above is given in Appendix 9, Table 23. The events per variable was 32.4 (389/12), showing that there was little risk of overfitting for this model. The following variables made an independent contribution in a multivariable analysis, with the following groups more likely to have a tendency to contact an ED for an adult (see Appendix 9, Table 23):

• sociodemographic characteristics – male, BAME

• programme theories – loss of functioning, specifically work (programme theory 4), prefer EDs because can get tests done quickly (programme theory 9)

• recursivity – think that if tests are done they were right to go to a service

• beliefs about overuse – if people do not think there is a problem with too much use of EDs and do not think people are reluctant to use an ED.

Four vignettes were included here, but the tendency to attend an ED was dominated by the painful rib vignettes in that around two-thirds of those with a tendency had ticked the rib vignette. There was only one statistically significant interaction between the samples completing the cough/rib/Saturday and those completing the vomiting/back pain/Wednesday vignettes in the final model. This occurred with ethnic group (p = 0.012). People from BAME communities were more likely to tick ‘clinically unnecessary’ options for the cough/rib/Saturday vignettes than white people. White people were more likely to tick ‘clinically unnecessary’ options for the cough/rib/Saturday vignettes than for the diarrhoea/back/Wednesday vignettes. People from BAME communities were slightly less likely to tick ‘clinically unnecessary’ options for the diarrhoea/back/Wednesday vignettes than white people were to tick ‘clinically unnecessary’ options for the cough/rib/Saturday vignettes. Interactions should be interpreted with caution.

When the binary comparison was limited to those ticking attending an ED compared with those selecting lower levels of care only, the results were very similar. All of the variables in the original model featured in this model except programme theory 4, loss of functioning. There were some additional variations by geographical region (no specific region was statistically significantly different): programme theory 1, pain is a sign of something serious, and that too many people call 999 when they do not need to.

In summary, the key findings were:

• There was evidence of the influence of preference for EDs, getting positive feedback about the validity of contacting a service if tests are undertaken by that service and beliefs that EDs are not overused.

Tendency to attend an emergency department for a child

In total, 33% (949/2906) of respondents ticked ‘go to an ED’ for any of the two child vignettes. In a univariate analysis respondents had a higher tendency to an ED for a child if any of the following applied (see Appendix 9, Table 24):

• sociodemographic characteristics – aged 55–74 years, male, BAME, Wales and Scotland

• resources – none

• health – none

• programme theories – previous traumatic event (programme theory 2), sleep loss (programme theory 4), loss of work (programme theory 4), not very likely to take medication (programme theory 5), prefer no appointments (programme theory 7), willing to wait in the waiting room if it means being seen that day (programme theory 7) and prefer an ED because get tests more quickly (programme theory 9)

• awareness of services – none

• recursivity – tests are carried out then think it was right to contact the service

• recent service use – used ED recently

• beliefs about overuse of services – believe too many people use ED when not needed.

There were some unusual findings:

• Households with children aged < 5 years had a lower tendency to go to an ED for children.

• In the responses to look up on the internet to find out what to do, the ‘missing’ category is statistically significant.

Based on a complete-case analysis of 2304 out of 2906 (79%) of the data set, the full model containing all of the statistically significant variables identified above is given in Appendix 9, Table 25. The events per variable was 44.4 (754/17), showing that there was little risk of overfitting for this model. The following variables made an independent contribution in a multivariable analysis, with the following groups more likely to have a tendency to contact an ED for a child (see Appendix 9, Table 25):

• sociodemographic characteristics – aged 45–74 years, male, BAME, Wales and Scotland

• programme theories – willing to wait in the waiting room if it means being seen that day (programme theory 7), prefer ED because get tests more quickly (programme theory 9).

Two vignettes were included here: the tendency to attend an ED was evenly distributed across the Wednesday and Saturday child vignettes. There were no statistically significant interactions between the samples completing the two child vignettes in the final model.

When the binary comparison was limited to those ticking attending an ED compared with those selecting lower levels of care only, the results were very similar. All the variables in the original model featured in this model. There were some additional variables of looking on the internet to decide what to do and thinking that too many people go the ED.

In summary, the key findings were:

• The attraction of the ED is prominent here for allowing same-day contact and access to tests.

• The high tendency in Wales and Scotland to attend an ED is difficult to explain.

Tendency to attend an emergency department (all)

In total, 42% (1216/2906) of respondents ticked ‘go to an ED’ for any of the six vignettes. In a univariate analysis respondents had a higher tendency to an ED if any of the following applied (see Appendix 9, Table 26):

• sociodemographic characteristics – male, BAME, social classes III manual/IV/V, from Wales or Scotland, were not in a household with children aged < 5 years

• resources – none

• health – none

• programme theories – loss of functioning (programme theory 4), preference for no appointments (programme theory 7), willing to wait in a waiting room to be seen that day (programme theory 7), preference for ED because can get tests quickly (programme theory 9), think ED doctors know more than GPs (programme theory 9)

• awareness of services – none

• recursivity – if test were done they were right to contact that service

• health literacy – none

• recent service use – recent and frequent use of ED

• beliefs about overuse of services – do not think that too many people use ambulance and ED when not needed.

There were some unusual findings:

• If they were not likely to take medication, then they had a lower tendency.

• If they were ‘missing’ for looking things up on the internet, then they had a lower tendency.

Based on a complete-case analysis of 2304 out of 2906 (79%) of the data set, the full model containing all the statistically significant variables identified above is given in Appendix 9, Table 27. The events per variable was 45.6 (959/21) showing that there was little risk of overfitting for this model. The following variables made an independent contribution in a multivariable analysis, with the following groups more likely to have a tendency to contact an ED:

• sociodemographic characteristics – male, BAME, Wales and Scotland

• programme theories – loss of daily activities (programme theory 4), willing to wait in a waiting room to be seen that day (programme theory 7), prefer ED because can get tests quickly (programme theory 9)

• beliefs about overuse of services – do not think that too many people use EDs when not needed.

All six vignettes were included here but the tendency to attend an ED was dominated by both of the child vignettes and the painful rib vignette. There were no statistically significant interactions between the samples completing the cough/rib/Saturday and those completing the vomiting/back pain/Wednesday vignettes in the final model.

The analysis above compared ED with ‘not ED’. The ‘not ED’ group included people who called an emergency ambulance as well as those calling less urgent options than the ED. Another analysis comparing only those ticking ED with those ticking the less urgent options gave very similar results. Age was included, with older people having a higher tendency, and programme theory 7, willingness to wait in a waiting room to be seen that day, was not included.

In summary, the key findings were:

• People who are attracted by characteristics of EDs and want to address loss of daily activities such as work and sleep have a higher tendency to contact an ED.

• Interestingly, negative views of GPs did not feature here.

• The high tendency to attend an ED in Wales and Scotland is difficult to explain.

Tendency to call a general practitioner

In total, 33% (961/2906) of respondents ticked ‘go to GP’ for the two adult illness vignettes. In a univariate analysis respondents had a higher tendency to contact a GP if any of the following applied (see Appendix 9, Table 28):

• sociodemographic characteristics – aged over 65 years, BAME, different regions/countries of Britain including Wales, Scotland and London

• resources – no car, no access to internet

• health – fair/poor health and with limiting long-term illness

• programme theories – likely to worry that pain was sign of something serious (programme theory 1), not confident in deciding to see a doctor or self-care (programme theory 1), loss of functioning for daily activities (programme theory 4), feeling overwhelmed when faced with health problems (programme theory 7), travel to ED difficult (programme theory 7), prefer no appointments (programme theory 7), willing to wait in waiting room if can be seen that day (programme theory 7), very likely to check with family or friends about what to do (programme theory 8), prefer ED for quick tests (programme theory 9), registered with a GP (opposite of programme theory 10)

• awareness of services – not confident in finding out when services are open, not confident in how to contact GP out of hours, would not look up on the internet about what to do about a health problem

• recursivity – if tests are done they feel right to contact that service

• health literacy – people with low health literacy

• recent service use – frequent ambulance use, recent use of GP

• beliefs about overuse of services – do not think too many people use GP when do not need to, think people are reluctant to use EDs.

Based on a complete-case analysis of 2264 out of 2906 (78%) of the data set, the full model containing all of the statistically significant variables identified above is displayed in Appendix 9, Table 29. The events per variable was 24.9 (746/30), showing that there was little risk of overfitting for this model. The following variables made an independent contribution in a multivariable analysis, with the following groups more likely to have a tendency to contact a GP (see Appendix 9, Table 29):

• sociodemographic characteristics – > 75 years old, BAME, Wales and Scotland and south of England

• programme theories – worry that pain is a sign of something serious (programme theory 1), feeling overwhelmed when faced with a health problem (programme theory 7), ease of access to GP (opposite of programme theory 10)

• recursivity – if tests are done then right to contact that service

• health literacy – people with low health literacy in terms of understanding information

• recent service use – recent use of GP but the variable does show consistent pattern.

Only two vignettes were included here, with a tendency to contact a GP dominated by the diarrhoea/vomiting vignette in that two-thirds of people with a tendency for ‘clinically unnecessary’ use of a GP ticked this vignette rather than the cough vignette. There were only two statistically significant interactions by vignette in the final model. The first was with age (p = 0.013) but none of the interaction levels was statistically significant. The second was with people who thought that having tests was a sign that they were right to attend a service (recursivity). People who experienced recursivity were much more likely to tick the cough/rib/Saturday vignettes than people not experiencing recursivity. People who did not experience recursivity were more likely to not experience it for the diarrhoea/back/Wednesday vignettes than for the cough/rib/Saturday vignettes. People who experienced recursivity were more likely to do so for diarrhoea/back/Wednesday vignettes than were those not experiencing it for cough/rib/Saturday vignettes. Caution is needed when interpreting interactions.

The analysis above compared GP with ‘not GP’. The ‘not GP’ group included people who ticked an emergency ambulance or attendance at an ED, as well as those calling less urgent options than ED. Another analysis comparing only those ticking GP with those ticking the less urgent options of another NHS service, NHS 111 or self-care gave a similar regression. Region was replaced by BAME and the belief that too many people use GPs.

In summary, the key findings were:

• Older people, people with low health literacy and who feel overwhelmed when faced with a health problem, and people who do not have access difficulties to general practice, have a higher tendency to use a GP.

Tendency to make any ‘clinically unnecessary’ demand for any service

In total, 66% (1841/2906) of respondents ticked a ‘clinically unnecessary’ option for any of the six vignettes. In a univariate analysis respondents had a higher tendency to do this if any of the following applied (see Appendix 9, Table 30):

• sociodemographic characteristics –aged > 65 years, male, BAME, social classes III non-manual–V, London or Scotland, not living in households with children aged < 5 years

• resources – no car, no personal access to the internet

• health – poor health, limiting long-term illness

• programme theories – worry that pain is a sign of something serious (programme theory 1), never had to decide about seeing a doctor (programme theory 1), loss of functioning in daily activities (programme theory 4), likely to take medication if in pain (opposite of programme theory 5), overwhelmed when have a health problem (programme theory 7), do not know if someone would care for them if ill (programme theory 7), find it difficult to travel to an ED (opposite of programme theory 7), prefer services with no appointments (programme theory 7), willing to wait in waiting room if seen that day (programme theory 7), check with family or friends for what to do (programme theory 8), prefer EDs for access to quick tests (programme theory 9), believe that ED doctors know more than GPs (programme theory 9), registered at GP (programme theory 10), family and work do not make it difficult to see GP (opposite of programme theory 10)

• awareness of services – not confident in finding out when NHS services are open, not confident in knowing how to contact GP out of hours, unlikely to look up on the internet what a problem is or what to do about it

• recursivity – if tests are carried out then this validates service use

• health literacy – lower health literacy levels for understanding information and ability to communicate with health professionals

• recent service use – frequent ED user, recent user of GP

• beliefs about overuse of services – too many people use ED and GP, think that people are reluctant to use EDs.

Based on a complete-case analysis of 2297 out of 2906 (79%) of the data set, the full model containing all of the statistically significant variables identified above is given in Appendix 9, Table 31. The events per variable was 22.6 (859/38), showing that there was little risk of overfitting for this model. The following variables made an independent contribution in a multivariable analysis, with the following groups more likely to have a tendency to make ‘clinically unnecessary’ use of services (see Appendix 9, Table 31):

• sociodemographic characteristics – aged 65–74 years, male, BAME

• health – limiting long-term illness

• programme theories – worry pain is a sign of something serious (programme theory 1), loss of functioning of daily activities (programme theory 4), overwhelmed when faced with a health problem (programme theory 7), prefer ED because get test quickly (programme theory 9), work and looking after family does not make it difficult to see a GP (opposite of programme theory 10)

• awareness of services – not confident in contacting GP out of hours, unlikely to look up on the internet what to do about a health problem

• recursivity – if tests are done then this validates service use

• recent service use – recent use of GP.

All six vignettes were included here but the back pain vignette did not feature strongly in this analysis compared with other vignettes. There was only one statistically significant interaction between the samples completing the cough/rib/Saturday and those completing the vomiting/back pain/Wednesday vignettes in the final model. People were likely to experience recursivity than not for the cough/rib/Saturday vignettes. People were more likely not to experience recursivity for the diarrhoea/back/Wednesday than for the cough/rib/Saturday vignettes. People were more likely to experience recursivity for the diarrhoea/back/Wednesday vignettes than not experience it for the cough/rib/Saturday vignettes. Caution is needed when interpreting interactions.

In summary, the key findings were:

• Many of the variables from the individual service analyses appeared within the univariate and multivariable analyses for tendency to contact any service.