Models of generalist and specialist care in smaller hospitals in England: a mixed-methods study

Overview of methods

We conducted a systematised literature review with the aim of providing context for our discussion around the patient, public and professional views of medical generalism in the acute setting (see Appendix 1 and Report Supplementary Material 1). We were particularly interested any definitions or paradigms of medical generalism underpinning current models of care in the UK.

Two stakeholder workshops were undertaken to explore how theories of medical generalism influence models of care.

Initial development of theoretical models of acute care

The literature review failed to find any consistent or shared definitions of generalism or what constitutes a generalist, with a profound separation between any theoretical models of medical generalism and how these are enacted within clinical practice. It was hoped that with the construction of this study this divide would be crossed, by devising a classification scheme for typology that was primarily based on the realities of day-to-day service delivery but informed by an understanding of how theories of medical generalism shape the ways of working and influence decision-making about patient care.

To explore how the theories of medical generalism might influence models of care, we convened two typology workshops (10 June 2016 and 14 November 2016). The first workshop was designed to explore with both medical generalists and medical specialists the boundaries between disciplines, the approaches to triage and how one judges how ‘generalist’ a medical service actually is. The second workshop was designed to explore the contexts in which models of care operate and which aspects of the wider system are driving them. As the aims of the two meetings were different, participants in the first group consisted of generalist physicians, as well as representatives from subspecialty societies (such as rheumatology and neurology); participants in the second workshop were selected for their ability to provide insights into the workings of either smaller hospitals and/or insight into service configuration at a national level.

At each workshop, group discussions were facilitated by Louella Vaughan, Candace Imison and Anne Marie Rafferty. Whole-group discussions were recorded and transcribed verbatim. Additional notes were taken by team members to capture smaller group discussions. Data were discussed by the team on consecutive occasions, with attention paid to emerging themes and points of divergence. All quotes extracted have been fully anonymised.

Theoretical models of medical generalism

The first workshop was useful in honing the theoretical construction of the emergency and acute medicine pathway. The team had initially assumed that patients would pass in a relatively uniform way between the ED, the AMU and the downstream medical wards (Figure 2). Intelligence from the participants suggested that models have become substantially more complex over the past 2–3 years, with the spread of services such as ambulatory/emergency care and frailty services, which has led to revision of the model (Figure 3).

FIGURE 2.

Assumed basic service model for the delivery of emergency/acute care.

FIGURE 3.

Revised service model (v3.0) for the delivery of emergency/acute care. AP, acute physicians; GP, general practice.

There was no consensus among participants, however, on how to characterise either the system components or the whole system as specialist or generalist.

The views expressed in the workshop confirmed that, rather than attempting to explore both service issues and professional issues in the telephone interviews, these should concentrate on capturing systems and processes of care, whereas the case studies should explore in more depth the issues that relate to professional identity and hospital culture.

Professional aspects of medical generalism

The first workshop also focused on professional aspects of medical generalism: who are the general physicians and what are their attributes? The following key themes emerged from the discussion.

Complex definitions

There was general agreement with the primary underlying assumption of the study: that more ‘generalists’ are required to meet the needs of the growing numbers of multimorbid patients. However, participants agreed that there was no universal definition of general medicine, that it was unclear which doctors deliver generalist care and that it was unclear what might actually constitute generalist care. Indeed, it was argued that attempting to nail down definitions may be counter-productive for the purposes of the study. The boundaries between specialist and generalist practice emerged as permeable and partially located at the level of the individual doctor, rather than being neatly and consistently drawn at system level:

General physicians have the use the expertise they have from where they recognise that they can manage the problem and when they recognise they can’t manage the problem they find someone who can.

Workshop participant

A complex mosaic of overlapping factors that might constitute the general physician as a professional emerged and are shown in Figure 4.

FIGURE 4.

The general physician as a professional: identified factors.

Contexts of models of care

The second workshop explored the extent to which the emergency and acute medicine pathway is shaped by factors at multiple levels. Examples of the resulting rich pictures are given in Figures 5 and 6.

FIGURE 5.

Rich picture produced for workshop 2, example 1.

FIGURE 6.

Rich picture produced for workshop 2, example 2.

Overview of methods

We outline here the methods used; a fuller description is contained in Appendix 2.

This study aimed to explore medical generalism through the perspectives of the patient, the professional and the service. WP1 was designed primarily to define and explore the service perspective, as well as elements of the professional perspective, namely how doctors define and experience medical generalist work.

We used a multistep approach to investigate models of care in smaller hospitals and understand how and why these models of care are deployed. This was carried out by:

1. identifying and creating organisational profiles of all smaller trusts in England (n = 69)

2. undertaking telephone surveys of all smaller trusts that agreed to participate in the study (n = 48 trusts; 50 hospital sites) with a view to map processes of medical care across the acute/emergency pathway

3. creating a typology of the models of medical generalism used by smaller hospitals, based on the telephone surveys

4. carrying out case study visits (n = 11) to examine in more detail the similarities and differences between main models of care identified by the typology and to evaluate the broader contexts in which the models of care sit, as well as the definitions, boundaries and meanings of medical generalism as theoretical concepts and lived experiences

5. undertaking a descriptive analysis of workforce, combining data obtained from the interviews and case studies with national workforce data from NHS Digital.

We considered the service perspective in two ways. First, we considered that the emergency and acute medicine pathway could be considered a closed system, boundaried by the patient’s presentation to and discharge from hospital, and therefore amenable to a general systems approach, which seeks to define and understand systems in terms of the functions and processes of each component within systems and the relationships between these. 96 Emergency and acute medicine pathways have also undergone radical change over the past two decades, with the introduction of AMUs, ambulatory care and the maturing of emergency medicine as a specialty. 76,97,98 It was therefore considered that emergency and acute medicine pathways in individual organisations were the result of serial rounds of innovation and improvement, mediated by contextual factors at multiple levels. 99 The Consolidated Framework for Improvement Research was adopted so as to better conceptualise and explore how and why hospitals operationalise their services and the contexts in which these services exist, particularly at the levels of the team provider, hospital and external landscape/policy. 100

Identification of smaller hospitals

NHS Improvement (formerly Monitor) has previously defined ‘smaller hospitals’ in England as providers with an operating revenue (income) of < £300M in the 2012/13 financial year. 88 Of the 142 general acute trusts, 75 of these in the 2012/13 financial year were found to fit into this category. Following review of financial information for 2015/16, six trusts were removed from the sample on the grounds that their operating revenues now exceeded that which might be expected from natural growth and were likely the result of major service reconfiguration. This left a sample of 69 smaller hospitals in England. This definition captures trusts that are predominantly single site, minimising the problem of attempting to distinguish Hospital Episode Statistics (HES) data at the hospital level in multisite trusts. It also ensures that there is a reasonable geographical distribution of sites, including hospitals in urban, provincial and rural settings.

Organisational profiles

Organisational profiles were constructed for all smaller trusts (n = 69) from a combination of publicly available data (e.g. trust and CQC reports) and hospital-level data that were collected by Monitor88 (later NHSI) and the Health and Social Care Information Centre (HSCIC), now known as NHS Digital.

Exploring models of care in smaller hospitals

Semistructured telephone interviews were used to gather data on the models of medical generalism deployed in smaller hospitals. Guides for the semistructured telephone interviews were constructed, following the work of Reid et al. ,101 to explore the following:

• hospital department characteristics across the emergency and acute medicine pathway (ED, AMU, ambulatory care, frailty services, downstream wards and any other relevant spaces)

• local triage rules for determining patient journeys

• construction of the acute medical take

• cover of the downstream medical wards

• networking arrangements.

Constructing the typology

As discussed in Appendix 3, the marked differences between organisations with regard to almost every aspect of service delivery across the emergency and acute medicine pathway made meaningful classification and comparison of the models of care difficult.

The classification scheme that was ultimately adopted focused on the AMU and the downstream wards, the components of the system through which the majority of patients with medical conditions flow. The classification also limited the analysis to the categories that best express splits between generalist medicine and specialist medicine:

• AMU – unit openness (closed/partial/open)

• AMU – patterns of consultant working [acute physician dominant (APD), mixed, specialist dominant (SpD)]

• wards – unit openness (closed/partial/open)

• wards – patterns of ward distribution within the hospital (some designated general medical wards, SpD wards only).

Using this scheme, distinct models were visible for the AMUs and the wards. However, there were still no consistent relationships between the models used on AMUs and the models used on the downstream wards. For this reason, some aspects of the analysis have been carried out using the typologies of AMUs and wards separately.

Case studies: investigating the typology

The purpose of the case studies was threefold:

1. to examine in more detail the operational aspects, such as the processes of care and staffing, with particular reference to the similarities and differences between models of care

2. to evaluate the contexts in which the models of care sit at provider team, hospital and external landscape/policy level

3. to explore concepts of medical generalism and how these may relate to the development and deployment of the different models of care.

Case studies were chosen so as to explore the different models of care that emerged from the typology. Care was taken to ensure that there was appropriate representation of trusts according to geographical location (urban vs. rural), size (small vs. ‘smallest’) and case mix. Three pairs of adjacent trusts were chosen, providing an opportunity to explore the contribution of external landscape and case mix to model development.

Case study visits were conducted over 1–2 days and involved the following activities:

• Semistructured interviews with key clinical, managerial and nursing staff. Topics included processes of care and triage, past and present organisational strategies, local networking arrangements, tensions between generalists and specialist workloads.

• Mapping patient flows by ‘walking’ through the emergency and acute medicine patient journeys. 102

• DCSs103 that allow insight into the appropriateness of inpatient care.

• Observations, with particular interest in unwritten rules, that influence how patients pass the emergency and acute medicine pathway.

• Staff focus groups (at half of the sites), exploring themes around processes of care, organisational culture and views of medical generalism.

• Document review, with a particular interest in standard operating policies for the AMU and downstream wards.

The case study interview guides and other materials can be found in Report Supplementary Material 2.

Testing the typology

There were two concerns about the typology: (1) whether or not clinicians and managers would recognise the models described and (2) whether or not the typology would be sufficiently pragmatic, as opposed to theoretical and abstracted, to be useful. To this end, the typology was presented, in various stages of development, for expert consideration by convened groups of senior managers and clinicians. It was presented twice to the New Cavendish Group and to the expert convened to look at case mix (see Chapter 5). It was also iteratively reviewed by the Study Steering Committee.

Construction of case studies

Case studies for each site were constructed, which were supplemented by material from the telephone survey and publicly available data such as trust annual reports. The studies aimed to combine detailed information about the processes of care at the unit level, while also providing contextual information about the hospital’s internal and external environments. Sankey diagrams, which depict flow through systems, were constructed for each organisation (Density Design Research Lab, Milan, Italy, URL: https://rawgraphs.io).

Framework analysis

Five separate sets of coding frameworks were used for all of the material generated by the case studies and relevant portions of the telephone surveys:

1. models of care – descriptions of mechanics of care were grouped by system component

2. the Consolidated Framework for Implementation Research (CFIR),100 which explores the contexts in which systems exist through five major domains – intervention characteristics, outer setting, inner setting, characteristics of the individuals involved and the process of implementation

3. medical generalism and specialism

4. smaller hospitals

5. workforce.

‘Pattern matching logic’ was used to look for emerging themes, with an emphasis on convergent and divergent evidence between data sources. 103 An iterative approach was taken, with regular discussions used to confirm important themes and guide further rounds of analysis.

Day of Care Survey

The DCS was analysed according to methods described elsewhere. 104

Description of participation

Table 2 describes the participation of NHS staff in the study.

What do smaller acute hospitals in England look like?

Using the definition of ‘smaller’ equating to an operating revenue of < £300M per annum,88 we identified 69 smaller trusts in England in 2015/16. Six of these trusts had more than one acute site. The geographical distribution of trusts is shown in Figure 7. Organisational profiles, detailing the number of beds, catchment population, distance to the next ED, ED attendances, medical admissions and CQC ratings, are given in Report Supplementary Material 3.

FIGURE 7.

Geographical distribution of the smaller hospitals in England.

Characteristics of the participating trusts

In total, 48 trusts, with 50 hospitals, agreed to participate in the study (including 70% of all smaller hospitals). The average population served by these was 300,000 people, compared with an average of 54,000 people for a ‘general hospital’ elsewhere in the European Union. The catchment populations varied from 140,000 to just over 500,000 (Figure 8), of which only 17 trusts had catchment populations of under 300,000. The average distance to the next nearest acute hospital was 17 miles; four hospitals were > 30 miles to their closest neighbour. The majority of hospitals were located in urban or well-populated rural areas.

FIGURE 8.

Trusts by catchment population.

There was wide variation in the level of ED activity across the sites, from just over 40,000 attendances per annum to nearly 150,000 attendances (Figure 9).

FIGURE 9.

Annual ED attendances 2015/16 by trust.

Overview of models of acute medical care in smaller hospitals in England

The raw results of the telephone interviews are in Report Supplementary Material 4. These results have been the basis for two publications. 67,87

Variation at every level

Our stakeholder engagement had suggested that emergency and acute medicine pathways in England have changed markedly over the past 5 years, with the expansion of ambulatory/emergency care and the introduction of frailty units (see Figure 3).

The reality did not quite match this theorised model. Although all hospitals by definition had an ED and downstream wards, there was marked variability in the numbers and types of other emergency and acute medicine components between hospitals (Table 3).

TABLE 3 - Distribution of components of emergency/acute care

Component of emergency/acute care	Percentage of hospitals (%)
AMU	100
Ambulatory care unit	96
Frailty unit	56
GP assessment area	26
Other assessment area	41

Within each of these areas, there was marked variability between organisations with respect to the size, function(s) and, for certain components, criteria for accepting patients.

The relationships between these components of the system also varied markedly. Most organisations had a mosaic of services at the front door (e.g. ED, ambulatory care and GP assessment), all providing some form of primary assessment function, with patients moving from these areas to the AMU and then to the downstream wards. In other organisations, there were two discernible trends: towards either ‘acute hub’ or ‘hyperstreaming’ models of care. The acute hub, as suggested by the Future Hospital Commission,8 looks to co-locate and streamline all services with a primary assessment function into a single ‘hub’ (Figure 10). Although no hospital had fully achieved this, four hospitals reported moving towards this.

FIGURE 10.

Schematic of the acute hub model of emergency/acute care.

By contrast, ‘hyperstreaming’ models had moved the bulk of the work outside the ED into a series of parallel places of assessment (Figure 11). Seventeen hospitals reported having more than four places for the initial assessment of patients, with one organisation reporting eight.

FIGURE 11.

Schematic of the ‘hyperstreaming’ model of emergency/acute care.

Although this schematic shows the AMU in parallel, the rules for where patients should proceed varied by organisation, with some hospitals requiring all patients to pass also through the AMU and others allowing patients, after primary assessment (regardless of location), to be admitted directly to a ward. Other organisations conceded that patient flow through the system was wholly contingent on the availability of ward beds.

Highly complex systems

The result of the multiplication and the diversity of services at the front door has resulted in highly complex flows through the emergency and acute medicine pathway. Figure 12 shows the Sankey flows for a smaller rural hospital with 350 beds (Sankey flows for all case studies are given in Report Supplementary Material 5). Patients usually undergo multiple moves, through up to five different components of the system before finally being discharged. Considering that each move between the different components (denoted by the solid black lines) will be accompanied by a change of treating team, this visualises the fragmentation and complexity of the emergency and acute medicine pathway used in many hospitals.

FIGURE 12.

Sankey flow diagram of emergency/acute care at a 350-bed rural hospital.

Acute medical units: definitions of generalist and specialist models of care

Two categories were chosen to classify the generalist nature of AMUs: the permeability of boundaries (Table 4) and the patterns of contribution of the medical consultants to the acute take (Table 5), a schematic of which is given in Figure 13.

TABLE 4 - Definitions of the permeability of AMUs

Type	Definition	Descriptive code	Type code
Closed	All patients on the AMU are under the care of the acute medical team	Clos	A1
Partial	Most patients remain under the care of the acute medical team, with the exception of a small number of specific patients. This applies to models where one or two groups of patients routinely have their care transferred to a specialist team (e.g. care of a patient with myocardial infarction is routinely carried out by cardiology)	Part	A2
Open	Newly admitted patients initially remain under the care of the acute medical team but are then triaged to a specialist team, which may include general or acute medicine	Open	A3

TABLE 5 - Definitions of patterns of consultant working on the AMU

SAG, specialist and generalist.

Type	Definition	Descriptive code	Type code
APD	Acute physicians provide the majority of input into the acute take. SAG contribution is limited to acute take on the weekends and/or overnight on-call. This does not include models where the SAG are physically present in the evening	APD	B1
Mixed	SAG consultants make a regular contribution to the acute take across the whole working week	Mix	B2
SpD	SAGs provide the majority of input into the acute take	SpD	B3

FIGURE 13.

Schematic for the patterns of consultant working on the AMU.

These definitions come with a number of important caveats:

• They are based on usual patterns of care, rather than behaviour in exceptional circumstances (such as a patient belonging to a specialist consultant needing to remain on a closed AMU owing to nursing needs).

• They do not capture all transfers of patient care between individual consultants, only those when the care has the potential to change from being generalist to specialist.

• The closed/openness of an AMU ignores the issue of who leads the acute medical team at any point in time.

As all patients on a closed AMU will remain under the care of a single physician who co-ordinates their care, these units can be considered to be the most generalist. Conversely, the system operating in open units, commonly known as ‘take triage’ (patients are assigned or ‘triaged’ to specialty services as part of the admitting process), were specifically created to bring specialists as close to the front door as possible, rather than patients being assigned to specialist teams only at the point of transfer to the wards. 106

Prior to the fieldwork, it was assumed the balance between acute physicians and specialists and generalists would correlate with how generalist the unit was in its approach to patient care. Although during the course of the fieldwork it held that hospitals with an APD model viewed their care as being more generalist, hospitals with SpD models often viewed themselves as being highly generalist.

Correlation between interviews and case study visits

There were two major discrepancies between the description of the models of care given in the telephone interviews and those that were observed during the field work. In the case of hospital A, the whole patient pathway had been changed during the intervening 6 months. Hospital K was classified as an A2 B2, but was operating an A3 B1 model.

For all other sites, the overall models of care observed agreed with the telephone interview descriptions.

Similarities and differences within models of care

One of the purposes of the case studies was to assess the extent to which the typology accurately captured the critical differences between models of care. We found that AMUs varied substantially more within categories than wards. Subtleties of consultant working, particularly continuity of care, and the ways in which different functions (e.g. primary assessment and ongoing care) were distributed within the components of the system accounted for the main differences within models of care.

Coherence of models of care

As with the interviews, the extent that models of care could be presented as coherent models across the whole emergency and acute medicine pathway varied. Staff in four hospitals viewed their models of care in this way: hospitals D, H, J and K. Hospitals F and I identified as having a clear approach to care on the AMU, but not across the whole system.

Hospitals that viewed their models of care as a coherent whole all operated different models of care, which does not match the assumption that coherent whole systems of care should all look the same. Staff all viewed their models of care as unique, and as an expression of their organisations’ overall commitment to serving the needs of the local community. Notably, all four organisations had stable leadership at both hospital level and team level [i.e. chief executive officers (CEOs) and key service leads had been in place for > 5 years], and all had substantially invested in programmes to imbed quality improvement and/or cultural change into their organisation.

Hospital-level factors

Consultant behaviour

Issues with workforce shortages were exacerbated by negative consultant behaviours. Two particular patterns occurred across organisations, which appeared to affect the model.

The first issue was specialist staff refusing to participate in the acute take, thereby ‘withdrawing’ their services from the front door. This was more common in large organisations that had sufficient staff to support the aspiration towards separate on-call for specialties, or in those organisations that were moving towards regional networked approaches for interventional work. Withdrawal of individual or groups of specialists from the acute take was considered to increase the pressure on those still participating in the on-call, which drives further flight, resulting in a ‘spiral of doom’ (Figure 14).

FIGURE 14.

Schematic of the ‘spiral of doom’.

The second phenomenon described we have termed 'fortressing', which was where staff actively limited their workload by tightly patrolling service boundaries. This was more often seen in organisations with more geographical places for the parallel processing of patients, and tended to occur when services were shoe-horned into small or otherwise inappropriate spaces. Hospitals B and E, for example, both moved from providing consult-type geriatric services (where roving teams reviewed all of the referred patients across the hospital) to frailty units, with a limited number of beds. The result was that rather than 15–30 patients having specialist input, the number of patient contacts was limited by the number of unit beds (6 and 10 beds, for hospitals B and E, respectively). This phenomenon was also seen in other places including the AMU at hospital E, where the acute physicians insisted that each of the four ‘sides’ of the AMU (10 beds per side) required its own consultant each day, and hospital B, where the cardiologists refused to provide input into patients not on their ward:

It was an attempt to get them to take ownership of their own area and to react to the needs of the service. And, unfortunately, what happened was that they took ownership of their own well-being and made it quite easy for themselves.

Senior clinical manager

Instances of fortressing and flight behaviours were not isolated. Instead, attempts had been made by clinicians to employ these strategies at least once at every organisation, with fortressing being described as present at four sites and flight at five sites. Although such behaviours were usually presented by consultants as rational responses to their increasing work pressures, they effectively created artificial workforce shortages. Moreover, these behaviours caused ripples throughout the rest of the consultant body, with other groups of specialists then attempting to deploy similar strategies. The end result, almost invariably, was a consultant body where relationships could be described as fragmented at best and openly hostile at worst:

[When] there were two of them on who were sharing the bulk – it was fine . . . I don’t think [specialist staff] loved it and embraced it, but I don’t think they were hating the concept of coming down [to the AMU]. Since the acute physicians retrenched into the aberrant model of looking after themselves, they all are really quietly pissed off because they’re coming on at four o’clock to an oncoming battlefield.

Senior clinical manager

Resistance to flight and fortressing was frequently discussed by senior managers. Managers in organisations where flight and fortressing had occurred were usually regretful that matters had been allowed to develop in that way. In organisations where the behaviour had been resisted, managers remained vigilant about their possible return.

External factors

External factors were less frequently mentioned as a determinant of model of care and were usually seen as negative or inhibitory rather than as a positive influence. There were few direct external drivers of model, but a number of external factors exerted substantial indirect pressures on systems and, hence, on models of care.

Other indirect influences on model of care

A range of other factors appear to influence models of care in less direct but still important ways:

• Change in hospital case mix – all organisations reported an increase in the number of frail elderly patients. Although attempts to improve care for this group of patients had been universally made, two hospitals had changed their model of care in response to this. One had instituted a parallel ‘elderly care’ take and the other was in the process of ‘right-sizing’ its downstream wards.

• Geography – the distance between individual service components, particularly the ED and the AMU, was a major factor in determining the distribution of medical work across the system.

• Junior medical staff – all hospitals reported severe shortages of junior-grade and middle-grade staff. Although this did not directly affect the patterns of consultant working, this was seen a major threat to the stability of models, with most organisations seeking alternative types of staff to augment the junior-grade and middle-grade rosters.

• Surgical services – the reorganisation of local surgical services, independent of STPs, was considered to affect the arrangement and even sustainability of medical services.

• Relationships with other hospitals – the quality of the relationships with neighbouring hospitals, particularly referral centres, was considered important in shaping local services. Movements towards networked regional services were generally seen as positive, although some viewed this as having potentially negative impacts, through loss of income and reducing the attractiveness of specialty jobs with an interventional component.

The development of models of care as a ‘journey’

Virtually all of the organisations described the development of their models of care as being a dynamic process taking place over a period of many years. In many places, development of models was evolutionary, but some described episodes of revolutionary change that were usually prompted by a change in clinical or executive leadership. In all cases, the ‘journey’ was shaped by the size, skill-set and co-operativeness of the available consultant workforce. This, in turn, was influenced by a whole constellation of other factors exerting influence at various points, sometimes being the catalyst to force change, other times influencing change already under way. This led to a sense, in most organisations, of change (and hence model of care) being reactive rather than carefully planned and executed.

The general medical patient

Medical generalism was most frequently framed in terms of ‘the general medical patient’. There was a universally held assumption that virtually all patients started out as general medical patients at the point of admission to the hospital (i.e. during the process of the acute take). Patients were then sorted into the categories of ‘general medicine’ or ‘specialty’ at any one of a series of time points: the first review by a senior medical decision-maker (registrar or consultant), the post-take review, the triage meeting, the point of specialty review or the discharge from the AMU to a downstream medical ward.

Positive definitions of the general medical patient were framed around their age, frailty, number and type of comorbidities, deprivation and social circumstances of the patient. However, definitions were not fixed and varied both within and between organisations. This, in turn, stemmed from the complexity and fluidity of these concepts themselves. That there is no clear or universally accepted definition of ‘frailty’ was frequently touched on:

Those frail elderly and the complex comorbid patients who aren’t necessarily elderly. I mean, it’s always difficult, how do you define ‘elderly’? I’ve always felt that we should have a biological definition rather than a chronological definition of who requires elderly care.

Senior doctor

Patients with intersections of needs, particularly if social needs were dominant, were almost universally seen as belonging to general medicine, regardless of other aspects of definitions:

Sometimes [the patients] haven’t got any acute medical problems, so it’s more regarding their social circumstances or families not coping with their care at home because of all the chronic comorbidities they’ve got.

Senior doctor

However, more frequently the general medical patient was framed in a negative sense, predominantly owing to them not having an easily definable, organ-based reason for their admission:

[General medical patients] don’t come in neat little packages of ‘I’ve got a gastroenterology problem’ or ‘I’ve got a cardiology problem’.

Senior manager

However, even in patients with organ-based disease, decisions around which patients were the preserve of the specialists varied and were dependent on factors such as the commonness of the disease, disease severity and expected LOS, with patients with an expected LOS of < 2 days usually remaining under the care of the acute physicians:

The one specialty that puzzles me is respiratory. Is it a specialty or is it a general thing? On the one hand, I think that 40% of the general medical take you could, if you wished, classify it as ‘respiratory’ – COPD [chronic obstructive pulmonary disorder], pneumonia or whatever. On the other hand, the number of people who actually need a specialist respiratory opinion is relatively small.

Senior clinical manager

The capacity of the specialists to provide care was also a factor in drawing boundaries. For example, in one organisation the lone cardiologist was always off-site on a Wednesday, meaning that patients who would be otherwise assigned to his care would remain under the acute team on that day. Capacity, in turn, was at least partially determined by the attitudes of individual clinicians towards service provision. Two organisations had only one respiratory physician; although in both of the hospitals this resulted in patients with chronic obstructive pulmonary disorder (COPD) and pneumonia consistently being labelled as ‘general medicine’, in one organisation the respiratory physician provided a daily round to all patients with respiratory conditions, regardless of their location, and in the other organisation, patients were cared for by teams other than the respiratory team with no additional specialist input.

Specialty patients were also relabelled as ‘general medicine’ when they were unable to be placed on the base ward of their team in partial and closed-ward models. Hence, a patient with a respiratory problem would be regarded as ‘general medical’ if placed on a gastroenterology ward. This also applied to patients if a particular specialty service was not available at the hospital; for example, almost all patients with neurological or renal problems became ‘general medicine’ by default:

If you’re in a gastro[enterology] ward, you’ll be looked after by a gastro[enterology] consultant: you may have a respiratory problem but you’ll be looked after by a gastro[enterology] consultant. If you then move wards and end up on a cardiology ward, you’ll then be looked after by a cardiologist. So we’ve got care, at the moment that is based on location of patient as opposed to patient need.

Senior clinical manager

Diagnostic uncertainty was viewed as problematic. As one of the tasks expected of the AMU was to assign a diagnostic label to the patient, those patients who left the AMU without a clear diagnosis were viewed as also general medical by default:

There are always patients where it could be anything . . . you are always going to get patients who fall through the cracks.

Senior doctor

Other negative definitions of the general medical patient could be considered as hostile, with the general medical patients being characterised as ‘the least interesting cases’ or ‘the ones that that we are stuck with’.

Labels were rarely assigned in a consistent fashion, even in single organisations. Most organisations attempted to sort the specialist from generalist patients clinically, with labels being assigned by either the take consultant or the take registrar; in one organisation, sorting was carried out by senior nursing staff on the AMU. Formal sets of rules were rarely used. Instead, there were often loose sets of understandings about what might be specialist (e.g. troponin-positive, anginal-type chest pain being the preserve of cardiology), along with calibrations for personal opinions of the assigning clinician and the knowledge of the receiving clinician’s capacity and preferences. However, pressures in the system meant that patients were often moved ‘to the next available bed’ without labels being formally assigned:

It depends upon the geriatrician.

Senior clinical manager

We’ve got some specialist ward areas that may or may not have specialty patients and gen[eral] med[ical] patients on them. We do have a gen[eral] med[ical] ward that’s called something else but, essentially, delivers the general medical service.

Senior clinical manager

The general physician

As with the general medical patient, definitions and meanings around the general physician were inconsistent and fluid. Importantly, few doctors characterised themselves as being solely general physicians, with discussions instead framed around acting as a ‘generalist’, which was one of a series of roles that a consultant physician could adopt:

When we’re ‘on’ each of gets rids of our regular hat . . . and all our other roles are cancelled for that day.

Senior doctor

Three factors emerged as defining the generalist role:

1. the place of working, with front-door working being predominantly generalist

2. the type of care delivered, with primary assessment and initial management functions being predominantly generalist

3. the label attached to the patient, with consultants adopting the generalist role when delivering care to a patient labelled as ‘general medical’ (regardless of place).

That these three factors were highly intermingled and difficult to disentangle was reflected in the fact that no interviewee articulated a consistent view of what constituted a generalist or generalism. However, it was apparent that the generalist skill-set required by the acute take was fundamentally different from that for managing patients on the downstream wards.

Distinctions between generalism and specialism were drawn on the basis of the skills and training of individuals, where older physicians or those with longer periods of training were more likely to identify with the role of the general physician than their younger colleagues, who had straight-through dual-specialty training. Notably, younger consultants did not dispute their ability to manage the take, but rather struggled with the needs of complex medical patients requiring ongoing care:

I wasn’t trained to look after these sorts of patients. It’s not the job I [trained for]. Even before I started here as a consultant it was already different.

Senior doctor

Although the acute physicians were considered to be generalists by their colleagues, the ability to provide competent care to patients, regardless of their location or their needs, was considered to be clear point of distinction between acute physicians and those who identified as ‘true’ general physicians.

Opinion was split as to whether or not geriatricians were ‘generalists’. As geriatrics was considered to be need-based rather than organ-based, it was viewed as some as being entirely generalist. Others thought that their training equipped them to care for a specific group of patients:

Geriatricians more than any other profession bring together the knowledge base that encourages and foster more generic identification of needs outside disciplinary-focussed care.

Senior clinical manager

I would not call a geriatrician a generalist. I’d call them a Geriatrician.

Senior doctor

Some viewed the essence of generalist working as being co-operative and collaborative, and focused on the co-ordination of care:

My tagline is that your role as a general physician is to make everyone else do their job, because it literally feels like you’re going down to the playground and dragging people to come home early to do scopes, to offer opinions, to get their arse into gear and actually do the job they applied to do.

Senior doctor

Systems as generalist or specialist

In discussions about generalism and specialism being properties of models of care or systems, there was a clear split between organisations who identified as having coherent models of care and organisations who did not identify as having coherent models of care. Interviewees from the former readily identified their organisations as being ‘generalist’, regardless of the model of care. Hospital K was one of the largest hospitals in the sample and operated a take triage system, properties otherwise associated with specialism, but it viewed itself as being highly generalist on the grounds that the acute work was equally divided among specialties on a daily basis. Interviewees at other organisations were able to locate their hospitals along a spectrum of ‘more generalist’, ‘hybrid’ or ‘more specialist’, but on prompting only. Clinicians at the smallest hospitals were more likely to identify their systems as ‘generalist’, regardless of the model of care.

Staffing as a major challenge

A key finding of the telephone interviews was the scale of the staffing challenge that was faced by trusts. Staffing problems for junior medical staff are severe, described to us as a ‘nightmare’. Nearly all of the hospitals that were surveyed were dependent on locum staff to support their senior medical rotas. Proportional dependence on locum staff seemed to be greatest in the smallest hospitals.

Hospitals were trying to address these problems using a variety of means, including redesigning job roles, changing skill mix, offering recruitment incentives including ‘golden handshakes’ and international recruitment. However, the scale of the challange meant that few of the case study sites had fully addressed these problems. The capacity to recruit can be affected by the location and the dynamics within some clinical teams, in some cases acting as either a deterrent or an enabler.

Variation

There were stark variations in the number of medical staff, and non-medical staff in advanced roles, who were working within each of the medical specialties. For example, the number of respiratory physicians varied from one to 16 and the number of staff in elderly care medicine ranged from zero to 14. This variation is not fully explained by the size of a hospital, the number of patients or the case mix of patients (see Chapter 5).

In addition, there is nearly a three-fold variation in the number of junior medical staff per consultant, with the medical specialties going from 0.9 junior medical staff per consultant to 2.75 junior medical staff per consultant.

Balance between generalist and specialists

Most trusts had a comparatively low number of consultant medical ‘generalists’, including general medicine, acute physicians and geriatricians. Only five hospitals had > 60% ‘generalist’ doctors, as shown in Figure 16. The balance was different for junior medical staff.

FIGURE 16.

Percentage of generalist and specialist consultants by trust, smallest to largest.

Management of the acute take

The number of consultants supporting the medical on-call rota varied significantly. On-call cover included specialists from respiratory medicine (90% of sites), geriatric medicine (84% of sites), endocrinology and diabetes (86% of sites) and acute medicine (65% of sites). A number of specialties have started to withdraw from the on-call rota. Some sites have insisted that their specialists do take part in the medical on-call rota.

Staffing ambulatory/emergency care

Ambulatory/emergency care was a service where generalists rather than specialists took the lead. Acute physicians provided clinical leadership for 75% of the ambulatory care services. In other sites this was provided by the ED, or exceptionally by specialty physicians.

Staffing the acute medical unit

The AMU was also a service where generalists rather than specialists took the lead. In 90% of sites an acute physician provided medical leadership for the AMU. The number of acute physicians employed by sites varied significantly. Many places had very few acute physicians in post (five or less), with heavy AMU commitments. One site had no acute physicians, preferring a model in which specialists provide cover to the AMU.

Ward responsibilities

Ward-based care was specialty led. In the majority of cases, patients would be managed by the specialist medical team responsible for their ward and would be reliant on the host ward calling in the appropriate advice. Only 12% of sites had specialists managing patients, irrespective of their location.

Outpatient and other commitments

We did not explore outpatient and other commitments in depth. It was clear from the interviews that there were some medical specialties for which the competing demands between inpatient work and outpatient work were a problem. The most notable was gastroenterology.

Support from other clinical staff

Increasingly, trusts are deploying staff in advanced non-medical roles; however, there was significant variation in the number of these non-medical roles across hospital sites (Figure 17).

FIGURE 17.

Maximum and average number of advanced non-medical roles by specialty.

Creating and testing the classification

In brief, creating and testing the classification followed the following stages:

1. defining a working data set of 1.9 million care episodes based on acute medical emergencies that were admitted under the care of a defined set of treatment specialties

2. creating an initial classification of broad diagnostics groups that sought to identify high-number common cases

3. testing the affect of a range of secondary diagnostic characteristics and prior history variables

4. working with a group of clinicians to test the face validity of the groups and use their input to refine the scheme

5. finalising the model and comparing the explanatory power in terms of the reduction in variance in length of stay, including a comparison to HRGs.

Describing inpatient case mix in acute medicine

The case-mix classification was then used to describe case mix at the hospital level, in terms of the number of episodes of care within a given time period, the number of admitting spells and the total bed-days. We also estimated the average number of admissions on any single day (based on average episodes of care per day) and estimated the number of beds occupied by any one case type. Other features of episodes were also noted, including the proportion of short-stay cases, where stays lasted 0 or 1 day, and the proportion of patients aged > 75 years.

We undertook a range of analyses to explore:

• Changes in case mix over time (see Report Supplementary Material 8, Section 5). Changes over time were examined by using the data for successive years, from 2012–13 to 2017–18.

• Differences in case mix between providers (see Report Supplementary Material 8, Section 4). Each of the 69 hospitals had its own profile of case types; however, we can also summarise similarities and differences into single metrics, such as the proportion of cases falling into the most common case-mix groups (CMGs). We also calculated some measures of case-mix complexity based on weighting activity in each case-mix group by the overall average stay length.

• Differences in underlying admission rates at the population level (see Report Supplementary Material 8, Section 4). We undertook this analysis to assess the degree of variation in the underlying needs for acute emergency medical care between geographic areas. 125 To do this, we used data from all hospitals at the hospital-spell level. The analyses used data from 2017–18 and related those to ONS population estimates from 2016126 of 326 local authority areas.

• Differences in average LOS in hospitals (see Report Supplementary Material 8, Section 5.5). We compared the aggregated LOS at the hospital level with the impact of case mix on this value. This used a technique advocated by Fetter et al. 111 in their original paper on the creation of HRGs, which was based on a technique described by Kitagawa. 127

• Pathways through the hospital (see Report Supplementary Material 8, Section 6). The analysis of the consistency of treatment specialties for the case types explored handovers between specialties within a given spell of care, focusing on a small number of tracer conditions. The pathways of care were defined using spell-level data from all of the smaller hospitals in the cohort. Pathways of care for individual trusts were examined to consider the extent to which they were similar or different from the pattern exhibited across the set of smaller hospitals as a whole.

Chapter 7 describes the patterns of mortality and re-admission that are associated with case-mix groups and explore the relationship to hospital typology.

Comparing the case mix according to organisation of general medical services

For this we looked at two key features:

1. The differences in case mix associated with acute medical typology (see Report Supplementary Material 8, Section 9). We looked for a major difference in case mix associated with the hospital typologies that were identified in Chapter 4. We considered a range of basic descriptors, including number of cases, lengths of stay and case-mix complexity.

2. The relationship between inpatient case mix and medical skill mix (see Report Supplementary Material 8, Section 9). In this section, we looked at whether or not there are observable relationships between the numbers of ‘specialist consultants’ within a hospital and the mix of cases seen in acute medical workload, using information on the medical workforces from NHS Digital. We used a variety of tests to explore the relationships for the overall proportion of generalists and four specialty groups: respiratory medicine, cardiology, gastroenterology and care of the elderly.

Further details of the methodology are given in Report Supplementary Material 8.

Basic case-mix classification

The case-mix classification was derived from a data set covering 1.9 million records from acute medical specialties in a subset of 69 smaller hospitals. The final classification consisted of 82 base CMGs, most of which were defined in terms of diagnostic criteria. The nine groups that described cancer workload were created on a more complex basis by combining information regarding prior episodes of care, treatment recorded and cancer diagnoses.

In addition to the base CMGs, additional flags were created to split these, where necessary, according to:

• A frailty flag. In our analysis of reduction in variance we had found that applying a basic flag indicating potential ‘frailty’ proved better than age at explaining differences in the LOS and was also felt to be more clinically meaningful. There are a number of approaches to measuring frailty. 128–131 We tested two alternatives,132,133 which produced similar results. We opted for the one that had slightly higher reductions in variance.

• A procedure flag. One of the issues raised in our workshops was that the clinical perception of a patient often revolves around whether or not there was some procedure undertaken, usually diagnostic. Empirically, this was seen to be a large influence on resource use.

• Cancer as a secondary diagnosis. One of the problems raised by our clinical groups was the challenge of distinguishing what were the most important diagnoses influencing care. We used the principal diagnosis to drive our schemes, although we recognised that in some cases that could potentially be misleading, such as a patient admitted with pneumonia but who also has lung cancer. Sometimes the principal diagnosis may not be the most important one influencing treatment and care. This was felt to be particularly problematic for people with a cancer diagnosis. Therefore, we added a flag to see if any of the secondary diagnostic field include cancer, as this might be driving an atypical pattern of resource use.

If all variables were used, there were a total of 328 permutations, which is still less than the number of HRGs. Table 12 summarises the reduction in variance in LOS that was associated with different combinations of variables. The descriptive analysis predominantly used the diagnostic groupings at the episode level, with secondary variables being used for sensitivity testing. For some analyses of overall hospital length of stay, we used classifications based on the first episode of the hospital spell.

Case-mix episodes in acute medicine

Table 13 summarises the 25 most common base CMGs seen across the subset of 69 smaller hospitals. Figure 18 relates these conditions to the use of occupied beds (average of average across the 69 hospitals). There was a high concentration of cases in a relatively small number of diagnostic groups. The most common grouping J3 Pneumonia (aspiration) and LRTI and pneumonitis due to solids and liquids accounts for 12% of all cases and approximately 16% of all bed-days. Approximately 80% of all care episodes and 75% of all bed-days fall into just 31 groups (full details can be found in Report Supplementary Material 8). Table 13 also includes estimates of the typical number of admissions per day linked to that group, and the estimated number of beds that might be occupied on any one day.

FIGURE 18.

Approximate use of occupied beds for acute emergency medical admissions (excluding short stay) for 2017/18. Full codes and code titles are presented in Table 13. LRTI, lower respiratory tract infection.

An average across all 69 hospitals suggests that the acute emergency medical caseload in a typical smaller hospital consists of:

• 78 episodes per day

• 266 occupied beds at any one time

• 145 occupied beds were short stays of < 2 days

• 121 occupied beds were longer stays (≥ 2 days).

In total, 27% of cases had some form of procedure and 31% of cases had an indication of frailty. Around 51% of all cases had neither a procedure nor a frailty flag. The presence of either a frailty flag or a procedure was associated with much longer lengths of stay. For example, for the simplest cases (i.e. no frailty flag or procedure) the average episode length was 1.8 days, whereas for the 8.6% of cases which had both a frailty flag and a procedure, the average episode length was 9.1 days. The average length of episode for cases with a secondary cancer diagnosis was longer than cases with no cancer diagnoses, but less than for episodes with a primary diagnosis of cancer.

In exploring the characteristics of the case-mix grouping, we looked at the level of variability between geographic areas and the extent to which such variability might be explained by local demographic factors (see Report Supplementary Material 8). At the CMG level, the groups with the least variability [standard deviation (SD) < 0.26] represent groups in which the underlying incidence of the disease was more constant and less influenced by differences in disease prevalence, the accessibility of services or the vagaries of hospital recording systems. Although some groups that varied considerably are clearly linked with differences in local population needs, such as sickle cell disease, others are harder to explain.

Changes in case mix over time

Overall, there was a 26% increase in the total number of emergency medical episodes between 2012–13 and 2017–18, from 1,703,783 cases to 2,155,126 cases; the average yearly increase in the number of episodes was 5% (range 4–6%). There was a 19% increase in the total number of spells across a comparable time-period. The number of short-stay episodes (0–1 days) increased by 34% and the number of long-stay episodes by 18%.

Length of stay declined at both episode level and spell level between 2012–13 and 2017–18. Average episode-level LOS declined by 0.71 days, from 4.12 days to 3.41 days. Average spell-level LOS fell by 0.73 days, from 6.52 days to 5.79 days. Table 14 shows the total number of episodes, spells and LOS, by year.

Some groups showed a particularly large change in LOS, including septicaemia and pneumonia, most probably related to coding practices. A few groups decreased in number, with the most marked changes in absolute terms for I2 Angina pectoris and dyspepsia, R1 Pain in throat and chest and G4 Transient ischaemic attacks and dizziness and giddiness. Some of these changes in presentation, such as for cardiovascular conditions, may be the results of deliberate changes in the referral pathways for selected conditions.

Analysis of changes in the distribution of case types at individual trust level showed that of the 69 hospital trusts, only three showed significant changes in their case-mix distribution over time (p < 0.01 was used to allow for multiple tests across 69 trusts). Although the case mix in diagnostic terms remained fairly stable, there were other indications of changes during this period. A number of groups showed a marked increase in cases aged > 74 years, despite the fact that the average age remained fairly consistent across the time period. The proportion of cases flagged as ‘frail’ also increased from 2015–16 to 2017–18.

Differences in acute medical inpatient case mix between trusts

We used a number of approaches to summarise the case-mix complexity within an individual hospital. One approach was to study the proportion of cases that fell into the 10 or 20 most common CMGs. Table 15 summarises the results, showing the trusts with highest and lowest values. Homerton University Hospital NHS Foundation Trust appears to be an outlier, which was a result of a high number of cases in one particular group (D2 Sickle cell disorders). If Homerton University Hospital NHS Foundation Trust is excluded, the proportion of cases falling in the top 20 CMGs shows little variation: the range is from 65% in Poole Hospital NHS Foundation Trust to 70% in Burton Hospitals NHS Foundation Trust.

The consequences of differences in a case-mix profile can also be expressed in terms of the affect of different case types on the demand for beds. We applied a simple case mix-specific weighting based on the ratio of the group average length of episode to the hospital-specific case mix, to calculate an expected number of bed-days that arise from the particular mix of cases. Values of this ratio greater than 1 indicate a case-mix profile that is expected to require more beds than average, and in these terms, be more complex. Note that this approach does not equate to a measure of complexity in clinical terms, for example acuity, risk of death or costs.

The results suggest that when a simple diagnostic standardisation is used, the complexity ratio ranges from 0.92 to 1.14, and most hospitals sit within a range of ± 5% of the average. Overall, these results indicate that there is relatively little variation. Table 16 shows the hospitals at either end of this range.

Differences in average length of stay in hospitals

The average length of stay in hospital is best summarised in terms of the whole spell length rather than individual episodes of care. The unstandardised average lengths of stay in 2017–18 across all hospitals was 5.2 days and ranged from 8.4 days (Isle of Wight NHS Trust) and 8.1 days (East Cheshire NHS Trust) to 2.8 days (Torbay and South Devon NHS Foundation Trust) and 2.55 days (Royal Bournemouth and Christchurch Hospitals NHS Foundation Trust). The proportion of short-stay episodes had a strong effect on average stay lengths. Figure 19 shows that the average length of stay excluding the short-stay cases was usually around 2–3 days longer.

FIGURE 19.

Average length of stay of spell with (vertical) and without (horizontal) short-stay cases (data from 2017/18).

To quantify the effects of case mix on the aggregated length of stay, we split the differences in the group-average LoS into three components: (1) the effects of case mix, (2) the effect of within case-type stay length and (3) the interaction between these (see Report Supplementary Material 8). For this analysis we also included frailty flags to indicate something about patients whose prior history indicated greater needs, which, in turn, may be a reflection of stay length.

Table 17 summarises findings for the hospitals that had the longest and shortest average LOS. It appears that in most of the hospitals, the within case-type LOS effect is greater than the case mix itself; hospitals tend to exhibit lower than average stay length across all or most case types. For most of these hospitals, the case-mix elements are also negative in effects – indicating that these hospitals have slightly fewer than expected case types where you might expect a longer LOS, that is a less demanding (in bed terms) case mix. The exceptions are Southend University Hospital NHS Foundation Trust (where the case mix is more demanding) and Royal Bournemouth and Christchurch Hospitals NHS Foundation Trust where there is little effect due to LOS.

Looking at the hospitals where the average length of episode is longer than elsewhere (see Table 17), it seems that the case-mix element is more marked in these hospitals, with effects on average lengths of stay of plus 1.2 days at the Isle of Wight NHS Trust and plus 0.99 days at George Eliot Hospital NHS Trust. However, the LOS effect was generally large and positive and, in most cases, greater in magnitude than the case-mix effect. Therefore, the implication is that although these hospitals may have a case mix that would suggest longer lengths of stay, the stay lengths within groups are still longer than average. This was a relatively small difference compared with the lengths of stay within case types.

Figure 20 shows the effect of the two elements plotting the size of each element for each hospital moving from shortest to longest LOS. Figure 20 shows that as you move from shortest to longest stay length hospitals, the case-mix elements generally stay in a band of ± 0.5 days; however, the LOS element swings more dramatically and is clearly far more important in ‘explaining’ differences in the aggregate stay lengths. Aside from the extremes, there are some trusts that appear atypical in that they have a positive case-mix effect but negative lengths of stay – these are Ashford and St Peter’s Hospitals NHS Foundation Trust, Mid Essex Hospital Services NHS Trust and Southend University Hospital NHS Foundation Trust.

FIGURE 20.

The LOS and case-mix components for individual hospitals standardised using CMG and frailty (sorted by aggregate average LOS: longest LOS on right).

Pathways through the hospital

Tracer conditions were selected to test and document pathways of care on the basis of sufficient case number and clinical knowledge of pathway variation. Analysis of pathways of care at an overall level showed that for all of our tracer conditions, the most common pathway of care was a single episode spell under general medicine. The results also show, as might be expected, that where patients are treated under geriatric medicine, average lengths of stay tended to be longer, most likely a result of more complex case presentation and/or increased levels of frailty as individuals age.

We suggest that there are three key points to note:

1. There was heterogeneity in the pathways within the same basic diagnostic group.

   We noted that there were no simple patterns describing the specialties within a diagnostics group. As an example, Table 18 shows the most common patterns for one CMG, I9 Heart failure and pulmonary oedema. Across all of the groups that we looked at, there was considerable variation in the pathways adopted (see Report Supplementary Material 8, Section 6). At a trust level, on average, 41% (range 15–95%) of spells for I9 Heart failure and pulmonary oedema were managed under ‘other pathways’, that is they were not categorised under the top five pathways of care.

1. There were large differences in patterns of bed use linked with the pathways.

   Table 18 shows that patients seen in a single care episode in general medicine had an average LOS approximately 5 days shorter than any of the other top five pathways of care. This would fit with a model whereby short-stay patients were retained under the acute physician along with some other patients who are kept on the AMU for clinical reasons (e.g. infection control). There are indications of patient differences linked to the pathway; for example, the average patient age was highest (86 years) under pathway 4 (general medicine – geriatric medicine) and was lowest (74 years) under pathway 5, where patients received a single episode of care under cardiology.

1. There were large differences between trusts in the range of pathways.

   When we examined the patterns at trust level, it became apparent that even within trusts the distribution of different pathways was complex. Figure 21 summarises the pathways used in different trusts for chronic lung disease and COPD, looking at the most common six pathways, with the remainder categorised as ‘other pathways’. Spells falling under ‘other pathways’ ranged between 6% in Rotherham NHS Foundation Trust and 91% in Royal Surrey County Hospital NHS Foundation Trust.

FIGURE 21.

Differences between trusts in pathways used for J4 Chronic lung disease, including COPD: trusts with the five highest and five lowest percentages of spells managed under other pathways.

Differences in case mix associated with acute medical typology

We compared the case-mix characteristics of hospitals classified using the four attributes of the typology (described in Chapter 7) to test for any major differences between hospital subtypes. The analysis was based on a subset of 46 hospitals for which we had a typology and data on case mix from HES. Table 19 summarises some basic characteristics of the hospitals according to the four dimensions in our typology.

Overall, there were no major differences in terms of the proportion of short-stay cases, the case-mix complexity (bed use) and the proportion of ‘frail’ patients across the types of model. There were some exceptions in which there were small differences. For example, the episode length was slightly longer for the small number of open AMU hospitals and the proportion of cases in the top CMG was slightly higher. There were also tiny differences in the proportion of older people, suggesting a more typical case mix; however, these differences were small.

Relationship between inpatient case mix and medical skill mix

In this section we looked at whether or not there was an observable relationship between the number of ‘specialist consultants’ within a hospital and the mix of cases seen in the acute medical workload (see Report Supplementary Material 8, Section 8). Chapter 5 provides an overview of skill mix. Note that in this analysis we separated the numbers of very short-stay patients from the longer stay patients, as these patients would in most cases be associated with AMU care only. This analysis was undertaken for four specialist consultant subtypes: respiratory medicine, cardiology, gastroenterology and elderly care medicine, and took into account the number of patients with a frailty flag. We also examined the proportion of frail cases and the number of generalists in acute medical staffing.

Overall, the analysis did not find a relationship between the case mix and the widely different patterns of medical staffing. The two examples below demonstrate these results (see Report Supplementary Material 8, Section 8, for further details).

Respiratory medicine

Although the number of consultants in respiratory medicine in the hospitals ranged from zero to 16, there was no clear link between the number of respiratory medicine consultants and the proportion of patients (excluding short stays) in the seven key CMGs who were linked with respiratory medicine (Figure 22).

FIGURE 22.

Percentage of spells in respiratory CMGs vs. the number of respiratory medicine consultants.

Geriatrics and frailty

We compared the prevalence of the frailty flag with the number of consultants in geriatrics (care of the elderly). At trust level, the proportion of cases flagged as at risk of frailty showed quite a large range from 14.4% to 39% of spells, and these equated to between 30% and 66% of acute medical bed-days. As shown in Figure 23, although the number of consultants described as geriatricians/elderly care medicine varied from zero to over 14, there was no relationship between this number and the proportion of admitting spells where the patient has been flagged as frail (or potentially frail).

FIGURE 23.

Percentage of spells flagged as frail vs. the percentage of consultant geriatricians in overall medical consultant workforce.

Developing a new classification

As part of this study, we have developed a new classification scheme to capture the case-mix dimension in relation to acute medical emergency inpatient care in small- to medium-sized hospitals. At its simplest, the classification scheme can operate on one level, driven by clusters of diagnoses, but these can be augmented by other information. It is worth noting that, for our general acute medical caseload, the resulting scheme outperforms HRGs (which has many more groups) in terms of the reduction in variance in lengths of stay. It also has a more manageable number of groups and allows flexibility in use.

The resulting classification seems to demonstrate validity in a number of ways:

• The recognition of certain diagnostic categories in describing the overall case mix fitted with the clinical perception of what constitutes key elements in the workload.

• Where differences in case mix exist, they seem to fit with expected patterns. For instance, prehospital triage will influence admissions for some specialist care (e.g. stroke).

• The scheme identifies expected differences in specific hospitals – for example, the importance of D2 Sickle cell disorders in a hospital such as the Homerton or differences between neighbouring hospitals such as Harrogate and Airedale.

• The scheme explains differences in the length of stay and the incremental benefits of including measures such as ‘frailty’ or process measures such as the presence of a procedure.

The scheme, like many other case-mix descriptors, does have limitations:

1. A reliance on existing computerised data sets means that the scheme is at the mercy of variable hospital information systems. Moreover, as we have observed, coding practices change over time, which can interfere with studies of change.

2. The level of clinical detail is limited to standard diagnostic and procedure coding schemes. Similarly, the level of detail on the care process is limited – we cannot say which patients went to an intensive care unit or a frailty unit.

3. The scheme does not deal with ambulatory care, which is of increasing importance, and could not use newly emerging emergency care data sets. Once established, these new data sets would help to develop the approach. We feel that this is a major weakness, which is driven by the lack of information about outpatient care.

4. Although the observed reduction in variance was better than HRGs when the full classification was applied, and produced values typically in line with the case-mix groupings, it still reaches only 30% of variance explained. This leaves a lot of unexplained variation in lengths of stay, which may be related to dimensions of case mix that we cannot measure with the data to hand.

Perhaps most importantly, the classification provided a more sophisticated way to use information about the same person, linked over time134,135 within the hospital spell or between hospital spells. One of the major constraints in case-mix classifications is the depth of information usually available and, therefore, it is important to make the best use of it by linking data between episodes. The ability to make better use of existing information is now possible through better computing power and data sets. It enables a better approach to case-mix descriptors that can use information about prior events to classify a patient on admission. In this case, we looked at the prior diagnoses linked to frailty or previous admissions for cancer. Basic descriptors of case mix were first used in hospital planning and finance over 30 years ago,136 and they have remained largely the same as static cross-sections of activity at one point in time;112 as we have shown, there is scope to do better than this.

What does this tell us about acute medical case mix in small- to medium-sized hospitals?

We have looked at acute medical emergency case mix from 6 years of data across 69 NHS trusts, a total of over 10 million patient episodes. The most striking features of our analysis were:

• The differences in acute emergency medical case mix between hospitals (with a few exceptions) are not that large – whether within our subset of smaller hospitals or across larger hospitals. The profile of acute medical emergencies across hospitals has more in common than it has differences. Therefore, we concluded that the myriad of different organisational forms for dealing with these are probably not a result of underlying differences in the presenting cohort of patients.

• There are large differences in bed use that cannot be explained by case mix. We noted that when using standard weighting, expected bed needs per cases varied by a small amount only (±10%) compared with the twofold to threefold differences in the average LOS.

• We did note the importance of short-stay cases in shaping the overall average lengths of stay. This is hardly a surprising finding but we also noted that across all cases, hospitals with a higher number of cases had shorter lengths of stay. The significance in terms of the resource use of short-stay cases is an indication of the way hospital practice is changing137,138 and presents a challenge for traditional approaches to describe hospital case mix and activity.

• Overall, we found that case mix was less important than within-hospital effects as an explanation of differences in aggregated LOS. The implication is that there are many different ways in which hospitals are managing throughput in the hospital and it is these options that drive the efficient use of one of the most critical of hospital resources, available beds. It is important to note that bed use and lengths of stay are highly imperfect as measures of efficiency, and even worse as markers of outcomes. However, it can reflect something important about how hospitals are run, and more work is urgently needed to understand this better.

• For selected case types there was huge variability in the pathways taken through hospitals. The variability in pathways was in some cases linked to subsets of patients within diagnostic groups, which may reflect the limited depth of the information we had. Overall, the impression was that for a given diagnostic category, the range of specialties employed within a hospital spell was extremely variable.

• The rise in activity for these acute medical cases seems especially high; although partly explained by increasing numbers of short stays, this is not wholly the case. Over the past 5 years, hospitals seem to have squeezed 20% more patients through the available beds.

What’s the link between the case mix and the organisation of acute medical care?

In this section, we have looked in cross-section at two key issues: (1) the typology describing arrangements in AMUs and wards and (2) the differences in the skill mix of medical consultants.

As Chapter 4 notes, there are major differences in the way acute medical services are organised; however, we were unable to see any strong relationships between the case types treated and these organisational factors. There was a weak relationship between the size of the hospital and the case mix, but otherwise no relationships were found between case mix and the organisation of care. Whatever is driving models of care in hospitals, it is not the case mix that presents to them.

With regard to medical skill mix, there are large differences between hospitals. Our analyses looked at a number of selected specialist areas (cardiology, respiratory medicine, gastroenterology and elderly care) and found no relationships between the number of consultants in these areas and the number of cases admitted to diagnostic groups linked to these specialties. Once again, the conclusion is that whatever is driving differences in the mix of consultant specialties in a hospital, does not seem to be the acute medical workload.

Understanding hospital activity

Some of the analyses that we have undertaken have been new and innovative ways of looking at characterising current hospital activity, for example the use of linked data over time and breaking down constituent care episodes within a hospital spell; much of what we have carried out is the type of basic case-mix analyses that has been common for many years. This suggests that there is still untapped potential to exploit HES data further to look at models of service delivery and support better ways of understanding and reimbursing care. This is symptomatic of a wider problem that our ability to analyse data has not grown to match the massive expansion in the data available, and there is a shortfall in the analytical capability required to support decision-making.

Overview of approach

We explored the relationship between the hospital staffing model and the patient-level costs by using costed emergency admissions data from the 2015–16 HES database119 for the 43 smaller hospitals from our survey. We constructed a data set that contained spell-, patient- and hospital-level variables, including staffing model typology, and regressed the spell-level cost variable against the medical generalism typology variables plus covariates.

Data and variables

Hospital-level characteristics

Hospital-level characteristics are presented in Tables 20–22. In terms of the medical staffing typology, the modal categories were: AMU1 partial, AMU2 mixed, Ward1 closed and Ward2 specialist ward (see Table 20). The mean proportion of the medical staff in each hospital who were generalists was 0.412 (range 0.223–0.833; see Table 21). Most hospitals were in ‘urban – town’ locations, were a ‘small’ size, were rated as ‘requires improvement’ in terms of their CQC rating and fewer than half of their medical staff were generalists (see Table 22).

Spell-level analysis

The distribution of costs per spell are shown in Figure 24. Across the 707,054 costed spells, the mean cost per spell was £2445 (SD £2921) and median cost per spell was £1624 [interquartile range (IQR) £437–3414]. The 1st and 99th percentile costs per spell were £168 and £13,809, respectively. We explored the lowest cost spells in further detail. For spells with at least one overnight stay (507,655 observations), the mean cost per spell was £3170 (SD £3033) and the median cost was £2530 (IQR £1051–4161) (see Report Supplementary Material 9, Figure 3). For spells with no overnight stay (199,399 observations), the mean cost per spell was £592 (SD £1413) and the median cost was £396 (IQR £301–507). Four hospitals accounted for 70% of the 1% lowest cost spells and three case-mix groups accounted for over half of the 1% lowest cost spells (‘pain in throat and chest’, ‘sickle cell disorders’, ‘poisoning’; see Report Supplementary Material 9, Table 5).

FIGURE 24.

Distribution of cost per spell values (707,054 observations). Mean cost per spell: £2445 (SD £2921). Median cost per spell: £1624 (IQR £437–3414). The 10th and 90th percentile costs per spell: £302 and £5695. 5th and 95th percentile costs per spell: £246 and £7718. The 1st and 99th percentile costs per spell: £168 and £13,809.

Unadjusted mean (SD) and median (IQR) costs per spell by selected categorical variables are shown in Report Supplementary Material 9, Table 7. By inspection, unadjusted costs per spell do not seem to vary appreciably by medical staffing typology or the proportion of the medical staff in each hospital who were generalists. ‘Fracture of femur’ was the case-mix group with the highest median cost per spell, and ‘sickle cell disorders’ were the least costly. In addition, spells involving a procedure incurred higher median costs on average than those not involving a procedure; spells involving frail patients incurred higher costs than those not involving frail patients; spells for patients with secondary cancer incurred higher costs than spells for patients who did not have secondary cancer. There was considerable variation in the median cost per spell by trust (median values ranged from £361 to £3137).

Results of the regression models (models 1 and 2) are shown in Table 23. Based on the predictive margins and 95% CIs around these, there is no significant variation in cost per spell depending on whether the AMU is ‘open’, ‘closed’ or ‘partial’ (AMU1), whether the AMU is ‘acute physician dominant’, ‘specialist physician dominant’ or ‘mixed’ (AMU2) and whether the ward is a ‘general medical ward’ or ‘specialist ward’ (Ward2). There is some evidence that if the ward is ‘open’ it has significantly lower mean costs per spell than if it is ‘closed’ or ‘partial’ (Ward1), that is the 95% confidence interval (CI) around the predictive margin for the ‘open’ category does not overlap those of the other categories. Similar results are obtained for both model 1 and model 2. In model 2 we also controlled for the proportion of the medical staff in each hospital who were generalists > 0.5; adjusted costs for this variable were similar in the ‘no’ and ‘yes’ categories.

The following covariates were also included in models (but not reported):

• case-mix groups

• procedure flag

• frailty flag

• secondary cancer flag

• patient age

• patient gender

• patient ethnic group

• admission source

• urban/rural indicator for patient’s small area of residence

• whether or not the patient died in hospital

• population covered by hospital

• hospital distance to nearest A&E

• trust operating revenue

• approximate number of beds on main hospital site

• hospital A&E attendances 2015/16

• hospital emergency admissions 2015/16

• urban/rurality indicator

• hospital size

• the CQC rating.

Standard errors were adjusted for clustering at hospital level.

The association between selected covariates and the adjusted mean cost per spell are shown in Report Supplementary Material 9, Table 8. As with the unadjusted analyses, ‘fracture of femur’ is the highest cost spell and now ‘calculus of kidney or urinary tract’ is the lowest cost spell. As with the unadjusted figures (see above), spells involving a procedure were more costly than those that did not; spells involving frail patients were more costly than those that did not; spells involving patients with secondary cancer were more costly than those that did not. Adjusted costs per spell do not seem to vary by whether or not the patient died in the hospital, conditional on the other variables included in the models.

When low-costs spells were removed, there was some evidence that the significantly lower costs for patients treated on ‘open’ wards remain, and that the trends with regard to the medical staffing typology variables were similar to the main analysis described above (see Report Supplementary Material 8, Table 4).

Summary of main results

In this study, we investigated the relationship between medical staffing typology in small hospitals and health-care costs. We found that costs did not vary by characteristics of the AMU (whether it was open, closed or partial; whether it was APD, specialist physician dominant, or mixed), by whether the wards were general medical wards or specialist wards and by the proportion of medical staff in the hospital who were generalist. There was some evidence that costs were lower in open wards compared with those that were closed or partial, but the reasons for this finding were unclear. It could be due to chance, as there were only three hospitals from our sample of 43 that were in this category (see Table 20). It might also be a true effect and one hypothesis is that ‘open’ wards facilitate continuity of care in a way in the hospital that ‘closed’ or ‘partial’ wards do not. 142 Further research to understand the reasons for this finding would be beneficial.

Strengths and limitations

This study provides, to our knowledge, the first quantitative analysis exploring the association between models of medical staffing in small hospitals in the UK and health-care costs. The analysis was undertaken using a large, comprehensive data set accounting for hospital variation in resource utilisation and unit costs, including a range of covariates to control for differences in patient case mix between hospitals. The medical staffing typology variables were derived from comprehensive assessments undertaken as part of the present study.

There are several limitations. Our costs are based on trust-level unit costs and rely on hospital recording of resource use, which may be inaccurate. In addition, we cannot exclude the possibility that unobserved confounding factors might affect the observed relationship between staffing typology and spell-level costs.

Overview

The aim of the staff focus groups was to explore the lived experiences of health-care professionals who were working in each of the different models of care and their attitudes towards generalism. The original intention was to conduct focus groups for one of each of the different models of care expected to emerge from the typology. As there were more models of care than anticipated, sites for focus groups were chosen to explore the more common models of care, as well as those that appeared novel. Given that the interviews with staff conducted as part of case study visits contained substantial data regarding the topics of interest, the protocol was amended to include the interviews from all study sites in the analysis.

Methods

Participants for the focus groups were recruited via the site liaison officer (see Appendix 2). Informed consent was obtained. The focus groups were moderated by Louella Vaughan and Candace Imison; other team members recorded non-verbal communication and other interactions. Prompts were used to facilitate discussions around the staff’s experiences of providing care, especially in a smaller hospital, and the competing demands of generalist and specialist care and the boundaries between these. The groups were then presented with a simplified version of the main models of care and asked to discuss where they thought their hospital might fit in the typology and whether or not the perceived strengths and weaknesses of models of care were a property of the model or the hospital in which it was located. The focus group guide can be found in Report Supplementary Material 4.

Given that the staff interviews also contained rich material on the perceived strengths and weaknesses of the models, the protocol was changed to permit analysis of these, alongside the focus groups. Analysis was conducted as outlined in Chapter 4.

Overview of the findings

Two assumptions underpinned this WP: (1) that models and systems of care would heavily influence the lived experiences of staff and (2) that staff would have defined views of the strengths and weaknesses of the models and systems in which they worked. Instead, we found that staff found it difficult to discuss models and systems of care and instead tended to focus on the experience of working at a smaller hospital. Moreover, explorations of lived experiences were almost solely discussed in terms of the extent to which clinicians felt that their own workloads were manageable and whether or not their training equipped them for generalist working. In Chapter 4, we noted that staff were more able to relate to systems as sitting along a spectrum from ‘more general’ to ‘less general’; therefore, what follows here is framed around that spectrum, rather than relating findings to the typology. We also touch on views of the hyperstreaming model of care (see The hyperstreaming model of care), as this was viewed by staff as affecting the system and patients in unique ways.

Strengths of the generalist approach

The main strengths of a generalist approach were framed around its ability to meet the needs of the local population and improve patient care.

Weakness of the generalist approach

A weakness of generalist approaches was that they were seen as delaying the delivery of specialist care to the cohort of patients who had specific organ-based problems:

I think that the downside is that when you then have to refer to a specialist you can build in sometimes a 1 to 2 days of delay for that specialist input.

FGP

Strengths of the specialist approach

Specialist approaches were seen to support the best evidence-based care for specific conditions that are associated with higher short-term mortality, such as myocardial infarction or gastrointestinal bleeding:

So a patient with a respiratory problem will get a better outcome and a shorter length of stay and lower mortality and morbidity if looked after by a respiratory consultant . . . there has been good evidence to say, ‘the earlier you get your specialist involved . . . the better your outcome’.

Senior clinical manager

Patients were also noted at times to have preferences for specialist care and to have an expectation that this should be provided regardless of the size of the organisation:

It can be a bit devastating when you’re doing a consultant ward round and the patient just says ‘Well, I only want to see the cardiologist . . . I was promised I’d see a cardiologist’.

Senior clinical manager

Weaknesses of the specialist approach

Problems with a dichotomised approach

A recurring theme was that highly dichotomised approaches to care were in themselves unhelpful and that the key to improving care lay not in changing models of care along the lines of more generalist or more specialist, but rather in improving collaboration between consultants within a single organisation:

I would like to not see that model of a specialist sat over there somewhere behind 12 beds, ‘these are mine’. I would like to see them here, right next to me, go ‘yes, let’s have a look at this patient’.

Senior doctor

Thinking about generalist and specialist care in smaller hospitals . . . actually all hospitals . . . we need to work as one service rather than split services with barriers in between.

FGP

The hyperstreaming model of care

The hyperstreaming model of care involves having multiple services at the front door, such as ambulatory care and frailty units, which can provide the same functions of the first assessment and initial management of unselected patients with medical needs (see Overview of models of acute medical care in smaller hospitals in England). This emerged as a distinct model of care in the interviews and case studies, with unique strengths and weaknesses.

The main strength of this model was in decongesting the ED by moving patients elsewhere in the system. This was seen as essential in helping organisations to achieve the 4-hour target, as well as in reducing admissions and improving flow through the system. It was also seen as providing patients, particularly the elderly, with better care environments.

Staff, however, described the model as having a number of flaws. Substantial problems were created through the stretching of limited staff across highly dispersed geographic areas. The inherent complexity and the dispersed geography meant that it was difficult for managers to have an operational overview of the system and to respond to variations in demand. Being off the ‘ED clock’ in parallel assessment areas meant that patients frequently experienced delays in receiving care and that less urgency was described in the running of these units in comparison with EDs. Moves between different areas were more common than might be expected and often difficult to carry out, leading to fragmentation and discontinuity of care. In the case of patients who were moved between areas, the standard practice of reclerking patients with each transfer also led to substantial duplication of work across the system. It was rare for all of the system components to sit within the same managerial structures, with little thought given to how the individual areas might interact at the system level or how the burgeoning complexity ought to be managed.

The strengths and weaknesses of smaller hospitals

Most staff found it difficult, if not impossible, to divorce their experiences of models of care from the overall hospital context. This was particularly the case in smaller organisations, where generalism was seen to be the natural default model of care. Hence, strengths and weaknesses of models of care were more often than not coterminous with the strengths and weaknesses of the smaller hospital. Notably, strengths were almost always framed as properties of the hospital, whereas weaknesses, with the exception of a relative lack of resources, were referred to in the context of the increasing external expectations to deliver the same standards of quality and to meet the same requirements of bigger teaching hospitals, while negotiating a treacherous political landscape. It was striking the extent to which particular characteristics were viewed as both positive and negative features of a smaller hospital. For this reason, results will be presented by characteristic, rather than as lists of respective strengths and weaknesses.

Management of workload and experiences of training

Discussions of medical generalism were substantially influenced by the personal preferences and attitudes of clinicians, which appeared to be heavily informed by the extent to which clinicians felt that their own workloads were manageable and whether or not their training equipped them for generalist working.

In hospitals where the ‘horrendousness versus frequency’ of general medical working was able to be managed, clinicians were substantially more likely to express positive attitudes towards medical generalism or to tolerate generalist working, despite that they may experience greater satisfaction from specialty working:

We just accept that as a respiratory physician . . . I may have to accept an elderly patient with complex geriatric-type problems and I’ve just to suck that up and deal with it because it’s a DGH . . . I think that most people get that.

Senior clinical manager

Instances in which the insufficient workforce meant that consultants were struggling to meet the targets for specialty outpatients or the delivery of interventional services, or the medical take was considered to be busy, generalist work was considered an unnecessary burden with consultants actively seeking to withdraw. Some argued that generalist work was ‘deskilling’ and that the task of staying sufficiently up-to-date in two specialties was difficult given the service demands:

The problem is, you know, on-call weekends are so busy that it’s just exhausting, physically and mentally exhausting.

FGP

Current models of training were seen as being particularly important in shaping attitudes towards general medicine. Experiences and role-modelling appeared integral to the shaping of future attitudes, both positive and negative:

The thing I’d enjoyed most about my training was being on-call and looking after people . . . when they’re so unwell.

Senior clinical manager

Generally speaking, these days, the physicians I think are trained to see themselves as specialists . . . the trainees see the general medicine as drudgery that they have to do.

Senior doctor

Conclusions

Both generalist and specialist models of care were seen as having advantages and disadvantages. More generalist models were seen as a more natural fit for smaller organisations. Of note, perceptions of models of care were difficult to disentangle from staff’s views of working in smaller organisations. Positive aspects of smaller hospitals, such as an enjoyable working environment, were located as being properties of the organisation, whereas negative aspects were more often attributed to external factors. Perceptions of workload and experiences of training were also critical in informing doctors’ opinions of models of care.

Methods

Patient focus groups were held at the same hospitals as the staff focus groups. Recruitment differed for the patient focus groups, with advertising material being distributed across a number of channels, including local patient and Healthwatch groups. Participants were paid in-line with Involve guidelines143 Focus groups were conducted in the same way as the staff groups and informed consent was gained for all participants. Topic guides and other materials can be found in Report Supplementary Material 4. Participants were first asked to recount their experiences of specialist and/or generalist care; they were then presented with outlines of the more generalist and more specialist models of care and asked to discuss how this might lead to different outcomes for different groups of patients. A discussion of how care might be better delivered in smaller organisations, such as the use of networks, was facilitated. Analysis was conducted by a single team member (HC) using pattern-matching logic searching for convergent and divergent themes. This was then reviewed by two senior team members (LV and CI).

Description of participation

Five patient and carer focus groups were held. The typology of the organisations and the number of participants is given in Table 24.

Findings

Further details of the findings are presented in Appendix 5.

Patients expressed an overall preference for specialist care, although when pressed, participants could see that patients with complex needs may be better cared for by generalists and that specialist care may not be universally possible in the context of the smaller hospital. Moreover, it was considered that the overall quality of care given by local hospitals was more important than the actual model of care.

There was an overall perception that specialist care was the gold standard of treatment. However, the choice between a generalist and a specialist clinician was seen by some as a matter of personal preference, rather than as a choice between superior and inferior options. For others, the overall quality performance of the hospital seemed a more significant factor than the role profile of individual clinicians, and as long as they were satisfied with the hospital, they did not see a need to express a preference for a specialist or a generalist clinician. Some wished that the local hospital could deliver that standard of care universally, yet it was tacitly understood that expert care for ‘more obscure, difficult things’ may involve a transfer to a tertiary organisation:

Our age group, that grew up with the NHS just coming and the doctors were all powerful. If they told you to do something, you didn’t think ‘do I want a specialist or a generalist?’.

FGP

Some participants were reluctant to support the notion of generalist roles being a preferred option in secondary care. Only when pressed were certain respondents able to accept that there could be a benefit in seeing a generalist clinician when a patient had five or six things wrong under different specialties. Where respondents were able to acknowledge a role for care co-ordination by a generalist clinician, more than one respondent appeared to view that this co-ordinating role should belong to their general practitioner rather than a hospital doctor:

. . . it comes down to the GP [general practitioner] who knows you and knows you as a person – she should be able to discern whether it needs just solely a cardiologist or a general physician that would take it on a broader level.

FGP

However, geriatric care was the one area where most respondents saw a potential benefit of a generalist role in secondary care, particularly where patients had dementia. As one respondent put it:

If they’re presented with half a dozen different doctors, they usually don’t understand what it’s all about and it’s confusing. They get conflicting messages and it’s not good care. Particularly for the elderly I think you need one person.

FGP

However, these views were not universal, and one participant expressed a preference for specialist care, even when presented with the hypothetical situation of being an older patient with multiple complex conditions.

Conclusion

Overall, patients expressed preferences for specialist care, which was viewed as being the ‘gold standard’. However, most were prepared to accept that generalism was useful for patients with complex comorbid disease. The overall quality of care at any hospital was considered to be more important than the type of doctor delivering the care.

Introduction

We conducted a DCE144 to examine preferences between different workforce models in small hospitals. The aim was to quantify the preferences of hospital doctors, managers, patients and carers for different workforce models; the relative importance of different attributes of these models; and how the preferences vary between different stakeholders.

A DCE is a quantitative technique for eliciting individual preferences. Based on the assumptions that a service model can be described by its characteristics (attributes) and that individuals’ valuations of the model depend on the levels of these characteristics, it uncovers how individuals value selected attributes of a service model by asking them to state their choice over different hypothetical alternatives. This allows the attributes of a service to be evaluated, as well as the trade-off that individuals are willing to make between attributes of said service.

Ethics approval for this study was granted by the joint chairperson of the University College London Research Ethics Committee on 8 May 2018 (project ID 13187/001). International best practice guidelines for DCE were followed for study design and analysis. 145

Materials and methods

Respondents’ characteristics

In total, 214 responses were received between June and August 2018 (182 responses from hospital doctors, 23 from members of the public and nine from managers). DCE questions were completed in full by 173 hospital doctors, 20 members of the public and nine managers. It was not possible to estimate a response rate for each group because the survey was sent via multiple overlapping distribution routes using convenience sampling and snowball sampling techniques. The analysis was a complete-case analysis using these respondents’ answers only.

Of the 173 doctors, 78 were female (45.1%) compared with 17 (85.0%) of the patients and public sample (N = 20), and five (55.6%) of the sample of managers (N = 9). Mean ages were 47, 52 and 49 years, respectively. In total, 17 of the patients and public sample (85.0%) had a degree, nine (45.0%) patients were full-time employed at the time of the survey. In the samples of doctors (N = 173) and managers (N = 9), most respondents were working full time: 138 (79.8%) and 8 (88.9%), respectively. A total of 68 (39.3%) doctors classified themselves as ‘generalist with some specialism’, 13 (7.5%) as ‘pure generalist’, 15 (8.7%) as ‘pure specialist’ and 67 (38.7%) as ‘specialist with some generalism’. There were doctors from 17 different specialties, the biggest group being geriatric medicine (n = 84, 48.6%), followed by acute internal medicine (n = 15, 8.7%) and gastroenterology and hepatology (n = 10, 5.8%). In the managers’ sample (N = 9), there were representatives from finance (n = 1, 11.1%), human resources (n = 2, 22.2%), clinical management (n = 1, 11.1%) and operational management (n = 4, 44.4%).

Among the sample of doctors, 68 (39.3%) were resident in urban areas, compared with seven (35.0%) of the patients and public sample and two (22.2%) of the sample of managers. From all three groups, there was a mix of respondents across different regions of England, with only three (1.8%), two (10.0%) and one (11.1%) respondents from London among doctors, patients and public and managers, respectively. Some of the respondents had been admitted to a hospital in the past year: seven (4.0%) doctors and two (10.0%) of the patients and public sample.

Simple attribute ranking

The responses to the ranking question that was posed before the DCE questions were examined (Report Supplementary Material 9, Figure 3 graphically displays the responses for each of the three groups separately). Attributes are ranked by the likelihood of being selected as the ‘most important’ factor. The kappa statistic overall was 0.2234; it was 0.2236, 0.2593 and 0.1481 for doctors, patients and the public, and managers, respectively, representing ‘fair’ agreement among rankers in most cases. 148

Using this method of ranking, the type of doctors in the AMU is ranked highly in each sample, except for managers, and the total cost is consistently considered to be the least important factor by each group. We observe some differences between groups: patients and public appear to consider the role of doctors in the ED to be the second-most important, whereas hospital doctors consider the organisation of hospital wards to be the second-most important.

Discrete choice experiment analysis

The regression results for all groups combined and by subgroups are shown in Report Supplementary Material 9.

Given the small sample size of the managers’ subgroup (n = 9), we do not discuss the results for this subgroup any further and instead focus on the results for the doctors and the public and patients subgroups. Both the doctors and the patient and public subgroups prefer that in the ED, doctors diagnose and treat patients instead of referring them to specialists; they also prefer that the AMU is run by specialists compared with generalists, and that there are condition-specific hospital wards compared with mixed wards for follow-up care. All subgroups prefer lower levels of total cost per patient stay. There were no statistically significant differences in preferences between the doctors and the patients and public subgroups (all p > 0.1).

The MRS show the relative importance of each attribute by enabling the comparison of different attributes on a common scale. Overall, doctors’ willingness to pay was higher than the patients and public. Doctors were willing to pay £2469 for ED doctors to work on patients instead of triaging them to specialists, whereas patients and public were willing to pay £1123 for this. Doctors were willing to pay £4609 for AMUs to be run by specialists with expert knowledge in specific clinical areas, instead of specialists in acute/general medicine with broad knowledge in most clinical areas; patients and public were willing to pay £2827. Finally, doctors were willing to pay £6545 for patients with specific conditions to be treated on specialist wards instead of on mixed wards, in which patients with a variety of conditions are treated on the same ward; patients and public were willing to pay £2797 for this. These MRS values reflect the ranking of attributes that can be derived from combining the estimated coefficient for each attribute and the range of levels. According to this ranking, for doctors the most important attribute was organisation of hospital wards (range 1.44), followed by the type of doctors in the AMU (range 1.01) and the role of doctors in the ED (range 0.55). For patients and public, the most important attribute was the type of doctors in the AMU (range 1.66), followed up by the organisation of hospital wards (range 1.64) and, finally, the role of doctors in the ED (range 0.66). 149

Further subgroup analyses were conducted, stratifying the doctors’ sample by role (generalist/specialist), age (< 50 years/≥ 50 years old), area of residence (rural/urban) and specialty (geriatric medicine specialists and acute internal medicine specialists). The results are presented in Report Supplementary Material 9. We observed significant differences in the preferences between generalists and specialists, in particular the trade-off between the types of doctors in the AMU is higher for specialists, as well as the trade-off between different types of the organisation of the hospital ward. Doctors who were working in geriatric medicine preferred hospitals in which the AMU was run by specialists, whereas for those working in acute internal medicine, this attribute was not statistically important (this could be due to the small sample size, n = 15). There were significant differences in the preferences for the organisation of hospital wards by age: for doctors aged < 50 years it was more important to work in a hospital with specialist hospital wards than for doctors aged ≥ 50 years.

The probability that respondents would choose a hospital with particular attribute levels compared with a least preferred model of staffing is presented in Report Supplementary Material 9. Compared with a hospital in which doctors in the ED refer patients to specialists, there are acute/general medicine doctors in AMU, there are mixed-condition hospital wards and there is a cost of £4000 per patient stay, respondents were more likely to choose a hospital in which doctors diagnose and treat patients in the ED (holding the rest of the attributes constant), a hospital in which the AMU is run by doctors with expert knowledge in specific clinical areas (holding the rest of the attributes constant) and a hospital in which there are condition-specific hospital wards (holding the rest of the attributes constant). If a hospital had all of these changes in the attributes, at the expense of having the largest cost, £4000 per patient stay, defined as the ‘best case scenario’, the probability that respondents would prefer to work or to be admitted to this hospital reached 95% and 98%, respectively.

Summary of the main results

In this study we explored the preference of hospital doctors, patients and public and managers for different workforce models using a DCE. We found that responses were consistent between the hospital doctors and the patients and public; both prefer that in the ED doctors diagnose and treat patients compared with referring them to specialists, that the AMU is run by specialists rather than generalists, and that there are condition-specific hospital wards compared with mixed wards for follow-up care. Both groups prefer lower levels of total cost per patient stay, and there were no statistically significant differences in preferences between these two groups.

Strengths and limitations

This study provides, to our knowledge, the first evaluation of individual preferences with respect to workforce models in small hospitals. We used a DCE that allowed the evaluation of the attributes of a service that respondents would prefer to receive and the trade-offs they are willing to make between these attributes. The analysis was carried out with responses from over 200 participants. Content validity of the DCE was obtained by grounding the attributes and levels included in the experiment on careful construction of the typology conducted during this study. The questionnaire was carefully tested and revised during piloting.

We acknowledge several limitations, the main limitation being the limited sample size for the patient and public and manager subgroups. In particular, the number of responses from the latter was very small and, therefore, meaningful analyses were not possible. In addition, DCEs elicit hypothetical choices and, therefore, might lack external validity if individuals do not make the same choices in real-life situations. The representativeness of the sample responding to the questionnaire might be limited, as the generalisability of the findings depends on individuals elsewhere having similar preferences. Although the selection of attributes included in the DCE was carefully considered, we acknowledge that there might be other factors affected by workforce models that are not included in our analysis that may also be considered important for individuals; however, the number of attributes that can be included in a DCE exercise is limited.

Mortality rates

The measurement and use of mortality rates following a hospital admission has been the subject of considerable debate in the UK over the past few years. 150–152 Our position is that some examination of what happens to patients in terms of survival is essential in a modern health service, with survival being the most basic of outcome measures. However, to come to conclusions about what is driving differences in mortality rates between clinicians, services, hospitals and areas, additional evidence is required, and there is a danger of jumping to conclusions too quickly. In this study, our concerns were establishing whether or not we could observe crude differences in mortality between hospitals and whether or not these could be linked to patterns of consultant working.

The analysis of mortality considered both in-hospital mortality, as recorded in 2015/16 HES data for the smaller hospitals cohort, and out-of-hospital mortality. Overall, the analysis of mortality was limited by the time frame for which corresponding out-of-hospital mortality data were available (up to August 2015 only). Therefore, the analysis focused on 12-week survival of patients who were admitted between 1 April 2015 and 31 May 2015.

Observations

Within our 3-month follow-up period we studied 151,430 patients who were admitted to hospital, of whom 16,130 (10.7%) died within 3 months. Of these, 7587 patients (5.0%) died in the first week after admission, 38% of whom died in hospital (Table 25).

Deaths were concentrated in a relatively small number of case-mix groups: 15 groups accounted for 70% of all deaths. This is not surprising, as although there are many different health problems and underlying diseases that might exist in an individual, there are a few common reasons that predominate as the final cause of death, especially in the acute hospital setting. It is probably the case that the information derived from diagnostic codes in these final episodes of care does not capture the underlying clinical complexity of an individual. Therefore, caution should be used in interpreting the quality of care provided or the avoidablity of the final outcome.

The crude death rates showed a three-fold range from 5% to 15%, although the variability was less when standardising for differences in age and sex. This was further reduced when the diagnostic CMG grouping was used to standardise between trusts (see Report Supplementary Material 11). However, few of these differences appeared to be statistically different from the group averages at the 1% level. This could arise from a number of reasons, including different recording systems for the denominator and differences in the presenting patients that are not captured in the diagnostic standardisation. It is worth noting that routinely used hospital indicators, such as the HSMR or the SHMI, include more sophisticated risk adjustment.

Re-admissions

The occurrence of an emergency re-admission has been widely used for some time as an indication of the quality of hospital care. Some health systems, including the NHS, have looked at ways to penalise hospitals that show high rates of emergency re-admission, on the basis that these are a result of a lapse in the quality of care. However, there are a number of studies that suggest that the reasons for an emergency re-admission are more complex and that, at most, only about 30% of emergency re-admissions could be related to lapses in the quality of care in the index episode. 137 For example, re-admission may be a fact of life for people with some types of chronic disease, with the rates of emergency re-admission giving a small glimpse of longer disease trajectories.

We were interested in the degree of variation between case types and between hospitals in the observed re-admission rates to see whether or not these could be related to the approach to medical generalism. The analysis of re-admission was based on emergency re-admissions following one of 1.2 million admission spells in 2015/16. We looked at three different time periods following the discharge date:

1. re-admission within 30 days

2. re-admissions within 90 days

3. re-admissions within 365 days.

The proportion of episodes leading to re-admission was fairly high at 14% after 30 days and 29% after 365 days (Table 26). Given that these are acute medical cases, it is perhaps not surprising that these re-admission rates are slightly higher than those recorded for a wider set of index episodes; for example, for the same year, Friebel et al. 153 observed a 30-day re-admission rate of 12.7% based on all emergency admissions.

Re-admission rates at the trust level show some variation between hospitals. For example, 30-day re-admission rates were highest at the Homerton (25%), which is probably linked to the high incidence of sickle cell problems in the admissions. A few other trusts showed crude 30-day re-admission rates of 15–17%.

Links between outcome measures and hospital characteristics

To look at the relationships between these outcome measures and the characteristics of the way hospitals organised acute medical work, we used the typologies that were described in Chapter 7. In this section, the mortality was derived from in-hospital deaths; this was necessary to increase the number of cases that could be examined.

In developing the models, we sought to test the association between an outcome model and hospital subtypes, standardising for a range of covariates that could influence the outcome. These included:

• Information about the patients, including case-mix markers of frailty.

• Characteristics of the provider hospital, including size and CQC rating.

• Characteristics of the local area, including a measure of deprivation, markers of urban/rural areas and the distance to hospital.

The models were developed in parallel with the economic models described in Chapter 6, and used logistic regression throughout with a wide range of covariates. Standard errors were adjusted for clustering at the hospital level. For each outcome we present two models, one including percentage of medical staff who are generalist > 0.5 and one not including it [the proportion (%) of medical staff who are generalist are not available for all hospitals, so the sample size is smaller].

Results

Table 27 summarises the model results when looking at in-hospital mortality. There was little difference in the adjusted probability of in-hospital mortality for all of the typology subtypes, and the 95% CIs overlap. Therefore, we conclude that there is no detectable association between the models used to organise acute medical care and the observed mortality rates, when a series of additional covariates are included. Both the models that include and the models that exclude the proportion of medical staff who were generalist produced the same results.

We also note that although a lot of predictor variables were included in the models, the overall adjusted R² value was 0.202, which indicates that only 20% of the variability in mortality was accounted for by these variables. Therefore, there is a lot of variation linked to other factors. For each of the typology categories, the adjusted probabilities of in-hospital CIs overlap as shown in Table 27.

The same approach was used to look at emergency re-admission rates for 30, 90 and 365 days. Table 28 shows the results for the model based on 30-day re-admissions. Once again, there were no discernible differences between hospital subtypes and re-admission rates at 30 days. The pseudo R² value for this model was lower than that for mortality, which indicates that there are higher levels of uncertainty about what drives differences in re-admission rates.

Similar results were observed for models looking at 90-day and 365-day re-admissions (not shown).

Covariates that were also included in models but not reported were the:

• case-mix groups

• procedure flags

• frailty flags

• secondary cancer flags

• patient age

• patient gender

• patient ethnic group

• admission source

• urban/rural indicator for patient’s small area of residence

• population covered by hospital

• hospital distance to nearest ED

• trust operating revenue

• approximate number of beds on main hospital site

• hospital ED attendances 2015–16

• hospital emergency admissions 2015–16

• urban/rurality indicator

• hospital size

• CQC rating.

Standard errors were adjusted for clustering at hospital level.

Discussion

The analyses covered in this section have been exploratory and were designed to test for any large-scale impacts that might be present and help to inform more detailed modelling. We noted that:

• The severity of illness that was present on these cases that meant a relatively large proportion of these patients died within 3 months, and even within the first week.

• Re-admission was a common phenomenon and for many patients acute medical episodes are just one element of a much longer and more complex disease trajectory.

• Although there can be some large differences in crude mortality rates between trusts, many of these differences are reduced using simple standardisation methods. There are, however, some residual values that look atypical and would merit further investigation.

• For the acute medical caseload, a relatively small number of the diagnostic categories can be used to capture the majority of deaths. This suggests that comparisons between providers using this standardisation should be treated with some cautions, as there will be other factors, not captured in the information we have, that may influence these local rates.

• Despite building complex models we could not detect any difference between the way acute medical services were organised and the outcomes in terms of in-hospital mortality rates or emergency re-admission after 30, 90 or 365 days.

The biggest limitation in our analysis, and our greatest challenge, has been obtaining permission to use the relevant data sets, despite that our organisation held these data for other projects. Throughout the course of this project, the national requirements on information governance have been subject to constant flux that has meant that it proved impossible, despite our very best endeavours, to obtain the data that we wanted. For this study it meant that the information on out-of-hospital mortality was limited to a relatively short window. Therefore, for some analyses, we restricted ourselves to in-hospital mortality only.

The implication of our findings is that the organisation of acute medical services does not have a dramatic impact on some basic outcome measures, such as mortality and re-admissions. Any claims otherwise should be treated with caution. It also suggests that explanation of the benefits or disbenefits need to consider other factors, such as process measures for staff continuity, or explore other outcomes.

Generalist and specialist models of care

At its conception, the key underpinning assumptions of this study were that smaller hospitals used a limited number of clearly defined models of care, which could be classified along a spectrum of generalist to specialist and would be amenable to a systems analysis approach. Instead, we found that no two hospitals operated identical systems of care and that systems were resistant to classification and comparison. This necessitated a switch in approach, so instead we concentrated on the two main locations of medical general working: the AMU and the downstream medical wards. We categorised AMUs according to patterns of consultant working (APD, SpD or mixed), whether or not units were heavily boundaried and the subsequent patterns of patient ‘ownership’ by medical teams (closed, open or partial). Downstream wards were similarly classified by their degree of closed/openness, as well as how hospitals labelled wards and distributed patients across them (general medical wards, specialty wards).

Classification of the 48 hospitals for which we had sufficient data revealed that there were 25 different models of care in operation. Correlation of the organisational profiles with the typology found that there were more generalist models of care in smaller organisations (< 300 beds). Beyond this, there were no associations found between models of care and other factors, such as hospital size, number of emergency admissions or hospital location. Patterns of working in AMUs varied more than patterns of working on downstream wards.

We further found that there were two trends in terms of arranging care at the front door. The ‘acute hub’ model looked to co-locate services with a primary assessment function into a single ‘hub’, whereas the ‘hyperstreaming’ model saw the bulk of the primary assessment work move out of the ED and into a series of parallel locations, such as ambulatory care, frailty units and other assessment units.

The telephone interviews and case study visits found that most models of care, however defined or arranged, were highly unstable and in an almost constant state of flux, with usual processes of care frequently breaking down in the face of internal and external pressures.

Patient perspective

Review of over 10 million acute medical emergency episodes of care at 69 NHS trusts over 6 years confirmed a rising demand in smaller hospitals. Overall, there was a 26% increase in the total number of episodes between 2012–13 and 2017–18, with an average yearly increase of 5%.

Analysis of the case mix for the same patient cohort found that most patients had conditions that could be considered to be largely amenable to generalist care. There was a high concentration of cases from relatively few diagnostic groups, with over 80% of episodes and 75% of bed-days falling into 31 groups; 31% of patients had a frailty indicator. Overall, there were more similarities than differences with regard to case mix across hospitals, the exception being smaller hospitals in urban areas.

There was no correlation between the case mix of hospitals and the skill mix of the medical staff within them. Telephone interviews and the case studies revealed that although attempts were made to meet patient need at the service level, such as the introduction of frailty teams or units, care was rarely orientated around patient need at the hospital level.

The DCE found that patients have a weak preference for specialist care. This was confirmed by the patient and carer focus groups, although participants indicated that the quality of care at their local hospital was more important than what type of consultants provided care, and acknowledged that smaller hospitals cannot necessarily be expected to provide highly specialist services.

Service perspective

Smaller hospitals have responded to the challenge of rising admissions by improving flow through their systems. Between 2012–13 and 2016–17 the number of episodes of care rose by 26% but the number of occupied bed-days rose by only 5%. The extra activity was accommodated by a reduction in the length of stay and an increase in short-stay cases. There was a three-fold difference between hospitals in terms of the shortest (2.8 days) and longest (8.41 days) average LOS.

Neither the model of care, as defined by the typology, nor the case mix accounted for the variability in the average LOS between hospitals (no associations were significant at p < 0.5). There were also no detectable associations between the models of care and the observed in-patient mortality rates or re-admissions at 30, 90 and 365 days. These findings imply that other factors acting at the hospital level are the critical determinants of outcomes. There were no significant differences in the costs of the models of care.

Telephone and case study interviews found that models of care were shaped by complex constellations of factors that interacted at multiple levels. Predominant among these were hospital size and location; the ability to retain and recruit staff; organisational culture and leadership; and often competing policy imperatives. Models tended to be highly reactive, rather than proactive. Where there was evidence of strategic planning, this tended to be driven by single individuals operating at the service level who were hindered or enabled by other factors at the hospital level. Hospitals with the most stable and mature models of care were those that planned at higher levels, had clear strategic aims to meet local community need and fostered collaborative working environments.

The work on pathways found that the journeys of patients with similar conditions through the emergency and acute medicine system varied markedly both within and between organisations. The telephone interviews and case study visits corroborated these findings. The interviews also found that systems were frequently overwhelmed by demand, leading to differences between models as conceived and models in action. It was also found that staff frequently adjusted the pathways of patient care according to complex calculations that were based on instinct, experience and intimate knowledge of their hospital system.

Both more generalist and more specialist models of care were viewed as having significant advantages and disadvantages by hospital staff. Overall, more generalist models were thought to be more appropriate models of care for smaller organisations. Staff viewed the size of the hospital as being a more powerful explanatory category for almost all aspects of service delivery and outcomes than the model of care per se.

Professional perspective

The qualitative work found that the preference of doctors for generalist or specialist working was contingent on the hospital environment and the heaviness of the burden of generalist working. Experiences of training were also important, with preferences for generalism being associated with longer periods of training; younger specialists frequently felt poorly equipped to care for older complex patients.

Although the majority of consultants who were interviewed accepted generalist working as an inevitable consequence of working in a smaller hospital, we found patterns of behaviours, particularly ‘flight from the front door’ and ‘fortressing’, whereby consultants actively sought to limit their exposure to generalist working and regulate their own workloads. These behaviours were viewed as highly destabilising and their presence was correlated with more negative views of organisational culture and leadership.

The preferences for withdrawal from participation in front-door services and for working only as a specialist were expressed more strongly in the DCE.

Smaller hospitals

Although this study was constructed primarily as an exploration of aspects of care within hospitals, it rapidly emerged that systems of care could not be divorced from their contexts. Staff considered their smaller hospitals to be categorically distinct from larger teaching organisations and ‘smallness’ to be the most powerful explanatory category of their experiences of delivering care.

The idea that smaller hospitals might be categorically distinct was supported by the explorations of case mix, which found that there were more similarities than differences across all 69 organisations. The exception to this was the few smaller hospitals in urban areas.

Most smaller hospitals viewed themselves as being under genuine existential threat, being understaffed, being under-resourced and almost constantly battered by ‘system shocks’, to the point where positive change was extremely difficult to enact. Although we did not measure resilience or any other measure of organisational climate, hospitals with skilled, stable leadership teams who had invested long term in cultural change were associated with environments that staff viewed as desirable to work in and more stable systems of care.

Concepts of medical generalism

No clear or consistent definitions of medical generalism were found in the literature. Instead, the term emerged as slippery and frequently negatively defined, for example as ‘the work left undone’ or a ‘gap’ to be bridged. Moreover, there was a major schism between the constructions of medical generalism as a philosophy of care or a discipline within medical practice, and the way generalist practice was enacted, especially where generalism was viewed as the solution to a particular problem within health care.

These inconsistencies and gaps were reflected in what we saw in practice. Although there was a broad consensus about which patients should belong to specialists, we did not find any consistent definitions in use nor was there any real agreement about the boundaries between specialism and generalism. Instead, there was often a substantial distance between what clinicians and managers thought systems of care should look like and the enaction of these systems of care. Within systems, there was evidence of the boundaries of medical working being almost constantly negotiated and renegotiated.

Doctors enjoyed generalist working more in systems that were geared to meeting patient need, regardless of the location of the patient or the label allocated to them, and where work was shared in ways that were transparent, equitable and collaborative. Some clinicians viewed these ways of working as being the definition of ‘true generalism’.

The mapping of emergency and acute medical services

To the knowledge of the authors, this was the first systematic attempt to map and compare emergency and acute services for medically unwell patients across the whole patient pathway in English hospitals.

This study revealed that emergency and acute medicine services were highly complex. National surveys of the acute care of specific conditions154,155 and research on emergency services125 have noted marked variability in the structures of acute care in England and the journeys followed by patients. However, the increasing complexity of front-door services in England has been most clearly highlighted by the annual Society for Acute Medical Benchmarking Audit (SAMBA),156 which also found an increasing array of parallel front-door services, with patients following highly fragmented pathways through systems, with multiple handovers, before arriving at a definitive place of care.

This contrasts with Reid’s survey on Scottish hospitals,101 where it was noted that although 65.5% of hospitals had ‘acute medical work’ happening outside the AMU, the only other locations for work were the ED and the downstream medical wards. By contrast, we found that 100% of hospitals had some form of the same work taking place outside the AMU and the ED, with most hospitals providing multiple parallel geographic spaces for the primary assessment of patients. This underlines the impact of central policy drivers that have sought to move acute work out of the ED in English hospitals.

The emergence of models of care

We expected that the establishment of models of care with hospitals would follow logical pathways, with a conscious choice of a coherent model thought to be suited to the hospital, planned execution and subsequent evaluation of implementation. Instead, we found that models were heavily contingent, driven by policy imperatives, hastily executed and rarely systematically evaluated. They were frequently cobbled together and the result of negotiations and compromises, which were based on available staff and resources. Models that were presented initially in the interviews and case studies as being stable were revealed as actually being fluid, dynamic and frequently overwhelmed by the combination of demand and inherent fragility. These findings point to the tension between the presentation of change within health-care systems as being ordered and planned, and the actuality of systems as being emergent and dependent. 157 Moreover, models of care, in many cases, were so tightly bound to their contexts that staff were unable to articulate models of care as independent artefacts.

This is particularly problematic in the NHS, given the implicit assumptions that components of care, such as EDs, AMUs and ambulatory care, are independent of microlevel factors and an increasingly top-down driven model of change. Moreover, although individual studies of AMUs and ambulatory care units have been found to produce better outcomes for certain groups of patients, the evidence base remains limited and there has been very little or no study of their impact at the system level. 76,97,158 The implication is that there is a major gap between the theoretical understandings of context, and its importance in the diffusion of innovation, and the non-linearity of complex systems159,160 and the reality of how this is conceived and enacted within the NHS at multiple levels.

The other gap that emerged from the interviews was the lack of capacity within smaller organisations for the rapid change that is demanded of them at almost every level: financial, managerial, geographical, staffing and other resources. Such pressures led to organisations ‘making do’ to meet external expectations, rather than embarking on programmes of sustained change.

Patterns of presentations and admission to hospitals

Patterns of presentations of acute disease across first-world countries are highly similar, with a predominance of cardiovascular and respiratory disease. 161–163 However, there have been few systematic attempts to identify how patient care might be better organised at the system level according to disease state. Kellett and Deane’s164 efforts to categorise medical admissions to a single hospital in Ireland similarly found that most of the acute work could be accounted for by relatively few conditions and was amenable to a generalist approach to care, with near complete overlap of the top 20 conditions between their cohort and ours. 164 The similarities between these two studies, despite geographical and temporal distance, suggest that acute medical work in smaller hospitals is predictable and that services should be able to be better matched to demand.

Despite similarities in patterns of presentation, patterns of admission to hospital were found to have changed over time, with increasing numbers of short-stay presentations, and found to exhibit marked variation between organisations. This is in line with other studies that have suggested that thresholds for admission have fallen in England over time165 and that admission appears contingent on other factors, such as the structure of front-door hospital services and consultant preference. 125,166

Association between outcomes and models of care and case mix

Variations in the quality of outcomes are a key concern to health-care professionals and policy-makers alike. Despite decades of research, consensus has not yet been unreached around the boundaries between ‘warranted’ and ‘unwarranted’ variation, and to what extent unwarranted variation is driven by differences in the underlying demographics of patients, by differences between organisations and their delivery of services or by the performance of individual clinicians. 167–169

A key finding of this study was that variations in LOS and other outcomes were not explained by either the typology or the case mix, implying that other hospital-level factors were responsible. We first discuss aspects of case mix and models of care with regard to other literature and the implications of our findings, and second explore the other potential explanatory factors.

Supporting and sustaining smaller hospitals

Interviewees were virtually united in the view that smaller organisations felt different from their larger counterparts. Although it was not a principal focus of the study, a number of findings are likely to be immediately useful in improving care in smaller hospitals, some of which have already been presented. 67

Training of doctors

The training of doctors and the extent to which they felt comfortable managing the current patient case mix emerged as an important theme in the interviews. There was virtually a generational split between the older doctors who had experienced longer periods of training, often with considerable periods in smaller hospitals, and the younger consultants who often had only 2 years of training in general (internal) medicine alongside specialist training, mostly in the context of larger teaching organisations. Moreover, although consultants of all ages spoke of the pressures of the acute take, it was the complex, older patients on the downstream wards that the younger consultants struggled with most. Both the stress of the acute take and the feelings of being unable to deliver adequate care were major factors in driving flight into specialty and the withdrawal of labour across the whole emergency and acute medicine pathway.

Surveys of specialty trainees have noted the perceived inadequacy of GIM training in its current form,221 something that the Shape of Training review222 proposed to rectify. However, the proposed changes have focused on broadening the trainees’ exposure to patients with multimorbid illness early in training (Senior House Officer equivalent) and on acute illnesses. 223 This does not seem fully to address the management of more complex patients across the whole of their hospital stay. Moreover, this training in the care of older, multimorbid patients is likely to be delivered by consultants who feel less well-equipped to provide care themselves for this group than the previous generation. Alongside this has been the trend to move trainees out of smaller organisations, despite research suggesting that doctors prefer to work in the types of organisations in which they train. The logical outcome of these two policies is that workforce shortages in smaller hospitals are likely to worsen rather than improve.

Meeting patient and carer expectations

We found that patients and carers tend to view specialist types of care as being the ‘gold standard’. However, at the same time, there were acknowledgements that smaller hospitals could not be expected to deliver the same care as larger hospitals and that the overall quality of care provided by a hospital was more important than the type of doctor or service delivering the care.

The fact that these findings should contain a degree of cognitive dissonance is congruent with the literature. A study224 of patient experience on AMUs in England found little variance between organisations, with perceptions of care being associated more with patient-level factors. From the perspective of generalist versus specialist care, the limited research has found that patients are happy with generalist care, so long as access to specialists, when required, is quick and easy. 225 Studies of preferences for the provision of rural care echo this: patients strongly prefer services to be local, so long as it is backed up by tertiary care. 226–228 Watson et al. ’s study229 of preferences of a patient population in rural Scotland found that patients wished for cutting-edge services to be provided locally.

These findings suggest that patients may find generalist models of care acceptable, despite the overall preference for specialist services, so long as they are constructed to meet patient needs, are of high quality and are backed by robust networked services.

Implications of sampling strategy

The Monitor report’s88 definition of smaller organisations was based on operating revenues of trusts, not hospitals. We were aware that a different sample would have emerged if we had used hospital size (either by beds or by ED attendances) as the inclusion criteria. We are also aware that we achieved a response rate of only 70% for the telephone interviews and purposively sampled for the case study visits, which introduced the possibility of selection bias at several points in the study. The strengths of this approach were that we looked at a broader range of organisations by size and location than if we had used other criteria. Even within this sample, we still found more similarities than differences between the case mix, and the qualitative work reached saturation in terms of themes, although we still may have missed some of the more unusual examples.

Dynamic nature of models of care

The original study design had assumed that systems were in evolution rather than revolution, with the quantitative comparisons between models of care to encompass a period of 5–6 years. Where revolution did exist, we had assumed that change would be easy to locate and amenable to capture using interrupted time series analysis. Instead, we found models of care that were highly unstable and changing rapidly, with staff frequently unable to pinpoint the exact dates of change. To reduce the likelihood of hospitals switching categories within the typology entirely, we limited the time frame to 1–2 years for some analyses. However, we are aware of at least one organisation switching models within this time frame, and the possibility of others, which limits the reliability and reproducibility of the results.

A priori assumptions about medical generalism

The study was predicated on a number of a priori assumptions about medical generalism and systems of medical care. We had assumed that the demarcations between generalism and specialism would be clearer and more universally agreed, with a limited number of well-defined models of care in operation, meaning that a systems approach would offer a mechanism for understanding and comparing organisations and their structures. The structure of the study, with its iterative approach to the emerging data, meant that we were able to adopt different frameworks for analysis in response to this. This does, however, mean that our conclusions are not definitive, but should rather be considered the starting point for further work.

Narrow scope of analysis

The framing of this study around medical generalism has led to a focus on doctors, particularly at consultant level, and the nature of doctoring. The original intention had been to explore generalism at the system level, which would have naturally encompassed other types of staff, particularly nurses and allied health staff, who are essential to the delivery of high-quality care. These voices were captured through the course of the study, although the analysis has been necessarily limited, owing to the complexity and scope of the material. Even when discussing issues around doctors and doctoring, the focus on generalism and smaller hospitals means that we may have missed other important factors, especially with regard to what influences the shape of work.

No comparisons with larger hospitals

The design of this study did not allow for comparisons between smaller organisations and larger organisations. Although the evidence suggests that smaller hospitals are distinct, this cannot be confirmed without a similar exploration of urban teaching organisations.

Interview focus of the case studies

The case studies did not include extensive non-participant observation or immersion and so we relied heavily on direct accounts and experiences of the systems. To have included this type of research would have been time-consuming and would have reduced the number of case studies that could have been completed in the time that was available. We tried to compensate for this by using the ‘walk around’ method of mapping the patient journey, talking to a broad range of staff across multiple departments and using patient volunteers to provide an additional perspective.

Lack of patient voice

This study was planned to have extensive PPI input at every stage. However, we found it difficult to recruit for the patient and carer focus groups, with some groups being very small with one or two participants only, and the open space event was cancelled because of a lack of interest. This does leave a sense of not having adequately captured patient voices. However, the analysis of the focus groups rapidly reached saturation of themes and there was broad agreement across organisations, both with the results of the DCE and with the general literature. This reassured us that while we may have missed unique perspectives, the overall views presented are very likely representative.

Representativeness of the discrete choice experiment and the response rate

The DCEs are based on preselected attributes and elicit hypothetical responses. We acknowledge that there may have been other factors affecting workforce models that were not included in the DCE, which limits the external validity in real life situations. The number of responses was also limited, meaning that the findings may not be generalisable to larger groups of doctors, patients or managers.

Uncertainty around economic costs

The hospital-level economic analysis was severely limited by the lack of data on hospital staffing costs, while the small number of observations meant that adjusted analyses controlling for observed characteristics between hospitals were not possible. For the spell-level analyses, costs were based on trust-level costs that may be inaccurate.

Implications for health care

1. National strategies to identify and remove unwarranted variation in the surgical and procedural realm have been successful in reducing cost and improving efficiency. 233 However, such strategies rely on an established evidence base and shared understanding of best practices. In this study, models were emergent and dependent on complex constellations of local factors, rather than driven by an evidence base or theoretical considerations. No model of care emerged as being better than any other, with outcomes appearing dependent on other hospital-level factors and the suggestion that the size and staffing of hospitals may affect their ability to implement standardised models. Furthermore, the ‘system shock’ of external political imperatives frequently undermined the ability of hospitals to enact change. This suggests that a ‘one size fits all’ approach may be less appropriate to guide best practice with regard to emergency and acute medicine in smaller hospitals. Empowering and enabling organisations to understand how to adapt their staff model to the local case mix and the requirements imposed by how their local microsystems operate, may be a more productive approach to improving patient care.

2. Overall, the cohort of smaller hospitals shared more commonalities than differences, with both organisations and staff seeing themselves as having an identity distinct from their larger counterparts. This suggests that treating smaller hospitals as a separate, unique category within the NHS and supporting their specific needs may be an effective strategy in ensuring their survival and improving patient care.

3. There was substantial variability in pathways of care for specific conditions and their relationships with LOS, both within and between organisations. This suggests that understanding pathways and improving flow for specific groups of patients may be more efficient than a wholescale change in the model of care.

4. Workforce gaps at the middle grade and consultant level were found to be a major problem across nearly all organisations. This suggests that there is a particular need for best practices around recruitment and retention to be identified and disseminated, with hospitals encouraged to shift to strategically planning their consultant workforce at the hospital level. With regard to the training and distribution of middle-grade doctors, the study suggests the imbalances in workforce distribution and improving training experiences outside major urban areas may need to be readdressed to ensure that there is an adequate workforce in both the short term and the long term.

5. The remit of the Shape of Training review7 was to look at potential reforms to the structure of postgraduate medical education and training across England to meet the needs of the increasingly older and multimorbid patient population. This includes management of patients across traditional inpatient–outpatient boundaries in integrated care systems. The evidence supports the contention that the current generation of consultants is not as well equipped as it ought to be in dealing with a generalist patient population. However, the finding that preferences for generalism were associated with long periods of training suggests that Shape of Training’s7 recommendations may still not be sufficient to produce a consultant workforce not only capable of, but willing to care for an ageing population.

6. ‘Fortressing’ and ‘flight behaviours’, whereby consultants actively sought to limit generalist working and regulate their workloads, emerged as a significant issue in this study, undermining the stability of emergency and acute medicine systems and creating artificial workforce shortages. The creation of more collaborative working environments, with the transparent and equitable sharing of work across the consultant body, is likely to reduce incentives for consultants to withdraw from generalist working. Alternative ways of providing consultant cover out-of-hours for relatively rare interventions, such as regional networks, need to be urgently explored.

7. The advent of integrated care systems offers an opportunity for the reimagining of the organisation of care across regions and between multiple providers. Much of the work to date has explored how specialists might better support community-based services. 8 This study suggests that rethinking how specialists might better support generalists in smaller hospitals and reshaping the boundaries between generalists and specialists might be equally useful. The same understanding of the local case mix, the demand for acute services and the organisational mechanics required for improving care at the hospital level, could also provide a platform for innovations in models of care that are based around skill mix, multisite working, networks for the delivery of specialist advice and interventional care, and standardising and smoothing pathways between providers and across systems.

Recommendations for further research

1. Mapping and understanding models of emergency and acute care across the whole of the NHS, including an exploration of the relationship of models of care with patient and hospital-level outcomes, and the models of care and the organisations in which they sit.

2. Studies on the deployment of the medical workforce across the NHS as a whole and within individual hospitals, and the relationships between workforce, models of care and measures of hospital culture. This should include an exploration of how skills gaps in doctoring impact on and are compensated for by others, particularly nursing and allied health staff.

3. Development of more sophisticated methodologies, particularly those of a rapid nature, to deal with the marked variation seen in health-care systems and testing of specific organisational and contextual factors that help or hinder system improvements.

4. Applied research into how more refined case-mix analyses may allow organisations to make better choices around the configurations of services, model future services and monitor subsequent change.

5. Better understanding of medical work, the decisions around its distribution, how this translates into differences in patient journeys and impacts on outcomes for both the patients and the system.

6. Analysis of what skills, experience and expertise are required to provide excellent care for complex comorbid patients and how the undergraduate and postgraduate curricula should be structured to ensure that the next generation of doctors is fully equipped to meet the challenges of the changing patient need.

7. Further research into what constitutes satisfying work for doctors and other health-care professionals, the relationship between work environment and both positive and negative physician behaviours and how hospitals environments can be shaped to support this.

8. Better understanding of how smaller hospitals might meet the needs of their local populations, recruit and retain appropriate staff and be supported to provide high-quality care.

9. Better understanding and further exploration of the costs attached to different models of care, particularly the nuances around the patterns of deployment of consultant staff.

Final conclusions

This study sought to examine the hypothesis that medical generalism may provide a way to deal with the rising numbers of patients who are complex and/or have multiple comorbidities, particularly in smaller hospitals. We found that the case mix of smaller hospitals was dominated by patients, with presentations likely to be amenable to generalist approaches to care. Although generalist models of care appeared a more natural fit for smaller organisations, there was no evidence in our study to suggest that any particular model of care produced better outcomes for patients. The willingness of medical staff to work in generalist ways was dependent on their training and their lived experiences of working within organisations. This suggests that matching hospital staff to meet patient need and the creation of more collaborative working environments is more likely to improve care than changing models. Therefore, although medical generalism may provide a way of caring for an ageing population, the delivery of high-quality care will be possible only if the right conditions are created to support generalist working, both within individual organisations and across the NHS as a whole.

Study Steering Committee

We would like to thank David Steel for his most excellent chairing of the Study Steering Committee and the support he has offered to the study team over the past 3 years.

We would like to thank the members of the Study Steering Committee who offered support and advice and assisted with the evaluation and interpretation of the content and progress of the study: Mrs. Marilyn Frampton (Governor, Kingston Hospital NHS Foundation Trust), Professor Karen Bloor (University of York), Professor Mary Boulton (Oxford Brookes University), Professor Tim Evans (NHS England), Professor Hugh Gravelle (University of York), Dr Jonathan Howes (Yeovil District Hospital), Dr Alasdair MacDonald (Launceston General Hospital), Dr Cliff Mann (NHS England), Professor Jane Sandall (King’s College London) and Dr David Ward (Hinchingbrooke Hospital).

We would like to note the sad passing of Professor Tim Evans in November 2018, shortly after the submission of this manuscript. Professor Evans' forensic eye was critical to the shaping of aspects of this work from its inception to completion. For this, and other acts of kindness, we are deeply grateful.

Patient and public involvement

We would like to thank our patient volunteers for their contributions to the conduct and interpretation of the case studies: Kenneth Greenfield, Joanna Parker, Frank Rust, Linda Samuels, Keith Spurr, Jenny Stiling and Malcolm Watson.

We would like to thank David Steel and Marilyn Frampton, also members of the Study Steering Committee, for providing a public and patient perspective on the design, findings and interpretation of the study.

We would like to thank Rachel Matthews (Patient and Public Engagement and Involvement Theme Lead, Northwest London CLAHRC), Rachel Downing and Francine Beadsworth (both formerly of the RCP) for their assistance with planning and executing the PPI engagement with the patient volunteers.

Other Nuffield Trust staff

We would like to thank the following Nuffield Trust staff.

Nigel Edwards (Chief Executive Officer) for his support of this project from its inception to completion.

April McMullen (former Team Assistant) for her administrative assistance with the qualitative portions of the study and ensuring that the case study visits were well organised.

Ben Willbond (Deputy Director, Finance and IT) for his oversight of the finances and data governance.

Rowan Dennison (Editorial and Content Manager), Meilir Jones (Comment Editor) and Paul Kirkup (Digital Officer) for their extensive assistance with the preparation of the final document and other publications relating to this study.

Leonora Merry (Deputy Director of Communications) and the rest of the Communications Team for their support in the production of and promotion of other publications relating to this study.

Eilis Keeble (Research Analyst), Theo Georghiou (Senior Research Analyst) and Sarah Scobie (Deputy Director of Research) for supporting the quantitative work.

Sophie Castle-Clarke (Senior Fellow) for support with the production of the final document.

Rachel Bleach for ongoing administrative support.

The New Cavendish Group

We would like to thank the members of The New Cavendish Group for their interest in and support of this project.

Participation in the study

We would like to thank all those who participated in the study. This study would not have been possible without your willingness to share your experiences and opinions openly and honestly. We are particularly appreciative of those who, despite working in highly pressured environments, assisted with the organisation and conduct of the case study visits.

Sampling strategy

The original intention was to undertake 2–3 case studies of each of the 4–5 different models of care expected to emerge from the typology. However, sites were primarily chosen because they were exemplars of the different AMU models. As there were fewer ward models, this approach captured most of them; only one site was selected explicitly on this basis. We sought to achieve appropriate representation of trusts according to geographical distribution (urban vs. rural), size (small vs. ‘smallest’) and case-mix.

To better explore the contribution of the external landscape and case-mix to the model development, we theorised that neighbouring organisations would probably be very alike in terms of case mix and probably similar in terms of regional political and other external pressures. We then included in our sample three pairs of adjacent trusts (one inner urban, one outer-urban and one rural), where each hospital employed different models of care.

Site recruitment

Recruitment took place in three rounds: after each round of the telephone interviews and then a final round after one hospital had declined to participate and another had failed to respond to the invitation. Initial invitations to participate in the case studies were made through either CEOs or medical directors (in cases where medical directors had been involved in the telephone interviews). Respondents were asked to nominate a study lead investigator, who could assist with co-ordinating study visits and obtaining local ethics approval.

Site visits

Each hospital site was visited over 1–2 days by between three and five team members and a lay researcher. Schedules were drawn up ahead of time, with local leads assisting with the recruitment of appropriate members of staff for the interviews and focus groups.

Interviews, observations and document reviews were carried out for each site. At half of the sites, separate focus groups were conducted for staff and patients. Topic and interview guides can be found in Report Supplementary Material 4.

Construction of case studies

Case studies234 for each site were constructed from the data collected and field notes, supplemented by material from the telephone survey and publicly available data, such as trust annual reports. The studies aimed to combine detailed information about the processes of care at the unit level, while also providing contextual information about the hospital’s internal and external environments. Sankey diagrams, which were developed to depict proportionate flow through complex engineering systems235 and are increasingly being applied to health care,236 were constructed for each organisation (https://rawgraphs.io).

The study team attending the visit debriefed on multiple occasions after the case study visit. The first was conducted as soon as possible after the visit and team impressions were recorded to be included in the studies. The input of the lay researchers was particularly important, bringing a unique perspective. The draft write-ups were then reviewed collectively as a team: key points which marked organisations as being unique were decided in consensus fashion.

Framework analysis

Five separate sets of coding frameworks were used for all of the material generated by the case studies and relevant portions of the telephone surveys:

1. models of care – descriptions of mechanics of care were grouped by system component

2. the CFIR,100 which explores the contexts in which systems exist through five major domains – intervention characteristics, outer setting, inner setting, characteristics of the individuals involved and the process of implementation

3. medical generalism and specialism

4. smaller hospitals

5. workforce.

Each framework was separately coded by different members of the junior research team (Silvia Lombardo, Silvia Machaquiero, Louise Taylor and Claudia Leone) and then reviewed by senior members of the research team (LV, CI and AMR). ‘Pattern matching logic’237 was used to look for emerging themes, with emphasis on convergent and divergent evidence between data sources. An iterative approach was taken, with regular discussions (weekly among the WP1 team and monthly with the wider research team) being used to confirm important themes and guide further rounds of analysis.

The Day of Care Survey

The DCS was analysed according to methods described elsewhere. 103

Acute medical unit

Wards

There was no relationship found between the overall models of care used on the AMU and the overall models of care used on the wards [χ² = 46.66; degrees of freedom (df) = 40; p = 0.21]. No relationships were found between the open/closed nature and the patterns of consultant working for either the AMUs (χ² = 4.02; df = 4; p = 0.40) or the wards (χ² = 1.45; df = 2; p = 0.49).