Demand management for planned care: a realist synthesis

Article history

The research reported in this issue of the journal was funded by the HS&DR programme or one of its preceding programmes as project number 11/1022/04. The contractual start date was in September 2012. The final report began editorial review in May 2014 and was accepted for publication in November 2014. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HS&DR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

none

Copyright statement

© Queen’s Printer and Controller of HMSO 2016. This work was produced by Pawson et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Research method: realist synthesis

Our method of collecting together and drawing lessons from primary research evidence is known as ‘realist synthesis’. It was developed by one of the current authors. 3 A comprehensive overview on methodological guidance, publication standards, training materials and a bibliography of earlier realist reviews is available as part of a previous Health Services and Delivery Research (HSDR) report. 4,5 Readers are referred to these sources for a detailed account of the method. In this report, we commence with a brief account of the core principles and of the basic steps of the realist approach, assigning the full details on the conduct of the complex search strategies to Appendix 1.

Realist synthesis avows membership of the ‘theory-driven’ school of evaluative inquiry. It differs from other forms of systematic review in spending a considerable amount of research time in attempting to tease out the ideas, the assumptions, the logic or the ‘programme theories’ that underpin the interventions under inquiry. All policy-making and all interventions begin with theory: accounts of the causes of the problem under scrutiny (diagnosis) and conjectures on what changes must be made to the system in order to alleviate that problem (remedy). This starting point establishes the fundamental question for the review. Did these theories come to fruition? Did the remedy meet the diagnosis? The underlying expectation, and one that holds especially true in the field of demand management, is that social and behavioural interventions tend to meet with partial success. This state of affairs provides realist synthesis with its fundamental task, namely to chart the many contexts and conditions in which interventions work (or fail) and, above all, to explain the reasons for the inevitable mixed picture. Better explanations provide the ammunition to make better policy. They make provision for improved targeting and implementation of interventions.

The conduct of realist synthesis follows an orthodox path, starting with a research question and then finding the means to answer it by searching, identifying, appraising, extracting and combining together materials from the body of existing research. There are significant differences from other review methods, however, most notably in that this sequence is iterative – it is travelled repeatedly. The learning from one tranche of studies provides provisional explanations and new clues that that can be further refined by focusing and refocusing the searching for crucial primary materials. Learning accumulates as the review travels around the research cycle.

We now set out the basic sequence of steps in a realist synthesis. 3 For each, we provide some forward glimpses into our review, illustrating its application in researching the field of demand management.

Background: what is demand management?

We turn next to our substantive topic, namely to review interventions aimed at curbing demand for planned health care. The problem is immediately familiar to anyone with the slightest association with health service provision. Medicine has been a victim of its own success. All advanced health systems face substantial increases in activity and costs, with a seemingly unstoppable rise in demand for planned care. A particular strain is often felt on the matter of referral management, where the patient is relayed from one part of the system to another, often without due care being given to the balance of resources across the system. These are perennial problems and this is our playing field. As per expectations, putative solutions have been contemplated time and again, exploratory interventions have been implemented over and again and they have been researched again and again.

The task of reviewing demand management interventions, however, does have its particular challenges, which governed our approach and which, in turn, structure this report. The distinctive feature of our chosen ‘intervention’ might be best explained by reaching out to the much contemplated idea of ‘wicked problems’. The basic notion is simple and sensible enough. Not all problems are the same, some are more intractable than others, and essential differences in terms of problem complexity should determine what we should expect by way of a solution and how we search for solutions.

Rittel is usually credited with introducing the distinction between ‘tame’ and ‘wicked’ problems into the planning literature and it is worth rehearsing the essential contrast as a prelude to our review. 7 Tame problems, the argument goes, (1) have a well-defined and stable problem statement; (2) have an easily recognised stopping point when it is clear a solution is reached; (3) have a solution that can be objectively evaluated as right or wrong; and (4) belong to a class of similar problems that can all be solved in similar ways. Wicked problems lie at the opposite poles, where (1) the problem is ill structured, arising from sets of interlocking issues and constraints; (2) there is no overall resolution, only a series of partial and time-limited solutions; (3) there is no stopping point, with each partial solution creating a new system with its own strains; and (4) each manifestation of the problem is different, requiring the production of a revisable and versatile set of solutions. The last point should perhaps be emphasised. Wicked problems are neither unfathomable nor insoluble; rather, they require iterative, adaptive, ongoing responses.

Our thesis in this opening chapter is that the problem of demand management is much better understood as a ‘wicked’ rather than a ‘tame’ issue and that this has significant ramifications for the conduct of an evidential review. We are aware, in using this distinction, that it is in itself an oversimplification. The literature on systems theory and complexity science, originating in contribution by Rittel and Webber,7 has since become byzantine (for an elegant, modern summary see Byrne and Callaghan8). Each of the conceptual anchorages has become contested, with, for instance, wicked problems now being apportioned to somewhat different species as ‘super wicked’, ‘chaotic’ and ‘messes’. Complexity theory itself has spawned many different subschools, each with its own way of describing complexity: general systems theory, morphological analysis, holism, chaos theory, actor–network theory, diffusion models, soft systems theory, agent-based modelling and so on.

Our starting claim avoids all such elaborations. It rests on a straightforward comparison that is difficult to contest – namely, that relative to most topics covered in health-care research the dilemmas of demand management typify all the features of a wicked problem. One characteristic in particular shines through. As we shall see, every reform that is forwarded tends to expose new aspects of the problem, requiring further adjustment to the understanding of the initial problem. Our review thus imitates something that we discovered in the primary testimony, namely that an understanding of the true nature of supply and capacity imbalance develops alongside, rather than prior to, the application of potential remedies.

Coming to its specific features, we shall see as the review unfolds that demand management interventions are ‘relatively complex’ according to the following typology:

1. Causal structure. There are many notable interpretations of the causes of rising demand for health care, emanating from quite different sources: the inability to cope with backlogs arising from fluctuations in demand; the blockages in treatment pathways caused by professional closure and turf protection; the diverse motivations for referral that extend beyond formal clinical need; innovations in screening and testing producing self-propelling diagnostic cascades; innovations in treatments and services leading to supply-induced demand; changing patient demographics, particularly in age, driving up need for care and services; and the informed patient and the patient choice agenda increasing demand for all services. These causes, moreover, inter-relate, and their effects can be multiplicative.

2. Contextual diversity. As with its root causes, the locations of the potential demand management reforms are also widely dispersed. Demand has to be managed for every single condition from abdominal aortic aneurysm to zygote intrafallopian transfer. As one moves through the alphabet of disease, and from physical to mental health, and from pre-natal to end-of-life care, it becomes clear that each sphere has quite different supply and demand profiles. The institutional contexts responsible for managing these movements are also diverse. Modern referral services rarely run directly from the GP’s surgery to the outpatient clinic – all manner of appointment, transfer and triaging services are positioned in between (choose-and-book systems, ambulatory monitoring devices, text reminders, community-based specialists, etc.). These managerial systems, moreover, are under the control of the wider apparatus of trusts, commissioning bodies and boards (national, regional and local). Quasi-autonomous non-governmental organisations (QuANGOs) such as the National Institute for Health and Care Excellence (NICE) and the Care Quality Commission as well as targets such as those contained in as the Quality and Outcomes Frameworks (QOFs) all set parameters in which demand and capacity is constricted. All of these features are subject to periodic, top-down reorganisation, continually reconfiguring the scope for adjusting demand.

3. Implementation diversity. Referral chains themselves are long and complex, capable of providing potential pinch points at any point in the existing or newly modelled pathways. Again, in the modern era, the key players in referrals far exceed the traditional axis between the GPs and the specialist. ‘Handovers’ are also the business of practice mangers, assistants, screening personnel, triagers, physiotherapists, radiographers, haematologists, geneticists, practitioners with special interests, appointment clerks and, of course, patients. The communications flowing across these channels are also diverse – referral letters, appointments, reminders, test results, patient records. The communication media are also manifold, with referrals being conducted on paper, by telephone, by text and on the internet. All of these implementation chains need to be reviewed with an eye on the ‘weakest link’ syndrome.

4. Programme-theory diversity. Perhaps because they echo the diverse range of causes, contexts and channels of demand, it turns out that interventions designed to contain demand not only are huge in number but also vary greatly in their basic ideas and programme theories. The procession of ‘approaches’ includes RMCs, administrative triage, joint working, service relocation, clinical triage, referrals to GPwSIs, financial incentives, audit and feedback, guidelines, queue sculpturing, behaviour change and so on. Another indication of complexity here is that the focus of the intended remedy varies from the macro to the micro, closing in progressively from shifts in overall strategy, to organisational remodelling, to change in individual’s role, to procedural modifications, to motivational change. A further wicked twist here is that many interventions, intentionally or otherwise, end up pursuing change at all of these levels and are thus dependent for their success at achieving change across this entire piste.

5. Outcome conundrums. Although the aim of our class of interventions is seemingly clear, namely to control demand, the intended outcome is sometimes obscure and often contested. The moment when demand is considered ‘unmet’ is difficult to gauge, the politicised wrangling about the measurement of NHS waiting times being a prime example. Issues of quantity and quality also intermingle awkwardly. This is particularly evident in the underlying demand management aspiration concerning the ‘appropriateness’ of referrals. One of the causes of excessive demand is often considered to be the tendency of some practitioners to make improper or premature referrals. But the reasons for referral are themselves complex and judgements on appropriateness/inappropriateness is heavily contested. Indeed, this polarity is a two-way street for, as we shall see, it is quite possible that the goal of manufacturing better considered referrals can lead to an increase in demand.

6. Intervention history. These interventions are not mounted from scratch, that is to say they are never implemented in the absence of existing demand controls. As soon as any service is created (be it for the provision of housing, hamburgers, hunting horns or health care) it has to face the universal issue of managing shortfalls or oversupply in those services. All services wrestle with the permanent challenge of balancing demand with capacity. Accordingly, our inquiry presupposes that informal demand management regimes, in varying shapes and forms and operating at varying levels of success, are generally in place throughout a health service long before its practices become proceduralised and recognised formally as ‘demand management programmes’. This previous history will always limit the effectiveness of ‘new interventions’.

7. Ever-present emergence. The idea of emergence is perhaps the central proposition of complexity theory. It refers to the idea that novel structures are created by the interaction of the existing process within a complex system. It relates closely to the aforementioned wicked issue that some problems have no stopping point, with each partial solution creating a new system with its own strains. The concept of ‘supply-induced demand’ speaks to this very point. A new practitioner role or patient pathway may alleviate demand or referral pressures within some previous service arrangement – in time, however, it may become so popular that it becomes a pressure point of its own.

Hopefully, the above seven dimensions of complexity provide some indication that the problem under scrutiny is indeed wicked. We cease our initial exposition of intricacy at this point, reflecting briefly on the issue of how complexity impacts on the conduct of a systematic review. The implications are considerable and it is instructive to demonstrate this with a brief comparison of how topics are delineated in other modes of systematic review: ‘Systematic reviews should set clear questions, the answers to which will provide meaningful information that can be used to guide decision-making. These should be stated clearly and precisely in the protocol’. 9 Many reviews are traditionally conducted under the assumptions that they are dealing with stable systems – given populations (P), given interventions (I), given comparators (C) and given outcomes (O). When agreed operational definitions of this ‘PICO’ formula are in place, the review proceeds under the assumption that it is dealing with interventions that share these homogenous, reproducible and exact parameters.

To be sure, the more formal traditions of systematic review have also turned their attention to complexity, led in the health domain by the Medical Research Council’s guidance on developing and evaluating complex interventions. 10 This document, however, admits a rather restricted view of complexity into the evaluation palate – in recognising that programmes may have multiple components and multiple outcomes. Programme implementation is still seen as something to be controlled rather being subject to professional judgement, constant negotiation and perpetual improvement regimes. Heterogeneity is seen as a blight rather than an inevitability. 11

To make the point in a less partisan manner, it should be noted that the inclination to limit the remit of inquiry is indisputably not restricted to the statistical tradition. All reviews, whatever their methodological complexion, are assisted if their scope can be reduced. Scope control is clearly evident in existing reviews conducted under the realist banner. The study by Wong and colleagues contains a catalogue of realist health-care reviews conducted up to 2012, with such titles as ‘Efficacy of school feeding programmes’, ‘Water, sanitation & hygiene interventions in reducing childhood diarrhoea’, ‘Role of district nurses in palliative care provision’ and ‘Effectiveness of ban on smoking in cars carrying children’. 4 Although the subject matters here are hardly ‘tame’, and although unforeseen complexity is discovered in each of these inquiries, it a reasonable inference to suppose that the overall scope of each exercise is better circumscribed than in contemplating the full, sevenfold ferocity of entire ‘demand management’ regimes.

Outline and design of the current review

We close this introduction with our plans for squaring this circle and limiting the scope of our investigation. We also set out the chapter-by-chapter structure of the report. All realist reviews involve ‘prioritisation’, a decision on which of the myriad potential programmes theories should be put to investigation. 3 Demand management reaches into every stitch of health service delivery: where should we begin to weave? In this respect it is appropriate to acknowledge some significant circumscription built in to the commissioning of this our review. Our remit from HSDR was to cover demand management for planned care rather than emergency provision. We were also commissioned to review referral management interventions, as they are located between primary and secondary care. At a stroke, this eliminates consideration of the formidable demand dilemmas involved at the very beginning and end of the care cycle. Intricate referral sensitivities permeate preliminary triage initiatives such as NHS Direct and the 111 Service and demographic change places an ever-increasing demand on palliative care. Although we are thankful that these interventions have been removed from our overflowing plate, it is the case that they too could benefit from rigorous secondary analysis.

So, how have we prioritised investigation of the residual, but still formidably complex, sectors of demand management? The most obvious strategy is to limit investigation to a subgroup of the familiar demand management interventions. Previous research has distinguished (at least in name) over a dozen different schemes and strategies. Lacking the resources to tackle them all we have focused on four: RMCs, GPwSIs, GP direct access to tests and referral guidelines. Our reasoning here is that these interventions span quite different theories of change – seeking to modify demand, respectively, by transforming organisations, inserting posts, swapping tasks and improving recommendations. We have attempted to build these subsections of the report in enough detail to supply some decision aids on the contexts and respects in which these different ideas work.

We have also striven to retain a ‘complexity lens’ throughout the report. Partly, this motivation emanates from broader ongoing changes in the focus of health service research. 12,13 The latter group of authors begin a recent paper thus: ‘Incremental approaches to introducing change in Canada’s health systems have not sufficiently improved the quality of services and outcomes. Further progress requires ‘large system transformation’ considered to be the systematic effort to generate coordinated change across organisations sharing a common vision our goal’. Our findings point repeatedly to fact that demand management involves whole-system transformation and our review supplies a modest test of this ambitious thesis.

The main consideration for using a ‘system complexity’ approach, quite literally, is that it has been forced on us in conducting our review. Repeatedly, as we scrutinised the findings of a particular inquiry of a particular demand management initiative in a particular corner of the health service, we have emerged with the interpretation that what transpired was more a matter of the interdependence of the intervention within a wider range of determinants rather than what might be considered as the specific action of each programme. Repeatedly, we came across promising, partial and time-limited solutions that created a new system with its own strains. Over and again, we discover demand management has no stopping point. Rather than just report this as a nervous tic of a conclusion, we have sought to explain why it is the case and structured our report to this end. The remaining coverage is as follows.

Queuing theory and the question of time

Queuing theory is a formal branch of mathematics, which perceives ‘demand’ as a queue for services. And, as anyone who has queued will recognise, ‘time’ is the core concern – the demand management dilemma is understood as the task of bringing into balance these two sides of an equation:

1. ‘demand’ = the time required to complete a service/procedure in any period

2. ‘capacity’ = the time available to complete the service/procedure per period.

Fluctuating demand is seen as the underlying problem – it is irregularity of demand that creates queues. To use a contemporary example – planes arrive at Heathrow in uneven batches (seasonal and daily) and the immigration officials there to greet passengers sometimes sit twiddling thumbs and sometimes are over-run. How can the problem be managed?

Figure 2 (reproduced from Pandit and colleagues17) demonstrates the problem of setting the appropriate service level in the face of fluctuating demand. The three horizontal lines in Figure 2 represent different ways in which capacity might be set and foretells of a delicate balancing act. Line 1 is set below the minimum weekly demand. Demand is never absorbed and backlog increases unceasingly. Line 2, set above maximum weekly demand, absorbs all demand, eliminates backlog but wastes considerable capacity. Line 3, the practical alternative, sometimes absorbs demand and sometimes increases backlog.

FIGURE 2.

Alternative strategies and probabilities of meeting fluctuating demand. Reproduced with permission of John Wiley & Sons, from Pandit JJ, Pandit M, Reynard JM. Understanding waiting lists as the matching of surgical capacity to demand: are we wasting enough surgical time? Anaesthesia, vol. 65, pp. 625–40, 2010. 17 © 2010 The Authors. Journal compilation © 2010 The Association of Anaesthetists of Great Britain and Ireland.

A basic and perfectly general insight can be drawn directly from the figure.

Spare capacity in one week cannot be carried forward to the next week. Time cannot be stored for future consumption or sale as can manufacture goods such as cars or washing machines. Furthermore it is not always apparent in any week that time is spare, since this is only known after the event. Therefore the occasions when demand outstrips capacity always contribute to backlog, but the occasions where capacity outstrips demand cannot always compensate and capacity is wasted

Reproduced with permission of John Wiley & Sons, from Pandit JS, Pandit M, Reynard JM, Understanding waiting lists as the matching of surgical capacity to demand: are we wasting enough surgical time? Anaesthesia vol. 65, pp. 625–4017

A potential paradox is raised, put mischievously in Pandit and colleagues’ subtitle – ‘are we wasting enough surgical time?’17 Some waste appears to be necessary to ensure that demand is met and this proposition provides us with one of the standard motifs of queuing theory, namely ‘what are acceptable levels of waste?’

Queuing theory is able to make useful predictions, some counterintuitive, on queue dynamics under more complex conditions – what happens if you increase the number of servers? What happens if service levels also vary? What happens if you create a waiting list ‘buffer’ to even out demand? Figure 3, which is reproduced from Pandit and colleagues’ figure 10, illustrates a standard response to fluctuating demand. The pictogram illustrates varying demand over a 4-week period. Rather than dealing directly with these demand contours, patients are placed on a pooled waiting list, which if operated flexibly and efficiently yields a constant demand in successive weeks. More sophisticated queuing models are able to incorporate and predict the smoothing effect of a range of further demand management strategies.

FIGURE 3.

Pooling strategy to smooth demand. Republished with permission of John Wiley and Sons, from Pandit JJ, Pandit M, Reynard JM. Understanding waiting lists as the matching of surgical capacity to demand: are we wasting enough surgical time? Anaesthesia, vol. 65, pp. 625–40, 2010. 17 © 2010 The Authors. Journal compilation © 2010 The Association of Anaesthetists of Great Britain and Ireland.

Critics have pointed out some limitations of queuing theory, namely that it suffers the characteristic limitations of a mathematical theory. 18 It is defined by those inputs (number of arrivals, service requirements, number of servers, etc.) and outputs (probabilities of delay, queue length, mean occupancy, etc.) that can be modelled. Such models cannot incorporate all of the real world sources of variation and uncertainly, which reach right down into the idiosyncratic dynamics of every single consultation and up into contractual negotiations on service provision. Unwisely, some strict rules on service levels have been adapted on the basis of elementary queuing theory. Bain quotes one oversimplistic edict that was carried directly back into emergency medicine in Australia: ‘Queuing theory developed by Erlang nearly 100 years ago tells us that systems are most efficient when they operate at 85% capacity.[19] This applies to the queues at the local bank waiting for the teller or at a ticket booth at the MCG (Melbourne Cricket Ground)’. 18

Practical resolutions to demand fluctuation involve close tracing of patient flows through every procedure and pause within a particular service. Ironing out demand generally involves close attention to each link in the chain (changing division of labour, changing appointment procedures and priorities, etc.). A glimpse of some specific strategies can be seen in the following ‘case study’ (Box 2) summarised from the NHS Institute for Innovation and Improvement. 20

Our purpose here is not to be drawn into the specifics of this case study, still less to comment on the effectiveness of the suggested improvements. The point is to establish that fluctuation of demand is all-pervasive and operates through all the micro process within a service. It is thus likely to remain pertinent to our review, with the following implications.

Box 3 provides a summary of this section.

Professional closure, informal control and turf protection

From mathematics we move to social science. The sociology of the professions has always had a fascination with the medical establishment’s ability to retain power, restrict entry and control clinical practices. 21,22 The continued social transformation of the medical profession has been explored in terms of the clinicians’ relationship with the state, with management, with patients, with the pharmaceutical industry and with evidence-based policy. 23,24 Here, we want to trace the path of this literature into the daily routine of health care and into the matter of referrals. How might professional power affect the issue of handovers? Here, we report briefly on three of very many case studies that contain the typical features of ‘turf protection’ and show how these go on to create significant demand dilemmas.

Foote and colleagues’ study of the management of ultrasound waiting lists examines a process beginning with the primary patient consultation with the referring GP, moving through radiological investigations and into the secondary care. 25,26 The study begins with the basic predicament, namely that waiting lists are effectively formal institutionalised agreements between hospital departments and referring clinicians ‘to attend to patients at a later date’. 25 Satisfaction with this concordat differs from stakeholder to stakeholder, and a range of strategies evolve to modify, manipulate or circumvent this bureaucratic pact. If they believed it was not in a patient’s interest to wait at the back of the appointed queue, some GPs were quite prepared to load the system at other points. They made attempts to (1) bypass the ultrasound service by making hospital outpatient appointments, (2) send patients directly to the service in the hope of on-the-day cancellations, and (3) queue jump by forwarding urgent requests.

The real turf wars commence with the arrival of another set of stakeholders. Radiologists, sonographers and departmental managers held a different set of priorities: ‘GPs and waiting patients were broadly interested in managing diagnostic uncertainty . . . In contrast, the ultrasounds service’s account of the waiting list centred on difficulties in allocating scarce capacity to patients who were likely to have abnormal pathology in order to minimise the impact of waiting lists on ill patients’. 26 An ‘all clear’ diagnosis thus represented a positive outcome for GPs, while a congregation of ‘all clears’ constitute an unmerited demand for the radiological service.

Much of the immediate contestation between these professional groups occurred on the matter of prioritisation and handling of the ‘urgent’ requests. Radiologists and managers were fearful that the increasing number of urgent requests meant that the regular waiting list grew ever longer and the attendant publicity grew ever darker. As with most other forms of referral, bureaucratic hurdles were constructed to ration urgent requests. GPs were required to forward detailed, urgent requests only by fax/telephone and were heavily discouraged from not simply marking a referral form as urgent. But even under this system the majority of ‘urgent’ notifications fetched up with ‘normal’ scans. The result was a form of stand-off. Given the potential medico-legal consequences of declaring requests as inappropriate, ‘radiologists rarely refused to scan GP requests and only passed judgement on the quality of the referral after the patient had been scanned’. That judgement was often brutal: ‘The ultrasound service referred to such requests as “rubbish” ’ though letting off steam in this manner did little to resolve the numerical log-jam. 26

A further internal fissure, with a considerable impact on demand from radiography, occurred between the professional groups responsible for its implementation: radiographers (clinicians) and sonographers (technicians). Ultrasound imaging occurs in real time, potential abnormalities being best uncovered in an intricate, immediate to-and-fro between the production of scans and their interpretation. Investigation and diagnosis are intertwined with a resultant challenge to the identities and the division of labour between the specialist and the technician. One way that the former group were able to maintain their expert role was through the widespread adoption of ‘second-look stenography’ or ‘double scanning’ – the radiologist rescans the patient to confirm the accuracy of the sonographer’s scan. The demand dilemma is clear: ‘Double scanning restricts session throughput (as patients may be scanned twice), creates session overruns (a single radiologist covers two ultrasound scanners) and makes session throughput vulnerable to the availability of a radiologist (as a radiologist must be present when patients are being scanned)’. 26

Abel and Thompson’s qualitative study27 sheds further light on the ways in which generalists may circumvent attempts by specialists to control their behaviour via guidelines. It also reveals the different perceptions of specialists and generalists regarding what constituted an ‘appropriate’ referral. The authors interviewed 15 GPs and 11 specialists in New Zealand to explore their use of surveillance guidelines for colorectal cancer. These guidelines were written by specialists and contain advice about when a patient should be referred to secondary care. The guidelines classify people into low-, medium- and high-risk categories and seek to control demand by stipulating that only high-risk groups should be referred. In and of themselves, the guidelines can be seen as attempts by specialists to exert control over the behaviour of generalists.

The authors concluded that specialists and generalists have different perceptions of risk by virtue of their clinical remit and relationship with the patient. Specialists, who are one step removed from the patient, managed risk ‘scientifically’, often based only on a referral letter and family history. They perceived that GPs referred people inappropriately, in ways that did not fit the guidelines, which they then assumed were ‘too complex’ for GPs. GPs felt that the formal guidelines often did not ‘fit’ the patient in front of them and made decisions based on their relationship with the patient and their clinical experience. They also took into account the patient’s anxiety about their condition, rather than statistical calculations of risk. Once again, the turf wars resulted in a stand-off. GPs also embellished their referral letters to ‘fit’ with the guidelines in order to get their patient referred at an earlier date; specialists saw these embellishments as ‘lying’.

Currie and colleagues’ qualitative study of the development of genetics services run by GPwSIs provides further insight into the turf wars. 28 The focus in this case is on the rift that occurs between specialists and generalists. Twenty-four interviews were conducted with three key stakeholder groups (GPs, GPwSIs and specialist geneticists). The GPwSI role involves a reconfiguration of the standard health service division of labour. This new group of specialists take referrals from their fellow GPs, offer diagnostic and some treatment services and provide leadership in primary care in the reshaping of services around particular conditions and disease areas.

In the cases examined, GPwSI services that gained ready acceptance were those designed to supplement the work of consultant geneticists. Those that were set up to substitute the work of the geneticists encountered greatest opposition from the consultants. New functions such as ‘raising awareness’ in an ‘educational role’ prompted support; the development of new clinics within primary care did not. Consultants continued to hold sufficient power to constrain the development of GPwSI services through defining what constituted specialist knowledge and controlling access to training, education and support required to set up the new GPwSI services. GPwSI always had to establish consensus with the specialist regarding the purpose of service. Thus, whether or not GPwSI services were successful depended on whether existing inter- and intraprofessional relationships supported or constrained their development. Demand and duplication dilemmas continued and were sometimes exacerbated in new forms of informal control in which ‘geneticists took on the role of appraiser, vetted all referrals to the GPwSI and filled a supervisory role’. 28

Although the case studies noted here refer to specific services dealing with specific conditions, the underlying dynamic undoubtedly will repeat itself in other parts of the health-care system. Note that our analysis in this section does not depend on the original research having tapped the precise extent of professional closure available to each subgroup. Accounts of baronial contestation will always be contested. The point is that, whatever the precise resultant, assertions of territorial expertise, in and of themselves, will shape patterns of supply and demand.

Box 4 provides a summary of this section.

Micro-dynamics in the decision to refer

Before corrections are applied to the referral process, it is wise to consider what we know about the inner workings of that process. Numerous studies have explored how generalists – usually GPs – make the decision to refer to specialist care. This work can be traced back to a classic 1981 paper on ‘referral thresholds’. 29 At issue is the identification of the exact point at which the GP decides that specialist help is required and, thus, referral is needed. There is a linkage to the above theories on turf wars, through the explorations of where a generalist’s remit ends and a specialist’s remit begins. Several studies have developed general frameworks describing spectrum of influences on GP’s referral decisions. 30–32

In this section we concentrate on studies in the micro-sociological or social psychological traditions, which attempt to model how GPs explore patients’ problems. 33 Often, this complexity is described using flow diagrams, such as Figure 4, of the interaction and decision-making process. 34 To illustrate these studies, we focus on three inquiries exploring how GPs make decisions to refer patients to psychiatric services. This is a domain with significant variation in referral rates stemming from the considerable uncertainty on whether or not, when and where a patient should be referred to secondary care.

Morgan tracked all new referrals to two outpatient psychiatric clinics over 6 months. 34 Of the 184 patients referred, 62 did not keep their appointment, 14 did not wish to be interviewed and 120 were interviewed to explore the history of their problems and the chronology of events before their referral. A total of 27 out of 31 referring GPs were interviewed to explore their views on the history and nature of the patients’ problems and how the referral came about. In addition, documentary evidence (referral letters, medical records) and clinical data recorded by the psychiatrist for each patient were scrutinised. In most cases, the decision to refer evolved out of a series of encounters between the GP and the patient and their family over a period of months, usually 3–5, but sometimes up to a year. This suggests, significantly, that referrals are best conceptualised as a process rather than as a single decision at a specific point in time.

Morgan identified three ‘patterns’ or types of referrals. The first category were patients experiencing problems that could clearly be labelled ‘psychiatric’ and beyond the remit of primary care – such as patients exhibiting violent, psychotic or suicidal tendencies. These patients were usually referred as soon as the problem was known. The second category, termed ‘elective’, was less clear-cut; in these cases, psychiatric disturbances were masked by physical symptoms, were transient or were accompanied by other events in the patient’s life. The referral was contingent on a series of events shown in Figure 4. 34 The GP first treated the physical symptoms and ruled out a physical cause for the problems, which were then recognised as ‘psychiatric’. The GP then provided reassurance and supporting treatment, which represented an attempt to ‘contain’ the symptoms and gave the GP time to ‘wait and see’. Persistent symptoms and pressure from relatives may undermine previous reassurance from the GP and lead to a referral, often in the context of difficult and unstable doctor–patient relations. In the third category, labelled ‘negotiated referrals’, the GP acted as an intermediary to arrange a psychiatric referral at the request of others – sometimes solicitors or hospital consultants but also patients and relatives with ‘ulterior motives’ (e.g. to gain better housing or sickness benefits, etc.).

FIGURE 4.

A micro model of the referral decision. Republished from Morgan D. Psychiatric cases: an ethnography of the referral process. Psychological Medicine, vol. 19, issue 3, 743–753, 1989, reproduced with permission from Cambridge University Press. 34

The severity of the patient’s symptoms was judged a weak predictor of which category of referral came into operation. The authors conclude that the process of referral is structured by social as well as clinical events. GPs sometimes struggle to focus on symptoms rather than the patients underlying problems, which are not always seen as within their remit. Sometimes GPs do not feel they have the time or skills to address them. Furthermore, GPs find themselves managing not simply the patient’s condition but also their relationship with the patient, and they may choose to refer when this relationship breaks down.

In a related study, Sigel and Leiper interviewed a purposive sample of 10 GPs to explore how they defined and detected psychological problems, how they understood psychological therapies and their decision to refer a patient to psychological therapies. 33 They report that GPs saw ‘containment’ of mental health problems as a fundamental part of their management of patient’s mental health problems. Containment included helping patients through their problems, monitoring risk of self-harm and facilitating access to specialist services. Some elements of the findings are strikingly similar to Morgan’s study:34 GPs referred when they felt they had reached the limits of their capabilities for addressing a problem. This often occurred when the patient had not improved after the GP had attempted a course of treatment or the GP felt that they did not have the time to address the problem appropriately (similar to Morgan’s ‘elective’ category). In other cases, GPs referred without trying to treat the problem, usually when they felt ‘out of their depth’ with problems such as psychosis (similar to Morgan’s ‘psychiatric’ category).

In an earlier study, Nandy and colleagues interviewed 23 GPs about their management of minor mental illness. 35 They identified two key referral strategies used by the GPs, namely ‘conduits and containers’, which can be thought of as ‘referral to’ and ‘referrals away’. Containment was adopted by GPs who saw minor mental illness as part of the remit of general practice, while the ‘conduit’ strategy was used by those who saw their role as diagnostic and that of others as being responsible for the management of patients. Similar to the previous studies,33,34 GPs tended to see the patient themselves and then refer when they felt they were not making progress. Containment behaviour was enhanced by an interest in minor illness and by confidence in dealing with mental health problems. ‘Referrals to’ describes referrals made for ‘proactive’ reasons, such as the skills of particular professional or because the patient desired to be referred. ‘Referrals away’ described referrals made for ‘reactive’ reasons and triggered by feelings of frustration, anger or irritation on the doctor’s part where the GP felt that they needed help in managing the patient.

Although these studies have focused on referrals for psychological problems, it is likely that a number of the issues identified in these studies are mirrored in other domains where clinical uncertainty exists about referral.

Box 5 provides a summary of this section.

Self-propelling diagnostic cascades

Cascade effects refer to, ‘a chain of events (which) tend to proceed with increasing momentum, so that the further it progresses the more difficult it is to stop’. 36 They occur throughout physical and social systems and even turn up in folklore. In one of Uncle Remus’s tales, Br’er Rabbit becomes the victim of a cascade in the form of the ‘Tar-Baby’. 37 In his ceaseless campaign to trap the wily rabbit, Br’er Fox places by the roadside a dummy child constructed of tar. The loquacious rabbit tries to engage with the uncommunicative baby, ending in a loss of temper during which he throws a punch. His fist becomes stuck, prompting even more punches and kicks, which eventually entrap him in a ball of tar. We omit the happy ending to the story, noting that in modern parlance, the ‘tar-baby’ effect refers to sticky situations that are aggravated by additional effort.

Mold and Stein introduced the term ‘cascade effect’ into the health-care literature via a treacly tale about a patient admitted for routine hernia repair. 36 His history of mild heart disease and anxiety about his precise coronary status led the surgeon to request a preoperative consultation. This was to include an exercise tolerance test, which it transpired could not be conducted for several hours. Meanwhile, the patient, waiting outside the test room, became agitated and suffered mild chest discomfort. Because of this condition, the cardiac physicians now responsible for his care advised against the exercise test and transferred the patient to the telemetry unit. The patient, now even more anxious, received medication and underwent further tests. Although his cardiac catheterisation eventually turned up with an acceptable result, the hernia repair operation was not performed because of the full theatre schedule. The result is a demand management problem – lost operating time cannot be recovered and the diagnostic mêlée delayed the procedure by 2 weeks.

The moral of the tale is simple. In becoming more thorough, the diagnostic pathway also becomes more technical, more formulaic and more time-consuming. This self-momentum may have unintended consequences – it is difficult to stop even when counterproductive. This scenario leads us to a more recent definition of the cascade effect, suggested by Deyo in what is widely regarded as the seminal inquiry: ‘With regard to medical technology, the term refers to a chain of events initiated by an unnecessary test, an unexpected result, or patient or physician anxiety, which results in avoidable adverse effects and/or morbidity’. 38 Note that is it not diagnostic procedures of themselves that are deemed faulty or time-wasting; the problem is that they that are often prompted too easily and once prompted they become self-propelling.

Following Deyo, research on the cascade effect has parcelled into two separate bodies of inquiry: one charting the psychological, institutional and cultural triggers that may prompt the overuse of screening and testing and the other describing the multiplicative, unintended chain of events that may follow from inserting new or additional diagnostic tests into the treatment pathway. Deyo’s account of the ‘triggering events’ that may fire misplaced effort is summarised in Box 6. 38 It is impossible to describe all of these scenarios in detail here; most can be deciphered from their titles. ‘Defensive medicine’ refers to the physician’s inclination to request multiple tests in order to avoid potential legal liability for a missed diagnosis or treatment opportunity. ‘Low tolerance of ambiguity’ by patients or doctors refers to idea that, for a widening range of conditions, modern medicine offers scientifically warranted guidance on the best course of treatment. The corresponding prerequisite is the availability of the battery of tests able to identify each condition, clearly and unequivocally.

The utility of any classification system can always be questioned, of course and we note that some of these issues such as ‘patient demand’, ‘defensive medicine’, ‘low tolerance of ambiguity’ and ‘overestimating benefits’ enter the literature as generic causes of demand and capacity problems rather than specifically as diagnostic cascade phenomena (e.g. patients may attempt to ‘demand’ particular clinics, clinicians and booking times – and not just tests). Other items listed in Deyo’s typology also seem to describe the consequence rather that the causes of diagnostic cascades.

Deyo’s second task is to demonstrate how testing becomes self-propelling and for cascades to have unintended effects. He reviews a range of common tests [scans, foetal monitoring, angiography, spinal magnetic resonance imaging (MRI), pulmonary catheters] applied across a range of conditions. For each condition he describes the initial test, its complexities, the uncertainties of diagnosis and the chain of consequences that may follow. We quote his account of just one procedure:

Electronic fetal monitoring for women in labor is another technology that may lead to cascade effects. These devices monitor fetal heart rate, and certain patterns are associated with a greater likelihood of fetal distress. However, the risks of the test were not well considered before its adoption into routine care. Use of such monitors requires the mother to be relatively inactive in bed and this may increase anxiety levels. The combination of inactivity and anxiety may slow labor and lead to interventions to speed up labor (e.g., by artificial rupture of the membranes). When labor is accelerated, the pain of contractions increases and pain medication or epidural anesthesia may be requested by the patient. The loss of amniotic fluid may lead to higher pressures inside the baby’s skull, which could lead to more abnormal readings on the fetal heart rate monitor. Use of epidural anesthesia may lower maternal blood pressure, similarly leading to more abnormal readings. Perhaps as a consequence of such events, Cesarean section rates are 40% higher when electronic monitoring is used rather than simple auscultation of the fetal heart rate.

Reproduced with permission of Annual Review of Public Health, Volume 23 © by Annual Reviews, http://www.annualreviews.org38

What of the research quality of such evidence? It does, of course, capture the inexorable logic of a cascade. We note, moreover, that in the dozen years since the publication of Deyo’s paper, technical advance will have added to rather than reduced the diagnostic permutations. But we also note that his conclusions on this as well as all other cascades are unremittingly negative and quite possibly selective.

Given Deyo’s blanket conclusions and some confusion over its causal status, should the idea of the clinical cascade be taken forward into our analysis of demand dilemmas? There is clear evidence of a steady increase in the demand for and costs of diagnostic services. NHS Scotland report that:

[b]etween ‘2003/04 and 2006/07 the number of patients who had CT [computerised tomography], MRI and ultrasound tests (radiology tests) increased by 38 per cent to almost 736,000. The number of endoscopy and cystoscopy procedures increased by ten per cent over the same period; almost 168,000 procedures were carried out in 2006/07 . . . there were almost 73 million clinical chemistry tests, a 50 per cent increase since 2003/04.

Review of NHS Diagnostic Services, Audit Scotland, 2008, licensed under the Open Government Licence39

In 2000, the Department of Health introduced a Cancer Equipment Programmes in England, spending ‘£407 million on new capital equipment, resulting in greater numbers of CT, MRI and linac machines to spur implementation of increased diagnostic and treatment capability’ (quote © National Audit Office 2011, Managing High Value Capital Equipment in the NHS in England). 40

There is also good evidence that, despite increases in activity and equipment, demand is not always met. For instance, in the midst of the expansion reported above, a ‘Scottish Government Diagnostic Collaborative Programme’ was introduced to service delivery:

All have worked to identify any bottlenecks in their radiology and endoscopy services and taken action to improve them – for example, reducing the administrative tasks carried out by clinical staff to free up more time for clinical work. All of the sample boards have extended their working day for radiology, endoscopy and laboratory services or increased the number of sessions by working over lunchtimes.

Review of NHS Diagnostic Services, Audit Scotland, 2008, licensed under the Open Government Licence39

These examples enable a clearer insight into the causal structure of the cascade effect. Ever-increasing diagnostic systems are set up in response to external demand from patients, doctors, outside agencies and financial incentives. The diagnostic system itself may then create new imbalances of capacity and demand.

What of Deyo’s thesis on the adverse effects of the diagnostic cascade? A more balanced and potentially productive analytical tool may be found in another classic paper, Diminishing returns on the road to diagnostic certainty. 41 Borrowed from economics, the ‘law of diminishing returns’ states that in all productive processes, adding more of one factor of production, while holding all others constant, will at some point yield lower per-unit returns. Applied to a diagnostic cascade, this leads to the question of whether or not the additional information given by an extra test leads to a genuine increase in diagnostic certainty at a reasonable cost. It asks of any test whether it involves a positive or negative return or, quite routinely, whether or not it is marginal? Difficult as it is to differentiate these outcomes, it is precisely this calculation that is involved in an evaluation of pre-operative laboratory testing,42 childhood cholesterol screening,43 screening for Down syndrome and hypertension,44 mammography,45 etc.

Box 7 provides a summary of this section.

Supplier-induced demand

It is often noted that the supply and consumption of health care bears little relation to the exchange of any other product. One of the key reasons why the health-care market is considered perverse is rooted in the phenomenon of supplier-induced demand or, in North American terminology, physician-induced demand. Supplier-induced demand occurs when there is asymmetry of influence and information between suppliers and consumers, allowing the former group to encourage the latter group into seeking more services than they would under conditions of fully informed (or perfect) competition. This underlying and deeply rooted asymmetry has been captured dramatically and in lay terms by Williams in the following description of the doctor–patient encounter, which:

ostensibly works on the principle that the doctor’s role is to give the patient all the information the patient needs to make a decision, and the doctor should then implement that decision once the patient has made it. I am sure that the reader would find the above statement closer to his or her experience if the postulated roles of the doctor and patient were interchanged so that the sentence would then read: ‘the patient’s role is to give the doctor all the information the doctor needs in order to enable the doctor to make a decision, and the patient should then implement the decision once the doctor has made it.

Reproduced from Journal of Health Economics, 7, Williams A, Priority setting in public and private health care, 173–183, © 1988, with permission from Elsevier46

The literature is crammed with examples of market distortions that stem from this potential deformation in the dynamics of demand. It should be said that supplier-induced demand is a contested theory, with claims and counter claims about both its source (what drives it?) and extent (how deep is the problem?). Before these complications are reviewed, some well-known studies providing basic examples of its presence are noted.

The most compelling evidence stems from studies in which physicians perform the same tasks under different payment regimes. Our first example is the dramatic result of a two-stage randomised trial. 47 Doctors working in a university hospital clinic were randomised to receive income in two modes: by salary or by fee-per-service. Patients attending the clinic were also randomly assigned, in this case to the different payment regimes. The key result was that for the same population of patients the fee-for-service doctors scheduled almost 30% more return visits that did those receiving a salary. Most of the discrepancy was attributable to a 50% increase in the scheduling of ‘well child’ visits, which the authors describe as having ‘doubtful’ medical worth.

Another significant supplier-induced demand outcome can be seen in the findings of a ‘natural experiment’ comparing the treatment of heart attack emergency admissions in public and private hospitals in acute care hospitals in Victoria, Australia. 48 The incentive structures are, of course, different. In the public sector costs are met by the state through taxation, no fee is passed over and there is no incentive for the doctor to promote services. In the private sector the patient covers payment through health insurance, full fees are earned and service promotion is the norm. The research compares usage rates of the (then) most expensive and up-to-date tests and treatments for acute myocardial infarction, namely (1) angiography tests (two forms), (2) coronary artery bypass graft surgery and (3) coronary artery revascularisation procedures. The social profiles and medical histories of patient groups entering the two systems were matched statistically and the treatments selected under the different regimes are reported in Table 1. The patterning is self-evident. Far greater usage of cutting-edge diagnosis and treatment is made in the private sphere, with some interesting variations in the mixed-mode central cells.

The authors of both studies47,48 deploy several interpretative caveats, most notably that the key inference is indubitably not about which patients receive the best care or best continuity of care. The crucial point is to understand who is driving the health-care choices. In the Australian case study, the patients are undergoing heart attacks and being treated in emergency conditions and, thus, unlikely to be exerting strong pressure on treatment modalities. In the absence of any convincing counter-explanations a powerful inference is thus drawn: ‘These patterns must therefore be driven by physicians’. 49

Another body of evidence for supplier-induced demand, incidentally the largest and most controversial, emanates from the ‘variations’ literature – the surprising but common observation that the use of services and expenditure on services varies significantly among apparently comparable populations and localities. The phenomenon is often traced back to a pre-war study which discovered a 10-fold difference in tonsillectomy rates between different UK school districts. 50 These variations were stubbornly persistent, with the Chief Medical Officer’s 1959 Annual Report continuing to note: ‘the tonsillectomy rate ranges from 0.5% in Merthyr Tydfil to 16.3% in Chester; in those aged 15 years, it ranges from 1.3% in Swansea to 36.5% in Kingston-upon-Hull’. 51 Given demographic similarities between the latter two cities (e.g. dominantly working-class populations, estuary-based industries) it is very hard to see why patient demand would drive the vast procedural difference.

Since these early sightings, the phenomenon of variations in health-care practices has been documented at local, regional, national and international levels, the best-known database being the US Dartmouth Atlas of Health Care (www.dartmouthatlas.org/). One notable and repeated finding is that variation is strongly associated with the level of uncertainty about clinical outcomes. This pattern is illustrated in some data (Figure 5) reproduced from a British Medical Journal (BMJ) paper by Mulley. 52

FIGURE 5.

Rates of common surgical procedures among Medicare patients for 306 referral regions. Reproduced from Inconvenient truths about supplier induced demand and unwarranted variation in medical practice, Mulley A, BMJ, vol. 339, p. b4073, © 2009, with permission from BMJ Publishing Group Ltd. 52

Variations are plotted of the rates of common surgical procedures among Medicare patients across the 306 US referral regions. The rather tight cluster on the frequency of hip repair fracture indicates broadly similar rates for this activity across all US hospitals and regions. This contrasts with radical prostatectomy, one of the many procedures selected with ‘very high variation’ between surgeons, between hospitals and between districts. Again, the straightforward inference is that this pattern is unlikely to be driven by variations in patient demand, not many laypersons possessing knowledge of the vicissitudes of percutaneous transluminal coronary angioplasty. Rather, for an explanation of the disparity one needs to look across to the supply side. The working hypothesis is that the lack of an overall professional consensus about the effectiveness of particular treatments renders them open to parochial interpretation and implementation – and it is local custom, practice and vested interests that drive the high variation.

In the hope that these brief illustrations establish some of the empirical credentials for supplier-induced demand, we move to explanations for its ubiquity. There are a number of potential explanations, three of which are highlighted here: (1) technical innovation, (2) income maximisation and (3) professional uncertainty. As always in social science, these hypotheses both bestraddle each other and jockey for position. The common point of origin is that all are based on the idea of ‘asymmetrical relations’ between doctors and patients: the difference lies in the understanding of what it is that drives the physicians to promote supply in order to induce demand.

The first explanation perceives that demand is driven by technological innovation, to which medical care is particularly susceptible. Cain and Mittman trace an early example: ‘In 1895, Roentgen discovered the X-rays during his study of light phenomena, and within six months the first diagnostic radiograph was performed’ (quote ©1996–2015 California HealthCare Foundation). 53 Rogers has produced the best-known analysis of the diffusion of innovation and health-care organisations provide the prototypical example. 54 His first ‘dynamic of diffusion’ is the idea of ‘relative advantage’ – the more potential value or benefit of the innovation relative to current practice, the more likely it is to be adopted. Another key driver is ‘observability’ – the more obvious the evidence of improved experience and better outcomes the more likely is the adoption of new technology. Staying with X-rays, the relative gain is indeed palpable – being able to examine the functioning of internal organs without cutting the body. The potential outcome is that organisations failing to chase such technical gains will be left behind and this pressure perpetuates the asymmetrical understanding of treatment.

Although the quest for relative advantage is self-perpetuating this does not mean, of course, that technical capacity can be built willy-nilly. Demand leverage is not simply a matter of ‘if you build it, they will come’. Relative advantage is sustained and supported by marketing strategies: ‘build it and do everything in one’s power to make them come’. Any health-care organisation that has made substantial investment in equipment, scanning, testing and operating facilities will also pursue an engagement campaign. Two examples may be found in quite different health-care economies: the advice to new London polyclinics on how to ‘portray’ their relative advantages55 and the ‘quick tips’ on technology diffusion prepared for the California Healthcare Foundation by the ‘Institute for the Future’. 53

The second explanation for supplier-induced demand rests on more brutal profit maximisation hypotheses and is often labelled the ‘target income hypothesis’: doctors are motivated to by a desired level of income and will modify diagnosis and treatment until patients respond in the right numbers at the right price. We have already provided some powerful randomised trial evidence for this at the level of the individual physician. 47 Perhaps the most notable institutional-level evidence for this theory comes in a US study of the relationship between physician financial incentives and rates of caesarean section delivery. 56 The paper exploits a historical change in the financial environment facing obstetrics and gynaecological services, namely the declining US fertility (births per 100 population), which fell by 13.5% between 1970 and 1982. They argue that such a significant reduction in demand increased income pressure and led these services to substitute normal childbirth with the more highly reimbursed alternative, namely caesarean delivery. Used in only 5.5% of births in 1970, caesarean delivery rose by over 240% over the subsequent 12 years, and became ‘the second most frequently performed major surgical procedure in the U.S., with a rate of 23.5 caesarean deliveries per 100 births’. 56

Clearly, the opportunities for the target income mechanism to come into play are limited by the reimbursement practices operating in particular health-care regimes, with fee-for-service and fee-for-target systems being the most open to excessive recommendations for expensive treatment. There are, however, apparent examples of this process in the ever-changing UK GP contract system. For instance, Hughes and Yule chart the effects of a highly specific change in the UK payment system for cervical cytology. 57 In the 1990s, per-item fees were replaced by target payments, with GPs being paid a graduated lump sum according the proportion of cervical smears conducted among the eligible women in their practice. Their economic model suggests that the new contract accounts for ‘a dramatic rise close to 50% in smear tests performed by GPs’. The implications of such a shift remain contested. Is this a case of GPs manipulating demand by providing more information and encouragement to use the service? Or is this a case of meeting real population demand as captured in government targets? It is safe to conclude that there is wriggle room in many reimbursement regimes for physicians to recommend treatments that meet institutional targets and target incomes.

The third body of explanation for supplier-induced demand is characterised as the ‘professional uncertainty’ hypothesis. This model is driven by the data on local, regional and national variations in rates of treatment for particular conditions described earlier. Mulley provides a vivid explanation of the underlying and somewhat hidden mechanisms responsible for the disparity:

In the United States, striking differences exist among hospitals in the numbers of consultations and days in the hospital intensive care unit in the last six months that cannot be explained by differences in patients.

Mulley, 200952

and

Clinicians become accustomed to standard clinical policies that have been shaped over time by local capacity and are often surprised to learn that they are radically different from those in other regions. Though invisible, these local policies profoundly affect behaviour.

Mulley, 200952

He goes on to quote evidence from a study in which a representative sample of US clinicians, presented with standardised patient vignettes, make inconsistent treatment decisions – but ones which are entirely consistent with local service intensity. 58

This brings us to a very different explanation for the asymmetrical relations between doctors and patients, namely that it is ‘Professional uncertainty rather than consensus about the scientific basis of clinical practise that is emerging as the dominating reality’. 59 As long as medical decision-making remains located in practitioner wisdom there will always be different preferences for diagnosis and treatment, and these preferences will become hardened unevenly in local institutional practices. The upshot is a different formulation of supplier-induced demand: ‘With such a state of medical uncertainly it is clearly impossible for patients to evaluate what doctors cannot properly evaluate, and the abdication of consumer sovereignty represents a rational response to this’. 49

Box 8 provides a summary of this section.

Demographics and demand

Demographic change is widely regarded as a fundamental force accelerating the demand for health care. Populations evolve and their changing composition affects overall demand as well as the need to service the particular requirements of different segments of the citizenry. Demographic change, as it impacts on health care, has been analysed by social class, race and ethnicity, locality and so on. The most profound transformation, and corresponding research effort, however, follows on the consequences of caring for an ageing population. We thus restrict this short summary to the issue of dotage and demand.

Populations are growing older, so much is certain, but people, including patients and physicians, are also busily redefining the process of ageing. Accordingly, there is a research literature on both the ‘demographic time-bomb’ and the ‘red-herring’ of ageing. Most proponents of the former, the so-called ‘agequake’ hypothesis, start with some basic trends and extrapolations. If one takes almost any modern society, it is clear that the baby boom has given way to grey growth. If we take the Scottish population as a random example and compare the magnitude of different age groups between 2000 to 2010, one notes a 7% reduction in the number of 0- to 15-year-olds, compared with a 13% jump in 60- to 74-year-olds and a similar spurt (14%) in the oldest segment (75 years and over). Other cohorts with longer life expectancy are building behind this advanced guard and if the trends are extrapolated it is plain to see, as in the US data in Figure 6, that ‘seniors’ are, and will continue to be, the fastest growing population sector.

FIGURE 6.

Population projections by age group (US). Reproduced from Dill and Salsberg60 with permission © 2008 Association of American Medical Colleges.

The subsequent bulge in health-care demand, so the hypothesis goes, will occur because this particular population makes the greatest call on medical services. Figure 7, taken from the same US report, examines the number of ‘office visits’ to physicians across time for different age groups. At a glance, one sees that the oldest bums occupy considerably more seats and, significantly, that the rate of increase in consultations over the 15-year period is much more marked in these groups. By contrast, it seems that over the decades 15- to 24-year-olds in the USA have been less inclined to go to see the doctor. Treating such raw numbers as secular trends has led to economists at the Association of American Medical Colleges to conclude that, assuming that future growth follows historical trends of an increase in visits per person of 1.5–3%, the expected physician shortfall will almost double by 2025. 60

FIGURE 7.

Office visits to physicians per year by age group (US). Reproduced from Dill and Salsberg60 with permission © 2008 Association of American Medical Colleges.

As well as such applications about ‘manpower’ requirements, this methodology is the basis of much cost estimation – discover the costs per capita of each age group and then extrapolate future costs allowing for a rising age profile and the uneven distribution of health spending per capita by age. This algorithm underpins many cost projections going back to the pioneering work in the UK by Abel-Smith and Titmuss, who predicted that NHS costs might increase, simply by dint of an ageing population, by 11.2% in the period 1951–71. 61 A 1991 report on Canadian health spending, based on the same premise, follows the same pattern, forecasting that demographic shifts alone would generate a doubling of health-care spending in the following 40 years. 62

Since these early days, this method and its associated conclusions have been called into question. The first caveat, which required no major conceptual reorientation, acknowledged that these predicted increases should be expected to bite unevenly across different medical conditions. Put simply, demographic change should be expected to focus demand on particular services – and, therefore, a net X% increase in forward projections should not be met by X% increases in services across the board. A US study provides a comprehensive dissection of potential costs and savings by medical condition. 63 A handful of estimates illustrate the point: in the period 2005–15, age profiling alone should increase the call on services for respiratory disease by 10.2% and for circulatory disorders by 11.8% and, conversely, may also be expected to lower demand on pregnancy services by 2.6% and on human immunodeficiency virus (HIV) by 6.0%.

We now turn from caveat to critique. The ‘demographic time-bomb’ argument is regarded in some quarters as an ‘illusion of necessity’. 64 If the health professions (doctors, hospitals, health services, insurance companies, drug companies, etc.) can attribute inexorably rising costs to an inevitable, uncontrollable externality such as the ageing population, then they, the establishment, are absolved of blame. More significantly, this political critique finds support in an empirical evidence base. From the mid-1980s ‘researchers began to examine the possibility that a person’s age might be a less reliable predictor of health care expenditure than their proximity to death’ (emphasis added), because ‘healthcare interventions are common in the last stages of life and . . . a high proportion of individuals who die are in hospital at that time’. 65

This hypothesis, that age is a demand ‘red-herring’, gained momentum in a study by Zweifel and colleagues. 66 This paper studied demand using longitudinal rather than cross-sectional data. We quote from the abstract:

The econometric analysis of health care expenditure (HCE) in the last eight quarters of life of individuals who died during the period 1983–1992 indicates that HCE depends on remaining lifetime but not on calendar age, at least beyond 65+. The positive relationship between age and HCE observed in cross-sectional data may be caused by the simple fact that at age 80, for example, there are many more individuals living in their last 2 years than at age 65. The limited impact of age on HCE suggests that population ageing may contribute much less to future growth of the health care sector than claimed by most observers.

Reproduced with permission from John Wiley & Sons, from Zweifel, P., Felder, S., & Meiers, M. Ageing and population health care expenditure: a red herring? Health Economics, vol. 8, pp. 485–96, 1999. 66 Copyright © John Wiley & Sons, Ltd

Since this time there have been many studies attempting to explore and further explain this pattern. Gray provides a useful overview of the longitudinal studies, from which we include two typical fragments of data (references not claimed for this paper). 65 An early study of Medicare expenditure in the USA discovered that members who died in 1984, while comprising 6% enrolees, accounted for 28% of total Medicare expenditure. 67 A Dutch study found a similar pattern, with hospital costs rising by 170% when moving from the second to the last year of life. 68 Studies of this ilk convince Gray to the extent that he declares that they ‘show quite conclusively that time to death is a better predictor than age of health expenditure’. 65

Prediction is one thing; explanation is another. There is a need for precise understanding of how time-to-death shapes demand. A study by Levinsky and colleagues provides great insight by breaking down the aggregate bulge in final year of Medicare expenditure into figures on ‘per head’ costs. 69 Quoting Gray again:

during this last year of life health care expenditure per person actually declined with age, a pattern that was found in different geographical areas, for both sexes, for black and white beneficiaries, irrespective of degree of comorbidity, in hospices and hospitals, and regardless of the cause of death.

Gray, 200565

Gray explains that this was because the probability of being admitted to hospital or intensive care or of receiving interventions such as cardiac cathertisation, dialysis, ventilators or pulmonary artery monitors decreased with age.

Here is a rather subtle explanation, not always appreciated in the ‘time-bomb versus red-herring’ battle. The evidence reveals, once again, that proximity to death drives health-care demand but it also shows that age imprints itself on the subsequent supply of services. There is little evidence that demand at this point leads to treatment to prolong life at all costs. Other studies have found that costs peak in those aged 70–79 years but decline in the ‘oldest olds’. 70

Further subtleties and conditionalities reassert themselves, moreover, if one examines studies that have pursued data that break down total health-care expenditure into different cost units for aged patients: ‘ambulatory care’, ‘hospital care’, ‘drugs’ and ‘long-term care’. Werblow’s investigation shows that the former three types of expenditure have very weak age correlations once time-to-death has been taken into account. 71 Long-term care costs, for both nursing home and home care are, however, strongly related to the age of the recipient. Demand is met, it would appear, by a progressive shift from acute to long-term care.

So, where do we stand? As a matter of HSDR commissioning, we are absolved from tackling many of the subtleties of long-term and palliative care. Our brief is to study demand and capacity issues in the transfer from primary to secondary care. From this perspective the distinction between the ‘age’ of patients and their ‘time to death’ is perhaps oversubtle. There are a demand bulges associated with both, and the distinction between them is likely to be irrelevant to the physicians charged with making referrals and largely beyond the subtleties of programmes that attempt to control that demand. We conclude that age demographics are not so much a ‘cause’ of rising demand for health care as a ‘site’ or ‘setting’ or ‘repository’. It is a ‘structural condition’, which acts as the carrier for more specific demand issues. And what the aged or those in their final years bring to the consulting room is, above all else, the challenge of comorbidity. It is the common factor in the studies above about what accounts for the high number of consultations as well as swelling treatment and care costs.

Typical and more precise evidence on the issue can be located in a study by Caughey and colleagues. 72 It was based on the Australian Longitudinal Study of Ageing, which followed 2087 randomly selected men and women aged 65 years, correlating mortality with their clinical characteristics. Participants with ‘3–4′ and ‘5-and-over’ diseases had, respectively, a 25% and an 80% increased risk of mortality in comparison with those with no chronic disease, after adjusting for age, sex and residential status. Comorbidity is what killed them and what carried them into the health-care system.

Box 9 provides a summary of this section.

The internet, the informed patient and demand inflation

It is increasingly the case that patients have access to the internet and, at the same time, the amount of health information held online is increasing exponentially. There are bespoke, national sites specifically constructed for purveying advice, such as NHS Choices (www.nhs.uk) and Patient.co.uk (www.patient.co.uk). Evidence compilers such as NICE produce versions of all of their guidance for patients, carers and members of the public. Health-care charities such as Macmillan Cancer Support provide robust and timely cancer information for cancer patients, their families and carers. Once inside the system, the patient will be directed to condition-specific information as obtained through the websites of the Royal Colleges and other professional associations. There is also private provision of information should one choose to visit the like of the PruHeath official website (www.pruhealth.co.uk). Local, informal support groups crop up for most conditions, which can be accessed from bodies such as the Hysterectomy Association. Even the most routine activity of ‘Googling’ information will produce detailed results for almost any health problem.

As a result, patients’ use of the internet as a source of information is also growing. One study indicates that in 1997, 41% of regular internet users in the USA reported having searched the internet for health information; this had risen to 50% by mid-2000 and then to 60% by the end of 2002. 74 Rates for the ‘general population’ are lower; a national survey of US citizens indicated that 31% had reported seeking health information online. 75 Younger women, people on high incomes and those who are better educated are more likely to search for health information on the internet. 74–77 As we have already seen, the demographic footprint is never straightforward.

How might all this activity impact on demand for health services? At least four theories can be distinguished: that (1) health information acts as a substitute for visits to health-care professionals and reduces demand; (2) health information acts as a complement to information from doctors and means that time with the doctor is used more efficiently and, thus, reduces demand; (3) health information acts as a driver of demand, the newly informed patient being stimulated to seek health-care services; or (4) internet-informed patients ‘time hog’ medical services, leaving less time for other patients, and, thus, increase demand.

The first hypothesis, substitution theory, argues that people access health information as a surrogate for physical access to health care. Patients use health information on the internet instead of accessing doctors, or use it to decide whether or not they do really need to see a doctor. Both impulses, the argument goes, reduce demand for health care. A typical study is an evaluation of the NHS Choices website. 78 The authors conducted two surveys: one of users of the NHS Choices website and another of people waiting to see their GP. The survey of ‘Choices’ users found that 59% reported using the website to inform their GP consultations. Of these, 33% reported that the website reduced their frequency of GP visits, while 3% reported that it increased their visits. Those who reported that the website reduced their GP visits tended to be young, female and infrequent users of primary care in good health. In other words, substitution hypothesis seem to apply more closely to people who do not make large demands on health services in the first place.

Furthermore, the survey of patients waiting to see their GP found that although 71% of patients used the internet to search for health information, only 8% reported having used the NHS Choices website. Of these, 18% reported that the online service reduced primary care consultations. Similarly, the majority fulfilling the substitution hypotheses were young, white women in good health who were infrequent users of primary care. The authors go on to make extravagant claims about potential financial savings to the NHS, using a dubious method of directly extrapolating from the survey sample into general population costs. More attention should have first been given to the response bias in ‘pop-up questionnaires’ and ‘snapshot surveys’. We take a different key message from this survey: that health information may reduce demand for health care among certain groups for certain conditions. Accordingly, the net impact on the internet’s capacity to reduce demand is conditioned by the relative contribution of these groups to the overall demand for health care. A subsidiary lesson is also drawn from the study: not all websites are the same in terms of traffic and style of usage, with the NHS Choices website being less visited than others. 78

Linked to substitution theories is the idea that health information acts as a complement to information from physicians and enables patients to utilise their time with the doctor more efficiently without having to be in permanent contact with the doctor. One study, through focus groups with patients with diabetes, heart disease and hepatitis C, found some evidence that patients used health information in this way: ‘The Internet was perceived to be particularly useful for confirming and expanding on information received without “bothering” the doctor’. 79 Again, we note the methodological limitations of the study. The data are limited to people who were suffering from a chronic condition, all of whom required regular contact with the medical profession. The evidence is also ‘prompted’ rather than ‘naturalistic’. Respondents were each allocated a computer with three internet interventions appropriate to their condition bookmarked, before they moved onto the focus group.

We move briefly to the rival set of theories, which suggests that the internet might act as a driver for demand for health care. There are a number of mechanisms through which this might occur. One is the ‘hypochondriac’ or ‘cyberchondriac’ effect, whereby otherwise healthy patients prone to hypochondria read about something on the internet, think that their minor symptoms represent a terrible disease and consult their doctor unnecessarily. A brief and distinctly anecdotal piece in The Wall Street Journal sets the scene. 80 It recounts the experience of Melissa Woycechowsky, who searched the internet to understand the cause of the numbness and tingling in her legs and spent many hours each day in a multiple sclerosis chatroom, becoming convinced that she had the condition. Several visits to the doctor and many hundreds of dollars eventually served to confirm that Woycechowsky suffered from hypochondria rather than multiple sclerosis.

A related hypothesis is that internet-informed patients may exacerbate or stimulate ‘demand-induced supply’ – as, for example, when patients with existing conditions read about a new drug on the internet and visit their clinician to ask for it, even though it may not be clinically appropriate to them. One study used surveys and econometric modelling to explore the effect of more informed patients on demand-induced supply. 81 Do physicians accede to such requests? The survey data used in this study were taken from the public-use version of the Community Tracking Study Physician Survey – though the specific questions put to the medics are not reported. Multiple regression analysis is taken to show that a limited group of physicians are so influenced – according to their ‘years of experience, race, international medical Graduate (IMG) status, Primary Care Physician (PCP) status, average weekly hours of work, ownership status, practice type, sense of control over clinical decision making, levels of influence’. 81 Again, methodological limitations are noted: general-purpose survey, self-report, absent questionnaire items and correlational evidence.

A yet further influence that might drive demand is a ‘clarity over confusion’ mechanism, whereby patients read the burgeoning materials on the internet but find the information confusing or conflicting or cannot see how the information relates to their specific woes and so consult their doctor for clarification. Two qualitative studies on small groups of self-selected physicians and patients found evidence that internet-related dialogue did feature regularly in consultations. 79,82 It was, however, the medium of conversation rather than its inspiration. Patients found some information on the internet confusing or did not trust it and visited their doctors for clarification. These studies are primarily interested in the quality and consequences of these exchanges and provide little additional evidence on prevalence.

So, is there any numerical evidence that internet use does increase health-care utilisation? One study analysed data from the US Health Information National Trends Survey between 2003 and 2007 and found that, holding age, race, sex and education constant, ‘e-health information seekers’ had 1.2 more health professional visits per year than ‘non health seekers’. 77 Further, frequency of usage also marks a difference in demand. Those who searched for health information once per week increased an individual’s visits by 1.8 visits, compared with only 0.5 visits for infrequent e-health seekers. This study also examined the hypothesis that internet health information may complement rather than substitute visits to the doctor. The author tested a rather complex scenario, as follows. At onset of disease patients are less knowledgeable about their condition and thus would be more likely to visit their doctor following encounters with internet health information. As they become more familiar with their condition they may substitute the internet for health visits. However, if their condition deteriorates, they may look for more information over time and this may drive further visits. Suziedelyte tested this hypothetical sequence, restricting her analysis to those diagnosed with cancer and comparing the increase in the number of visits between those diagnosed ‘less than 2’, ‘3–5’, ‘6–8’′ or ‘more than 9 years’ ago. Her results indicated a small variation in visits that did correspond to the U-shaped prediction. 77 There are signs of a pattern here but the progress of cancer is undoubtedly too complex to be tested using this data set. Some cancer ‘sufferers’ may be entirely cancer free 3 or so years after diagnosis while others may have experienced stasis, slow deterioration, recurrence or death. Disease trajectories do not march to a uniform tempo.

Fang and colleagues hypothesised that internet-informed patients may act as ‘time hogs’ and place additional demands on doctor’s time, either through wishing to display their own knowledge to the doctor or by trying to secure more information from the doctor, which in turn leaves less time for other patients. 83 They tested this theory by applying an econometric model to the 2000–1 Community Tracking Study data. 84 This is a survey of US physicians’ perceptions of their services, including questions on the percentage of patients they regarded as consumerist, on the quality of care provided and on time constraints. The authors found that physicians reporting more consumerist patients were substantially less likely to believe that they could provide high-quality care in the appropriate time periods. Again, although there is some support for the presence of internet-inspired time-hog, the study suffers clear methodological drawbacks (opaque questions, self-report by one party on another, correlational evidence).

In summary, there is little doubt that the internet-informed patient is real and little doubt that, collectively, these patients’ new knowledge is affecting medical communication and, therefore, demand. However, the research paints a profoundly mixed picture of the impact of this transformation. Whether such information acts as a driver of demand or a substitute for services depends on (1) the quality of the information – poor-quality information is likely to increase demand while high-quality information may reduce it; (2) the characteristics of those absorbing the information and their existing rate of demand for health care; and (3) the point during the patient’s journey in which the search for information takes place. The ‘informed patient’, it seems, faces both ways, sometimes elevating demand and sometimes stunting it.

These counterbalancing forces should not, however, be understood as a neutral causal affect. At quite another level, the ‘informed patient’ has become an institutional reality. Since 2006, patients in England requiring a referral to a specialist have been entitled to a choice of four or five providers. Since 2007, an attempt has been made to construct an ‘extended choice network’, enabling patients to choose treatment from any accredited NHS hospital. These reforms have been underpinned with the roll-out of ‘Choose and Book’, an e-booking software application enabling patients who need an outpatient appointment to choose the hospital, the date and the time of their appointment. Similar patient-centred reforms are in train in many other advanced health-care systems. 85

How might this change in institutional apparatus impact on demand management interventions? First of all we should make it clear that our remit does not include the evaluation of the extended choice agenda, although this has been included in previous studies. 86,87 We do, however, discover patients and their volitions in many of the demand management subsystems already reviewed.

In terms of those causes located within the internal dynamics of the consultation, we do need to recognise the potential role of the patient within the growing culture of ‘shared decision-making’. It is well known, of course, that many patients still prefer to delegate treatment and referral decisions to their patients, with one study demonstrating that the desire for involvement in decision-making varies by type, severity and longevity of a patient’s condition. 88 Elwyn and colleagues, moreover, paint a picture of the complexities involved in trying to accommodate informed choice within treatment decisions in a model which we have summarised in Box 10. 89 The predicament for our review, therefore, is to recognise that some parts of the referral system are working to install these changes, while other elements are attempting to streamline consultations in the name of demand management.

Patients are also present at quite a different level in the service delivery apparatus of modern medicine. They are represented at committee level in all of the key decision-making bodies. Since January 2003, all NHS bodies have had a legal duty to involve and consult the public about the running of local health services. One example is the system of 74 Clinical Reference Groups (CRGs). CRGs are a primary source of clinical advice on the specifications, quality standards and commissioning policies for each specialist service, with each CRG appointing up to four patient and carer members. 90 In terms of our inquiry, we can also locate the patients’ voice in the construction of several of the interventions set up to manage demand such as referral guidelines and RMCs. Accordingly, one of the many tasks for our review is to assess the significance of these contributions.

Conclusion

These conclusions look forward to the main work of the project, which is to review the evidence on the effectiveness of the many schemes and strategies which have been devised to manage demand and make referral more effective. Our basic logic in this chapter is that knowledge of the symptoms comes before assessing the cure: we need to understand the nature of demand dilemmas before assessing any reforms. We should also remind readers that the above effort should be considered a rapid review, designed to carve out the main contours of demand and capacity imbalance rather than to inspect every local nuance. So, what have we learned?

The first lesson is that the roots of the problem are stunningly diverse. They range from the micro to the macro, from local to population levels, and from the immediate to the historic. Work routines and treatment pathways can fall in and out of rhythm. Power imbalances may sway services levels and slant rates of provision. Personal and emotional factors can shape the decision to refer. The technical apparatus of medical investigations can be self-propelling. Modern medicine and mercenary medics may induce demand. Demographic change can shape the profiling of demand. Populations grow more health literate and this can have a numerical effect on demand.

Box 11 provides a summary of this section.

The second conclusion is that these roots are bound to overlap and entangle in any particular corner of the health service. It is easy enough to imagine a scenario where all or most of these problems collide. An elderly patient under treatment may find himself in a queue-inducing population bulge, his comorbidities may prompt uncertainty and thus cascade testing, treatment decisions may be coloured by emotion and familiarity, his family may pester on the basis of cyber-health wisdom, his GPs may be confronting or indeed managing quotas and targets, his clinicians may have professional interests to defend and procedures to control, he may be from Hull (or Hell or Halifax) where a procedure is under- or overused or, indeed, he may become so agitated in working and waiting through this minefield that he develops some new disorder. Meeting this, or the endless equivalent configurations of pressure points, demonstrates the colossal difficulty of the task of demand management and affirms our initial contention that it may be considered one of the ‘wicked problems’ of contemporary public policy. 7

Our third and final conclusion is to cast some critical self-reflection on our sevenfold typology of ‘causes of the rise in demand for health care’ in Box 1. We have just noted that these features may overlap and intersect, and this is the configuration we wish to stress when it comes to studying how we might review their potential resolutions. So, for emphasis, it should be made clear that Box 1 is not simply a listing of ‘causes’ understood as direct pushes or elevators of demand. Nor should our brief evidence reviews be considered a matter of ascertaining that each push was of sufficient magnitude to consider it a worthy opponent in need of management. Instead, we have tried to spell out their precise aetiologies structures and discovered these are not the same across the seven features.

For instance, we have argued that demographic age is not in itself a generator of demand. ‘Proximity to death’ is regarded by some as a better indicator of increased call on services – but this gauge might too be erroneous as when services to the very old are withheld if considered fruitless and expensive. Thus, instead of ‘age’ being a direct cause, we have interpreted it as a ‘gathering point’ for comorbidity and consider this conjunction of clinical conditions to be a more direct press on demand for services. In the preceding section we argued that increasingly well-informed patients are not in themselves a direct cause of elevated demand. Health literacy may drive demand but may also apply a break. More fruitfully, we might consider that it is reforms in the overall management apparatus that increasingly recruit patients and that changing norms about their inclusion in treatment decisions are the crucial developments. Can these new privileges be accommodated in restrictive demand management regimes?

Another indication of causal complexity of the rising demand for health care is the recurrence of common issues across the typology. One feature occurs and recurs through these accounts: the issue of ‘clinical uncertainty’. It might be considered a key and eighth demand dilemma in Box 1 but its capacity as a common contribution is what marks it out. The issue raised its head most clearly in item 5 within our discussion of supplier-induced demand. Recall Wennberg and colleagues’ study, which showed that procedures with the highest variation are those in which there is more uncertainty in the evidence base about the medical benefit of the procedure. 59 But uncertainty, it transpires, is everywhere. We noted in the section on the micro-dynamics in the decision to refer (item 3) that social and emotional factors often squirrel their way into the decision to refer. Partly this is because there is inherent uncertainty in the clinical consultation. 91 Uncertainty is certainly a driver in the clinical cascade: the quest to bring finality to demanding diagnostic decisions, despite the law of diminishing returns. Uncertainty certainly underpins the turf wars, with generalists inflating demand by using referrals in seeking to reduce uncertainty and specialists seeking to control their services by reducing what they see as speculative referrals. Uncertainty certainly increases with comorbidity, in which treatment decisions are chosen on a selective rather than all-embracing basis. Uncertainty is certainly what drives patients to the internet, now established as a universal source of second opinion.

We conclude that our typology, imperfect and uncertain as it is, provides a useful entrée to systematic review. We hope to have captured some, if not all, of the obstacles curbing the demand for planned services, and to have understood something, if not everything, about how the drivers of demand intersect and combine together. Sun Tzu, the father of military strategists, once famously declaimed that ‘if you know your enemies and know yourself, you will not be imperiled in a hundred battles’. 15 A careful appraisal of some the causes of health care demand may allow us to recognise the perils and the potential of some of the solutions.

Introduction: eliciting and selecting the underlying programme theories

With this chapter we enter the first main phase of our review. All policy-making begins with theory: accounts of the causes of the problem under scrutiny (diagnosis) and conjectures on what changes must be made to the system in order to alleviate that problem (remedy). In realist terminology, these foundational ideas are known as ‘programme theories’. The process of locating and articulating the proposed diagnoses and planned remedies that have gone into the making of an intervention is known as ‘theory elicitation’.

Demand management is a perennial problem and so its principal ideas have been contemplated time and again. And, as we shall see, the result is a large family of interventions underpinned by a welter of interrelated ideas. The purpose of this chapter is to provide a thorough review and a detailed description of these intrinsic programme ideas. Note that the chapter travels no further than an exposition of these key concepts and conjectures. This is an ideas exercise; indeed, given the formidable ambitions for demand management, it is a ‘big ideas’ chapter. The assembled conjectures will go on to provide the foundation stone for further analytic work: they act as hypotheses to be tested. The basic objective of the upcoming phase of realist synthesis and subsequent chapters is to trawl the empirical evidence to gauge how, where, why and to what extent each of these theories has proved fruitful in practice. But that is for later.

Here, the key task is to hunt down the demand management programme theories. This was accomplished in the following stages, the chapter’s sections mirroring this running order. Our preliminary task was to take stock of the field – to draw up an initial list of the interventions that have been applied in the name of demand and referral management. The purpose here is simply to scope and survey the field; we identified and classified the basic array of programmes without yet penetrating to their underlying ideas. Our attempt to map the coverage of the demand management interventions ends in a simple typology breaking down the programme theories according to what they perceived to be the main axis of change. Demand management schemes have been designed to operate at all levels from the macro to the micro. As already noted, these range from attempts to control demand by relocating the entire apparatus in RMCs to interventions seeking to change individual decision-making through financial incentives. To capture this range we made an initial sift of programme theories into five types: those seeking change at the (1) strategic, (2) administrative, (3) role, (4) procedural and (5) motivational levels.

For each class of programmes we then searched for sources that discussed and debated the underlying theories. As our focus is on underlying policy thinking, we directed our attention, self-evidently, to the ‘ideas literature’. In other words we searched for sources in the so-called grey literature (planning documents, guidance materials, discussion documents, proposals, rationales, policy expositions, professional journals and critical debate). By and large our subject matter here is not located in the academic journals and we focused our search on documents and expositions produced by the health-care bodies, professional associations, institutes, agencies, services, QuANGOs, etc.

The next task was to extract, capture and compile the constituent programme theories. In essence we are collecting explanatory propositions. In a planning document an author might opine that intervention A works by doing B. In a further policy appraisal we might note claims that A will not work because of stultifying force of C. Debate is often enjoined in further critical opinion as when it is hypothesised that A will work in respect D but not if the plan is to achieve E. There is no shortage of such material. Demand management has a considerable pedigree, and we expected and found and were able to catalogue a constant refrain of proposal, critique and counter-critique.

The next stage is known as theory selection. Thus far we had charted five broad theories about where change might manifest itself and scores and scores of subtheories about how these improvements might be achieved. It is impossible to review every idea that has ever dawned about an intervention: it is necessary to prioritise and seek out patterns within programme theories. Because we were investigating interventions with a considerable pedigree, one pattern became abundantly clear, namely the constant adaptation and revision of the original theories. We have thus presented many programme theories to capture this evolution, beginning by unearthing the founding ideas, and then charting some of the emerging critical material, and then moving to some of the prosed remedial improvements.

This idea of programme drift or implementation shift is routine in the policy literature. What we had not anticipated was that there might be an overall pattern to the transformation of ideas about demand management. Our inspection of the underlying programme theories reveals a trend for the incorporation of ‘whole-system’ thinking into demand management thinking. By this, we perceive a move away from the ‘single lever’ models which situate change at the strategic, administrative, role, procedural and motivational levels. This understanding mirrors much contemporary of thinking about change and innovation in organisations, which also lays stress on the idea that organisations are layered and that change is meditated by intercommunication between these layers. This idea, which we refer to as the ‘depth ontology’ of organisational change, provides us with a middle-range theory that finds its way into all subsequent stages of the review.

This brings us to the final section of the chapter. Remember that all of the above analysis lies in the domain of programme theory. To that extent it should be viewed as speculation and conjecture. We do not know as yet whether or not these policy revisions lead to improvement in demand outcomes. What can be said, however, is that that reformulation of the problem – namely, ‘have the adaptations worked?’ – captures the optimal strategy for reviewing this particular class of interventions. The concluding section considers how and how not to mount a review of a moving target. It outlines the strategy we use in the next phase of the review: what evidence will we need to test these adapting programme theories, where can it be located and what will be the shape of our emerging findings?

Taking stock: mapping the field of demand management interventions

Just as the causes of demand inflation are many, the efforts to stem the escalation are numerous. The purpose of this section is to gain an immediate measure of our review domain and to map the main types of interventions that have been implemented in the name of demand management. We are aided in this task by the production of several existing typologies and we commence by reproducing three of them as Figure 8 and Boxes 12 and 13. 93–95

FIGURE 8.

Distribution of different approaches to referral management. Reproduced with permission from The King’s Fund from Imison C, Naylor C. Referral management: lessons for success. London: The King’s Fund; 2010. 93 Hospital Episode Statistics data copyright © 2015. Reused with the permission of the Health and Social Care Information Centre. All rights reserved.

These three typologies provide a blunt introduction to the complexity of the field. Favourably, we note many elements in common, the same interventions cropping up across the three classifications (guidelines, incentives, feedback, service regrouping, etc.). Contrarily, when it comes to the placement of the intervention types and their subspecies, the three figures diverge considerably. For instance, Coulter94 regards ‘financial incentives’ as the umbrella term for different funding interventions. Imison and Naylor93 perceive ‘financial incentives’ as one type of peer review and audit. In Roland and colleagues’95 model ‘financial incentives’ become a form of professional behaviour change. The most telling pointer to the complexity of the field, however, is the upwards-pointing arrow in Figure 8, which is an indication of the ‘different degree of intervention’. By this, the authors refer not just to the amount of effort and expenditure but to the fact that a composite programme, such as the introduction of a RMC, is likely to subsume many of the more targeted interventions. A RMC may well utilise guidelines, incentives, joint working, etc.

Without exploring other versions of these classifications (e.g. Pencheon96) it can be seen that there are many equally logical ways of cutting the demand management intervention cake and that they are unlikely to match the complexity and variability of programmes ‘on the ground’. Armed with this insight we sought to commence our exploration of programme theories with a relatively simple typology that distinguished the main ‘axis of change’ targeted by the programme: what is it that is supposed to change as a result of the intervention? Five different objectives are identified in Figure 9.

FIGURE 9.

Change objectives in demand management interventions.

Our model responds to the idea, self-evident in the previous typologies, that change is envisaged at quite different levels from the macro to the micro. We make no claims for the originality of this idea, of course, the notion of ‘vertical stratification’ or ‘depth ontology’ being much utilised in the literature on organisational change. 97,98 These terms refer to the idea that organisations are layered and that change is meditated by intercommunication between these layers. By ‘strategic change’ we refer to reorganisation of significant parts of the entire system; ‘administrative change’ anticipates the introduction of new teams or units to carry out specific tasks; ‘role change’ envisages a change in the division of labour between individuals or between teams; ‘procedural change’ focuses on changing the way tasks are performed; and ‘motivational change’ seeks to stimulate increased enthusiasm for performing a particular task in a particular way. As we are about to see, all of these objectives are envisaged in the entities we call demand management programmes and it is a simple enough task to fit the interventions compiled in Figures 8–10 into the model in Figure 9 (and vice versa). Our interest in Figure 9 is to discover how the respective changes are understood and to trace the evolution of each programme theory.

Demand management interventions

As noted, demand inflation and referral improvidence are perennial health service problems, which have gone on to inspire a medley of putative solutions. In this section we describe a spanning set of the major reforms. For each, we highlight typical examples of major UK NHS interventions, extracting brief illustrations of the core programme theory. Our immediate task is to show that the initial modes of change envisaged sit squarely within the stratified model provided in Figure 9. When planning change, programme architects grasp immediately for reforms of strategy or administration or roles or tasks or motivations.

However, we also take advantage of the longevity of demand management/referral management campaigns. A notable feature of our scrutiny of programme theory is that we came repeatedly on ‘second thoughts’. Realist synthesis is a theory-driven approach and so engages in a systematic search to track down the assumptions, working ideas and plans that contribute to the building of any intervention. A typical feature of such programme architecture is that there is never just the single blueprint. Plans change. Programme designers adapt their ideas in the face of criticism and by dint of their own and others’ experiences. Accordingly, the search for a programme theory rarely supplies an agreed model. More often it reveals a history of planning. In terms of long-running interventions, such as demand management/referral management, that history tends to build a picture of the adaptions that need to be made to foster successful change. Each of the following sections begins by unearthing the founding ideas, charting some of the emerging critical material, and then moving to some of the proposed remedial improvements.

Administrative change

The second proposed measure for improving demand management/referral management makes the case for transplanting one agency to do the work of another. The basic working hypothesis is that a fresh institution or agency, B, is better placed to carry out a particular function than the old regime, A. Specific responsibilities are handed over rather than the entire care process (which is the domain of level 1 change). Day-to-day improvements are deemed to flow by dint of better specialist management – the new body produces a better division of labour, better cost containment, better information flow, better training and so on. RMCs or referral management gateways are the perfect exemplification of this mode of change. A new body is transplanted into the midst of the care pathway to assume responsibility for core referral decisions. We borrow some promotional material from Fortis Healthcare,111 which provides a vivid picture of the intended transplant and its interconnections (Figure 10).

FIGURE 10.

Prototypical structure of a RMC/referral management gateway. Reproduced with permission under the Open Government Licence 3.0. 111

Given the substantial reorganisation involved, it is perhaps not surprising that there are many different (and still evolving) subtypes. We provide further details on these variations on a theme in Chapter 4; here it suffices to point out that RMCs vary markedly in terms of the personnel responsible for making referral decisions. Much ink has been spilt describing the potential of these agencies in accompanying plans, proposals and position papers. In these narratives RMCs are envisaged to improve the management of demand by queue sculpturing of flows and blockages, with the better use of information for planning, by keeping a close eye on cost containment, with better support for patient choice and with the provision of education and training for key players in the system. Some brief and typical examples of the working ideas (programme theories) behind these claims can be seen in the following examples (Box 15).

BOX 15 - Referral management centre programme theories

From Somerset Local Medical Committee’s 2004 ‘position paper’ on referral management centres

Not all referrals require a consultant opinion. Some can be dealt with by, for example, a specialist physiotherapist or a nurse practitioner. Yet others may be better handled by a GPwSI working for the PCT [primary care trust]. Clinical assessment of referrals should mean that the most appropriate person sees the patient – quickly, close to home, and at the lowest reasonable cost.

Reproduced with permission from Somerset Local Medical Committee112

From Billy Fashanu’s comment on Elwyn and Davies’ 2006 British Medical Journal paper

Better still, if the referral management system allows for closer access to hospital specialists guided by an agreed pathway, then the patient can be directed to the right services at the right time, hopefully in a more cost effective and clinically efficient manner (common evidence based pathways can be agreed by all clinician and issues of legality addressed as a group).

Reproduced with permission from Billy Fashanu113

From Somerset Local Medical Committee’s 2004 ‘position paper’ on referral management centres

Referral within the NHS has hitherto been chaotic. Without a checking system it has been impossible to make sure that a referral is despatched, that the hospital has received it, and that an appointment has been sent in response. An RMC can use simple IT to log and monitor all referrals. The patient will know that the referral has gone because RMC staff will contact them, and the PCT [primary care trust] will be able to check the progress of referral and initiate action if waiting list targets look like being exceeded.

Reproduced with permission from Somerset Local Medical Committee112

From Department of Health case study: Fortis Healthcare

Doctors get feedback about their referrals. If there is something missing, the doctor is asked for further detail. If the Gateway panel deem the referral unsatisfactory (ie, the referral is not reasonable), it is returned. This means that the doctors learn from the process and the quality of referrals improve. Monitoring referrals by type help the Gateway team to identify areas where GPs may require extra training.

Contains public sector information licensed under the Open Government Licence v1.0111

Our purpose here is not to test the veracity of these claims but simply to identify the purported mechanisms for change. One sees immediately that RMCs are expected to work through a range of other processes: changing responsibilities, devising agreed pathways, improving information systems, employing feedback and training. In other words organisational change (level 2) actually works though role, procedural and motivational mechanisms (levels 3, 4 and 5).

Perusing exactly the same literature reveals another feature of these particular programme theories, namely the accumulation of critical voices about why their expectations may not come to pass (Box 16).

BOX 16 - Referral management centre programme theories (negative)

Excerpt from Nowottny, ‘Gatekeeper job is one for trained GP’, Pulse, 10 August 2011

Generally, the decision to refer a patient is a carefully considered one, the product of a GPs experience and training, and based on one or more face-to-face consultations. For an anonymous nurse working to a set protocol in a remote referral management centre to reject that carefully considered decision out of hand is professionally frustrating, to say the least. The primary concern, of course, is patient safety. Both GPs and consultants told Pulse they feared under-qualified medical staff were inappropriately rejecting referrals because they do not conform to inflexible algorithms.

Reproduced with permission from Pulse magazine114

Somerset Local Medical Committee’s 2004 ‘position paper’ on referral management centres

GPs value their personal knowledge of a consultant’s skills, practice, and personality when making a referral. This will be lost if referrals are directed generically or to a different provider. Patients also usually prefer to see the same specialist if they have previously been treated for a related condition.

Reproduced with permission from Somerset Local Medical Committee112

Again, our purpose is not to begin the process of gathering material to adjudicate between the various claims and counter claims here, but to show that working programme theories are always mobile. The first batch of theories envisages change in the decision-making apparatus to ensure that patients are directed to ‘the right person in the right setting at the right time’. The second set of theories (some, interestingly, from exactly the same sources) foresees that some of the key personnel may doubt that the new incumbents can ever be that ‘right person’. Further iterations, what is more, take us to a yet more discriminating ‘compromise theory’ and begin to include notions of trust, shared ownership and gradual development (Box 17).

BOX 17 - Referral management centre programme theories (refined)

Excerpt from Wootton and Whiting, ‘Creating ownership of a referral gateway’, Pulse, 3 September 2012

With the exception of one or two practices, the vast majority of GPs from the 103 practices in Manchester now use the gateway . . . a number of factors we believe have led to this almost universal acceptance. The first is that this was not something introduced suddenly and on a whim. It was developed over a long period of time and after plenty of discussion and collaboration with primary care through local meetings, the CCGs [Clinical Commissioning Groups] and the LMC [local medical committee]. We showed GPs it would be effective through the smaller pilot at the beginning. One very important factor in setting up the gateway was the use of local GPwSIs and senior, well-known GPs. It meant the other GPs respected the decisions made and the feedback they were receiving. Because the advice is provided in a very professional and supported way, the GPs saw it as more of a CPD [continuing professional development] tool than a criticism of their decisions.

Reproduced with permission from Pulse magazine115

This brief history of RMC provides salutary lessons for the reviewer. Programme theories are not respectful of programme labels, and nor are they likely to remain passively stable.

Common denominator: a unifying programme theory?

So what have we gleaned from this close inspection of the programme theories underpinning the many decades of efforts to contain demand for health care? Unsurprisingly, we learn that the initiatives vary massively in order to take care of considerable differences due to medical condition, due to the complexity of patient pathway and due to existing modes and levels of primary and secondary provision. The potential causes of demand inflation, discussed in the previous chapter, sit unevenly across these assorted clinical and service domains. Accordingly, we learn that interventions aiming to improve GPs’ understanding of electrocardiogram (ECG) testing have little in common with those designed to improve referral letters or with those seeking to reorganise booking services, and so on.

We have, however, found a serviceable starting point to analyse these differences using a typology (see Figure 9), commonplace in the organisational studies literature, which differentiates the ambitions to produce change along a continuum ranging from the macro to the micro. Along with other authors we have been able to make a useful demarcation of demand management interventions into five main modes: strategic change, administrative change, role change, procedural change and motivational change.

However, by investigating these programme theories in a longer time frame, we notice a recurrent pattern under which they deform and reform. This should hardly come as a surprise. It is not the job of practitioners and managers to maintain interventions in their pristine purity. Their job is to make improvements work ‘on the ground’ and they work under the constant refrain of ‘innovation’ and ‘modernisation’. Programme adaptations are thus myriad and tailored to local circumstances. But we have also been able to demonstrate that refinements are not simply piecemeal – administrative minds do tend to think alike. Accordingly, the prevailing motivation for the modifications has been to add supplementary procedures that operate across the levels of change (schematically on can say that ‘level 3′ plans are often buttressed by auxiliary idea at levels ‘1’, ‘2’, ‘4’ and ‘5’, and so on).

It is useful to consolidate this proposition by returning to the generic literature on social and organisational change. We have already drawn on the idea of ‘vertical stratification’ to buttress out original fivefold typology (vertical stratification referring to the ontological levels in Figure 9 rather that a hierarchy of ‘seniority’ that occurs in most organisations). This same literature also draws attention to the idea that change is mediated by the intercommunication or intersection between these layers. We provide a highly abbreviated tour of this recurring theme using two sources, one elderly and purpose-built, the other modern and metaphysical. Suitably ransacked, these authorities provide the basis for a simple model of the key, interrelated layers of change in health-care systems that will form the basis of our subsequent analysis.

A pioneering attempt to describe the corporate layer cake occurs in a book aptly titled Social Change in Complex Organisations. 108 Sitting at the conceptual apex is something these authors refer to as ‘centralization’ of the organisation. This is the platform for core decision-making; overarching strategies and systems are established here. Next comes ‘formalization’ – as the authors explain, ‘organizations need daily guidelines for their operation; these guidelines are furnished by rules, the repository of past experiences’. At a level down, we have ‘stratification’, by which Hage and Aiken mean the division of labour between members and the degree of differentiation in their roles. The next two strata are labelled ‘production’ and ‘efficiency’. These carry similar meaning referring to how organisational goals are embodied in everyday procedures and work routines. Finally in the model we reach the level of the individual member and the idea that change inheres in their ‘morale’, their levels of loyalty and their team spirit.

All of these layers (systems, rules, roles, routines, morale) wrap and unwrap as institutions evolve. And it is this image which provides Hage and Aitken’s central thesis on the methodology for studying change: ‘A temporal perspective, one in which the focus of attention is not only on the interrelationships of parts of an organisation, but more importantly on changes in these interrelationship is ultimately necessary for a more complete understanding of organizational behaviour’ (p. 28). 108

A rather grander attempt to understand the anatomy of societal change can be found in ‘realist’ methodology, most especially in the works of Archer. 145,146 Her theory of ‘morphogenesis’ attempts to answer the age old sociological chicken-or-egg question about what comes first in propelling social change – is it ‘structure’ or is it ‘agency’? Put simply, her answer is ‘chickenegg’. People’s immediate actions are shaped within social structures in which they sit: communities, organisations, legal systems, power relationships, etc. However, in a longer time frame, these structures themselves change as a result of the activities and choices of the historically situated individuals who make them up. Social change, in short, occurs through a never-ending cycle: ‘structural conditioning’ shapes ‘social interaction’ which in turn shapes ‘structural elaboration’, and so on, and so on.

It takes more than structures and actors to make the Archerian world go round, of course, and in the detail of her work other layers of social reality are identified: different ‘strata’ that are also real and are also causally powerful. Thus, for instance, she makes the distinction between ‘primary’ and ‘corporate’ agents: individuals with momentary, specific leanings and people with roles and positions who punch more heavily in their world shaping power. Similarly, structures have ‘first-’ and ‘second-’ order shaping powers such as that between a government and any administrative agency or QuANGO it may create. Further, there are distinctions between structures in terms of how they exert influence, through the ‘material sphere’ of power and resources versus the ‘cultural sphere’ of information and know-how.

Many other authors have tried to uncover these vital constituents of social change. We rehearse the myriad distinctions no further, however, for what is important is the overall recipe rather than the precise ingredients. Despite the difference in their longevity and ambition both models under discussion have a similar cadence. They stress that it is the interdependency of the strata which enables and constrains change. They agree that each stratum has causal potential and change can be instigated from any level. They posit that sustainable change requires harmonisation through the levels and that there is, of course, every reason to suspect that change provokes disharmony in institutions that are deeply stratified. Finally, these and similar models concur in supposing that change is something that no one steers. It happens perpetually and of its own accord whenever people and groups reflect on their own position.

This latter scenario raises an interesting conundrum: what happens if someone tries to steer change? What happens when there is a formal intervention seeking system improvement? The answer emerging from our longitudinal study of demand management programme theory is that, regardless of where it begins, the progress of any such intervention will also involve a considerable journey as all the other layers absorb the change in interlocking and temporally complex ways. A basic knowledge of system change tells us that what is provoked at one level may be prevented at another. Accordingly, researching service improvement really requires an evaluation of the entire perambulation of a programme theory around a system. With this in mind we submit that an unorthodox but equally useful enterprise for research synthesis is to explore the interconnectivity of different schemes through the different strata. This also supplies us with a broad hypothesis for reviewing specific interventions, namely that they are more likely to work if they gain traction at the different levels. An initial approximation of this conjecture strategy is provided by adapting Figure 9 to Figure 11.

FIGURE 11.

Interconnected levels of change.

We close our review of demand management programme theories at this point and proceed to part 2 of the report, in which we review the available primary literature using the vantage point of Figure 11. Attempting to incorporate this ‘whole-system’ perspective with a systematic review places interesting demands on the exercise. From its origins in clinical research, systematic review began with the idea of examining all studies on one type of intervention. Increasingly, the situation where the reviewer examines many studies of several different families of interventions is becoming the norm. Now we are confronted with the additional challenge that each family should be considered a shifting lineage. As we noted in the introduction, we entered a busy domain with a colossal amount of primary research but also several existing reviews. It is useful to ‘compare notes’ on the most useful way of synthesising evidence on such a moving target. As the implications of such a comparison are largely methodological we locate these reflections in Appendix 2 and more directly to an explanation of our chosen approach.

Introduction

This chapter reviews the primary evidence on the effectiveness of RMCs. Their defining feature is a matter of the ‘level’ or ‘scope’ of the intended reform. RMCs operate within the organisational stratum and involve the insertion of a new institution, centre or unit into the patient pathway to take over aspects of the referral decision-making process.

The chapter begins with some terminological ground clearing and a brief history. RMCs are closely related to a whole family of similar gateways and assessment units and here we draw some simple boundaries around this phase of the review. Even within this relatively well-defined sphere we note heterogeneity. These new centres have been constructed with significantly different ambitions with respect to exactly which ‘aspect’ of the referral process they seek to control and we highlight some important contextual differences.

The bulk of the chapter is given over to reviewing the primary research on RMCs. The evidence base is not abundant and we comment on how we analyse a body of research mostly comprising descriptive case studies. The analysis follows the typical formula for realist synthesis. The basic RMC programme theory is elicited; the ‘ideal outcomes’ in terms of referral restructuring are formalised. The empirical evidence is then used to inspect departures from this basic model and two significant unintended consequences are unearthed in the primary literature. A summary explains the vital configuration of conditions that may contribute to the effective operation of RMCs.

Referral management centres and triage: operational logistics

Referral management centres do not appear explicitly in any policy documents emanating from the Department of Health, and nor do a closely related family of bodies variously labelled referral gateways, clinical assessment services (CASs), single point of access services, triage centres, and so forth. These units have largely been developed piecemeal by individual PCTs and later passed down to CCGs. A common ambition of improved steerage of patients through the referral pathway from primary to secondary care is clear to see. However, Imison and Naylor highlight that referral management schemes implemented by PCTs in their research had a subtly different range of different ambitions, with supporting the redirection of referrals to out of hospital assessment and treatment services and seeking reductions in the overall rate of referrals being the most common. 93 Davies and Elwyn148 also note that RMCs vary in whether their remit is (1) to determine if patients could be diverted to a different service or specialist (such as a GPwSI) with greater capacity, (2) to ascertain whether or not the referral contains all the necessary information to permit further care or treatment to be undertaken or (3) to judge whether the referral was clinically necessary or the patient should undergo further investigation in primary care.

Referral management centres and triage operate as an intermediary between primary and secondary care. They are often set up to manage referrals for conditions where there are a number of different possible referral destinations and treatments, depending on the severity of the patient’s condition. This proliferation of channels means that GPs may find it difficult to keep abreast of the different options and may also not know which service is most appropriate for the patient’s clinical situation. RMCs may also offer a ‘single point of entry’ for the different referral pathways and thus seek to take over decision-making from the GP about the severity of the patient’s condition and who is the most appropriate professional to see that person. For example, GP referrals for mental health services are often channelled through a ‘single point of access’ service, which triages referrals and directs them to the most appropriate service for the person’s needs. They may also offer ‘direct access’ to some services without having to see a consultant in secondary care. For example, in many musculoskeletal care services, in order to obtain a referral to a physiotherapist, it was necessary to be initially assessed by an orthopaedic surgeon. However, in many places across the country, services have been reorganised so that the physiotherapist is now the first person to see the patient. They treat those patients within their specialist remit and refer those that require surgery on to the orthopaedic surgeon. Although all these services operate with slightly different desired outcomes in mind (improved access, reduced waiting times), a key underlying aim is to ensure patients see the right person, at the right time, in the right place.

Accurate data on the prevalence of RMCs are difficult to obtain. In 2007, Coleman and colleagues conducted a survey of 152 PCTs to explore developments in practice-based commissioning (PBC), of whom 101 (66%) responded. 149 Of these, 26 PCTs (36%) had put in place new referral management arrangements and 31 (43%) were developing them. Of the 26 who had referral management procedures in place, only three were using units formally recognised as RMCs. Imison and Naylor’s93 analysis of Choose and Book data from 2009 (which is not representative of all referrals as a significant percentage of GPs still use paper referrals) suggests that 91% of PCTs have triage or assessment services that divert referrals that historically would have gone to secondary care.

Programme theories: how are referral management centres intended to work?

The perceived root of the demand management problem is the generic tendency in the existing patient pathway for unnecessarily large groups of patients to be referred directly from primary to secondary care (Figure 17). RMCs and triage seek to modify the patient pathway by acting as a block both to prevent inappropriate referrals and to triage referrals and sort them into different case-mix categories (e.g. ‘less complex’ to ‘intermediate’ to ‘complex’ patients). The underlying programme theory is that this enables patients to be redirected to the most appropriate services with the necessary skill set to meet the patients’ needs, such as those run by GPwSIs or other Allied Health Professionals. Consequently, this means that patients are diverted away from consultants in secondary care. The desired outcome, as depicted in Figure 18, is that some patients (e.g. less complex) can be referred back to primary care for the GP to manage, the ‘intermediate’ patients can be directed to GPwSI or other Allied Health Professionals in alternative community services and the more complex patients are referred on to the consultant. Referrals for procedures on the ‘effective use of resources list’ are blocked. These are procedures that are deemed to be cost-ineffective, according to systematic reviews of the evidence, and which many CCGs have decided that they will not fund.

FIGURE 17.

Existing elementary pathway and patient volumes. Cons, consultants.

FIGURE 18.

Revised ‘ideal’ pathway. The size of the arrows indicates approximate volumes; the pattern indicates case complexity. Wavy line, existing patient volumes; spots, less complex patients who can be redirected to the GP; horizontal stripes, intermediate patients who can be redirected to community services; diagonal stripes, complex patients for the consultant; solid black, referrals on the ‘effective use of resources list’/require more tests or information. Comms, community services including GPwSIs; cons, consultants; RMC, referral management centre (and synonyms).

Note further that this model assumes a deficiency in the existing system. Triage in its various forms already occurs informally and unevenly in the interface between primary and secondary care. RMCs are perceived as improving rather than introducing the triage process. They are perceived as an attempt to ensure that triage happens, as GPs or consultants are assumed not to be undertaking this process effectively (as depicted in Figure 17). They are also understood to have an educational function, being seen as a means of providing GPs with feedback on their referrals, with the idea that GPs will gradually modify their referral behaviour. Some direct testimony on these various claims has been reported in Chapter 3.

Figure 18 and the accompanying key describe the desired outcomes of RMCs in more detail. If this redistribution of case mix works, the ideal is that fewer patients see the consultant and, consequently, referrals to secondary care go down, without increasing the overall volume of referrals. This also may reduce waiting times to see the consultant. Those who may otherwise have been referred to secondary care are sent to alternative community services. Through having some referrals rejected and receiving feedback, GPs assimilate what constitutes an ‘appropriate referral’ and modify their referral behaviour in the future.

As is self-evident in Figure 18, these are ambitious reforms requiring a high degree of co-ordination of a range of different practitioners at the interface with different institutions. With ambition comes contention and realist synthesis seeks to examine how and to what extent plans come to fruition. One way of achieving this aim is to consider the programme theory alongside the critical ideas that have been raised in the face of such reforms. The empirical literature can then be scrutinised in an attempt to adjudicate between the contending claims.

One critical voice argues that RMCs can be seen as part of the infrastructure through which attempts are made to impose evidence-based practice on clinicians. Harrison and Wood coined the term ‘scientific bureaucratic medicine’ to describe the policies, institutions and organisations responsible for driving forward the implementation (or imposition) of evidence-based practice. 150 They argue that evidence-based practice is a means of managing demand in the NHS through limiting the availability of interventions to those which, according to ‘scientific’ evidence from RCTs, are cost-effective. Thus, ‘scientific’ knowledge about the effectiveness of interventions is summarised and codified into protocols and guidelines (the ‘bureaucratic element’) which guide the user on particular forms of action. In RMCs, decisions about the appropriateness or otherwise of a referral are commonly made through the application of clinical guidelines or referral criteria to referral letters. RMCs also act as a means of preventing referrals for procedures that have been shown, through evidence-based research, to be of ‘limited clinical value’ and have thus been placed on the ‘effective use of resources list’. These are procedures that PCTs and later CCGs have decided they will not fund because the evidence suggests that they do not provide sufficient value for money.

The co-option of clinicians to work within RMCs could constitute a strengthening of the power of the medical profession, as they may impose their values on how decision-making about referrals is made and how guidelines are applied. However, it may also represent a weakening of medical power if these clinicians take on the values of NHS managers to effectively police the activity of rank and file of GPs. Which of these outcomes emerge is likely to depend on how GPs, consultants and GPwSIs are involved in the ways in which the RMC is set up and the control they have over the means through which decision-making process about referrals are made. It is on this matter we need the decisive evidence.

Figure 18 also represents a considerable degree of ‘boundary work’ on the part of different professional groups. By dint of their specialised knowledge and control over accreditation, professional groups gain power that allows them to define turfs, control jurisdictions and exclude outside groups from participating in a rigidly defined range of specialised activities. From this perspective professions can be seen ‘as convex bodies with secure heartlands deep behind the boundary territories’. 151,152 It follows that understanding a particular activity, such as the right to make referrals, is first of all matter of seeking out the controlling power base. But these controls and their boundaries are not cast in stone. As Abbott puts it, ‘Professions only come into existence when social actors contest and create boundaries. A profession is a set of turf battles that are later yoked into a single defensible position in the system of professions’. 152 This alerts us to the need for our review to locate empirical inquiry that has examined boundary walls and boundary work.

In this light, GPs have expressed a number of objections about RMCs. We focus on two key criticisms here. First, GPs express concerns regarding the clinical governance of RMCs, specifically whether or not non-medical personnel have the skills and expertise to make triage decisions about patients. RMCs require GPs to refer patients to them in order to divert patients to alternative services. If GPs hold concerns about the triage process, they may not refer patients to RMCs and find ways of avoiding or circumnavigating them. Second, if GPs do refer to an RMC, they may not take the time to consider the options for the best management options for the patient. Over time, this may lead to a deskilling of GPs as they lose experience in selecting referral routes for certain groups of patients. A further possibility, especially where alternative community services are set up to manage patients, is that GPs may offload unwanted patients to these services. Hughes notes how professional groups with a high standing retain the desirable and interesting work and delegate the more unpleasant, uninteresting or ‘dirty work’ to those with lower standing. 153 This may particularly apply if RMCs or triage are seen as a route to offloading less complex or less interesting work to other Allied Health Professionals with less power or lower standing than GPs.

One final observation to note in Figure 18 is the assumption that both RMCs and triage services have the option of redirecting patients to alternative community services. It goes without saying that this would not be possible if such alternative services did not exist. The devil is clearly in the detail. In some areas, alternative services, including those run by GPwSIs or Allied Health Professionals, have been set up alongside and in conjunction with RMCs. In other areas, GPwSIs and independent clinical assessment and treatment centres (ICATs, which are run by the private sector) pre-existed the RMC, with the RMC being set up in order to make better use of these services. The precise pre-existing configuration of the local health economy is, thus, likely to influence the outcomes of any attempts to redirect patients within it.

We now move to review the evidence on RMCs and triage to test this bundle of competing theories. The synthesis is structured in recognition of three core aims and their potential intended and unintended outcomes.

Evidence review

Most of the evidence evaluating the fortunes of RMCs in reducing and/or redirecting referrals away from secondary care comes from relatively small ‘case studies’. Although there is considerable discussion in academic journals, much of the research literature emanates from ‘professional publications’ and the ‘grey literature’. One reason for this particular profile is that RMCs and triage have never carried the status of a ‘formal’ programme initiated and backed up by a succession of central policy edicts. Instead they represent ad-hoc, if much borrowed, responses to a more general policy imperative to reduce the number of referrals from primary to secondary care. As such, they have been devised, implemented and evaluated separately and locally, with relatively few attempts to ‘compare and contrast’.

Such an evidence base meets with some consternation in the methodological literature. For obvious reasons, this is a RCT-free zone, making a necessity of the idea of secondary analysis of mixed-methods primary data. More problematic is the usage of case studies that verge on promotional material. Especially if the evaluation comes from in-house, there is a well-recognised tendency for the production of ‘good news’ stories. Our response, following Pawson,154 is led by the idea that there can be useful evidence nuggets in studies that are otherwise indifferent (or worse). Accordingly, we do not simply import the material from these case studies: we interpret it. In the material below, the approach is most clearly evident in the discussion of the Manchester RMC in we combine the insight from several different sources that trace its history.

Intended outcomes: patients are redirected to alternative services, referrals decrease, general practitioners change their referral behaviour

Unintended outcome 1: short-circuit

General practitioners short-circuit the triage/referral management centre and refer straight to secondary care

Much is to be learnt about the importance of clinical engagement in the success of RMCs by exploring what happens when it is not there. We found a number of examples in which GPs did not engage with the triage/RMC process at all and simply referred directly the consultant, thus short-circuiting the RMC (Figure 19).

FIGURE 19.

Short-circuit. Comms, community services including GPwSIs; cons, consultants.

CRG Reasearch/Cardiff University158

This report provides a number of case studies of RMCs, one of which briefly describes the RMC set up in Hounslow. The focus of the Hounslow RMC was to reduce the quantity of referrals to secondary care through triaging referrals against agreed referral guidelines and returning to practices those referrals that do not meet the guidelines. The RMC intended to enlist GPwSIs and physiotherapists to undertake the triaging. However, the PCT changed their mind about this practice because GP practices had expressed concern about the use of triaging by GPwSIs and physiotherapists and, consequently, were refusing to use the RMC. In addition, a local shortage of GPwSIs to undertake the triaging in the local health economy brought further challenges to the successful implementation of the RMC. In this set of circumstances GPs appear to choose not to refer to the RMC, resulting in short-circuiting rather than a remodelling of support and substitution.

The authors of the report conclude that when referral management is understood as resting wholly within the remit of professional management, there is a substantial risk that usual practice will find a way to bypass the procedures introduced. However, when clinicians are involved in the decision-making to set up a RMC and see referral management as a process that can support or improve their decisions, simplify the pathway and make the patient’s journey more efficient, referral management will be supported by clinicians. This is another small-scale study with few reliable outcome data. It does reveal another small nugget of evidence adding further support to the developing theory, namely that if clinicians are expected to behave and take on the values of managers and have no power to shape the nature of RMCs, they will find ways of circumnavigating them. RMCs under bureaucratic control will trigger short circuit rather than support or substitution. If RMCs are shaped in a way that represents professional values and agreed outcomes, clinicians will engage with them.

Papers on the Manchester Referral Management Centre

• CRG Research/Cardiff University, Evaluation of Referral Management Pilots in Wales, Final Report. 158

• NHS Manchester, Agenda item 15, paper Number L, Individual funding requests (Non-commissioned Procedures) and Referral Gateway Initiative, NHS Manchester Board Meeting, 7 July 2010. 159

• Introducing a GP-led referral gateway, Pulse, 18 May 2011. 160

• Gateways using nurses to screen referrals, Pulse, 10 August 2011. 161

• GPs axe remote referral gateway run by private firm, Pulse, 2 March 2012. 162

• Referrals gateway worked well, Pulse, 20 March 2012. 163

• Creating ownership of a referral gateway, Pulse, 3 September 2012. 115

• Harmoni website NHS Manchester Referral Management Centre. 164

Further evidence about the importance of clinical engagement of local clinicians to the success of RMCs also comes from the evolution of the RMC set up in Manchester. Evidence about the development of the RMC in Manchester has been collected from a number of sources including the above report from NLIAH on referral management pilots, a number of Pulse articles, websites from the company Harmoni and NHS Manchester and papers from NHS Manchester Board meeting from July 2010 (listed above). The result is a useful example of a blow-by-blow history of the service, which traces the rise and fall of different approaches to referral management in one location. Such historical detail is rarely reported in journal papers, which largely focus on the immediate impact of programmes, rather than providing longitudinal and contextual detail on their implementation. With its emphasis on the fickle fortunes of programme theories, realist synthesis provides a useful means of corralling together different accounts of the same programme. The customary caveat must also be added. This small section follows ‘the story’ of only one RMC, and caution must be exercised in generalising these findings.

The NLIAH report indicates that in 2006, the initial focus on referral management in Manchester was referral tracking and information gathering about the volume of referrals to inform planning to meet capacity by the PCT. 158 This took place with 11 RMCs, which covered 14 PCTs. No referral diversion was taking place, although some RMCs had also introduced referral letter templates in order to gain rapid access to chest-pain clinics. These did not relate to or require compliance with clinical criteria or guidelines.

The report goes on to describes that, outside the daily remit of RMCs, a number of ‘Tier 2′ services had developed whereby GPwSIs assessed some categories of referrals, such as those in areas of low urgency referral impact [ear, nose and throat (ENT) and dermatology] to determine whether or not they could be managed by an alternative provider, such as a GPwSI in a relevant field. GPs, however, did not receive feedback on these referrals. In addition, a number of ICATs had been developed to provide a diagnosis in advance of attendance in outpatients, where, normally, diagnosis would occur within secondary care followed by a further appointment for treatment. The ICATs were run by a multidisciplinary team who confirmed a diagnosis and then directed patients to a range of providers including those in the private sector. Here we see how services within a health economy are often set up in a patchwork manner and there may not be co-ordination in how the services are intended to complement each other.

Two Pulse articles115,160 and information from the Harmoni website and NHS Manchester’s website then chart the further history of RMC activities in Manchester. By 2009, the 14 PCTs had merged into three PBC consortia. The three PBC consortia in Manchester had a 7-year contract with the ICATs described above, which cost £7M per year, but it is described in the sources as being ‘drastically underused’. As this was a block contract (the PCTs had to pay for it whether or not it was used), the PBC consortia were effectively wasting capital while hospital outpatient referrals were rising unchecked. The PBC consortia’s goal was to ‘reduce hospital outpatient activity levels by using alternative services’160 – attempting to shift demand away from the overused hospitals towards the underused ICATs. The commissioners initially tried to address this via a peer review of referral scheme by feeding back hospital referral data to GP practices but ‘while we identified learning opportunities, it seemed to do little to actually change referral behaviour’ and ‘generally outpatient referral activity hadn’t reduced’. 115 Lack of significant change here is attributed to the data being 2–3 months out of date.

Accordingly, the three PBC consortia tried a different tack. Local clinicians in South Manchester developed a referral gateway, which was piloted in four practices in South Manchester. Two senior GPs triaged referrals and a central booking team managed data flow and the Choose and Book function. Initial outcome data ‘showed great potential to affect outpatient activity’. 160 The implication of these reports are that the initial success of the pilot lay in the fact that local GPs, administrative staff and GPwSIs were involved in setting up and running the RMC. All three PBC consortia then agreed to develop a city-wide referral gateway. Each PBC consortium voted to use part of the local incentive scheme funding to pay for the gateway. 115

The three PBC consortia then met to agree how the referral gateway would work – these included how to give advice back to GPs, governance arrangements and who would triage referrals – Allied Health Professionals and consultants were considered. The PBC consortia took their plans to the LMC (a body that represents the views of GPs to the PBC consortium/PCT) who rejected them the first time. The LMC was ‘not convinced that the scheme would cut referrals massively or save any money. It also felt very strongly that this could not be mandated and GPs should be free to make direct referrals if they chose to do so’. 115 All three boards wanted local Manchester GPs to provide triage support in order to gain the good will and engagement of practices at the initial stage. 160 Owing to ‘problems with internal staffing’, the PBC consortia enlisted a private company Harmoni to run the gateway. The clinical triage work was subcontracted to the local out-of-hours provider, ‘with the proviso that Manchester specialists did the triaging’. 160

Wootton and Whiting115 explain in their article how the referral gateway operated. GP referrals were initially checked by the RMC/triage team for completeness and to determine that the referral was not for a ‘non-commissioned’ procedure; that is, procedures that were not funded by NHS Manchester unless clinically appropriate. These included procedures such as alternative or complementary medicines, cosmetic surgery not related to breast reconstruction or trauma and tattoo removal. If data were missing or the referral was for a non-commissioned procedure, an electronic advice note was returned to the practice. The referral then underwent clinical triage by a team of GP triagers who determined whether the referral could continue, be diverted to an alternative service or returned to the GP with advice on management and further tests. A paper regarding the referral gateway tabled an NHS Manchester Board meeting on 7 July 2010159 expressed the hope that feedback from the gateway would enable GPs to understand which procedures were on the non-commissioned list and, therefore, would reduce their referrals for these procedures in the future which, in turn, would result in a decline in the number of referrals returned to GPs by the referral gateway.

The referral gateway cost £220,000 to set up. 164 In terms of referral outcomes, it was a success. After 6 months, there was ‘a 1.2% overall reduction in outpatient activity instead of the forecast 3.8% growth’. 160 After 18 months there was a ‘12–15% reduction in outpatient referrals and an increase in ICAT usage from 42% to 79%’. 115 Wootton and Whiting also report that during the 18-month pilot period, the gateway produced and estimated cost saving of £4–6M. 115 The Harmoni website reports that the RMC returned 7% of referrals to GPs. 164

However, other articles reveal that the RMC underwent further changes in its configuration, despite its apparent success in curbing referrals. A Pulse article on 2 March 2012 reported that the Manchester CCGs had opted to bring the referral gateway ‘in-house’ and had not renewed their contract with the private provider Harmoni. 162 Two main reasons cited are costs and that the referral gateway did not have the required local knowledge of Manchester services. 162 Two main reasons cited are costs and that the referral gateway did not have the required local knowledge of Manchester services. 162 The secretary of the Manchester LMC and board member of the CCG said outsourcing was expensive and some minor issues occurred owing to lack of local scheme knowledge and tier two services. Most local GPs are happy back in-house with local people involved. 162 A NHS Manchester spokesperson was cited as saying the decision was based on cost and also a need to employ local people and help the local economy. 162 However, to counteract this implied claim that part of the reason why the CCG did not renew its contract with Harmoni was because it did not employ local people, yet another Pulse article asserted that the location of the referral gateway in Southampton had ‘no impact’ on its functioning. 163

Different accounts of this period in the development of the Manchester referral gateway make different claims about who undertook the triage and where this took place. It is worth dwelling on the ambiguity here as it relates to general concerns expressed by GPs about the way RMCs are run. These were well rehearsed in Chapter 3, but to reiterate, GPs have argued that ‘For an anonymous nurse working to a set protocol in a remote RMC to reject that carefully considered decision out of hand is professionally frustrating, to say the least’ and that ‘Nurses assess patients according to rigid criteria and do not have the experience to make flexible decisions in the same way a doctor can’. 114 We can surmise that the growing complexity of the system, led to concerns about who exactly was undertaking the triage. This might have influenced GPs’ support of the gateway (and perhaps their willingness to refer to it) and might have influenced the decision to bring the gateway back in-house.

For example, the paper prepared for the NHS Manchester board meeting on 7 July 2010159 indicates that although the initial checking of procedures would be undertaken by Harmoni, based in Southampton, the clinical triage would be carried out remotely by experience local GPwSIs working in defined areas. Indeed, a Pulse article from 2012 declaring the success of the Manchester RMC argued that ‘One very important factor in setting up the gateway was the use of local GPwSI and senior, well-known GPs. It meant the other GPs respected the decisions made and the feedback they were receiving’. 115 Although this might have been the aim, an article in Pulse based on a Freedom of Information request discovered that ‘two podiatric nurses and a non-medical prescriber with a diploma in podiatric medicine’ were used to triage GP referrals for vascular surgeons. 161

This case study illustrates how local approaches to referral management change are often in partial flux. To summarise, we see that policy-makers adopt new solutions when initial approaches fail (e.g. peer review) and that they attempt to roll out successful initiatives (the pilot RMC set up by clinicians) on a larger scale. However, in doing so, changes and adaptations to the original approaches are made; in this case a lack of local staffing infrastructure meant that the service was no longer run by local staff, and so the private company Harmoni, based in Southampton, ran the referral gateway. The local involvement of GPs, GPwSIs and administrative staff had contributed to the success of the initial pilot in terms of both clinical engagement and familiarity with the local availability of services. Scaling up meant that these two essential features were lost. The LMC also saw the RMC as an attempt to curb the clinical autonomy of GPs and wished to defend their right to refer directly to secondary care. Although the scheme successfully reduced referrals, after 18 months the configuration of the gateway was changed again to bring the gateway back ‘in-house.’ According to some reports, this was to tackle the problem that those undertaking the triage lacked local knowledge about available services.

The above exposition probably reflects the limit of secondary analysis in piecing together a reliable account the history of one RMC. For instance, one member of our advisory group with knowledge of this particular RMC also suggested, contrary to some of the published accounts, that a further reason why the RMC had to be reconfigured was a lack of engagement by local GPs because they did not know the GPs who were conducting the triage and questioned their capability for doing so. The pocket history, however, does provide value in pointing to the systemic character of organisational change. It illustrates something we characterised earlier as the ‘punch-bag’ problem of demand management: solving one problem at one level has a tendency to create an unintended outcome at another.

Box 30 provides an interim summary.

BOX 30 - Interim summary

Comparing the case studies reported to this stage reveals some incertitude about outcomes but also something of the pattern to confusion. In order to be ‘successful’, RMCs require collaboration between NHS managers and clinicians to agree on the purpose of the RMC and its remit and functioning. In the positive accounts described above, clinicians were able to exert considerable control over the development of the guidelines that determined how referrals were dealt with and the governance structures for ensuring this was done safely. In terms of the theory under consideration, RMCs were successful because this level of clinical engagement allowed clinicians to maintain their power over how the RMC is run. A key element of this was the use of experience GPs to conduct the triage. In turn, the perception that the RMC is being managed according to clinical standards of governance leads to more widespread support and trust in the RMC on the part of local GPs and consultants. Finally, the mechanism through which the RMC was deemed to work was not simply through the blocking or redirection of referrals but through GPs receiving feedback about their referrals and learning about what constitutes an appropriate and the availability of local services. This led to GPs reflecting on their own practice changing their own referral behaviour. The role of peers in delivering this feedback was seen as crucial to its success.

Where RMCs were not ‘successful’, there is evidence that this was due to them being perceived as attempts by the NHS managers to curb clinical autonomy. In terms of the underlying theory on scientific bureaucratic management,150 a RMC can be viewed, and was viewed in some quarters here, as an attempt to enforce managerial values on the medical profession. This suggests that if clinicians are not able to exert significant control over the remit and functioning of RMCs, they will not garner the trust of local clinicians. A further key issue in this case study appeared to be whether or not local GPs were involved in the triage process. This may have been important in ensuring that, and fostering trust that, the triage process was being conducted safely. Further, RMCs were not successful where it was perceived that a single interest dominated the triage decision-making – recall the CAS in Harrow, where the initial pilot study was not successful because the consultant maintained control over the triage process and few patients were referred on to the GPwSI.

Unintended outcome 2: triage/referral management centres increase referral volumes

A second pattern discernible in another series of case study reports is that RMCs/triage may have no effect on, or even increase rather than decrease, referral volumes. A function of RMCs and triage is to divert referrals to alternative community services, such as GPwSIs or those run by Allied Health Professionals. However, these services may simply provide additional capacity, rather than reducing referrals.

The lack of change or increase in the volume of referrals can be a consequence of patients who would not normally have been referred finding their way into the system, as GPs pass on ‘dirty work’ to other professionals. Alternatively, it may simply be a result of previously unmet demand that can now be addressed as access to the service has been improved. The result is that there is no change to the case mix of patients reaching secondary care. This is depicted in Figure 20, in which the total number of referrals entering the system increases, as does the number reaching alternative community services. The total number of referrals to consultants may decrease. A number of studies in our review show how this pattern of outcomes may arise. Here, we review a number of studies that explore this outcome in more detail.

FIGURE 20.

Increase in overall volume of referrals. The size of the arrows indicates approximate volumes; the pattern indicates case complexity. Wavy line: existing patient volumes; spots: less complex patients who can be redirected to the GP; horizontal stripes: intermediate patients who can be redirected to community services; diagonal stripes: complex patients for the consultant. Comms, community services including GPwSIs; cons, consultants.

Imison and Naylor93

Some of this text is reproduced from The King’s Fund from Imison C, Naylor C. Referral Management: Lessons For Success. London: The King’s Fund; 2010. 93

Here, we draw on the authors’ quantitative analysis of referral rates across PCTs and interviews with key respondents to give some insight into the extent to which RMCs in particular had achieved their outcomes and also some possible unintended outcomes of these and triage services. They compared changes in referral rates and referral volumes using Choose and Book data from 2005 to 2009 between the 22 PCTs with the highest proportion of referral activity going through some form of referral management activity (at least 30% of Choose and Book referrals in these practices went through triage or to a GPwSI) and PCTs nationally. This analysis indicated that there were no significant differences between the 22 PCTs undertaking a substantial amount of referral management activity and PCTs nationally. The authors report that ‘PCTs that had introduced full referral management centres were picked out, but were no more or less successful than others’.

They also used Hospital Episode Statistics data to analyse percentage change in volume of first attendances between 2005 and 2009 and found ‘no discernible difference between the pattern of activity for PCTs with active referral management and those without (Figure 21).

FIGURE 21.

Percentage change in volume of first attendances referred by GPs, all PCTs (excluding outliers) 2005–9. Reproduced with permission from The King’s Fund from Imison C, Naylor C. Referral management: lessons for success. London: The King’s Fund; 2010. 93 Hospital Episode Statistics data copyright © 2015. Reused with the permission of the Health and Social Care Information Centre. All rights reserved.

They also examined the distribution of GP practices by referral rate in 2008–9 and found no significant difference in the distribution of referral rates at practice level between PCTs undertaking referral management and all practices (Figure 22).

FIGURE 22.

Distribution of GP practices by referral rate 2008–9, top 22 PCTs undertaking significant amount of referral management and against all practices. Reproduced with permission from The King’s Fund from Imison C, Naylor C. Referral management: lessons for success. London: The King’s Fund; 2010. 93 Hospital Episode Statistics data copyright © 2015. Reused with the permission of the Health and Social Care Information Centre. All rights reserved.

Qualitative data from interviews with stakeholders from PCTs indicated that participants felt that referral management schemes generally had been effective in diverting activity into alternative out-of-hospital services but were far less effective in reducing the overall demand and improving the overall quality of the referral process. Some reported that reducing waiting times and increased access to services was encouraging GPs to refer more and that referral to triage services might mean GPs were less likely and less needful of ‘thinking for themselves’ and more likely to refer onwards than to manage patients themselves. There was also a belief that new out-of-hospital triage services had increased awareness of specific conditions and, thus, identified unmet needs in the patient population (a phenomenon discussed earlier as supplier-induced demand). For those PCTs where referral management schemes were not mandatory, respondents believed that the ‘worst’ or ‘highest referring’ GPs could refer outside the schemes and directly to the trust, evidence of the unintended consequence discussed in the previous section (see Unintended outcome 1: short-circuit).

General practitioners with special interests: policy, promise and pitfalls

The development and formalisation of the innovative role of ‘Specialist GPs’ is usually traced back to the 2000 NHS Plan,14 which called for a thousand new appointments. The initial designation itself was subject to a degree of inter-professional wrangling and has since been replaced with the more precise ‘general practitioner with a special interest’. The significance of the new role was further underpinned in a parallel Department of Health policy of organisational decentralisation promoted in Shifting the Balance of Power,173 which recommended increases in provision in the community and the local health-care economy. Two further 2002 documents,174,175 produced jointly by the Department of Health and the RCGP, marked the professionalisation of this new function, setting expectations for duties, recruitment, activities, competencies, facilities, accountability and payments. A rather wide range of GPwSI functions was envisaged (Box 33), listed in a RCGP information sheet (updated 2006) as follows:

Another crucial facet of the GPwSI initiative is the very wide clinical terrain in which these individuals have been expected to operate. These are listed below in a table reproduced from a ‘snapshot’ in a 2006 RCGP information sheet that lists current and planned services (Figure 23). 177 More recent figures on the extent of GPwSI provision do not appear to be available (a request to the RCGP information service on this matter drew a blank).

FIGURE 23.

Primary care trusts with current and planned GPwSI services (%) by clinical area. Reproduced from Pinnock and colleagues176 under the Creative Commons Attribution Licence Attribution 2.0.

Another notable variation in the development of GPwSI provision was spatial: where do these practitioners tend to operate? Locality is not insignificant as the ability to professionalise is partly territorial, whether one substitutes or supports depends on whether one’s potential partners are at arm’s length or cheek by jowl. Rosen and colleagues report on a variety of operational bases: (1) GPwSIs providing services in outpatient departments, (2) those offering supplementary procedure in community settings and (3) those developing services and education in management centres. 177

These initial snapshots of duties, clinical specialisms and locations already carry some status as ‘evidence’. These bureaucratic domains have quite different patient pathways, staffing structures, organisational logistics and decision dynamics, and these contingencies alone carry fair warning against the notion that GPwSI interventions could and would be rolled out in similar fashion. One size is very unlikely to fit all. We will return to evidence on the further implementation of the scheme in the ensuing years, and complete these introductory remarks simply by noting key criteria on which the scheme has become evaluated:

‘The stated aim of government in formalising GPwSI services was to shift the balance between primary and secondary care for reasons of lowering waiting lists, increasing cost effectiveness and patient accessibility’. 178 These key policy ambitions with regard to demand management are articulated further (especially in points 1, 2 and 3) in the following summary (Box 34), paraphrased from the Department of Health policy document. 179

Programme theories articulate the rationale behind an intervention: the reasons why it should work. All of the above claims envisage GPwSIs as having the capacity to redirect patient flows, change the skill mix, moderate service overuse and underuse, and so on. Whether or not these functions come to pass depends on the pre-existing systems into which they are dispatched, and the critical literature also spells out reasons why complex services are reluctant to change. Here we articulate three potential pitfalls to stand beside the policy promise, leaving the main body of the review to investigate how the balance falls in practice.

Substitution or support

Figure 24 presents a highly simplified diagram that begins to explain the intended purpose of role substitution (DM in this and subsequent figures refers to a ‘decision-maker’). A new post (or service) DM_2.5 is introduced, drawing some of the decision-making mandate from the previous incumbents DM₂ and DM₃. The responsibilities associated with the new roles and the qualities of their incumbents are assumed to be the underlying mechanisms that will generate a more efficient harnessing and smoothing of demand. Our investigation can thus be thought of as an exploration of the functions and functioning of ‘DM_2.5s’ in their GPwSI guise.

FIGURE 24.

Existing system and intended change.

The figure (upper section) represents the patient pathway and the decision-making responsible for referrals along that chain. DM₂ represents more junior and more generalist DMs; DM₃s are more specialist and senior. The basic assumption (lower section) is that DM_2.5s can take on those tasks that by a process of accretion have fallen on DM₃s but which do not really require high and specialist levels of expertise. A newly trained and accredited DM_2.5 can thus take over this portion of decision-making, with benefits ensuing because of task appropriateness, economies of scale, remuneration efficiencies and a consistency of approach. A subsidiary theory, much in play in the introduction of the GPwSI role, is that a DM_2.5, often with a similar background as a DM₂, will be in a better position to understand and to translate between the concerns of the generalist and the specialist.

A basic familiarity with the critical literature also provides a pointer to the main area of contestation, namely whether the new roles fetch up in substitution or support? A new role is devised but does it take on substantial decision-making duties or does it remain subordinate to the previous decision-making channels? As in the previous chapter, we will use the ‘substitution or support’ motif in assessing the impact of role change. Dozens of primary studies make reference to this benchmark in relation to quite different substantive domains (e.g. Zwijnenberg and Bours180) and there is a clear imperative to investigate this distinction in our GPwSI review.

Figure 25 provides a visual reference to this key benchmark, though note that we add a third potential outcome: ‘short-circuit’. Here we refer to another unintended consequence of the new role. Existing referral pathways and systems are rarely singular, especially under the current movements towards ‘informed patient’ and consumer choice policies such as ‘Choose and Book’, which are intended to widen rather than tailor patient pathways. In some cases these alternative routes (both formal and informal) enable DMs to bypass the new role altogether and this further eventuality can also be monitored in our review.

FIGURE 25.

Intended and unintended decision-making structures. Substitution: the new DM_2.5 role is established with real decision-making powers independent from DM₂ and DM₃. Support: the new DM_2.5 role is established, which provides intermediary support but remains under the control of DM₂ and DM₃. Short-circuit: the new DM_2.5 role short-circuited with DM₂ utilising alternative pathways to refer patients to DM₃.

The anatomy of decision-making

Many different modifications are intended to follow with the introduction of GPwSIs, but one key assumption is about the handover of decision-making. The ‘decision-making’ literature is voluminous: conceptual distinctions are drawn between rational and irrational choice, bounded rational choice, interpretative repertoires, tacit and explicit decision-making, evaluation, judgement, discretion, intuition and so on. For a more concrete illustration it is useful to examine the everyday stock of decisions made in the work environment. Thompson and Dowding, for example, produce a 10-fold typology covering the typical array of decision type in the routine duties of nursing. 181 Nurses’ decisions involve (1) diagnostic judgements on a patient, (2) gauging whether or not they are stable, (3) evaluating for any deterioration, (4) predicting upcoming problems, (5) choosing between interventions, (6) targeting who will receive the most benefit, (7) choosing the optimal timing of a procedure, (8) determining which patients and their families need support, (9) deciding how to reassure patients and (10) organising handover to other staff.

All of these distinctions, theoretical and practical, point to the fact that decisions are never singular, and they suggest that decision-making has its own internal architecture. In trying to assess the extent of change introduced with the introduction of GPwSIs it is crucial to understand more precisely what kinds of decisions are changed (or are reluctant to change) as these new roles and regimes are introduced. Clearly, we cannot operate with the close focus of Thompson and Dowding’s model, which uses fine-grained and role-specific distinctions. We require a more abstract classification system that establishes some core differences in decision-making responsibilities. Figure 26 is a simple hierarchical model that we will use in gauging the ambit of role change in the interventions under review. Decision-making ranges from the strategic to the routine. Role-change interventions that achieve ‘substitution’ involve climbing such a ladder of decision-making, whereas others will end in ‘support’ and would only cover the lower-order functions. These lower-order functions carry a name in the literature namely the ‘dirty-work’ hypothesis182 and a crucial point of investigation in what follows is to examine to what extent GPwSIs have to settle for enacting rather than making decisions.

FIGURE 26.

The DM_2.5 decision hierarchy (ladder of decision-making).

Professional control and boundary work

Much individual and institutional capital goes into development and maintenance of existing role hierarchies and changing the health-care division of labour will always be politically contested – medicine often being regarded as the archetype of profession power. The introduction of GPwSI services involves adjustments to existing decision-making structures and this in turn invariably involves the attempt to manipulate professional boundaries. Much prior scholarship has gone into what it is that make such boundaries porous or unbridgeable and we need to apply these theories in understanding the elbow-room provided to GPwSIs.

Studies of how professions come into being and are able to exert control over significant national agencies and agendas are one of the mainstays of the social science literature. Here we call mainly on a basic structuration theory, usually traced back to the work of Abbott in two influential contributions. 153,154 By dint of their specialised knowledge and control over accreditation, professional groups gain power that allows them to define turfs, control jurisdictions and exclude outside groups from participating in a rigidly defined range of specialised activities. From this perspective professions can be seen ‘as convex bodies with secure heartlands deep behind the boundary territories’. 154 It follows that understanding a particular activity, such as the right to prescribe medicines, the right to make referrals and so on, is first of all matter of seeking out the controlling power base.

But these controls and their boundaries are not cast in stone. As Abbott puts it, ‘Professions only come into existence when social actors contest and create boundaries. A profession is a set of turf battles that are later yoked into a single defensible position in the system of professions’. 154 This alerts us to the need for our review to locate empirical inquiry that has examined boundary work as well as boundary walls. We depict this line of inquiry in Figure 27. As is evident in the Care Closer to Home ‘manifesto’, role-change policies are intent on creating a new decision-making apparatus (depicted in the centre of the diagram). Whether the new structure materialises, whether there is a real change in decision-making responsibilities, depends in part on the dynamics of professional closure (depicted in the periphery of the figure). The solid arrows represent pre-existing lines of professional control – here we examine the staying power of direct, authoritative and established lines control over a particular decision. The striped connectors represent the boundary work that occurs under ‘reprofessionalisation’ – how are alliances drawn and redrawn with the introduction of new roles? The role of alliance-making, as we shall see, turns out to be crucial in the development of the GPwSI role.

FIGURE 27.

The shaping of decision-making across professional boundaries.

General practitioners with special interests evidence synthesis

We now move to the main body of the GPwSI review, which unfolds as follows. We begin by analysing the main ‘outcome studies’ covering materials on referral volumes and pathways, accessibility, waiting times and patient profiles associated with the introduction of GPwSI services. There have been surprisingly few such investigations, perhaps because the dislocated and amoebic nature of the GPwSI role makes the isolation of cause and effect patterns particularly difficult to pin down. The studies that do exist reveal a distinctly assorted and irregular picture in relation to changes in demand-and-referral regimes.

We then proceed to examine primary materials that may help to explain the mixed messages from the outcome literature. These commence under a subheading ‘process and context studies’. Our strategy here is to use the studies to interrogate the hypothesis just raised on substation or support, on decision dynamics and on professional alliances. We also encounter a substantial amount of evidence on the impact of reforms in wider NHS policy, and system complexity and heterogeneity.

The primary inquiries in both sections are presented and analysed study by study. We make brief references on the rationale for study selection and on their research quality but concentrate on the detailed findings emerging from each piece of research. Each primary study, of course, has its own set of concerns. We extract from each study only those items that contribute to our conceptual framework (ignoring, for instance, material on GPwSI remuneration and field training). Some studies have findings that reach across our analytic framework but, with a few exceptions, we stick to the one-study-at-a-time format, which we consider helpful in building the explanatory narrative.

Conclusion: emerging patterns

Putting the outcome and process studies together reveals something of a service delivery melee – but there is some order within that complexity. Our first conclusion takes us back the matter of what exactly is being evaluated in this research on the GPwSI role. GPwSI services do not represent an orthodox ‘planned programme’. There was never any point in the above history when a distinct ‘GPwSI intervention’ was specified and designed. As with much policy-making, it all began with a vision (thou shalt create a thousand GPwSIs), a rough specification of their role and the introduction of accreditation and training paraphernalia.

Research synthesis, of course, is much simpler if there is a designated series of programmes implemented to an agreed blueprint. GPwSI services, by contrast, originate at the confluence of two grand policy imperatives. The first is a breaking of the ranks of GPs. For many years there has been restlessness with the GP ‘generalist’ role and a fragmentation into an internal hierarchy beginning with the distinction between GP principals and salaried GPs. As noted in several papers above, this move to increasing specialisation began long before the RCGP call for a thousand GPwSI posts. A parallel and more momentous policy shift occurred under the broad rubric of ‘care closer to home’. In the same time period, fundamental changes in health-care management through PCTs and CCGs sought to redraw professional boundaries and power structures. PBC allowed for a more local approach to initiate and support services, moving them to locations ‘nearer to the patient’.

The resultant vector was the GPwSI, a notional function with uncertain powers. As we have seen, GPwSI posts have been deposited unevenly across every corner of medicine, with a professional remit ranging across clinical practice, education, leadership, management, research, quality assurance and public health. GPwSIs can be considered quite properly as a portmanteau profession launched, from the point of view of the reviewer, in a perfect storm of heterogeneity. Against this background it is not possible to issue an overall verdict on their contribution in solving demand and referral overload.

It is possible, however, to discern within the primary literature certain configurations into which the GPwSI role has settled. Several of the primary studies above describe a ‘default’ arrangement in which referral patterns remain relatively unchanged owing to residual pressures in the system. This occurs when (1) the consultant (DM₃) remains in relative control of referrals and the protocols that govern them, and when (2) the GPwSI (DM_2.5) settles for a role that concentrates on the support functions in managing the subsidiary apparatus and focuses on prescribed subset of ‘simpler’ patients, and when (3) the GP (DM₂) retains referral habits either by short-circuiting the process with direct referrals to secondary care or by concentrating on patient reassurance regardless of demand implications. Under this scenario there are no gains in demand management.

Such a pattern contrasts with a second configuration, often won under many years of adaptation and compromise, in which the division of power, motivation and expertise is more perfectly spaced, and in which (1) the consultant (DM₃) retreats to the specialist role in the treatment of complex patients and become advisor to rather than supervisor of the system to identify them, and (2) the GPwSI (DM_2.5) deals with an agreed and intermediate case mix and works in a (physical and social) space that is independent from the GPs and consultant surgeries, and (3) the GP (DM₂) works to protocols delivered via the GPwSI educational and management functions. This scenario represents a system-wide reconfiguration and is most likely to occur under the remit of the triaging and gateway systems reported on in the previous chapter. Referral outcomes are similar to best practice in RMCs; that is to say referral accuracy may improve, waiting times may shorten and patient satisfaction may increase, but the overall level of referral activity is likely to increase. Improvements at many points across a system tend to prompt supply-induced demand across the board.

The question arises about the best way of getting from scenario 1 to scenario 2. In many of the cases reviewed, this journey has involved a reflective process of ‘muddling through’, with one after another of the systems strains being identified and dealt with. Box 18 provides an agenda for such an exercise as perceived though the eyes of GPwSI services in the Rosen and colleagues’177 study. We borrow a term here from Charles Lindblom’s classic article ‘The science of muddling through’. 198 Instead of comprehensive analysis of every policy option, a much more constrained process of ‘successive limited comparison’ is really how policies are developed, insists Lindblom. While we accept this non-pejorative assessment, it still begs the question of whether or not GPwSI services could be developed in a less piecemeal fashion.

We close with a tentative solution being developed to handle these situations in which there is a grouping of semi-independent bodies, each with different agendas and capabilities, which are set some goal that involves collaboration. The interdependencies in these networks are so complex that it is impossible to design and control them centrally. Rather than dealing with a thousand post-hoc improvements, it is sometimes possible to incorporate ‘group model building’ into the development stage of interventions. Group model building is a form of decision-making that involves diverse stakeholder groups working with a facilitation team to solve a focused problem within a complex system. The classic components involve testing different scenarios and strategic options, modelling their outcomes (intended and unintended) from different perspectives and testing to find points of compromise between different ‘scripts’. Recent examples in devising new systems of obstetric care199 and responses to domestic violence200 have been published. This approach captures perfectly the underlying dilemma of introducing new roles, such as GPwSIs, into complex referral systems. Whether or not it is able to shift solutions from ‘ex post’ to ‘ex ante’ remains to be seen.

Introduction

We turn in this chapter to a review of our third form of mandate change, in which systems and organisational structures remain intact but the responsibilities and remits assigned to a given role are shifted. An example in this family is GP direct access to diagnostic tests and services.

The notion of providing GPs with direct access to clinical tests has been around for half a century. We open the chapter with a very early sighting in a study by Clarke and Rickards. 201 They audited a 3-month period of GP open access to an X-ray service from a Sheffield practice with 4000 patients (Table 3). During the 3 months, 29 patients used the direct access service and subsequently only five (17%) were referred to hospital or outpatient services. The authors calculated a cost saving to the NHS of £108.13s.9d.

Diagnostic tests such as medical imaging are a key feature of modern medicine; it has been suggested that 75–80% of medical decisions are based on a test result. 202 Many diagnostic tests and services such as imaging and endoscopies have conventionally been situated in secondary care and often the specialist in the outpatient clinic is the gatekeeper to these tests. Traditionally, the services of radiology and other diagnostic services have tended to be available only through hospital infrastructure. Indeed, a special article in the Lancet in 1962 suggested that there needed to be a physical redesign of hospitals to bring diagnostic services to the front and thereby enable speedy access for patients from outside the hospital. 203 Direct access to tests is, therefore, an extension of GP responsibilities into a remit that previously required specialist referral.

General practitioner direct access to tests has a relatively long history within the sphere of demand management interventions. A review of open-access pathology services as long ago as 1973 suggested that ‘use of pathology should be a significant feature of modern general practice’ and that the policy of the then Department of Health and Social Security was that GPs should have free and open access to the pathology services provided in hospitals. 204 A joint working party of the Royal College of Radiologists and the RCGP suggested in 1981 that direct access to radiological services was essential for primary care physicians. The report outlined that GPs should have a similar right of access as a consultant, although provision should be determined locally bearing in mind resources and ensuring that proper indications for testing are observed.

More recently, the Department of Health’s 2008 policy document Care Closer to Home205 advocated radical service redesign, with an emphasis on devolving aspects of a patients care pathway from secondary to primary care. The coalition government’s 2010 White Paper Equity and Excellence: Liberating the NHS206 aimed to shift decision-making closer to the individual patient and devolved power and responsibility for commissioning services to local consortia of GP practices and included specific targets for introducing patient choice for diagnostic testing by 2011.

This chapter first outlines the programme theories that underpin how GP direct access to tests is purported to work in managing demand for specialist services. It then offers some potential impediments to enacting procedural change as envisaged through the programme theories. The now familiar ‘turf war’ issues arise here in battles over claims to expertise and into whose remit the ‘dirty work’ of organising necessary diagnostic tests falls. The main body of the chapter is the evidence review, which is organised around a typology of the functions of diagnostic tests.

Programme theory

The principal idea behind allowing direct access to tests is to enable GPs to carry out tests required for the ruling in or out of diagnoses for certain conditions, or the initiation/maintenance of treatment, without the need to referral to secondary care. The underlying theory is about creating a more efficient division of labour, whereby the GP can use the test results to decide whether the patient requires management or treatment by a consultant or can be managed in primary care. There are a multiplicity of tasks and procedures to be implemented and managed along the patient pathway from primary to secondary care. The basic demand management programme theory here is about ensuring that the right person conducts the right task in the right sequence at the right price. The basic premise is that a GP is able to select an appropriate test given the patients presenting symptoms and to interpret the results accurately in order to make further management decisions without an initial referral to a specialist.

Diagnostic tests are routinely used in modern medicine to assist in diagnosis and establish appropriate management plans. Without direct access to diagnostic tests, a GP must make a referral to a specialist in secondary care for a patient to undergo the test. If the test excludes any serious disease pathology, or indicates a condition that can be managed in primary care, the patient is then discharged back into the ongoing care of the referring GP. With GP direct access to tests the specialist referral can be avoided in some patients and for others the results of the test can ensure the referral is to the most appropriate consultant. The premise is that a better, more efficient division of labour is achieved by shifting the point at which some decisions are made to earlier in the patient pathway, in primary as opposed to secondary care.

General practitioner direct access to diagnostic tests is thought to improve demand management by:

1. Preventing inappropriate referrals: patients are often referred to secondary care for a diagnosis as the GP is uncertain if their symptoms are indicative of serious pathology. Direct access to tests can reduce the number of referrals, as it allows the GP to make the diagnosis/establish a treatment plan and then manage less complex patients in primary care. Only those patients with serious pathology, those requiring surgical intervention and those with complex management issues need to be referred.

2. Earlier diagnosis and reduced waiting times: a number of patients will require referral to secondary care subsequent to the test. A further benefit of GP direct access is that waiting times to the diagnostic test can be reduced as the need for a specialist appointment to obtain the test is bypassed. Furthermore, treatment can be initiated sooner, even if that treatment needs to be managed by a specialist, as the results of the test will be available for the initial consultation.

3. Reducing costs: there are expected cost savings from reducing the number of referrals to secondary care and in reducing the number of consultations necessary if referral is required.

The adjustment may be relatively minor, involving the customisation of responsibilities for tasks rather than the creation of new roles or organisational units. The basic rationale is depicted in Figure 28 and a value chain for the role of diagnostic tests is shown in Figure 29. However, further examples of specific patient-care pathways utilising direct access given in Figures 30 and 31 demonstrate how complex direct access to testing is in reality, with a myriad of diagnostic and treatment decisions to be made along the pathway.

FIGURE 28.

Direct access to tests.

FIGURE 29.

Pathology value chain. This information is licensed under the terms of the Open Government Licence (www.nationalarchives.gov.uk/doc/open-government-licence/version/3/) with permission from David Laszlo Partnership from David Laszlo Partnership and the Department of Health Aligning Diagnostics, London, Department of Health. 207

FIGURE 30.

Birmingham South Central clinical pathway for headache. Reproduced with permission from Birmingham South Central CCG. 208

FIGURE 31.

Birmingham South Central clinical pathway for 24-hour ECG. Reproduced with permission from Birmingham South Central CCG. 208

Impediments to procedural change

Inevitably, in practice there are barriers to such shifts in decision-making responsibilities. The basic programme theory rests on the assumption that the GP is best placed and has the appropriate skills to decide when a test is necessary and to make appropriate management decisions on the basis of the results.

However, some argue that GPs do not have the appropriate skills to make informed decisions regarding the need for a test and an unintended consequence is that some patients will receive tests inappropriately and demand on diagnostic services will, therefore, increase. 209 Once again, we may expect ‘turf wars’ at play here; specialists lay claim to the expertise and clinical skills that can diagnose without excessive testing.

Perceived expertise is also a potential issue in the interpretation of the results of a test. Turf wars abound here, too, in that the specialist may again lay claim to the expertise necessary to make appropriate management decisions rather than relinquish control of this to a generalist. The specialist, therefore, controls access to diagnostics and guards their professional boundaries closely. In some cases it may be that the GPs themselves feel that the interpretation of the results lies outside the limits of their expertise and an unintended consequence is then that direct access is bypassed and referral to secondary care remains the pathway of choice.

As in the previous chapters, there are also potential impediments concerned with the nature of the duties which change hands. There are some tests that can be perceived as ‘dirty work’. No one camp wishes to lay claim to the work as within their specific boundaries, although the test remains a necessary stage in the patient care pathway. For such unwanted labour, an unintended consequence may well be a bypassing of direct access by the GP and referral to secondary care. However, specialists, in turn, may block referrals in the absence of the necessary tests and the patient wait may be extended. As per usual, we enter the review against a background of contested ideas and seek evidence to shed light on the hypotheses presented.

Evidence synthesis

An important proviso in reviewing the evidence on direct access to tests is that ‘all tests are not created equal’. Notwithstanding the differences between imaging, pathology, endoscopy, etc., the function of a particular test in a particular context differs significantly. We begin the evidence synthesis with an outline of typologies of the functions of tests and organise the subsequent review according to one typology, as this provides more explanatory power than reviewing the evidence by test or by condition.

In discussing the types of tests available, a common approach is to distinguish between the technologies or sample source underpinning a test, for example endoscopies versus CT scanning or blood tests versus urine tests. A more useful typology in the sphere of demand management is one which distinguishes between the function of a test. For example, the AHRQ Methods Guide for Medical Test Reviews differentiates three types: those that screen for the likelihood of a disorder; those that diagnose current disease; and those that monitor progress and/or assess response to treatment. 210 The Department of Health strategic guidance for aligning diagnostics outlines five functions of tests: screening for clinical risk; providing reassurance; supporting diagnosis; monitoring treatment effectiveness; and guiding the selection and/or titration of therapy. Such a typology can, therefore, make distinction between a CT scan for chronic headache whose aim is to reassure an anxious patient that they do not have serious pathology, and a CT scan in a patient with focal neurological symptoms where there is suspicion of cerebrovascular insult.

We present here an adapted typology that we then use to structure the review of the evidence. In this typology we make a distinction between diagnostic tests that identify the rare cases of serious pathology (ruling-out tests) and those that aim to provide a definitive diagnosis where one of a number of positive results are expected. Both screening tests and those that monitor treatment may be regarded as clinical indicator tests.

Typology of functions of diagnostic tests

1. A ruling-out test: these are tests where a low diagnostic yield is expected but a positive result suggests serious pathology; the expectation is that most patients would receive a negative result. The rule-out test is necessary to provide reassurance that serious pathology is not present.

2. A differential diagnosis test: these tests would generally have a high diagnostic yield if employed appropriately; the expectation is that most patients will receive a positive result as a test outcome. However, the diagnostic outcome may be one of a number and the actual diagnosis would drive subsequent management.

3. A clinical indicator test: these tests serve to identify at which point on a clinical trajectory the patient is and the results would guide the initiation of treatment or a change of management.

Ruling-out tests

One might expect that direct access to ruling-out tests is a prime candidate for a successful demand management intervention. In these cases a negative finding is just as important as a positive result. The key expectation in terms of demand efficiencies is that the majority of patients will have a negative scan and, therefore, a referral to secondary care can be avoided. This scenario, moreover, provides the patient with considerable reassurance. The literature gives us two examples in this modality: imaging for patients presenting with chronic headache and endoscopy for patients presenting with dyspepsia. In both cases, the test is used to rule out serious pathology, such as a brain tumour in the case of headache and gastroesophagial cancer or peptic ulcer disease in the case of dyspepsia.

As we will see, these referral efficiencies do not always materialise. Some evidence indicates that a number of patients with negative results still receive a referral to see a specialist. There is also debate about whether or not GPs have the skill to interpret such tests and whether or not they go on to request further tests that the experienced consultant might regard as invasive and unnecessary. We explore these unintended consequences and counter theories via a series of primary inquiries.

Simpson and colleagues211

This paper reports the results of a database review and GP questionnaire of direct-access CT for chronic headache to assess rate of significant pathology, outcomes, cost benefit and GP satisfaction. All patients who had been referred from primary care with a headache for longer than 3 months for a brain CT between 1999 and 2007 were identified from a database. The referring GP was sent a questionnaire along with the original radiology report to assess impact on treatment and referral and GP satisfaction with the service. It is unclear from the paper if the review was prospective or retrospective; it may be that GPs were asked to make comment on referrals from up to 8 years ago. All scans that reported abnormal findings were reviewed by a consultant neurologist who classified the findings into those that were likely to be incidental to the presenting symptoms and those that were potentially causative of the presenting symptoms. The subsequent management of all those with potentially causative abnormalities was reviewed.

During the 8-year study period, 4404 direct-access CT scans of the brain were performed. Of these, 461 (10.5%) had abnormalities reported with these thought to be potentially causative in 60 (1.4%). The authors report subsequent resource utilisation only on the 60 patients where potentially causative abnormalities were found and an additional 17 patients who had normal scans but returned for further imaging. It is not clear if the remaining patients were then managed in primary care. However, later in the paper, as part of the economic estimation, the authors cite that 14% of the 4404 patients went on to have a consultant outpatient appointment; this would amount to 616 patients, suggesting that some patients with normal scans still had a specialist referral.

There was a poor response rate to the GP questionnaire: 986 out of 2998 usable questionnaires were returned (32%). When asked what the management would have been without direct-access CT, 90% of the GPs would have referred to a specialist clinic with the majority of these referrals to either neurology (45%) or general medicine (38%). The authors thus cite an 83% reduction in the total number of referrals for specialist opinion. This, however, was calculated from the poorly responded questionnaires, so it is by no means certain that this figure would apply to those referrals where the GP did not respond.

The authors used these data coupled with audit data of imaging referrals from a headache clinic to suggest that direct access demonstrated a cost saving of £86,681.81 across the 4404 patients.

This study suggests direct access to CT scanning may reduce referrals to secondary care by ruling out serious pathology at an earlier stage in the patient pathway. However, there is no direct comparison with actual behaviour in the absence of direct access and the findings are based on hypothetical, post hoc, GP-reported assessments of management that may have been gathered some years after the actual test. Therefore, the reported effect should be treated with caution.

In this study, it appeared that the only task transferred to the GP was the decision to offer the test. All scans were reported by a consultant neuroradiologist and so the pre-existing professional boundaries between specialist and GP remain intact. CT scans are often ordered to provide patient reassurance and the majority of scans in this population are expected to be normal. However, in this study, some patients with normal scans still had a subsequent specialist appointment, suggesting that perhaps the scan did not always provide reassurance. It was unclear from this study if these referrals were made at the request of the GP, the specialist consultant or the patient, and so we do not know who needed the reassurance.

Taylor and colleagues212

This paper was a before-and-after comparison of GP direct access to brain MRI for patients with chronic headache. Outcomes for the first 100 direct investigations were compared with the last 100 referrals to neurology for the condition before the introduction of the direct-access service pilot. It should be noted here that the two groups of patients might have been quite different and this, rather than the presence or absence of direct access to brain MRI, could have affected how they were managed. Indeed, they differed in terms of mean age and gender split.

The authors reported that:

The main concern of neurologists prior to initiation of this scheme was that incidental findings would be investigated unnecessarily, with increased cost and possibly increased morbidity. This was thought to be due to the limited exposure of GPs to neuroradiological reporting. To address this risk, a standardized reporting style was introduced, which has been adhered to in practice

Reprinted from Clinical Radiology, vol. 67, Taylor TR, Evangelou N, Porter H, Lenthall R, Primary care direct access MRI for the investigation of chronic headache, pp. 24–7, 2012,212 with permission from Elsevier

Here, we see that neurologists express concerns about the ability of GPs to interpret tests, as GPs do not have a neurologist’s specialist experience. As result of this lack of experience, neurologists perceive that GPs may then order further, potentially invasive, investigations unnecessarily. To avoid this potential unintended outcome, in this study, all reports sent to the GP were expected to have a clear summary of the findings and guidance with regard to how to act on the results. In terms of the theory under test, the GPs were provided with support in the form of a standardised report to assist them with interpreting the test results.

The GP direct-access cohort had an onward outpatient clinic referral rate of 17%. This was twice as high as that advised in the reports accompanying the scan results; however, the referrals were not all to neurology clinics and included, for example, referrals for sinus problems or to a general headache clinic rather than to a neurologist. Furthermore, in one case, where the report gave advice that onward referral was indicated, no subsequent referral or appointment could be identified.

The authors conducted a crude cost-effectiveness analysis that suggested GP direct access resulted in a 46% cost reduction, compared with referral to a neurology clinic. This was mainly due to the cost saving of an outpatient appointment in 83% of the cohort, as it was assumed that all those referred would still have a MRI. However, this may not be a correct assumption, as there are no data to establish how many referrals to the neurology clinic undergo a MRI.

This study suggests that direct access to MRI for chronic headache seemed to reduce the need for referral; however, the results of a before-and-after study should be treated with some caution. In terms of the theory under consideration, neurologists perceived that GPs did not have the necessary experience to interpret the test accurately and may unnecessarily investigate incidental findings. To avoid this potential intended outcome, the neurologists retained some control over which patients were referred and how the results were interpreted by developing a ‘clearly defined referral pathway’ and including a guidance statement with the results. In this study, these strategies appeared to be moderately successful in these aims.

Thomas and colleagues213

This paper reported on a prospective observational study of primary care access to head CT for chronic headache. All GPs in an area of Scotland were given open access to head CT, subject to specific referral criteria:

• aged 18 years or over

• normal neurological examination

• symptoms consistent with chronic, unchanging headache.

All referrals over a 1-year period were included and a follow-up questionnaire was sent to the referring GP with the scan report to ascertain the impact of the result on subsequent referral decisions. Outpatient clinic databases were checked for 12 months to record any subsequent referrals.

During the study period, 232 patients were referred for brain CT out of an estimated 18,702 headache consultations, giving a referral rate to open access of 1.24%. Of these 232 referrals, 215 were scanned (11 patients cancelled/did not attend appointment, and six were not performed by radiologist as forms were incomplete/the patient did not fulfil the study criteria). Of the 215 scanned patients, 30 were referred to a neurology clinic because of headache [note that almost half of these (n = 12, 40%) had the referral made at the same time as the scan request].

Completed questionnaires were returned for 189 (88%) of the referring GPs and reported that the scan had stopped a referral to secondary care in 167 (88%) of the patients. As the scan caused a referral in 10 (5%) of patients, the authors give an estimated predicted reduction in clinic referrals to be 83% if GPs had open access to CT scans for chronic headache.

The authors noted that a number of referrals were made at the same time as the request for a scan and suggested that in cases where the GP is unsure of the diagnosis or in need of advice about management then direct access to CT will not alter referring behaviour. However, for many patients a scan is simply about reassurance, and in these cases a referral to secondary care is not always necessary:

The primary reason for arranging neuroimaging must remain its clinical indication, but the results of this study suggest that open access scanning may give GPs more confidence in managing the small minority of patients for whom a normal scan will provide reassurance and prevent secondary referral.

In terms of the theory under test, the low referral rate for open-access imaging per headache consultations found in this study would suggest that GPs are not overusing the direct-access service. It may be that the clear guidance on who to refer and the gatekeeping role of the radiologist ensured that use was controlled. The results suggest that clear referral criteria may help GPs to decide which patients are appropriate for referral to direct-access CT scanning and, as a result, the test results can provide the necessary reassurance to the patient. Consequently, in these circumstances, direct access can reduce the number of referrals to secondary care. However, short-circuit may still occur if either the GP or the patient is not reassured in the absence of a specialist opinion.

Benamore and colleagues214

This study did not exclusively examine the effectiveness of a rule-out test as the focus is on access to CT of the brain regardless of the function of the scan in the diagnostic decision pathway. However, in the study results there are indications that point to unintended consequences even when the results rule out significant pathology.

The paper reported on a retrospective scan and case note review of all primary care direct-access referrals for head CT scans to a local hospital radiology service over 8 years (1995–2003). All scan reports were reviewed and classified as normal, incidental or significant findings. All reports were cross-referenced for referral to secondary care and the case notes of those with incidental or significant findings were reviewed for secondary care attendance and outcome.

Over the time period, 1645 scans were performed, although only 1403 were available for review (85%). Headaches were the commonest reason for scan (n = 654, 46.61%) and, in keeping with the test performing a rule-out function in this context, the least likely to have significant findings on scan (18/654, 0.02%). However, when the referral rates were examined the scan result was a poor discriminator of who received a subsequent referral; not all those with significant findings were referred (13/18, 72.2%, were referred), a significant proportion of patients with normal results still received a referral (78/513, 15.2%), and a similar proportion of those with incidental findings were referred (22/123, 17.9%).

It is difficult to delineate how direct access might impact on referrals from this study; there is no assessment of expected referral patterns in the absence of direct access or what one would expect to happen given the scan results. In this study, it was unclear if the referring GP was given any supplementary detail except the scan results. The categorisation of significant, incidental or normal was not performed as a routine part of the result reporting, suggesting the GPs were left to interpret the scan results themselves. The authors do suggest that referrals may be generated by findings that a specialist would classify as incidental.

The authors noted that for some presenting symptoms (notably headache) a scan is primarily for reassurance against serious pathology as the expected diagnostic yield is low. They suggested that ‘It is reasonable to assume that the initial CT, generated by primary care, failed to provide the desired reassurance and potentially generated referrals from incidental findings’.

In terms of the theory under test, this study suggests that when GPs are not provided with guidance on how to interpret scan results, the meaning of the scan results are not always clear to the GP and, consequently, the scan itself may not provide the necessary reassurance to either the patient or the GP. This may then lead to GPs failing to refer patients with significant findings or referring those with incidental findings.

Wong and colleagues215

This paper reported on the evaluation of an open-access endoscopy service for patients presenting with dyspepsia in Hong Kong. Primary care physicians could use the open-access service but could also still refer direct to the specialist clinic. An extra morning session was established each week for open-access requests so that waiting times could be compared with the usual route of specialist referral. Waiting times and endoscopy results were compared for the two groups.

During the 3-year study period (1996–9), 978 referrals were made to the open-access service. The mean waiting time from referral to test was 6 weeks. For those referred to a specialist, the mean time to first consultation was 17.5 weeks and a further 4.5 weeks’ wait subsequently for the test.

The results of all patients during an 11-month period were assessed and there were no significant differences in diagnostic yield between the open-access tests and those from specialist referral except for an increased incidence of oesophageal varices (a known complication in patients with liver disease) in the specialist group, which may be explained by the preference to refer patients with liver disease to a specialist rather than go through direct access to the test. Table 4 from the paper is reproduced here as illustration.

TABLE 4 - Endoscopic diagnosis in open-access cases and specialist referral cases (for the period November 1996 to December 1997)

Peptic ulcers included duodenal and gastric ulcers.

‘Others’ included esophagitis, esophageal varices and other minor abnormalities.

Reproduced with permission from John Wiley and Sons, from Wong BC, Chan CK, Wong KW, Wong WM, Yuen MF, Lai KC, Hu WH, Lau GK, Lai CL and Lam SK. Evaluation of a new referral system for the management of dyspepsia in Hong Kong: role of open-access upper endoscopy, Journal of Gastroenterology & Hepatology, 15(11), 1251–6, 2000. 215 © 2000 Blackwell Science Asia Pty Ltd.

	Open access (n = 367)	Specialist (n = 967)	p-value
Number of normal results (%)	297 (81)	756 (78)	NS
Number of peptic ulcersa (%)	66 (18)	162 (17)	NS
Number of gastric cancers (%)	2 (0.5)	1 (0.1)	NS
Number of othersb (%)	2 (0.5)	48 (5)	< 0.005

The authors examined 166 of the referrals in detail (they did not state how they chose these). Of these 166 referrals, 126 attended for endoscopy as scheduled and 107 of these (88%) responded to a telephone interview at least 8 weeks after their endoscopy. All patients reported that they had received a rapid and definitive diagnosis from the endoscopy. Twelve per cent of the interviewees reported receiving a referral to a specialist clinic or an admission to hospital for management following their scan report and a further 12% were followed up by their family physician. The remaining 76% reported no further contact with services following the results of their test. The authors suggest that open-access endoscopy ‘gives a rapid and definitive diagnosis, which satisfies both patients and family physicians’.

Of interest is the 24% default rate, with 40 patients not attending for the direct-access test. When the patients who did not attend were contacted, the reasons given were:

• good response to empirical treatment since the last doctor consultation (40%)

• symptoms got worse and attended private hospitals or private doctors immediately (32.5%)

• went abroad (12.5%)

• a fear of endoscopy (2.5%)

• other personal factors (12.5%).

Direct access in this instance reduced the referrals to secondary care and reduced the time patients waited to receive an initial result. The authors noted that this is not merely about diagnostic yield as ‘a negative endoscopic finding is as important as a positive one in the management strategy of dyspepsia’. In terms of the theory under test, the results of the endoscopies in this study suggests that the case mix of patients referred via the two routes was not significantly different; however, many more patients were referred directly to a specialist than to direct access. Furthermore, even when patients were referred to the direct-access clinic, one-quarter did not turn up for their appointment and one-third of these patients sought advice from a specialist, thus short-circuiting direct access to the test. This suggests that GPs, patients or both would prefer access to a specialist rather than GP direct access to a test.

Todd and colleagues216

This paper reported on a case note review of direct-access referrals for endoscopy for the 12 months of 1995 in one endoscopy unit in Scotland. The authors identified all diagnosed cases of Barrett’s oesophagus, gastric ulcer and colonic polyps from the endoscopy reports and reviewed the case notes to ascertain GP referral behaviour. These conditions were chosen as specialist assessment and treatment may be required.

During the study period, 1451 procedures were performed under direct access and 89 of these were diagnosed with one of the identified conditions. The referral rates for all conditions were ascertained.

It is not possible to glean from this paper what impact direct access to endoscopy had on overall demand for secondary care; only the referral behaviour following a positive diagnosis on the test was investigated. The results showed that not all patients with positive diagnoses receiving a referral (Table 5).

TABLE 5 - Referral rate according to condition

As compared with adenomatous polyps.

Reproduced with permission from Sage Publications from Todd JA, Zubir MA, Goudie BM, Johnston DA, Scottish Medical Journal, 45, pp. 49–50, copyright © 2000 by (Sage Publications). Reprinted by Permission of Sage. 216

Referral rate according to condition
Condition	Number of cases	Number of cases referred	Referral rate	p-valuea (chi-squared)
Barrett’s oesophagus (all types)	25	14	56%	< 0.001
Barrett’s oesophagus (with intestinal metaplasia)	18	10	56%	0.08
Barrett’s oesophagus (with gastric metaplasia)	7	4	57%	0.25
Gastric ulcer	18	10	56%	0.08
Non-adenomatous polyps	24	3	12.5%	< 0.001
Adenomatous polyps	17	15	88%

The paper did not explore the reasons behind the differences in referral behaviour for the identified conditions. The authors did suggest that the results are reassuring as ‘the overwhelming majority of patients with adenomatous colonic polyps are referred to a specialist’. However, they also highlighted that ‘a substantial minority of patients with Barrett’s oesophagus and gastric ulcers are not referred to a specialist’, even despite local policy in place for the management of ulcers which suggests a specialist referral.

No advice was given to the GP about appropriate management of the patient following the endoscopy results and the results suggest that the test result on its own is not sufficient for GPs to manage patients appropriately. The authors suggest that:

It may be that general practitioners preconceived ideas of the importance of a condition affect their decision to refer more than the existence of guidelines. We believe that the existence of guidelines alone is not enough, education is a vital partner to ensure good practice.

In terms of the theory under test, this study suggests that direct access to a test does not necessarily allow a GP to identify those patients for whom a referral is appropriate. Even where there are guidelines in place regarding management of particular conditions, if the results are received without advice then GPs may not always follow the appropriate guidelines.

Morgan and colleagues31

This study did not explore direct access to tests but can help to shed some light on why some patients may receive a referral even when a test has ruled out serious pathology. The paper reported on a qualitative interview study exploring GP beliefs about patients presenting with headache and the reasons for their referral behaviour. GPs were purposively sampled according to referral behaviour in the preceding year (from no referrals for investigation of headache to three or more referrals). Data were collected from 20 GPs and analysed using a framework approach. The authors reported that recruitment was stopped at 20 GPs as data saturation was reached.

Quotes from the GP participants suggested that some patients would be reassured only if they were seen by a specialist:

. . . ultimately they say that’s all very interesting and thank you very much for the time you’ve taken, but I still want to see someone . . .

GP0604

However, results from the study also suggested that some GPs may be better at instilling confidence in a patient:

I think part of my ability to help people is [a] projection of my confidence, personality, holistic approach, whatever . . . to give them confidence in me so when I say: ‘it’s fine’ they believe me.

GP1704

In addition, it may be that some GPs have a lower threshold for managing clinical uncertainty and may make referrals to manage their own anxiety.

I’m, I would say, cautious I suppose. I tend to err on the side of caution and refer more . . . that’s my nature.

GP0607

. . . I might refer people just for my own anxiety and fears of missing something, even if they’re not anxious. So, really, I suppose it depends on lots of factors.

GP1803

This study suggested that referral despite a negative test result may be explained by patient pressure and the perceived confidence in the opinion of the GP. In these instances a negative result from a ruling-out test may not be sufficient to provide the reassurance needed to satisfy the patient that there is nothing seriously wrong. It may be that for some patients only a specialist can provide the necessary reassurance.

The study also suggested that GPs may vary in their tolerance of diagnostic uncertainty. In these tests a negative result can rule out serious pathology; however, this can mean that presenting symptoms remain unexplained and GPs may refer for a specialist opinion to ensure that they have not missed an important diagnosis.

Ridsdale and colleagues217

This was a cohort study by the same research group as that in the above study; it compared 488 patients who consulted their GP for headache over a period of 7 weeks but were not referred and a cohort of patients consulting their GP for headache over 1 year who were referred. The two groups completed questionnaires measuring the disability experienced due to headache, anxiety, depression, illness perceptions and frequency of consultation in the past 3 months.

The authors found that both groups experienced severe disability and that headache significantly affected their functioning. There was no significant difference between the two groups on measures of disability, anxiety, depression or satisfaction with care. Referred patients had consulted more frequently in the 3 months prior to their appointment and were more likely to link more of their symptoms to headaches. They were also more likely to have stronger emotional representations of their headaches and to worry more and were made more anxious by their headaches. The authors argued that referral was not related to the clinical severity of headaches but was associated with higher consultation frequency and patients’ anxiety and concern about their headache symptoms.

This study did not explore direct access to testing but the results suggested that patient anxiety does influence referral decisions and this may help to explain why some patients with a negative result on a ruling-out test still receive a referral. Box 40 provides an interim summary.

BOX 40 - Interim summary

A ‘rule-out test’, such as CT/MRI scanning for headache, is used to rule out the possibility that the patient has a serious pathology. A negative result is expected for the majority of patients and it is often ordered to provide patient reassurance. The theory is that GPs can decide when and for whom this test is appropriate and can interpret the test results in order to reassure the patient that they do not have a serious pathology and confirm that specialist referral is not, therefore, necessary.

In terms of the theories under test, the evidence clearly shows that direct access to rule-out tests can reduce referrals to secondary care. Such tests allow GPs to identify patients without serious pathology that do not require referral to a specialist and in the majority of cases a referral is thus averted. There was no evidence of overuse of rule out tests in GP decision-making. However, there were some inconsistencies where a minority of patients with a negative test result still received a referral to a specialist and some patients for whom the results indicated a referral was warranted did not receive a referral.

The evidence further suggests that the utility of rule-out tests in demand management can be maximised if results are accompanied by guidance from a specialist on interpreting the results and clinical advice on management given the results. Where these guidelines are not present, GPs may investigate incidental findings unnecessarily or fail to refer when necessary. In terms of the ‘turf war theory’ under test, defined referral criteria and guidance on management attached to the test results ensure specialist control over referral and management decisions, which, in turn, ensures that the appropriate patients are referred to secondary care.

However, some studies showed that even when a scan suggested no serious pathology, a small percentage of patients were still referred to the specialist. Even with clear guidance, some patients received a referral to a specialist following a negative result on a ruling-out test. In cases such as these it is likely that a negative result does not provide the necessary reassurance to the patient and/or the GP.

Differential diagnosis tests

The principal demand reduction theory underlying direct access to differential diagnostic tests is similar to that of the rule-out test. The idea is that through bypassing the need to see a consultant, patients can have speedier access to the test. The results of the test then enable GPs to distinguish between those who can be managed in primary care and those who require an onward referral to secondary care, thus reducing the number of referrals to secondary care.

Differential diagnosis tests sit in the same place as a rule-out tests in some typologies, as they are both diagnostic tests. However, there are important differences between ‘ruling-out tests’ and ‘differential diagnosis tests’, which have significant consequences for their impact on demand management. Compared with ‘ruling out tests’, the expected diagnostic yield in a differential diagnosis test is higher, as the test is performed in circumstances in which a clinical history is suggestive of significant pathology and the test serves to confirm (rather than rule out) this suspicion and/or to distinguish between different definitive diagnoses. Accordingly, direct access to a differential diagnosis test may not reduce referrals to secondary care if most patients undergoing the test turn out to require a referral. In this situation, direct access can reduce waiting time from presentation at the GP to diagnosis and treatment and reduce the number of specialist appointments necessary as the results are to hand at the first outpatient appointment.

As ever, these chains of events are the subjects of debate. One potential unintended consequence may occur if there are not agreed indications between the specialist and the GP for the test in the diagnostic pathway. In this context, direct access to a differential diagnosis test may increase demand unnecessarily for the test and result in a patient receiving an avoidable test.

We move to consider what the evidence has to say about various theories and counter theories with the examples of primary research on MRI for suspected internal derangements of the knee and MRI for the investigation of back pathology: Brealey and colleagues218 and three DAMASK (Direct Access to Magnetic resonance imaging Assessment for Suspect Knees) studies. 219–221

These papers reported on the same pragmatic multicentre randomised trial of GP direct access to MRI versus referral to orthopaedic surgeon for patients presenting with suspected internal derangement of the knee. All patients presenting to practices where referral to secondary care was considered were randomised within practice to either the intervention (direct-access MRI) or an orthopaedic surgeon appointment. In the intervention arm, the MRI was used by the referring GP to inform diagnosis and plan subsequent management. For those who were sent for MRI, a provisional referral to the orthopaedic service was also made to ensure that the waiting time from the GP consultation to the orthopaedic appointment was similar for both arms.

To note here is that this design is not pragmatic; it may not mirror actual practice, as it is unlikely that GPs would directly access differential diagnostic tests and refer to a specialist at the same time. Indeed, the study design, by making a provisional referral for all those in the direct access arm, implicitly acknowledges one of the potential unintended consequences of direct access, that patients may wait longer to see a specialist, as the GP may wait for the results to be available before making a referral.

Included in the trial was an educational intervention to support the dissemination of the clinical guidelines with the aim of improving GPs’ knowledge of MRI use. GPs were incentivised to attend training, although attendance was not a condition of participation in the trial. A further educational message was attached to the radiologists report to remind the GP that subsequent management should depend on both the MRI results and the clinical findings. Thus, both referral and management guidance were provided to GPs.

Only 44% (285 out of 647) of practices approached to take part accepted the invitation. This low participation rate perhaps suggests that the majority of GPs preferred to make their own decisions about which referral pathway they chose for patients, rather than have those dictated by the process of randomisation. Those participating were likely to be further from a hospital. Over a 2-year period 553 patients who were eligible and gave consent were recruited from 163 practices, 279 were randomised to receive the intervention (MRI) and 274 were randomised to the control (referral to orthopaedics). Follow-up was 85% at 24 months.

The authors reported minimal difference between the groups in terms of orthopaedic consultations; 230 (82%) patients in the MRI group and 236 (86%) in the orthopaedic group. However, this adds little explanatory value as the control arm was orthopaedic referral where the majority of the 14% who did not have an appointment were patients who failed to attend following allocation. The authors noted in the discussion that although GPs had the option to cancel the orthopaedic appointment on the basis of the MRI results, they rarely did so with only 27/279 (10%) appointments cancelled on the basis of the scan results. In terms of the theory under test, the design of the study meant that it was difficult to draw conclusions about whether or not direct access to MRI reduced referrals to secondary care.

The referring GP was asked to record the main diagnosis and how confident they were in this diagnosis at baseline and these questions were repeated on receipt of the radiologists report in the MRI arm and the letter from the orthopaedic specialist in the referral arm. Table 2 from the paper is reproduced here (Table 6) to show that there was no difference in the likelihood of change in diagnosis between the two arms, although GPs reported an increased confidence in their diagnosis in the MRI arm. The study was not designed to shed light on why this might be. We could hypothesise that GPs in the MRI arm were perhaps more likely to attempt a definitive diagnosis prior to the sending patients to the MRI as they would have to interpret the MRI scan and make subsequent decision, rather than the consultant. Receiving confirmation about their initial diagnosis was perhaps more likely to increase their confidence. In contrast, perhaps GPs in the consultant arm felt less pressure to make a definitive baseline diagnosis as they expected this to be made by the consultant.

TABLE 6 - Changes in GP diagnoses and diagnostic confidence

The χ² test.

The Wilcoxon-signed rank test for within-group change in diagnostic confidence between trial entry and follow-up.

Reproduced with permission from the RCGP, from DAMASK. 219

	Original randomisation		Between-group changea
	MRI referral	Orthopaedic referral
Diagnosis altered, n (%)
Yes	170/279 (61)	165/266 (62)	χ2 = 0.07, p = 0.79
No	109/279 (39)	101/266 (38)
Change in diagnostic confidence, n (%)
Increased	168/264 (64)	80/249 (32)	χ2 = 51.43, p < 0.001
No effect	77/264 (29)	129/249 (52)
Decreased	19/264 (7)	40/249 (16)
Within-group changeb	p < 0.001	p = 0.006

This study showed that direct access in this context had little impact on referral decisions, as the majority of patients received a referral following their MRI. It also pointed to some important differences in subsequent treatment, which give us some interesting evidence towards exploring the unintended consequences of direct access. There was a difference in the number of patients having an arthroscopy (a form of keyhole surgery used to both diagnose and treat joint problems): 40.1% in the direct-access group compared with 28.1% in the orthopaedic group. Interestingly, this was classified as a diagnostic arthroscopy in 39.5% of the patients in the orthopaedic arm, compared with 18.5% in the direct access arm. It is not reported in the papers how many patients in the orthopaedic referral also had a MRI scan; however, the economic evaluation gave a much lower mean resource use cost for MRI in the orthopaedic referral group – £128.96 versus £326.79 in the direct-access arm – which suggests that a significant number of patients referred direct to orthopaedics did not have a MRI scan as part of their diagnostic testing.

This suggests that a MRI in the case of suspected internal derangement of the knee is not necessarily the diagnostic test of choice by the specialist, which may explain why more arthroscopies in this arm were classified as diagnostic. It also shows that patients referred directly to a specialist had fewer arthroscopies and fewer MRI scans than those in the direct-access group. In terms of the theory under test, this implies that specialists are more likely to have the clinical skills necessary to reach a differential diagnosis without the need for scans or tests and that referral first to a specialist may prevent the use of unnecessary scans. In contrast, a clinical pathway where a MRI is performed first and then followed by a specialist referral may mean that some of those initial MRIs were performed unnecessarily.

Even though the study attempted to control for wait for orthopaedic appointment, the time from randomisation to first appointment (median 93 days, range 60–183 days vs. median 79 days, range 54–168 days) and subsequent treatment (median 337, range 219–458 days vs. median 294 days, range 174–468 days) was longer in the direct access to MRI group. Interestingly, there was a small but significant improvement in health-related quality of life (HRQoL) in the direct access to MRI group.

In this study, direct access to MRI seems to have had little impact on referral decision-making, as most patients also had a specialist appointment. Unlike in a rule-out test, the test had a high diagnostic yield, and although it may have distinguished possible diagnoses and subsequent treatment, in most cases, a positive result necessarily required specialist referral. Indeed, the authors note ‘It is likely that MRI confirmed what GPs already knew, which was the need for a referral to an orthopaedic specialist.’ However, an unintended consequence of direct access was to increase the usage of MRI scanning, as not all referrals direct to a specialist underwent a scan. In terms of the theory under test, it suggests the direct access to MRI scanning may increase patient waiting times and result in unnecessary tests, as GPs do not have the clinical skills to definitively diagnose joint problems. For the GP, a MRI scan may simply confirm their original diagnosis and that referral to a specialist is required. In contrast, specialists have the clinical skills to diagnose joint problems in some instances without the need for tests and direct referral to a specialist may, thus, prevent the use of unnecessary tests.

Watura and colleagues222 (plus responses and rejoinders in the same journal)223–225

This was a brief paper that reports the results of an audit of direct-access MRI imaging of the knee to one radiology department in a Welsh hospital. A direct-access service was introduced in January 1993 and the audit compared referrals via GP direct access up to May 1994 with referrals made via orthopaedic outpatients in the year preceding the introduction of the service.

The diagnostic yield from the scans did not differ significantly between the two groups, although the scan result alone was a poor discriminator of who had a subsequent referral to the orthopaedic clinic. The results table from the paper is reproduced here as Table 7 for illustration.

TABLE 7 - General practitioner vs. orthopaedic referrals for MRI

NS, not significant.

Reproduced from Magnetic resonance imaging of the knee: direct access for general practitioners, Watura R, Lloyd DCF, Chawda S. BMJ, vol. 311, p. 1614, © 1995222 with permission from BMJ Publishing Group Ltd.

Patients	General practitioner referral	Orthopaedic referral	p-value
Number referred	165	470
Number (%) males	115 (70)	331 (70)	NS
Mean (range) age (years)	37 (9–83)	32 (2–84)	< 0.001 (χ + 2 test)
Mean wait for scan (days)	19	14	< 0.001 (Student’s t-test)
Findings, n (%) of patients
Cruciate/meniscal tears	73 (44)	213 (45)	NS
Followed up	65 (89)
Referred to orthopaedic clinic	55 (75)
Normal	43 (26)	134 (29)	NS
Followed up	37 (86)
Referred to orthopaedic clinic	5 (12)
‘Other’ abnormality	49 (30)	123 (26)	NS

The authors suggested that direct access to MRI reduced the number of referrals to specialists, as most with normal scans or minor abnormalities were not referred; 76 out of 165 (65%) were subsequently referred to the orthopaedic clinic. In addition, they suggest that direct access did not increase the workload of MRI as all patients, if referred to the orthopaedic surgeon, would have had a scan performed anyway. An additional bonus, from the authors’ perspective, is that the availability of the scan results at consultation for those who were referred allows for more streamlined planning of care.

In terms of our programme theory, the suggestion is that direct access allows GPs to make decisions higher up the decision hierarchy. However, the scan result is a poor discriminator of who got a referral; not all abnormalities were referred and some normal scans received a referral. This suggests that referral decisions were not made solely on the basis of the scan result.

The authors assumed that all referrals would have a MRI scan anyway. The DAMASK trial,219–221 discussed above, shows that this is not necessarily the case; MRI utilisation in those referred directly to a specialist was lower than that in the direct-access arm.

The role of direct-access MRI in this instance is shown to be the subject of disagreement when we investigate the responses to this paper on the letters pages of the journal. This debate is useful evidence to show that real divisions exist between generalists and specialists in their understanding of the value of direct-access MRI in the diagnosis of knee derangement. Responses from orthopaedic specialists suggest that they have access to knowledge and skill sets that allow them to use such tests more selectively than a generalist.

We believe that magnetic resonance imaging of the knee should be used selectively. Patients who have obvious clinical symptoms and signs of a meniscal tear do not need such imaging before having an arthroscopy. Imaging may, however, be useful to avoid a diagnostic arthroscopy . . . The selective use of magnetic resonance imaging helped us avoid unnecessary operations, plan surgical procedures and save money. These benefits would be lost if demand for inappropriate scans increased. We do not believe that magnetic resonance imaging should be used as a blanket test; it would be a shame if such imaging replaced the need for accurate clinical skills

Reproduced with permission from Jeremy Southgate, from Southgate J, Thomas N. Direct access magnetic resonance imaging of the knee for GPs. Magnetic resonance imaging should be used selectively. BMJ 1996;312(7034):849; author reply 50225

An orthopaedic consultant with an interest in the knee should . . . sort out most patients with knee pain on clinical grounds . . . for some clinical diagnoses . . ., magnetic resonance imaging has little proven clinical value. Not all general practitioners will understand the implications of the findings of magnetic resonance imaging.

Mackenzie et al. 224

One consultant orthopaedic surgeon suggested that the assumption in the paper that all specialist consultations would necessitate a MRI was misplaced and, therefore, the cost savings of open access were overestimated. The author also pointed to the importance of clinical examination skills and, therefore, a need for more specialist orthopaedic skills. 223

In terms of the theory under test, these letters point to ‘turf wars’ between GPs and specialists over who possesses the skills and knowledge necessary to make a diagnosis in this context. Indeed, the original authors’ reply to these letters highlighted the issue of professional boundaries: ‘It would be absurd to suggest that general practitioners are capable of prescribing expensive drugs with complex interactions and side effects but not capable of selecting patients for magnetic resonance imaging of the knee and using the results appropriately’. 226

Chawda and colleagues227

This was a brief report of an observational study of the MRI service at Cardiff Royal Infirmary for patients referred for a lumbar MRI for sciatica or suspected spinal claudication. All hospital outpatient referrals (i.e. those initiated by consultants) and those directly referred by GPs were reviewed for a 2-year period (January 1993–December 1994). GPs were contacted in writing and by telephone to document subsequent management. This study design, using retrospective data collection, is open to the possibility of recall bias on the part of the GP and, as such, the findings should be treated with caution. Nonetheless, the papers provide further evidence with which to test the theories under investigation for a different condition.

Over the time period, GPs referred 459 patients to the direct-access service, compared with 435 patients who came for MRI via hospital outpatients. The differences in diagnosis and subsequent management between the two groups were shown in Table 1 in the paper, which is reproduced here as Table 8.

TABLE 8 - General practitioner vs. hospital outpatient referrals and GP patient management following MRI result

NS, no significant difference.

t-test.

Chi-squared test.

Reproduced with permission from the RCGP, from Chawda and colleagues. 227

	GP referral for MRI, n (%)	Outpatient referral for MRI, n (%)	GP vs. outpatient referrals, p-value	GP management following scan
				Hospital referral, n (%)	Conservatively managed, n (%)
Sex: male	244 (53)	234 (54)	NS (a)
Female	215 (47)	201 (46)	NS (a)
Mean age (years)	42	44	p < 0.025b
Mean waiting (days)	19	13	p < 0.001b
1a. Disc herniation corresponding with symptoms	182 (39.8)	165 (37.9)	NS (a)	113 (72.9)	42 (27.1)
1b. Disc herniation not corresponding with symptoms	46 (10.1)	55 (12.6)	NS (a)	20 (47.6)	22 (52.4)
2. Spinal stenosis	20 (4.4)	20 (4.6)	NS (a)	11 (57.9)	8 (42.1)
3. Spondylosis/degeneration	114 (24.9)	77 (17.7)	p < 0.01 (a)	27 (27)	73 (73)
4. Normal	57 (12.5)	68 (15.6)	NS (a)	17 (34)	33 (66)
5. ’Other abnormalities’	38 (8.3)	50 (11.5)	NS (a)	12 (38.7)	19 (61.3)
Total	457	435		200	197

There was no significant difference between the groups in the percentage of scans reported as normal: 12.5% from GP referral and 15.6% from outpatient referral. There were no significant differences in the percentage of specific classifications of abnormalities reported except an increased incidence of spondylosis/degeneration in the GP referrals. Overall, this suggests that the case mix of patients referred for MRI by GPs and specialists in this context was broadly similar.

The results of the GP follow-up (86.7% response rate) suggested that 49% of those referred for MRI were conservatively managed subsequently in primary care. However, the test result was not a good discriminator of who was subsequently referred for specialist care, which suggests that the test alone was not driving the decision regarding subsequent management.

The authors concluded that ‘prompt investigation allows GPs to make appropriate and informed management decisions. For those patients who are referred, the availability of the MRI scan result at consultation permits rapid assessment and avoids a follow-up appointment to discuss results, as would be necessary if the MRI were requested at the first outpatient attendance. The availability of a scan report could also assist hospital specialists in prioritizing outpatient appointments and reviewing the urgent cases more quickly’ [reproduced with permission from the RCGP, from Chawda, S. J., Watura, R. and Lloyd, D. C. (1997) Magnetic resonance imaging of the lumbar spine: direct access for general practitioners, British Journal of General Practice, 47(422), 575–6.]

In terms of the theory under test, direct access to MRI of the lumbar spine appears to have some impact on subsequent referrals, with the GP using the results to direct subsequent management, which suggests that they are making decisions which may have previously been deferred to specialist via a referral. This may be because there is genuine uncertainty on the part of GPs and specialists prior to use of the test and the test clearly differentiates between diagnoses that can be managed in primary care and those that need specialist management. The authors also commented on the clear guidelines for the use of MRI in patients with acute back pain and concluded that ‘direct access for MRI of the lumbar spine is a viable proposition if driven by protocols and used appropriately’.

Anderson and colleagues228

A study of diagnostic accuracy in 2-week-wait colorectal cancer referrals also suggested that specialists may have access to more sophisticated diagnostic tools which can aid in establishing diagnoses. In this study, the diagnostic accuracy of GP referral letters was compared with the accuracy of the patient’s first surgical assessment. Surgical assessment showed significantly greater diagnostic accuracy and this was explained almost entirely by the use of a rigid sigmoidoscopy during rectal examination; rigid sigmoidoscopy made a significant contribution to managing 12% of referrals. The authors reported that rigid sigmoidoscopy is standard practice for outpatient assessment of suspected colorectal cancer but is rarely performed by GPs. They suggested that a referral for surgical assessment was, therefore, preferable to a straight-to-endoscopic test pathway, as some patients may be denied the opportunity for immediate diagnosis. Box 41 provides on interim summary.

BOX 41 - Interim summary

The theory underlying direct access to differential diagnostic tests is that through bypassing the need to see a consultant, patients can have speedier access to the test. It is expected that the results of the test enable GPs to distinguish between those who can be managed in primary care and those who require an onward referral to secondary care, thus reducing the number of referrals to secondary care. However, our evidence synthesis suggests that direct access to tests such as MRI or endoscopy does not reduce referrals to secondary care and may result in unnecessary tests being performed and increased patient waits for treatment. This is because specialists have the necessary clinical skills and/or access to alternative and preferred diagnostic techniques that enable them to reach a differential diagnosis without the need for the direct-access test. Direct access to MRI appears to confirm what GPs already know: that the patient requires referral to a specialist.

In contrast, there is some evidence to suggest that direct access to MRI for suspected lumbar sciatic or claudication does enable GPs to distinguish between patients who can be managed in primary care and those who require a specialist referral, and did reduce referrals to secondary care. The difference in the outcome patterns between the two tests may be explained by the extent of agreement between specialists and GPs on the role and value of the test in the patient pathway and whether or not the test is designed to provide a differential diagnosis between conditions that require management in primary or secondary care. Although MRI scans and endoscopies can be interpreted by GPs, with guidance, they are not the initial tests of choice for specialists in diagnosing internal derangement of the knee or the presence of bowel cancer. In contrast, MRI for back problems appeared to be used with similar frequency by specialists and GPs in reaching a differential diagnosis in patients with back problems.

Furthermore, while MRI for knee problems may enable GPs to reach a differential diagnosis in these cases, the vast majority of these diagnoses are still perceived by GPs to require management by a specialist. In contrast, MRI for back problems does appear to distinguish between diagnoses that can be managed in primary care and those that require specialist referral. This evidence synthesis clearly shows that not all tests are created equal and that the function of the test in the patient pathway and differences in the practices, skills and experience of GPs and specialists influences whether the test produces intended or unintended consequences.

Clinical indicator tests

A clinical indicator test encompasses screening for disease, monitoring the progress of disease and/or monitoring an individual’s response to treatment. Although these tests are used at different points in the patient’s disease trajectory, the common theory underlying their use is to identify where the patient is on this trajectory in order to guide subsequent management and treatment. As screening tools, these tests aim to inform a decision about whether to intervene now or whether to continue to monitor the patient, and who is best placed (the GP or the consultant) to undertake this intervention. They are used when there is a high suspicion of a particular diagnosis from the patients’ signs and symptoms and the test is used to confirm this and also to quantify the stage or severity of the condition in order to identify the patient’s location on a disease trajectory.

By providing GPs with direct access to these tests, it is assumed that GPs have the necessary skills to decide when and for whom such a test is warranted and then have the skills and confidence to interpret the results and decide which patients can continue to be managed in primary care and which patients require referral. As a result, more patients will be managed in primary care without the need for a specialist opinion and only those requiring specialist management are referred to secondary care.

As ever these programme theories have been the subject of debate on now-familiar grounds. For instance, a potential unintended consequence of providing GPs with direct access to clinical indicator tests is that GPs misinterpret the test results and refer patients inappropriately or fail to refer patients when a referral is warranted by the test results.

The literature gives us a number of examples in this typology: bone mineral density scanning for suspected osteoporosis, cardiac arrhythmia monitoring and echocardiography for suspected heart failure.

Dhillon and colleagues229

This paper was a RCT of GP direct access to dual X-ray absorptiometry (DXA) scanning versus hospital referral for specialist consultation in women with suspected osteoporosis and its impact on clinical decision-making. DXA scanning is used to diagnose osteoporosis or to assess the risk of osteoporosis developing. The results are used to guide the initiation of therapy. The authors stated that, at the time of writing, access to NHS-funded DXA scanning and osteoporosis services in the UK was variable. At the time of the study, access to DXA scanning was only via referral to hospital specialist. Direct access to DXA scanning would allow GPs to use the results to manage subsequent treatment and care without the need for specialist referral.

No referral guidelines were given to participating GPs; the authors assumed that they would follow one of the established guidelines, such as those published by the National Osteoporosis Society. GPs in participating practices selected patients who they felt warranted DXA scanning, and completed a structured referral form. On this form they were asked to state their management plan if access to a DXA scan was not available. Participants were then randomised to the study arm – direct access to DXA scan with a brief scan report and one of four management recommendations – or the control arm of DXA scan and routine rheumatology clinic consultation.

To note here is that this is not a comparison of usual care versus direct access; it is not clear if hospital referral would be sought in the absence of direct access. Many of the patients in the study had commenced treatment for osteoporosis on the basis of clinical history alone and it is unclear if specialist referral would be sought for all patients even without access to DXA scanning. Furthermore, the study design does not compare direct access to the scan with specialist referral. The aim of the study was to test the effect of two different routes to DXA scanning; all patients in the study received a DXA scan so the comparison is between whether this is accompanied by a scan report with management recommendations to the GP or a specialist consultation.

A total of 330 participants were randomised: 165 in each arm. The groups were similar at baseline in terms of demographics, reasons for referral, pre-referral treatment, health state and GPs’ anticipated post-scan management. Significantly more participants in the control arm were recorded as having clinical risk factors and/or prior low-trauma fracture. On scan, there was no significant difference in bone mineral density (BMD) between the groups.

In the control arm, the GP received a detailed clinic letter giving the DXA result and a written opinion on management given by the specialist who saw the patient in the clinic. In the study arm (direct-access route), the GP received the DXA result and one of four management recommendations:

• lifestyle advice

• consider measures to prevent osteoporosis

• measures to treat established osteoporosis

• specialist referral.

The primary outcome measure was change of management following the DXA scan. Sixty per cent of all study participants had their management changed after the scan report; the proportion was similar across the arms. At 1-year follow-up there was no significant difference between groups in outcomes. The main difference was in waiting times, as those in the control arm had a longer time from referral to GP receipt of the report.

Resource use was assessed at 1-year follow-up and there was a significant difference in subsequent referrals to a specialist, with 17% in the direct-access arm and 9% in the control arm receiving a referral to a specialist rheumatology service. It is not clear if the referrals in the direct-access arm followed the advice with the scan results and this was not reported in the study.

The authors concluded that ‘DXA scanning influences GPs’ management of osteoporosis in two-thirds of patients’. However, there was no difference in GP management depending on whether the GP received the scan results and management plan or whether the GP received the scan result and the patient also saw a consultant. They also noted that ‘The clinical outcome is similar following direct access or the current model of hospital referral, but direct access is more economically efficient’.

In terms of the theory under test, in this study, the GP management of patients with suspected osteoporosis following direct access to DXA scanning accompanied by a management plan from a specialist was similar to that when patients were referred to a specialist and scan. However, any necessary changes to treatment could be enacted earlier by the GP following direct access, as the results were available more promptly than from the specialist referrals. In addition, direct access had an impact on specialist resource use, as only 17% of patients referred to direct-access DXA scanning had a subsequent referral to a specialist. This is in comparison with the control arm (referral to a specialist), where all patients had an initial consultation and a further 9% had subsequent appointments with a specialist.

This suggests that direct access can both speed up patient access to appropriate treatment plans and reduce the resource use of specialist clinics. It is difficult to quantify any potential savings in resource use of specialist clinics as it is not clear that all patients would have been referred to a specialist had the study not been recruiting. In this study, the referring GP did not have to interpret the results: the scan report was accompanied by a suggested management plan that stipulated if treatment should be commenced and identified which patients required specialist referral.

Sim and colleagues230

This paper was an observational study of the utilisation of an open-access service to bone density scanning (DXA) over a 12-month period. All referrals during the study period were assessed against agreed clinical referral criteria, supplied on a referral form completed by the GP at the point of referral, although even those who did not fit the agreed criteria still underwent the DXA scan. Following the direct-access scan a report was sent to the referring GP including comments from a consultant specialising in osteoporosis – this report gave a clinical judgement based on the DXA results and other clinical information available on the direct-access request form. The report gave management recommendations in one of four categories:

1. normal

2. borderline BMD with no treatment required

3. osteoporosis and treatment required

4. significant abnormal DXA result; refer to bone clinic.

All GPs who used the open-access service were sent a questionnaire regarding the first patient they referred during the study period.

During the study period, 586 patients were referred to the open-access DXA service. Of these, 26 were excluded from the study as the referral was for monitoring of treatment. Of the 560 included patients, 473 (84%) of referrals were judged as appropriate against the clinic referral criteria, suggesting that the GPs were reasonably good at following the referral guidelines. However, a number of referrals that did not match the criteria were also diagnosed positively from their scan, indicating that where GPs did not follow the guidelines this may well have been for good clinical reasons. In total, there were 229 (41%) of patients who had osteoporosis diagnosed by DXA. Table 2 from the paper is reproduced here as Table 9 to illustrate the results.

TABLE 9 - Diagnosis of osteoporosis in referrals that met the referral criteria and those that did not meet the referral criteria

Reproduced with permission from Sim MFV, Stone M, Johansen A, Ho P, Pettit RJ, Evans WD. An analysis of an open access general practitioner bone densitometry service, International Journal of Clinical Practice 2004;58, 300–5. 230 © 2004 Blackwell Publishing Ltd.

	Osteoporosis cases diagnosed and treatment required	No osteoporosis
Referral according to CRC	212	261
Referral which did not fulfil the CRC	17	70
p-value by χ2 test is < 0.001
Sensitivity of CRC in predicting osteoporosis = 212/(212 + 17) = 93%
Specificity of CRC in predicting osteoporosis = 70/(70 + 261) = 21%

The consultant recommendations with the scan report identified 19 (3%) patients who required referral to the outpatient clinic, with the remainder suitable for management in primary care:

• 324 (58%) normal

• 7 (1%) borderline BMD with no treatment required

• 210 (38%) osteoporosis and treatment required

• 19 (3%) significant abnormal DXA result; refer to bone clinic.

Questionnaires were returned by 119 (77%) of the referring GPs, who reported that in 101 (85%) of the cases they would have referred to hospital outpatients if direct access had not been available. The scan and report identified 76 (64%) patients for whom treatment was not indicated. The actual GP management of these patients corresponded to the recommendations. In the case of confirmed diagnosis of osteoporosis, 37 of the 43 diagnosed patients (86%) were identified as suitable for management in primary care. The reports from the direct-access scan recommended referral to a specialist in only six patients (5%). However, the results show that 10% of patients had a referral, with six patients for whom management in primary care was indicated being referred for a specialist appointment. Table 3 from the paper is reproduced here as Table 10 for information.

TABLE 10 - Recommendations of DXA reports and GP action

HRT, GPs were able to stop HRT in two patients because of their normal DXA.

Reproduced with permission from Sim MFV, Stone M, Johansen A, Ho P, Pettit RJ, Evans WD. An analysis of an open access general practitioner bone densitometry service, International Journal of Clinical Practice 2004;58, 300–5. 230 © 2004 Blackwell Publishing Ltd.

DXA reports (n = 119)	Specialists’ recommendations (%)	GPs’ action upon the DXA report (%)
Normal DXA report, patients reassured	75 (63)	75 (63)a
Borderline osteoporosis, monitor patient with repeat DXA later	1 (1)	1 (1)
Osteoporosis confirmed, treatments initiated as recommended	37 (31)	31 (26)
Referral to metabolic bone clinic	6 (5)	12 (10)

In terms of the theory under test, this adds some explanatory value to the RCT cited above,231 as the authors of that RCT suggested that many of the patients would have been referred to a specialist in the absence of direct access and, therefore, direct access can reduce the referrals to secondary care. In this study, direct access to DXA scanning enabled GPs to distinguish between patients who did not have disease and patients who had disease but could be managed in primary care, and thus reduce referrals to secondary care. Perhaps the reason for this was that the scan result was accompanied by guidance on the meaning of the scan result and guidance on management from the specialist.

In terms of professional boundaries, there was minimal transfer of decision-making responsibilities between the specialist and the GP beyond the request for a test itself. In this example the decision-making remained with the consultant specialist, albeit without him or her seeing the patient directly:

A DXA report with comments by a consultant specialising in osteoporosis was sent to the referring GP. The diagnosis on the report is a clinical one made by the specialist based on the DXA result [and] age, sex and any other relevant clinical information available on the request forms.

This suggests that the clinical information provided by the GP on the referral form was also important in enabling the consultant to formulate a diagnosis and management plan along with the scan result without seeing the patient. There is also some evidence of short-circuiting, with GPs still making referrals for a minority of patients with a positive diagnosis, even though the consultant has indicated that this could be managed in primary care. The study made no attempt to explain the source of this unintended effect.

Stock and colleagues232

This paper reported on a study that did not examine the effect of direct access to DXA scanning on referral behaviour, but is useful for exploring the impact of the format in which GPs receive test results. The authors hypothesised that a low use of DXA scanning among American primary care physicians may be because they had difficulty comprehending the often brief technical reports.

All primary care physicians ordering DXA scanning over a 7-month period were enrolled, and when one of their patients underwent the procedure, the physician was randomly assigned to receive either the usual short technical report or a long narrative clinical report for this and all subsequent requests. The clinical reports were written by a specialist endocrinologist. Of note here is that primary care physicians in this study included general internists, family practitioners and gynaecologists.

Physicians were interviewed by telephone after receiving at least two reports in their assigned format. Once physicians in the short-report arm had been interviewed, they received long reports for any subsequent scans and were interviewed again after having received at least two long reports. The interviews explored the physicians’ understanding of bone densitometry and asked specific questions about the individual patients. The questions asked were not specified in the paper, although a standard pro forma was used.

A total of 68 physicians were enrolled, although 11 of these were excluded from the analysis as either they had moved out of area before follow-up or they did not meet the study protocol (the paper does not state what this protocol was). The final sample included 57 physicians (35 in the short-report arm and 22 in the long-report arm) who between them ordered 894 tests over the study period.

On interview, physicians who received the long report were more likely to understand the role of bone mineral density in the diagnosis of osteoporosis: 86% of physicians in long-report arm compared with 30% in the short-report arm. The authors report trends towards fewer formal referrals to specialists in the long-form arm (3% of patients vs. 11% of patients), although this did not reach statistical significance. All of the 20 physicians who received the long reports after initially receiving the short report preferred the long-report format.

In terms of the theory under test, this study suggested that physicians may find technical test results difficult to interpret and that this may impact on subsequent patient management. Physicians preferred clinical reports and this seemed to have had some impact on referral behaviour.

Skipsey and colleagues233

This paper reported the outcomes from a pilot study of an open-access service for cardiac arrhythmia monitoring from a regional centre in Scotland. The aim was to improve access for patients to monitoring and reduce unnecessary referrals to the cardiology clinic in low-risk arrhythmia patients. ‘The definition of “high risk” was left to the discretion of the primary care physician or the consultant cardiologist’, so there did not appear to be any guidelines to facilitate the referrals.

Data were collected over a 29-month period (2008–10) on all consecutive patients referred for cardiac arrhythmia monitoring. Patients came from two sources: GP direct access and ‘redirected’ consultant referrals where the GP had made a cardiology referral but the consultant identified the patient as eligible for the monitoring without clinical review.

A total of 280 patients were referred for monitoring, although 41 were excluded from the analysis (39 cancelled/did not attend and the equipment failed in two cases). Of the 239 patients evaluated, 165 (69%) came through direct access and 72 (31%) came via the ‘redirected’ route. However, it is not known how many GP referrals were made directly to specialists during this period and, of these, what percentage of patients were then referred for arrhythmia monitoring.

Of the 239 patients, 230 (96.2%) were returned directly to be managed by the GP. Nine patients were reviewed by the consultant cardiologist and three of these were returned to GP care without clinical review. However, follow-up data showed that 50 (21%) of the patients had a subsequent cardiology outpatient appointment. It would seem from the paper that the length of follow-up was not consistent for every patient (days to follow-up ranged from 216 to 1119) which would suggest that the data on subsequent outpatient usage should be treated with a little caution, although it is perhaps likely to be an underestimate. Of these 50 patients, 27 of them were returned to GP care after one clinic appointment. Figure 2 from the paper, reproduced here as Figure 32, illustrates the patient flow in the study.

FIGURE 32.

Patient outcomes following arrhythmia monitoring. Reproduced with permission from via Medica from Skipsey, DA, Dawson FM, Breen C, Leslie SJ. Evaluation of a direct access cardiac arrhythmia monitoring service. Cardiology Journal 2012;19:70–5. 233

The authors conclude that:

there were no adverse events and thus the direct access service appeared to achieve its aim of safely reducing unnecessary referral of low risk patients with palpitations to the cardiology clinic. However, despite the availability of the service, many GPs still referred low risk patients directly to consultants.

The authors advocated both greater promotion and more effective ‘policing’ of this service in order to ensure that the benefits of the service were fully realised.

The authors note that 19 patients were referred to clinic after initially being cleared for GP management. They suggest that, as all these patients were discharged after a single clinic visit, no serious conditions were missed through the provision of open access.

In terms of the theory under test, this study suggests that despite the provision of direct access, there was evidence of GPs bypassing the service by referring directly to clinic. The authors suggested that better promotion of the service may increase the usage. However, the paper noted that no guidelines were given to aid the GP in deciding which patients were ‘low risk’ and, therefore, eligible for the service. It could be that without clear guidelines some GPs did not feel confident to make that decision and preferred to defer to the clinic. The results also suggested that many of the GPs made subsequent referrals even though the test had indicated the patient could be managed in primary care. The study made no attempt to explore why this might be the case.

van Gurp and colleagues234

This paper reported the results from a prospective cohort study in the Netherlands. GPs were given direct access to echocardiography and asked to complete a standardised request form for all patients, including an indication of management if direct access was not available. A telephone call to the GP following receipt of the results collected information on subsequent management. As no control group was utilised, the findings of the study need to be interpreted with caution, as it is not clear how patients would have been managed in the absence of direct access. Estimates of how many referrals to secondary care were avoided were based on GP response to a hypothetical question of how they would have treated patients without direct access. Despite this, the study provides a further layer of evidence to test the theories under review.

A total of 164 patients were referred to the direct-access service for echocardiography for suspected cardiac conditions. Of these, 156 patients consented to be included in the study, although one of these was ineligible as he was under age. The results are reported for only the 105 of these patients who had a complete referral form and interview data.

If the GP had not had access to the service then the referral request indicated that 97/105 (92%) would have been referred to a specialist. As a consequence of direct access, only 36 (34%) were actually referred.

Our study shows that open access echocardiography may lead to significantly less referrals to the cardiologist (34% vs. 92%, p < 0.001). GPs were able to manage more patients in primary care (p < 0.001) and in most patients (82%) GPs found the echocardiogram of benefit for decision making.

van Gurp and colleagues234

In relation to the theory under test, this study suggested that direct access to echocardiography reduced the number of referrals to secondary care. In terms of professional boundaries, consultants retained a significant amount of control over the decision-making process. Although GPs maintained responsibility for the management of patients, all tests were evaluated and reported on by a cardiologist, with those assessed as needing referral given specific advice to do so by the evaluating cardiologist. Indeed, in the discussion section, the authors noted:

At first the evaluating cardiologists did not . . . add any advice to the echocardiography results . . . Because other studies have suggested that GPs strongly appreciate the specialist’s advice, the cardiologist was asked to provide advice with every echocardiogram.

van Gurp and colleagues234

This suggests that direct access to echocardiography may reduce referrals to secondary care and perhaps works best when the test result is accompanied by management guidance from the specialist to support GP decision-making. The paper did not explore the impact of increasing the amount of information given to GPs with the test results and so we are unable to comment on how much advice might be optimal.

van Heur and colleagues235

This paper reported on a retrospective review of management of patients following direct-access echocardiography for suspected heart failure; for these patients, echocardiogram was the gold-standard diagnostic test.

Participating GPs received short training on indications and restrictions for echocardiograms and on interpreting the results. The reports contained a summary of the results from a cardiologist and, if abnormalities were noted, a suggested management plan of four options:

• refer the patient to the cardiologist

• start or change medication

• repeat the echocardiogram within a few years

• consider endocarditis prophylaxis.

All echocardiogram requests and reports during a 4.5-year period (2002–7) were reviewed. In addition, a questionnaire was sent to all GPs who referred to the service to establish subsequent management.

During the study period, 625 patients were referred and GP questionnaires were returned for 517 (82.7%) patients. GPs varied in their adherence to the cardiologist’s advice; Table 11, which is a reproduction of Table 4 in the original paper, gives details.

TABLE 11 - Management by GPs after open-access echocardiography

GP, general practitioner; LVD, left ventricular dysfunction.

Relevant echocardiographic diagnosis: at least one relevant abnormality, i.e. valve disease or left ventricular dysfunction or left ventricular hypertrophy or pulmonary hypertension or septum defects.

Reproduced from van Heur and colleagues235 under the terms of the Creative Commons Attribution Licence 2.0.

Indication	A	B	Management by GP
	Patients with a relevant echocardiographic diagnosis,a of whom the GP returned the questionnaire on management	Advice of cardiologist to refer patient (% of A)	GP followed cardiologist’s advice to refer patient (% of B)	GP referred patient without advice of cardiologist [% of (A – B)]
Whole group, n = 625	397	114 (28.7)	81 (71.1)	47 (16.6)
Cardiac murmur, n = 368	237	73 (30.8)	56 (76.7)	26 (15.9)
Dyspnoea, n = 198	128	43 (33.6)	30 (69.8)	16 (18.8)
Peripheral oedema, n = 105	75	35 (46.7)	27 (77.1)	12 (30.0)
Other, n = 137	84	19 (22.6)	8 (42.1)	12 (18.5)

The authors stated that the purpose of open-access echocardiogram is to lower the threshold for diagnostic testing in patients with suspected heart failure. They suggested that the referral rate in this study was low, compared with expected numbers based on population data. The authors suggested that this may be because GPs are too strict in their test ordering behaviour, or alternatively it may be that patients in whom heart failure is suspected are being referred to the specialist when they would be suitable for open access.

The authors concluded that ‘open access echocardiography is a popular service to detect patients with heart failure and patients at risk for developing heart failure. Overall, GPs used the open access echocardiography service efficiently (i.e. with a high chance of relevant pathology), but efficiency decreased slightly over the years . . . Further specification of the indications for open access . . . might improve the service and make it clearer for GPs when to use it’ (reproduced from van Heur and colleagues235 under the terms of the Creative Commons Attribution Licence 2.0).

In terms of the theory under consideration, this study suggested that the specialists retained much of the control over the decision-making and management of patients referred for echocardiography. Although the GPs received training in interpreting the results, the results were accompanied by a management plan, which suggests that decision-making following the test remains under specialist control. However, this advice was not always adhered to by the referring GP. This may point to evidence of a lack of consensus between GPs and consultants as to which patients can be managed in primary care. Although it is unclear from this study if all patients would have received a specialist referral should open access not have been available, the results do suggest that the service was useful in identifying those patients who could be managed in primary care and it is likely that this reduces the demand for specialist referral. However, evidence remains of short-circuit, with a number of patients receiving a subsequent specialist referral despite the advice with the results that this was not necessary.

Fuat and colleagues236

This paper reported the results of a qualitative study using focus groups to explore the beliefs, current practice and decision-making of GPs in the diagnosis and management of suspected heart failure. Participating GPs were purposively sampled to recruit a range of participants over a number of variables including age, sex, ethnicity, practice size and geography. Data were collected from 30 GPs over four focus groups and analysed using an adapted grounded theory approach.

The analysis identified three themes in the data: (1) uncertainty about clinical practice; (2) lack of awareness of relevant research evidence; and (3) influences of personal preference and local organisational factors. Although this study was not about direct access to diagnostic testing, the results do provide some explanatory value to our programme theories. The GPs discussed the difficulty in making a diagnosis in primary care, particularly as the patients tended to be elderly with multiple comorbid conditions. At the time of publication, access to diagnostic tests such as echocardiography was variable, although two-thirds of the GPs in the study had access to this test. The quotes from the GPs expressed ambivalence towards the value of echocardiography in making a diagnosis in primary care, as they did not feel confident in interpreting the results. Some GPs would, therefore, not use direct access even when this was available.

The problem with echocardiograms is that I really just don’t understand them. I don’t think of myself as being really that old, I mean I’m 43 . . . and when I went through my post-registration years echocardiograms just weren’t around . . . I just don’t know where I am with them. When does an ejection fraction of such and such per cent stop being reasonable and start being a problem?

I would rather refer than do an echocardiogram, the interpretation of which I am not confident with.

Reproduced from Barriers to accute diagnosis and effective management of heart failure in primary care: qualitative study, Fuat A, Hungin APS, Murphy JJ, BMJ, vol. 326, p. 196, © 2003, with permission from BMJ Publishing Group Ltd236

The authors reported that GPs were less likely to use direct-access echocardiography when the reports they received were technical and lacked a clinical opinion.

In terms of the theory under test, GPs expressed difficulty in interpreting the results of the test and, therefore, direct access may not enable them to make appropriate decisions as to which patients can be managed in primary care and which require specialist referral. This can result in short-circuit of direct access, with some GPs preferring to make a referral to a specialist to make a diagnosis. The results resonate with those from Stock and colleagues232 on DXA scanning in that GPs state a preference for clinical reports that give guidance on management decisions. If the results of the test are accompanied by guidance on management in the light of the results then short-circuit may be minimised and referrals for specialist opinion may be reduced. Box 42 provides an interim summary.

BOX 42 - Interim summary

Clinical indicator tests are used to identify the location of a patient on a particular disease trajectory in order to inform a decision about whether to intervene now or whether to continue to monitor the patient. This information is also used to decide who is best placed (the GP or the specialist) to undertake this intervention. By providing GPs with direct access to these tests, it is assumed that GPs have the necessary skills to decide when and for whom such a test is warranted and then have the skills and confidence to interpret the results and decide which patients can continue to be managed in primary care and which patients require referral. As a result, more patients will be managed in primary care without the need for a specialist opinion and only those requiring specialist management are referred to secondary care.

Our evidence synthesis suggests that providing GPs with direct access to clinical indicator scans can enable GPs to distinguish between patients requiring specialist referral and those who can be management in primary care if specialists provide GPs with referral guidelines and guidance on subsequent management. As such, specialists retain a significant amount of control over the decision-making in the patient pathway by setting guidelines and management plans for GPs to follow. Without these guidelines, GPs may bypass direct access to the test and refer directly to the specialist because they do not know which patients can be referred to direct access or they feel unconfident in the interpretation of the test results.

However, even in the presence of guidelines, some patients may still receive a referral when the results suggest this is not necessary. This may be because of a lack of consensus between GPs and specialists regarding which patients are suitable for referral to secondary care.

Introduction: the guidelines industry

This is the fourth of our ‘empirical’ chapters. We move from those demand management programmes utilising role or mandate change and enter the procedural domain. Our focus here is on the attempt to use formal guidelines in the attempt to curb demand and reduce inappropriate referrals. This topic has a rather different research history and we have amended our reviewing strategy accordingly.

Our first ‘historical’ observation is to note that, unusually, there is some consensus about the utility (or perhaps futility) of using guidelines in the role of demand dampener. Imison and Naylor’s report comes to the unequivocal conclusion, from which few researchers have demurred, that the ‘passive use of guidelines’ has failed as a tool for reducing demand. 93 More anecdotal evidence on the matter was voiced in Chapter 3 by the practitioner who asserted that they ‘usually end up either in the bin or are added to the guideline mountain in the corner of the clinic’. 136 This provides a helpful push for our review towards the question of what constitutes the ‘active’ use of guidelines and how they can be incorporated into wider systems of demand management.

The second matter of record brings us to our reference to the guidelines industry. There are guidelines for every condition from abdominal pain to the zoster virus. There are guidelines for all sectors: clinical practice guidelines, public health guidelines, technology appraisal guidelines, self-care guidelines. There are guidelines for every point in the patient pathway: diagnosis, screening, referral, treatment, withdrawal from treatment. Production takes place at every level from cottage industry (e.g. guidelines for local practices and emergency teams) to national function (e.g. guidelines issued by Royal Colleges and National Institutes) and to global enterprise (e.g. guidelines from World Health Organization taskforces). The delivery formats are diverse, covering everything from the one-page poster to hundred-page manuals and, latterly, to e-guidelines. And yet there is a gap in the glut. Relatively few of the guidelines restrict themselves to demand management per se. As noted, they tend to the comprehensive, offering advice on all aspects of screening, diagnostics, testing, treatment and then referrals. The significant corollary is that the primary research on their utility has not tended to confine itself to demand issues and also has leaned towards consideration of their uptake in general terms.

Our third point of ground clearing is to note that the topic, as just conceived, has gathered its own research industry. Many hundreds of studies have amassed assessing guideline ‘X’ in order to discover the extent of compliance and the reasons why the advice is and is not followed. Investigation on this scale results inevitably in the need for synthesis and so the primary research is rapidly followed by many systematic reviews of guidelines from ‘A’ to ‘Z’ – trying to understand, in broader terms, what works in guideline production and to unearth the all-pervasive facilitators and impediments to their implementation.

The methodological point of this brief history is that our review is preceded by several significant reviews and that much is to be gained by conducting a review of reviews. To this end, the chapter has a different format. We offer (1) a very brief look at a review in the Cochrane tradition; (2) a closer examination of the mainstream method in this area – primary studies and their reviews in the ‘barriers and facilitators’ tradition; and (3) a rapid realist review in which we concentrate on how guidelines may take on a more active role within a whole-system approach to demand management.

Approach 1: Cochrane orthodoxy – guidelines as interventions

With the reminder that clinical guidelines have been in use and under research for decades, we begin by reproducing the conclusion of a recent study following ‘strict Cochrane methodology’:

Our results reveal that there are only a few rigorous studies which assess the effectiveness of a strategy to implement clinical guidelines in Europe. Moreover, the results are not consistent in showing which strategy is the most appropriate to facilitate their implementation. Therefore, further research is needed to develop more rigorous studies to evaluate health outcomes associated with the implementation of clinical guidelines; to assess the cost-effectiveness of implementing clinical guidelines; and to investigate the perspective of service users and health service staff.

Reprinted from Health Policy, vol. 107, Brusamento S, Legido-Quigley H, Panteli D, Turk E, Knai C, Saliba V, et al. Assessing the effectiveness of strategies to implement clinical guidelines for the management of chronic diseases at primary care level in EU member states: a systematic review, pp. 168–83, 2012, with permission from Elsevier237

The review unearths 2562 potential ‘records’, of which 2535 are rejected (on the usual grounds of failure to meet the methodological gold standard established for clinical trials). 237 Although the remnants do not carry all the complexity described above, it turns out that the chosen 27 are still massively heterogeneous (local vs. national guidelines; single vs. multifaceted interventions; delivery variation – feedback, workshops, outreach, computerised decision systems, etc.). Unable to perform meta-analysis with such raw materials, the study settles for six dense pages that merely describe and summarise the ‘eligible’ studies. The methodological error, of course, is to assume that guidelines, which carry the imprint of every part of a health-care system, can be regarded as a form of treatment applied to institutions, which can be turned on and off and outcomes compared. Guidelines do not impact on health providers like a pill on patient. As we shall see, guidelines have meaning and impact only in the varied, shifting and uncontrollable reactions of their intended stakeholders. It may be concluded that approach 1, as is often the case, tells us much more about the dictates binding the particular review methodology than it does about the effectiveness of complex programmes.

It also goes without saying that the more-research-is-needed conclusion is of no use whatsoever to policy-makers. Just as misguided is the idea that, had enough gold-standard RCTs been unearthed, a mean effect calculation on their effectiveness would make a difference to the DMs. Health-care systems are awash with guidelines and the time has long passed for some aggregative verdict on whether or not they work.

Approach 2: thematic reviews – facilitators and barriers to guidelines

The second approach to research review considered here, which is sometimes characterised as a ‘thematic’ or ‘narrative’ perspective, can be considered the orthodox approach to research synthesis in the particular domain of guideline investigation. It finds widespread application because it responds directly to the available raw materials. Rather than finding them wanting, it builds foursquare on the research designs used in the primary studies. That mainstream research strategy is the self-styled ‘barriers and facilitators’ approach. The basic design may be summarised as follows. A particular guideline is chosen for investigation, as is a selected set of its intended users. They are then faced with a broad set of questions on their familiarly with, experience of, attitude towards and confidence in the said guideline. The questions are delivered in interviews, focus groups and surveys, though it is probably true to say that the majority are derived from small-scale case studies – so the questioning is predominantly non-directive and the responses are, in the main, discursive. Responses, usually via transcripts, are then subjected to thematic analysis. Initial codes are generated, candidate themes are extracted, data reduction and simplification is considered, overarching frameworks are compiled and member checking is conducted to support the veracity of the results.

The analysis is usually presented in the form a framework or typology. If surmounted, the extracted themes become ‘facilitators’; if they present a stumbling block to implementation, they become ‘barriers’. Although they have this common origin the ensuing frameworks vary greatly in the pattern of types and subtypes derived in the analysis. We present a brief synopsis from four characteristic studies in Table 12, noting a significant proviso that these represent drops in the primary research ocean. We have no space here to describe the eventuating themes in any detail – hopefully, their meaning will be reasonably self-evident. In the original studies the meaning, distinguishing features and rationale for each type is elucidated in greater depth. Illustrative, from-the-horse’s-mouth quotations are provided in the exposition of each theme. For example, consider the two following miniature testimonies. In respect of a widely cited theme on the need for ‘clarity’ in guideline presentation a GP complains: ‘a complicated piece of paper, it’s no use to me. I’m a simple man and I need to have simple ideas’. 239 On the idea that ‘patient pressure’ may encourage the physician to ignore guidelines, another GP explains that radiography referrals for back pain are in high demand as a source of ‘illness legitimation’. ‘The patient can come home and say, “I had an X-ray” and then everybody will realise I have a pain in my back.’238

Here, then, are the bountiful raw materials for another type of review, one that will go on to compile, translate between and synthesise the extracted types and subtypes. Because they represent the daily struggles of actual users, these primary studies present rich practical insight of the fortunes and foibles of guideline implementation. But does this approach constitute valid and reliable evidence? We have no space to conduct a methodological evaluation of even the handful of inquiries above. By way of a ‘composite quality appraisal’ we offer a reminder of the classic challenges of such qualitative, thematic analysis. They tend to be localised case studies and carry problems of generalisability. Especially if conducted in focus groups, they face problems of panel selectivity and ‘chatty bias’. The precise analytic details of code extraction and theme development are rarely reported and so there is the possibility that the chosen ensemble corresponds to some pre-existing conceptual framework preferred by the researchers. There is also the possibility of the imposition of a broader ideological or professional opposition to guidelines – perhaps noticeable in the main title of the Rashidian and colleagues’ study, ‘Falling on stony ground?’. 239 We do not suppose that these challenges disqualify such primary research from secondary review, merely that they should be reflected in any subsequent analysis.

We arrive at the main business of this section – to provide an account of two recent reviews attempting to synthesis the entire barriers and facilitators literature. 241,242 It should be noted that there are other reviews of this ilk, with a little history going back to 1999. 243 The research strategy is complex. Basically, it is a cross between a systematic review (scoping and definitional work, searching for studies, quality appraising, extracting findings) and a thematic analysis (to be precise, a secondary thematic analysis of the various primary thematic analyses). The essential aim is to provide a master or meta-framework that captures and combines all existing frameworks. Cochrane and colleagues call them ‘barrier groupings’ (see Box 43). 241 Gagliardi and colleagues call it a ‘final conceptual framework’ (see Table 13). 242 Box 43 contains some additional information on ‘frequency’, that is the number of times a particular theme has been discovered in the primary literature. Table 13 provides a different set of themes and subthemes but also includes a useful third column providing brief examples explaining the coverage of each theme.

A conclusion that hits the eyeball is that there are encouraging similarities but also significant differences between the meta-frameworks. Familiarity with other reviews in the barriers tradition, as above, confirms this volatile picture. Our focus here is to examine the practical purchase of such reviews and this quest now seems to turn on a prior question of deciding between frameworks. Why might the DM follow one typology rather than another? Is it feasible to seek the optimal framework?

There is, of course, a large methodological literature on frameworks, classification systems, typologies, taxonomies and so forth. 244 And within this there is the classic discussion on the different types of ‘validity’ – ‘face’, ‘content’, ‘criterion’, ‘construct’ and so forth. All of the frameworks under discussion (primary and secondary) have evident face validity. They originate in the lived experience of guideline users – so that notion, alas, is no help in deciding between them. It transpires that most methodological authorities place faith in ‘construct validity’ (there are various antonyms). Simply stated, this postulates that the worth of a measure is extrinsic to the measure and belongs to the inferences drawn using the measure. If one tests a theory using a particular measure and that theory is corroborated then this also adds validity to the measure. A measure becomes validated over time – the more theories it supports and the more those theories are successful, the stronger the faith in the measure: ‘The best construct is the one around which we can build the greatest number of inferences, in the most direct fashion’. 245

If one follows this advice (and we do) this allocates the assessment of the validity of any guidance framework to the utility of the model rather than the construction of the model. And this returns us to the central question of the chapter: how are users supposed to make use of all of this evidence on guideline effectiveness? We begin by considering the expectations on this score of one our review teams. In the second half of Gagliardi and colleagues’ paper there is a ‘test’ of 20 existing specialist guidelines (on the management of diabetes, hypertension, leg ulcer and heart failure) to see if they fit with the ‘extended model’ emerging from the typological review. 242 Results are disappointing. The various guidelines under scrutiny are largely adequate in terms of presenting ‘graded evidence’ but few examples contained ‘additional features that could improve guideline usage’. In other words existing guidelines fare much better in their coverage of items in the upper sections of Table 13.

A curiously limited notion of guideline validity is implied in this test. What the Gagliardi review seeks to inform is the business of guideline construction. The evidence on why potential users of guidelines follow or fail to follow them is returned on itself in order to improve the presentation and content of guidelines. A comprehensive, master framework is devised in the expectation that future guidelines should conform to that rubric. A grammatical shift from noun to adjective is introduced to establish this goal – guidelines vary in their ‘implementabilty’ and this evidence-endorsed template will help get their content right. The guideline industry is certainly promulgated and perhaps regulated as a result of such frameworks. Powerful collaborations have gathered promoting these ‘international tools for the rating and assessment of practice guidelines’ such as the AGREE template (www.agreetrust.org/).

But all of this is indeed a curious interpretation of construct validity and guideline utility because the main empirical lesson from the ‘barriers and facilitators’ investigations is that guidelines only have paper authority. The reason why guidelines fail is little to do with their content and format (their implementabilty) but mostly due to complex decision structures in which they are embedded (their implementation). However perfect their presentation, however comprehensive their coverage, however true to template, there is no reason to suppose that guidelines will be followed. This is the lesson that emerges from all empirical studies. What matters is the reception that awaits guidance when it has left the page and enters the clinic.

This brings us to a second, and perhaps more commonplace, expectation about how the thematic analysis on guideline effectiveness might be put to use. In this version, the review is said to provide the most authoritative checklist of barriers/facilitators to guideline implementation in expectation that it presents a ‘to do’ list for policy-makers, managers and practitioners. The role of barrier and facilitator reviews, on this view, is to provide strategic overviews; they are the design tools itemising what is required in a comprehensive planning process. In the context of our review on the utility of guidelines for demand management it would be possible to present a similar framework of ‘dos and don’ts’ as advice to the policy-maker about to set in train some new guidance. The obvious question is raised: are these ‘barrier groupings’ an effective planning tool?

In any review of the literature one encounters critical appraisals. We thus noted that the checklist perspective has come in for its fair share of criticism, most notably in an article with a telling title: Is the metaphor of ‘barriers to change’ useful in understanding implementation?246 This paper takes us back to the raw materials similar to those noted in many of the primary studies on guidelines. Prompted by questions about why guidance might be ignored, a GP in the Checkland study replies: ‘But if anybody thinks that things arrive here, somebody has the time to look at it and then spread it as useful information that everybody else thinks sensibly about, they’ve got another thing coming’. 246 Now ‘removing’ this barrier to change is usually discussed in terms of simplifying guideline presentation, providing guideline summaries, improving the channelling of recommendations and so on. The authors, however, submit the rival interpretation that the ‘time pressure barrier’ is simply an underlying organisational reality and that the source of the problem and its solution may lie there.

The moral of the tale is that barriers are not something to be ticked off and torn down one at a time. What the ‘barriers’ primary studies are actually describing are personal, social and institutional interrelationships. Barriers interlock because a change in one part of a complex system will always trigger change in another and then another. This suggests a rather different role for reviews of guideline effectiveness. What the synthesis should be studying and explaining is why some barriers are more intractable than others and why solutions always have emergent effects. Solving barrier A might exacerbate barrier B, solving barrier C might create unintended consequence D, introducing facilitator E might be crushed by impediment F. We rather suppose that real DMs inhabit this alphabet soup of system strains and we turn next to our review that attempts to decipher the interconnections.

Approach 3: a rapid realist review

We devote the remainder of the chapter to exploring the potential of a realist synthesis of the primary research on guideline effectiveness. The first question, as ever, is to discover the programme theories that underpin the development of guidelines and what is an appropriate way to test them. We start with an enormous advantage. The vast literature on barriers and facilitators provides us with a wealth of provisional programme hypotheses, starting at square one with the elementary idea that a guideline that is simply stated is more likely to be understood and to find use. Our model thus starts with familiar ingredients (thanks be to all that typological work). But we do not seek to repeat reviewing history by working sequentially through all of the themes and subthemes of guideline implementability already identified. The vital empirical test, therefore, is not about whether or not a particular application of a guideline fulfils the checklist of facilitators and avoids all barriers. The checklist approach, as we have just seen, atomises guidelines into separate themes without appreciating that they are inextricably linked. Our review treats guidelines as a complex, multicomponent system thrust into a pre-existing, complex health-care system. The goal, as in the previous chapters, is to examine the tensions and harmonies with the emerging system (Figure 33).

FIGURE 33.

A phases and interdependence model.

Our model thus begins by identifying and compressing ‘guideline theory’ into four broad ‘if–then’ phases, which must be met for successful outcomes. The first concentrates on guideline production and considers the qualities of its production and of its producers. Only if these are sufficiently authoritative will guidelines reach square one – they get ‘on the table’. The recommendations then pass into potential usage and to the next set of theories. Only if there is a close fit between the paper advice and the circumstances of the presenting patient will the guidelines reach phase 2 – the point of applicability. However, even if there is good initial ‘fit’ with the patient’s condition the guideline may or may not be followed. All depends on the intricacies of phase 3 and whether or not there is consonance between its recommendations and the experience, custom and practice of the clinical user. Even if they survive this internal dialogue, the guidelines still may or may not be followed at phase 4. Recommended actions may have to overcome significant external resistance from further stakeholders, most notably from the patient and then from other resource-constrained providers in the system.

To repeat, these ‘phase theories’ are simple distillations of material from the barriers and facilitators literature. The difference is that we treat them as part of an interlocking system (represented by the linkages in Figure 33). In classic system terminology – the whole is more than the sum of the parts. Each phase represents a necessary condition to reach the next, but, as every manager knows, there are harmonies and tensions throughout the system. Success in one phase can actually make it more difficult to achieve another. Success in one phase can drive unforeseen consequences into another. Success in one phase may be squashed in another.

The focus of our rapid review thus turns to system functions and dysfunctions (or ‘system strains’ for short). Our mental image of the policy-maker is not one of the master planner, designing a guideline blueprint from scratch; rather, it is of the manager coping with many existing guidance systems and attempting to integrate them into wider care regimes. They will face a daily diet of system strains – is there an evidence base that might provide decision support in these circumstances? We provide examples of four system strains:

1. the tension in using simple guidelines for complex comorbidity

2. the tension between (inter)national credibility of and local control over guidelines

3. the tension between patient choice and top-down guidelines

4. the tensions involved when there are competing guidelines and contending targets.

What follows are some indicative data from a realist review; we extract only a few key findings and discuss their implications.

System strain 1: simple guidelines versus comorbid patients

A recurrent refrain in all the aforementioned research is that guidelines are more likely to be implemented if the presentational format is straightforward, intelligible, comprehensible, uncomplicated and so on. The beginnings of a system strain lurk here when through the surgery door walks patient George whose ills are far from uncomplicated. Several studies have explored the problems of fitting a patient to guidelines when that patient has comorbidities and, more especially, if that patient happens to be frail or elderly.

Primary studies group, first, into those examining the texts of existing guidelines within a clinical domain and assessing the extent to which they deal with comorbidity. Typical results in the study of guidance on chronic conditions are as follows:

Half the guidelines addressed treatment for older patients or for patients with one comorbid condition. But only one addressed treatment for older patients with comorbid conditions.

Vitry and Zhang247

Of the 20 guidelines, 17 (85%) addressed the issue of comorbidity and 14 (70%) provided specific recommendations on comorbidity. In general, the guidelines included few recommendations on patients with comorbidity (mean 3 recommendations per guideline). Of the 59 comorbidity-related recommendations provided, 46 (78%) addressed concordant comorbidities, 8 (14%) discordant comorbidities, and for 5 (8%) the type of comorbidity was not specified.

Reproduced from Lugtenberg and colleagues. 240 © 2011 Lugtenberg and colleagues. This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited

A basic tension is thus recognised in the literature, which can then be explored for potential resolutions. The simplest adaptation is to increase the elderly, comorbid patient’s exposure to multiple guidelines and several studies cover the potential costs and unintended conflicts of following more than one guideline simultaneously. Caughey and colleagues used treatment-dispensing data showing that 16% of people with diabetes being treated for other conditions received medicine with adverse effects on diabetes. 248 Hughes and colleagues249 mounted a simulation exercise on the consequences of following the explicit recommendations of two or more guidelines demonstrating a significant hike in the treatment burden, especially on self-care regimes in an elderly comorbid population ill equipped to meet such demands. Boyd and colleagues250 carried out an early exercise following a hypothetical comorbid patient who, following all relevant guidelines, would be prescribed 12 medications at a cost of US$406 per month, some with possible adverse effects.

Neither is this system strain resolved by simply increasing attention to comorbidity within condition-specific guidelines, as recommended in both the primary and the secondary research. 240,251 This suggestion would create an elemental strain on the evidence base, unlikely to be resolved, which reserves a place of honour for clinical trials in which patients with multiple morbidities are generally excluded. It also sends us full circle to the original conundrum – addressing comorbidity within a guideline would inevitably increase the complexity of guidelines, which is already an established standard deterrent to their usage. 252

Even these simple nuggets of evidence show that guidelines are constantly being made and remade under system strain. A further recent refinement to guideline logic suggests a shift from ‘disease specific recommendations’ to ‘patient centred guidelines’. 253–255 To be more precise, such guidance places attention on carefully selected subgroups, for example ‘decision making on care of the elderly with condition X’. Guidelines here focus much more on choosing and prioritising treatment and so make useful ground in reducing the tension inherent in following multiple combinations of condition specific guidance. However, end-of-life care has complexities of its own which, if guidelines are to be comprehensive, will then be drawn into the equation. Consider the key issue of life expectancy and the following recommendation from Durso: ‘Estimate the patient’s approximate life expectancy compared to the median for individuals of that age-sex cohort by considering the presence or absence of unusually good or poor health and function’. 255 We pass no comment on whether or not this meets the simplicity rule, noting only the existence of further such algorithms in similar guidelines. 256

At this point the evidence trail grows thinner and much of the literature on these niche ideas may best be described as proposals, prototypes and product descriptions. But we can summon other aspects of the evidence base already amassed to note basic tensions that are likely to surface in relation to these more specialised guideline configurations. There is already a problem with physicians being bombarded with guidelines on X, Y, Z. If, thanks to patient complexity, these become subdivided into guideline X₁, X₂, Y₁, Y₂, Y₃, Z₁, Z₂, with supplementary decision rules on navigating to the appropriate pathway, there is an obvious emergent further problem afoot in the realm of guidance fatigue. 171

In conclusion, we observe that both practitioners and researchers are alert to this particular system strain. The chain of adaptation and readaptation teaches us that comorbidity represents an ineluctable limitation on the implementation of guidelines. However, with the move from ‘disease specific’ to ‘patient specific’ and ‘decision specific’ guidelines it is also possible to discern more positive and more subtle efforts to strike a balance between the rule and the recipient. Work is only just beginning to make progress on which decisions cause the most consternation and how to direct these nuggets of information on treatment alternatives to the deluged physician.

System strain 2: the tension between (inter)national credibility and local control over guidelines

As noted in Box 43 and Table 13, the ‘credibility’ of guidelines is often cited as a key facilitator in influencing their uptake. Credibility, however, may be nurtured in quite different quarters. One programme theory here begins with the notion that guideline development requires major methodological resources. Accordingly, guidelines endorsed by national or global professional organisations are seen as more trustworthy by the clinical community and are, in turn, more likely to be implemented (e.g. The Royal College of Physicians’ national guideline for stroke257). However, nationally developed guidelines may lack applicability and relevance to local contextual factors. Programme theory 2 thus posits that involving local practitioners is the key source of credibility because the guidance will include intelligence on the prevalence of the condition in the local community, on the local availability of services and resources, and on current inter- and intraorganisational relationships (e.g. the Oxfordshire Regional Genetics Service referral guidelines258).

Here we have the beginnings of a classic system strain, sometimes termed the ‘glocalism’ paradox. To address this tension, the local adaptation of nationally developed guidelines has been widely advocated as a potential solution. Programme theory 3 thus seeks a ‘best of both worlds’ solution and considerable resources have been expended in pursuing it. The best known of these initiatives is the international ADAPTE collaboration (www.g-i-n.net/working-groups/adaptation/history). The task of ‘customising’ a guideline so that the global becomes the local is itself a feat of organisation. Decisions need to be made about topics, organising committees, source documentation, consultation rules, format and promotion. The ADAPTE process has a whole series of phases and modules, numbering 24 steps in all. 259 At the other end of the scale, adaptation can be ad hoc, locally initiated and focused on particular units with identifiable users. 260

While all of this is proof positive for our thesis that reviewers of the guidelines literature must anticipate a moving target, it opens up another system tension in need of explanatory synthesis. There is copious material comparing guideline content before and after implementation. Some of this considers whether or not adapted guidelines are of the same standard as those with a national pedigree. Rowe, for instance, suggests that locally developed guidelines are poor in coverage according to AGREE standards. 261 By contrast, Shye and colleagues suggest that the process of local adaptation does not seriously distort the clinical validity of the original guideline. 262 Other studies seek to unearth subtle, qualitative differences between the national and local instruments. Sometimes, the adapted guidelines have almost identical coverage and content to the originals. 263 Other research notes the adaptations are real: specialising in topics with local priorities,262 having shorter and more accessible formats262,264,265 and, perhaps most significantly, providing additional information on the availability of local services and thresholds for referral. 264

There are also many process evaluations, which trace the practical steps and resources involved in guideline adaptation. Sometimes guidelines are adapted through a series of short workshops or meetings with local primary and secondary care clinicians. 262–265 For example, Capdenat St Martin and colleagues report on a process that took 2 months from initiation to completion. 260 Liaw and colleagues report a case where the adaptation was undertaken through 12 hours of group discussion. 265 By contrast, other studies have followed the processes involved using formal templates, such as ADAPTE, described above, which attempt to standardise the process of guideline adaptation. 266,267 Key evidence emerging here suggests that full-blown formal adaptation is a resource-intensive and costly business – involving a need to revisit the original evidence, identify additional research evidence to support local content and provide methodological support to critically appraise the final product. 259

The major issue lurking within the ‘glocalism’ issue is, of course, whether locally adapted guidelines improve outcomes. Do users pay more attention to and act on guidelines that have been locally adapted? Yet another outcrop of studies tackles this problem and we begin a mini-review by noting a considerable methodological difficulty. The basic design involves manipulating a situation whereby identical groups of practitioners are exposed to existing and to adapted versions of the guidelines and then observing differences in their understanding and action. The whole of the facilitators and barriers literature teaches us that achieving this clean difference will be demanding because the actuality of ‘exposure’ to a guideline is so diverse. For illustration, we examine findings from two such studies.

Silagy and colleagues263 conducted a cluster RCT on guidelines for stroke prevention and on urinary tract symptoms. Nationally and locally designed guidelines were allocated randomly to two divisions of general practice in Adelaide, South Australia. The local guidance included additional information regarding the availability of divisional resources and was presented with more ‘user-friendly’ design and formatting. Pre-trial measures were collected from a postal survey of 200 GPs from each division, covering knowledge, action and reported practices in respect of the two conditions. This was repeated 3 months after the dissemination phase. Dissemination, identical for both versions, included mailing shots, newsletter articles, prompt sheets, educational workshops and web-links. Considerable change in GPs’ usage of guidance was noted across the before-and-after measures – observable, disappointingly, in both arms of the trial. The authors concluded that although they found significant changes in knowledge, attitudes and reported practice as a result of guidelines dissemination, the local adaptation of the guidelines did not produce any additional effect. They argue that efforts to promote guideline implementation should focus on multifaceted dissemination strategies, rather than local adaptation per se.

A pioneering French study by Capdenat Saint-Martin and colleagues also throws light on mechanisms through which guideline adaptation may work. 260 The background here was the perceived overuse of preoperative tests for anaesthetic risk. National guidelines had shown that they were costly, had highly restrictive diagnostic use and did not add significantly to the safety of operations. Capedenat Saint-Martin’s team thus undertook a process of local adaptation of these guidelines at the hospital level (15 surgical wards). The organisational structures (decision trees) responsible for ordering the tests were mapped and team discussions of the new recommendations were set up in the appropriate branches. Before-and-after measures were taken of referrals for such preoperative tests and in the targeted low-risk groups requests fell from 80% to 48%. The question of attribution raises its troublesome head. For the triallists, we lack a control group here and simply cannot say that the adapted guideline bore responsibility for the change. The French team drew a different, system-based lesson. The people responsible for adapting the guidelines were the same people who organised the implementation of the new guidelines: ‘We think that the main contribution of this work is the linking of the process of local adaptation to an analysis of the organisational aspects of the practice and the emphasis we placed on the organisational aspects of change’. 260

Many other studies of the impact of local adaptation have followed but these two studies above provide indication of the direction of travel. Whether guideline content is de novo or de integro seems to be far less crucial than its passage from the text to consulting room, through the conversations between its users and onto the external constraints on its usage. The take-home messages seem to be (1) adapt guidelines to kindle interest rather than to impart new knowledge and (2) the more local the adaptation, the greater the number of interested parties who will be drawn directly into the scheme.

System strain 3: the tension between authoritative, population-based guidelines and discretionary, individual-based patient choice

Clinical guidelines are derived from the assimilation of evidence at the level of the population. The potential risks and benefits of referral or treatment are expressed in terms of population estimates; recommendations are general and focused on the average patient. By contrast, GPs and patients make decisions on an individual basis and patients vary in terms of their preferences for referral and treatment. There is a considerable anecdotal literature in the practitioner journals about ‘turning a blind eye to guidelines’ by referring the worried well patient onwards for further advice or by complying with their world-weary wishes to forego any recommended interventions. 268 This system strain also occurs in macro policy. Recent years have seen an introduction of procedures giving patients more choice on the mode, location and timing of treatment. 269 Clinical guidelines, paradigmatically, seek to impose order on treatment modalities.

The first and most widely evaluated solution to this tension is to involve patients in the process of developing the guidelines. It is hypothesised that incorporating patient knowledge into the guideline development process will make the guidelines more applicable to everyday practice and increase their legitimacy and, thus, improve their uptake. However, the process through which patients are involved is a matter of debate. Van Wersch and colleagues270 compared a number of different methods (guideline group membership, one-off meetings, workshop series and consumer advocates) and concluded that no single method alone would be sufficient to optimise patient involvement. Providing intensive support to patients or having a separate working group feeding into the main guideline group may succeed in ensuring that key patient concerns are addressed within the published guideline. But this approach is resource intensive and, accordingly, some patients report that their input feels marginal and tokenistic. 271

Van de Bovenkamp and Trappenburg’s review272 and also a later empirical study273 highlighted crucial practical impediments to including the patients’ voice in guideline development. Patients may have difficulty understanding medical jargon and, consequently, provide little input into the process. Restricting involvement only to patients who are able to understand the scientific literature may call into question the representativeness of this input. Training patients to understand the medical literature enlists them as fellow ‘academics’ and may distance them from their experiential knowledge that is the basis for their involvement. They may become professionalised and fail to represent true patients ‘in the wild’.

The degree of fit between the patient perspective and those of other stakeholders involved in guideline development has also come under empirical scrutiny. 274 Case scenarios of patients with knee and urinary tract symptoms of different severity were presented to a guideline development group consisting of surgeons, GPs, health-care professionals and patients. The research examined which patients were deemed appropriate for referral with the idea of testing whether or not the ensuing guidelines would be consensual. Interestingly, patient and surgeon viewpoints came significantly closer to those of the GPs in some instances. Overall, however, the results are described as a ‘juggling act’, with different stakeholders providing discordant ratings across the differing scenarios.

The fundamental and perhaps irresolvable problem with involving patients in guideline production is that patient’s experiential knowledge, which is inherently individual, subjective and fluid is incommensurate with guidelines based on knowledge considered to be objective, valid and scientific. Accordingly, policy and the research literature has gone in search of ways of ‘reconfiguring the tension’ by locating the patient perspective at other points in the guidance machinery.

An alternative strategy is to have explicit statements within guidelines that advise on making room for individual patient preferences. Van der Bovenkamp and Trappenburg argued that this can be accomplished by incorporating sections on doctor–patient communication within the guideline or by having a patient ‘version’ of the guideline to support their active involvement in treatment decisions. 272 Clarke’s research pinpointed a specific niche in which the two activities may be combined. 274 They argued that guidelines could identify and differentiate levels of symptom severity, as patients’ preferences for referral depended on the severity of their symptoms. 274 We were unable to discover any empirical evaluations of these specific manoeuvres but they are likely to reintroduce a strain already discussed. Guidelines that come in competing versions or are laced with too many contingencies and caveats are less likely to be followed.

Van der Weijden and colleagues’ solution was the collaborative development of clinical guidelines with patient decision aids. 275 The latter already have considerable authority and autonomy and are aimed directly at patient concerns. Their development, moreover, has been every bit as rapid as that of clinical guidelines. 276 Van der Weijden and colleagues set out a framework to specify another niche condition in which a marriage of these distant cousins might be fruitful. 275 Where there is low uncertainty about the factors that influence decision-making, such as agreement about treatment risks and benefits, it is possible to make strong recommendations on the ideal course of action in the guidelines. In these circumstances, decision aids can be used in parallel to support adherence to the guideline and behaviour change. However, some decisions entail a high level of uncertainty, for example because there are two or more treatment options of equal attractiveness regarding the outcome. For these ‘utility sensitive’ decisions, guidelines should make only conditional recommendations, which should be mirrored in decision aids that recommend ‘careful deliberation’. 275

Again we are presented with an attractive, but largely untried, programme theory. Several stringent tests, however, have already been suggested in materials reviewed to this point. The patients’ use of decision aids and the physicians’ use of guidelines would require considerable co-ordination. Shared decision-making in routine clinical practice is in its infancy and usage is deeply stratified. Han and colleagues, for instance, reported low levels of shared decision-making in prostate-specific antigen screening with ‘significant variations by black race, Hispanic race, higher education and health insurance’. 277 Although co-ordination may be relatively straightforward for individuals faced with a single condition, it would fall under the standard problems in dealing with comorbidities or multimorbidities. The ‘simplicity rule’ is once again under threat. Guidelines, like hymn sheets, work better in the singular. We are driven to a similar conclusion – solving a tension in one part of the system may overlook or even give rise to different system strains. We should anticipate that the patient’s consent to guidelines will remain a matter for piecemeal solution in another niche condition, namely in the doctor’s surgery.

System strain 4: tensions resulting from guideline oversupply – a new guideline can swamp a routine system

It is well established in the barriers-and-facilitators literature that the ‘prominence’ of a guideline can influence its uptake. Guidelines with the highest profile are much more likely to be implemented, with the accompanying caveats already noted about their requisite applicability and synchronicity. Coiera suggested that we now have a ‘clinical information marketplace’, where evidence-based recommendations are in competition for the clinician’s attention and, with the proliferation of guideline production, we are in a situation of information oversupply; ‘the amount of information that can be accessed or “consumed” is fundamentally limited by human attention’. 278 Another paper in the barriers tradition focuses on a significant aspect of the way in which GPs consume guidelines, namely that they often do so ‘haphazardly’ or ‘opportunistically’. 279 Especially in preventative care, where there are no clear daily routines into which guidelines on health promotion and disease prevention can be inserted, the response to the marketplace may be capricious and piecemeal.

One gathering and somewhat paradoxical response to information overload is that the ‘guidelines on guidelines’ also come replete with advice on how to position clinical guidance institutionally and how to increase its visibility in the daily business of the clinic and the surgery. We address this issue of guideline rivalry with a brief review of the evidence on a highly visible episode. A guideline will have obvious prominence if it is introduced as a national priority and accompanied by strong ideological wind. From a systems perspective, however, the key issue is not about tactics for assembling the requisite political leverage. The new guidance will always sit alongside a range of other clinical decision supports and so the key question is about displacement effects of new mandates.

In 2000, in response to comparatively unfavourable mortality rates from cancer, the UK government introduced the ‘2-week wait’ rule for suspected-cancer referrals, accompanied by widely distributed guidelines for urgent referral across a number of specialities. Several studies went on to identify problems with the 2-week mandate. Secondary care services, working under a pre-existing guidance regime, were often ill prepared for the change in referrals and suffered from ‘system swamping’. Malik and colleagues tracked all referrals to a bone and tissue soft tumour service over a 2-year period. 280 They also reviewed the case notes of those referred under the guidelines for urgent review by the GP:

40 patients were referred under the guideline between January 2004 and December 2005. Ten of these patients (25%) had malignant tumours, compared with 243 of 507 (48%) of those referred from other sources. In 9 of the 40 cases, the patient did not meet the criteria for urgent referral. Although this target has focused attention on shortening the time to diagnosis and treatment, prioritising patients referred from general practitioners has the potential to disadvantage those with malignant tumours referred from other sources

Reproduced under the terms of the Creative Commons Attribution Licence from Malik and colleagues280

This pattern recurs in other cancer services. Jones and colleagues reported the results of two separate audits of referrals for patients with suspected colorectal cancer, which suggested that the 2-week waiting standard was being met at the expense of a substantial increase in waiting times for routine referrals, even though fewer than one-third of patients were referred through the 2-week wait. 281 Potter and colleagues reported on a prospective cohort study of referrals for suspected breast cancer in Bristol. 282 Over a 6-year period (1999–2005), the referral patterns had changed, with a higher proportion referred as urgent using the 2-week-wait system. Over the same period the odds of receiving a positive diagnosis in the urgent referral group fell and the odds increased in the routine referral cases. Moreover, although the target wait was achieved for the urgent referrals, the waiting times for routine referrals were extended.

How is all this to be explained? Cornford and colleagues provided an explanation with masterful brevity: ‘Patients want early referral to assure them that they do not have cancer, specialists want early referral for patients with cancer, and GPs want both’. 283 To address this rather more systematically, it is useful to return to Figure 33. We discover that concomitant guidelines can carry subtly different agendas. The impact of any guideline depends on whether or not it squares with GP presumptions, with patient proclivities, with specialist preferences and with further resource constraints. Different mandates may thus be seized on, sometimes opportunistically, sometimes ruthlessly and sometimes unintentionally by different stakeholders, and the result may be broader system strain. Such system dysfunction may result even if the original objectives are quite benign. In a further paper, Cornford and Oswald suggested that the 2-week rule was flawed from the start, being based on two conflicting agendas: the need to limit referrals to high-risk patients but also the need to increase referrals to achieve early detection. 284

It is not all bad news. Barwick and colleagues285 provided an update to the study by Malik and colleagues. 280 Over the years there had been a gradual accommodation to and a subsequent increase in admissions under the 2-week rule. A higher proportion of them were also considered appropriate, given the resulting detection of a malignant tumour. The folk explanation here is that change takes time to bed down. The systems explanation is that the change in guidance reverberated across the entire system and thoroughgoing reorganisation was required to bring compatibility to the different referral routes. The referral centre in question went on to introduce a multidisciplinary team co-ordinator, who monitored care across intakes, who established a one-stop shop for access to ultrasound and biopsies and who was empowered to ‘stop the clock’ on patients who ‘clearly do not have a malignancy’.

The insertion of the 2-week rule is perhaps an acute example of how guideline overload may lead to competition for resources among stakeholders. Cookson and colleagues, however, suggested that it may be the norm: ‘seemingly robust professional advice is distributed to clinicians, which, if adopted, will distort resource allocation and waste scarce resources’. 286 A high-quality guideline may attempt to make suggestions on workflow changes and resources allocations but, in zero-sum games, the costs involved in implementing a new process will always generate costs elsewhere in the system. Competition and displacement effects are the norm and the guideline implementation challenge is to anticipate them.

Conclusion

Here we attempt to draw together the key implications of the above analysis, which we separate into three paragraphs: substantive, methodological and procedural.

Given the tumultuous increase in their usage, guidelines should be regarded as part of the fabric of service delivery rather than as separate interventions with unique objectives. When a new guideline is introduced, or when an old one is updated, it sits alongside a range of other organisational controls, rules, norms, customs, practices, targets and guidelines. Guidelines will always be a constituent part of a system of governance and their destiny rests on how well they are absorbed into that system.

There has been an equally tumultuous increase in the research on guidelines and it is important to find the appropriate means of synthesising the burgeoning evidence. We have argued that it is futile to try to answer questions about whether or not guidelines work. The only reasonable answer is that ‘it all depends’. We have also pointed to the limitations of trying to seek a master framework, listing all of the factors, on which success depends. There are scores and scores and scores of barriers and facilitators that help and hinder guideline implementation, but these do not resolve into some sort of winning formula because the factors identified are always interdependent. Dealing with barrier A will always have effects, anticipated and unanticipated, on enabler B, and so on. This interconnectivity does not leave systematic review with the task of describing unending, unforeseeable change. History does repeat itself and a raft of discernible system strains can be detected as guidelines are introduced. There are oft-repeated tensions between semiautonomous stakeholders. There are inevitable strains between blanket recommendations and limited resources. There are persistent frictions between national ambitions and local experiences. System change does have a pattern and in the paper we have described a method for analysing how some of the more familiar strains evolve and resolve.

This brings us finally to the business of research utilisation. Evidence is supposed to inform policy and practice and this ambition necessitates a realistic understanding of the roles of the policy-maker, the manager and the practitioner. Our understanding, in this domain, is that the time has long passed for high arbitration about whether or not to have guidelines. DMs are already awash with guidelines and so they are not sitting, Pilate-like, awaiting the definitive verdict about their effectiveness. We also presume that very few DMs work in splendid isolation and have the task of implementing an entire guideline system from scratch. They are already inundated with guidelines and they do not operate by ticking off ratified checklists about how they should be managed. We do suppose that the key business is system improvement. Policy-makers, managers and practitioners are awash with guidelines. Accordingly, the key task is to dovetail the latest manifestation into an existing organisational structure. The really difficult activity is to engineer and re-engineer guidelines so that they work smoothly with the pre-existing system. System technicians want to know how to get the grease to the squeak – they want to know about system strains and how to resolve them.

Lessons learned

We commence with some brief remarks on our topic, our study and the nature of the lessons learnt. The evidence amassed in this report confirms our preliminary hypothesis that demand management constitutes a wicked problem, which has defied clear and easily reproducible solutions. In scouring the evidence base we have unearthed an ever-growing list of contingencies that have to be overcome if such schemes are to succeed. We consider that the task of the impartial reviewer is to be the bearer of reasoned judgements, regardless of whether they constitute good or bad news. And, in this respect, the negative inference to be drawn from our review is that there are no bullets (magic or silver) to be fired that have the omnipresent capacity to slay rising demand. There are no preferred interventions that have the capacity to outperform all others. There is no best practice manual to be found out there or in this report to guide health-care personnel on when, where and how to make referrals. With equal emphasis it is worth recalling a general principle mentioned earlier, namely that the designation ‘wicked problem’ does not signify a hopeless, unfathomable, insoluble task. What this means is that there are many, diverse, hard-won, local and adaptive solutions. The key evidence in this review comes in the form of detailed expositions of the immense difficulties and occasional, bespoke successes, in bringing into equilibrium the interlocking systems on which sustainable change depends.

How can this highly nuanced evidence be marshalled for benefit of future policy-makers and practitioners? Hopefully, we have presented evidence enough to show that our recommendation cannot be neatly parcelled up into a league table that differentiates intervention winners and losers. Some previous reviews of demand management interventions have attempted to work to this formula, seeking to discover whether or not programmes of type A work better than type B or type C, and so on. We present evidence in Appendix 1 to show that these reviews always end in indeterminacy rather than an inventory of programmes to be cherished and a hit-list of those to be chastised. In exactly the same way, it would have been nonsensical for us to conjure up claims that ‘referral management centres’ work better than ‘GPs with special interests’, or that ‘guidelines’ are inferior to ‘direct access to tests’ and so on. The proper answer to the ‘what works’ question is always ‘it depends’, and the job of research synthesis, we submit, is to deepen our understanding of the contingencies. And this is what we have attempted in the main body of the review.

So, how does one convey the lessons learnt from a patchwork quilt of success and failure? Another approach, much more consonant with a complex and mixed picture of outcomes, is to present a listing of the ‘common denominators’ of success, those features of implementation that make a difference. The logic here is captured by adapting the old adage, ‘It ain’t what you do, it’s the way that you do it’. Although this proposition underscores perfectly the predicament of demand management, these laundry lists of implementation ‘must dos’ have a tendency to the tautological and the banal. It is indeed the case that we have shown that demand management requires ‘good communication’, ‘strong leadership’, ‘careful forward-planning’, and so on. We also know that there is evidence aplenty to show that the way forward lies in ‘local’, ‘adaptive’ solutions. Alas, this advice applies to almost everything under the sun and supplies little more than a moral message to the practitioner.

Another familiar approach to research recommendations is to present vignettes of successful ‘case studies’. These feature, for instance, in the NHS Institute for Innovation and Improvement’s document, Improving Patient Flows in the NHS: Case Studies in Reducing Delay. 287 What the case studies are able to show is the complexity of problem and response. There are many similar instances, recounted in the main body of our review, in which groups of practitioners have fashioned and refashioned an intervention to provide important local, adaptive solutions. Again, there is a drawback. The highly specific nature of the ‘exemplary cases’ may have a tendency to draw a different, localised retort – namely, that it is not really like that in my bailiwick. My clinical domain, my patient pathway, my management structure, my funding streams are all different.

Accordingly, what we have tried to present in these conclusions is a mixture of these two approaches, grasping their advantages and skirting some of the difficulties. Rather than present self-evident lists of common factors that lie behind intervention success, we have generated a series of propositions, drawn from our analysis, that explain why particular processes or features contribute to successful outcomes. In other words, we attempt to convey the core explanations embodied in the solutions and it is these design principles that may be transferable. In order to make this exercise reasonably economical we have restricted ourselves to 12 design principles, two drawn from each analytic chapter in the report. We also add vignettes, miniature case studies from the previous reviews showing how each principle may be realised. What this mix of design principle and local practice attempts to portray are some common predicaments that apply in common demand management interventions, pointing to the strains that have to be overcome and the interdependencies that have to be forged. We hope that these learning points echo with the struggles of practitioners as they design their own referral systems.

Design principles in demand management

In order to preserve the evolution of our own argument, we present our key lessons in the same chapter running order as in the review. It will be recalled that Chapter 2 presented the background to the demand management problem and Chapter 3 presented an overview of potential solutions. The remaining chapters, the evidence chapters, reviewed the primary studies on a selection of key interventions. Each chapter posed the same question – can the particular intervention ‘control and shape demand?’ – and each chapter examined the many contingencies that have to be battled through to achieve this desired end. Readers are referred back to the conclusions of each chapter for our detailed results. Here, we reduce our findings to a brief set of summary propositions, which explain the design principle that needs to be in place in order to meet the policy aspiration. These key design ideas are summarised in Box 44 at a level of abstraction that allows for transfer and application across wide range of service remodelling. They are followed by key vignettes, which are intended to exemplify practical applications of the each principle. They are also presented on a chapter-by-chapter basis.

Chapter 2 gathered evidence on the manifold interconnected processes that generate demand for health care. It presented us with two daunting hypotheses. The first is that multiple, intertwined problems are unlikely to yield to singular solutions, however well aimed. The demand for health care may be regarded as a swelling punch bag. Landing a blow in respect of one problem may simply be absorbed as other vicissitudes gather. Our second observation is that health service provision is itself a magnet for demand. Many of the solutions to demand inflation involve the creation of new roles and services. The solution always has the potential to attract further demand (Box 45).

Chapter 3 examined the programme theories that underpin the small army of interventions that have attempted to quell the inflation in demand for services. Interventions are theory incarnate. Practitioners do not simply implement programmes. They think them through. Much is to be gained by tracing this perpetual quest to improve programme effectiveness. When the sprint of implementation turns to the middle distance and then to the marathon we noted the tendency of programme theories to become whole-system theories. The programmes that appear to survive are the ones which adopt a solution, and then adapt that solution in the face of its untended consequences, and then adapt the adaptation, and so on (Box 46).

Chapter 4 examined the theory that RMCs and other centralised triaging services could bring efficiencies to demand management by rationalising decision-making along the habitual, heterogeneous referral pathways that exist presently between primary and secondary care. RMCs only function to this end in the presence of a cluster of conducive conditions. Major administrative reforms generate power struggles and there is need for extensive collaboration between NHS managers and clinicians to agree on the governance structures. Continuity in the form of the recruitment of experienced local GPs (rather than adjunct professionals) to help manage the triage service is also found to be important. All of the protocols and guidelines that define the RMC’s remit and functioning should be codeveloped rather than imposed (Box 47).

Chapter 5 examined the theory that an intermediate professional role, the GPwSI, blending the expertise of the GP and the consultant, could manage demand in the system by siphoning off and dealing with an intermediate case mix appropriate to their medical experience and proficiency. This ambition failed to realise and the new role descended into an administrative support function in situations where the consultant remained in relative control of referral decisions and the protocols that govern them. The intermediary function also failed when GPs retained referral habits either by maintaining direct referrals to secondary care or by using the new GPwSI pathway to offload many cases for purposes of patient reassurance. The new function comes to fruition only when there is a steadily negotiated division of labour, recruitment strategy, case mix and physical spacing (and political truce) (Box 48).

Chapter 6 examined the theory that providing GPs with direct access to tests can enable GPs to distinguish between those patients who can be managed in primary care and those who require referral to secondary care. GP direct access to tests, designed to ‘rule out’ serious pathology, or ‘clinical indicator tests’, designed to identify where patients were in a disease trajectory, led to greater efficiencies in demand management than did GP direct access to tests designed to provide a differential diagnosis. This is because ‘rule-out’ tests and ‘clinical indicator tests’ provide a clear distinction between patients requiring specialist care and those who can be managed in primary care. In contrast, differential diagnostic tests often differentiate between different diagnoses that all require referral to secondary care. Furthermore, there is greater consensus between consultants and GPs regarding the timing and function of rule-out tests and clinical indicator tests in the patient pathway. The differential diagnostic tests that GPs are provided with direct access to are not the diagnostic tests of choice for specialists, who may utilise their clinical skills or other tests to reach a diagnosis. However, the improvements in patient flow resulting from GP direct access to rule-out or clinical indicator tests are realised only when there is clear guidance from specialists indicating which patients should be referred and how they should be subsequently managed (Box 49).

Chapter 7 examined the role of guidelines in moderating demand, the evidence revealing, repeatedly, an elementary design principle that they work more effectively if they are adapted to local circumstances. The process of adaptation, however, requires far more than the rewriting of guidance to a local rubric. The guidelines need to be mapped onto local organisational structures. A division of labour should be agreed on which team members are responsible for which aspect of the guidance. The people responsible for adapting the guidelines should be the same people who organise the implementation of the new guidelines. Without this level of accommodation the documentation is likely to be added to ‘the guideline mountain in the corner of the clinic’ (Box 50). 134

Group model building

This coda provides a modest recommendation for a tool that may be of assistance in future demand management interventions. The idea follows logically from all of the previous analyses, though it was not formally part of our investigations.

There is a remorseless order to all the ventures we have reviewed attempting to manage demand and rationalise referrals in health-care services. Regardless of the point of origin – be it a new guideline or a fresh post or a novel procedure or a management remodelling or a government imperative – the change reverberates around the whole service. The system is thrown out of equilibrium, with both intended and unintended outcomes. Often the positive and the negative occur simultaneously, with the gains in speed and accuracy of referrals accruing in one patient pathway being matched by overload and imprecision along another. What then happens, post intervention, is that key personnel respond to the resultant of the various intended and unintended transitions. In other words there is always an adaptive and gradual response: the new guideline is ignored until it is embedded in job descriptions; the new role becomes low-grade support until local political conciliation is achieved; the new centre is bypassed until its providers and recipients become involved in its management. These adaptations are endless. Sometimes they occur so distantly from the original initiative that it is hard to envisage them as part of programme implementation. Nevertheless, we have evidence aplenty to show that improvement only occurs thanks to this perpetual process of ‘snagging’ and ‘desnagging’ across the entire referral system.

How can this insight be translated into a way of building programmes whose fate is always determined in the wider health-care system? One starting point is to revert to a ‘system change’ perspective in evaluating any intervention. The idea is neatly encapsulated in a quotation from a recent paper on ‘complex responsive processes’. 288 This is defined as ‘a perspective in which organisations are regarded as patterns of conversations between independent individuals . . . Learning can be understood as the qualitative change in conversational patterns’. Although we might prefer to label the active ingredients here as ‘groups’ and ‘negotiations’, the point holds that it is the interaction between different sectors and segments of heath service delivery that seals the fate of a particular change initiative.

This insight has been much seized on. Occasionally it has involved throwing out the baby with the bathwater. The recent NHS Energy for Change initiative recognises, quite correctly, that many change initiatives simply fade away and that they work in solo or solitary settings, and they meet with determined defiance attempts to spread and sustain improvement in quality. The system-level solution in this consultancy report is to create a widespread change culture, which is to be fed by the development of ‘organisational energy’. This goal, in turn, is met by applying a fivefold framework that will nurture such energy for change across the entire system. The development plan is to measure and then generate more energy across its five domains of social, spiritual, psychological, physical and intellectual activity. 289

Although we have encountered much fading away and plenty of stubborn resistance in our review, we have no evidence pointing to the fact that energy depletion is the core of the problem. The demand management problem in every case is a practical issue – disagreement over the implementation of an intervention that is supposed to change the way a service is run. In all the would-be examples of demand control it is the fate of specific ideas (programme theories) that enter the boiling pot of negotiation. And insofar as they emerge constructively from those negotiations it is because the underlying idea has been transformed progressively into something that is more workable. System-wide solutions are about concrete ideas and practices and not abstractions such as energy diffusion.

The real system problem that we have encountered, over and again, is well summarised in a paper recognising that large institutions contain networks of semi-independent organisations. 290 The paper argues that the interactions and interdependencies within these networks are so complex that it is virtually impossible to control and design them centrally. There is no single command and control structure, with the result that management imperatives always generate intended and unintended effects as they are negotiated and renegotiated. Top-down change most often resolves itself into bottom-up, emergent solutions. This pen-picture is a perfect summary of the predicament of all the demand and referral initiatives we have reviewed.

It should be emphasised that such analysis is not a defeatist cry. It acknowledges that solutions, gains and positive outcomes can follow major interventions but recognises that those solutions are adaptive and local. Our evidence is particularly pertinent to this final adjective. Unintended consequences may well flow because of local funding problems, condition-specific patient pathways and context-specific peculiarities in line management and, indeed, due to that particularly awkward individual who finds himself landed with unwelcome responsibilities. Negotiated solutions follow a similarly tortuous and provincial path.

Accurate as is the picture painted by this scenario, it has a wait-and-see quality that exponents of evidence-based policy may find disappointing. Is there any way of speeding and strengthening the emerging solutions? Is there any way of avoiding false trails and cul-de-sacs? One emerging and increasingly celebrated strategy for harnessing and quickening these emerging, bottom-up solutions goes by the name of ‘group mode building’,200 and we close the report with a brief example and a preliminary assessment.

Group model building, having found much use in the business sector, has more recently been applied in health-care settings. 291 Reminiscent of the initiatives reviewed in this study, Pieters and colleagues291 study the redesign of the care process in the attempt to produce ‘more efficient and patient-friendly flows’. The system in question, the Dutch obstetric service, is characterised as having a large number of semiautonomous professional stakeholders, harnessed together in diverse and ever-changing constellations. The system operates to a basic model, which differentiates pregnant women into three risk groups – each one cared for in different locations and by different groups of professionals. As in the systems we have studied, the responsible networks are numerous (radiographers, laboratory technicians, general nurses, gynaecologists, assistant gynaecologists, midwifes, obstetric nurses, maternity nurses and paediatricians) and their functions are spread over a considerable period (antenatal, intrapartum and postpartum care).

Relatively poor perinatal and maternal mortality rates, together with a perceived lack of co-ordination, co-operation and exchange across these stakeholders, drove the need for reform. This led to a voluntary, exploratory excise, conducted in the action research tradition. The group model building element has the task of creating the ‘bottom-up emergence of successful network relations’ to carry forward the reforms. Simply described, it consists of an iterative succession of workshops, conducted within and between the various professional groups listed above. Over time the focus of discussion ranges from problem analysis, to defining potential improvements, to trouble-shooting those improvements, and then to the first stages of implementing the emerging schema. Discussion centres around the production of ‘process maps’ incorporating the roles and responsibilities of the various stakeholders (the complexity of such maps has been illustrated via several examples in earlier chapters). The process is orchestrated by a facilitator who leads, records, draws out models (literally) and acts as intermediary between the various groups.

The idea of the exercise, in realist terms, is to gain a sound understanding of the programme theory underlying a potential improvement – as seen through the ideas of each grouping. They are thus urged to ‘map’ interventions according to the following typical format (Box 51).

The succession of meetings begins to explore the consensus (or lack of it) between respective maps. Pieters and colleagues291 describe how a particular map was perceived as privileging the perspective of the gynaecologist and how it generated suspicion from other groups and how, in turn, the comparison with other maps led to potential compromises that could be further articulated and tested. The humble facilitator is charged with mapping the emerging maps (hence the term ‘group model building’). The process ends with a range of ‘improvement proposals’, the expectation being that, as they had been through this process of road testing (on the local byways to boot), the chances of successful implementation are enhanced.

Clearly such a process chimes with our findings about the emergent, local adaptation of demand management interventions. It is, of course, a paper-and-pencil simulation of a genuine implementation process. The rotation of workshops is a mere simulacrum of the real hide-and-seek of interdisciplinary working. Nevertheless, group model building has the potential to shift both from an ex-post to an ex-ante as well as from a top-down to bottom-up appraisal of service reorganisation. At its best it can anticipate some of the problems that that reveal themselves only after ambiguous interventions are launched on unsuspecting audiences (c.f. the ambivalent beginning of the GPwSI interventions).

And at its worst? There are caveats, of course. Group model building has its roots in management consultancy and the intricate, participatory research processes described above can be reduced to standardised, off-the-peg events. Some consultants use patented and expensive software tools and handbooks to facilitate workshops and map making (others such as Scriptapedia are available under a Creative Commons Licence). The other obvious consideration is that group model building, itself a form of intervention, will work only in certain hands and in certain circumstances and in certain respects. Much depends, for instance, on the skill and experience of the facilitator. Above all, the success of the entire exercise depends on the willingness of all stakeholders to collaborate. In this respect, group model building might be said to be a perfect imitation of the real thing. As we have seen in the body of the review, turf wars are a perpetual problem in service reorganisation and may end up as the block over which everything stumbles.

Given the need for delicate relationship building, there is a danger then that group model building becomes regarded as ‘more art than science’ and in this respect we note welcome efforts to evaluate and review the effectiveness of the approach. 292,293 It is of interest to note that they use a realist evaluation and a prototypical realist syntheses approach in making these assessments.

Our conclusions thus end with a ‘modest recommendation’. The main body of our review has explored many interventions mounted in the name of demand and referral management. The evidence has pointed to the many, many different nuances required in their implementation and the need for careful targeting to the appropriate stakeholders and institutional spaces. One of the elemental lessons learnt has been about the importance of local collaboration. Group model building offers considerable potential in realising this goal.

Recommendations for future research

Our synthesis has identified some of the contingencies that influence the success of demand management interventions. We concluded that demand management interventions require local, adaptive solutions in order to make them work. Implementation requires stakeholders to take account of the whole system within which any collection of interventions is placed. With this in mind, we caution against future research endeavours that attempt to identify preferred interventions which have the capacity to outperform all others. However, here we make a small number of recommendations about the possible approaches future research may take.

As we anticipated in our original proposal, the literature on demand management interventions is uneven; the guidelines literature is expansive and spans the full range of research methodologies, while the literature on RMCs and the peer review of referrals is much smaller and is often limited to case studies or observational studies. Although our synthesis has demonstrated that important lessons can be learned from this literature, more detailed mixed-methods studies comparing the implementation of different configurations of demand management interventions in different local contexts would be valuable. Such studies should also take a systems approach and consider how such interventions influence and are influenced by the entire system within which they are placed. Such studies should also recognise that, in the real world, demand management interventions are rarely implemented in the singular and consist of a configuration of different interventions (e.g. peer review, guidelines and triage). What is needed is a closer examination of the processes through which these configurations are adapted to local circumstances. This would enable a deeper exploration of the contexts within and mechanisms through which such interventions become integrated into local practice.

There is also scope for further evidence synthesis. We examined the fortunes of GPwSI services and whether or not they provided support of substitution for the work of consultants. The substitution theory is ubiquitous in health and social care and a realist synthesis to compare the fortunes of different practitioners placed at different professional boundaries (e.g. nurses/doctors, dentists/dental care practitioners, radiologists/radiographers and so on) would be valuable to identify the contexts and mechanisms through which substitution, support or short-circuit occurs. While developing our programme theories for RMCs, we also touched on the literature exploring whether rule-based systems embodied in protocols offer a support or substitution to clinical expertise, for example, such as those protocols followed by nurses working in NHS Direct or NHS 111 services. A realist synthesis of this literature would also be useful to explain the different outcome patterns observed in the extent to which these services are able to curb or redirect demand for primary and emergency care. We would argue that the systems theory developed in our review (see Chapter 3) offers a useful ‘reusable conceptual platform’11 to explore these questions.

Prompt 1: what (exactly) is the problem?

A problem well put is half solved.

John Dewey294

Faced with soaring demand and rising waiting lists, as well as all of the accompanying angst, you may consider that ‘the problem’ is obvious. It lies with nagging patients or officious clinicians or obdurate bureaucrats. Or, given the elephantine nature of modern health services, the underlying cause might seem more amorphous: the glitches are inherent in ‘the system’. Given, however, that a problem well understood is a problem well on the way to resolution, a close consideration of manifold causes of demand inflation is a key first step. The literature is replete with explanations. Which of the following apply in your bailiwick?

1. Irregularity of demand? Are queues unpredictable, with services sometimes over-run and at other times having spare capacity?

2. Professional closure? Is ‘turf protection’ a problem with opposition about referral and treatment priorities across different stakeholders?

3. Personalisation of referral decisions? Are clinical referral decisions sometimes clouded by the inevitable intrusion of social and emotional judgements?

4. Self-propelling diagnostic cascades? The first step in referral decisions is often the diagnostic test. These have multiplied in number and sophistication. They are often ‘nested’ – thus automatically propelling patients along the referral chain.

5. Supply-induced demand? New treatments and services roll constantly off the production line. It is generally in the physician’s interest to promote them and the manager’s job to provide access, thus creating additional demand that then may need to be checked.

6. Changing demographics? Populations evolve and their changing composition has a direct effect on demand. The so-called ‘demographic time-bomb’ of ageing is the best known example but there are numerous other bulges as populations and provision shift.

7. The internet and the informed patient. Patients’ expectations and medical knowledge have grown rapidly in the ‘information age’. Concomitant innovations in policy such as ‘shared decision-making’ and the ‘choice framework’ may also fan demand.

Some of these causes may simply not apply within medicine’s many specialities, but our experience is that most practitioners display an instant sense of recognition when presented with this list. Research shows, moreover, that these multiple drivers of demand often interlock and reinforce each other. In the design of any new demand and referral management schemes it is important to:

Think through the full array of underlying causes. Use the sevenfold typology above to prompt analysis of the particular bulge in demand that you may be confronting. The immediate challenge is likely to have deeper roots.

Prompt 2: what are the options?

My washing machine overwhelms me with its options and its sophistication.

Uma Thurman295

Health services are a mite more complex than washing machines, with the options for reform multiplying and evolving rather than being fixed and pre-programmed. It is, nevertheless, important to think though all of the purported and existing options mounted in the name of demand management as the basis for planning and preparing for any fresh challenge.

Research reveals the remarkable array of schemes, interventions and strategies mounted in the name of demand management. Box 52 gives a brief list of options:

Is it possible to digest the lessons from this little lot without, Thurman-like, being completely overwhelmed? It is useful first of all to think through the basic mechanics of the proposed reforms: what exactly is it that comes under change? Beginning with this simple question reveals quite different starting points. Ambitions vary from the macro to the micro: some interventions seek to introduce new centres to administer the entire referral apparatus; some introduce new roles and key personal with specific responsibility for managing demand; some set in place new procedures and guidelines in the hope of standardising referrals; and some offer incentives to improve to motivation of staff towards specific referral targets. There are powerful advocates of all of these measures.

Research reveals a vital lesson in relation to these varied ambitions, captured rather well in the aphorism ‘you can’t have one without the other’. Sustainable change requires co-ordinated action at the macro, meso and micro levels. Starting at the top, the evidence shows that organisational change is resisted unless new roles are properly negotiated, procedural modifications are agreed, and staffs are motivated to work under the new arrangements. Starting at the bottom, newly incentivised staff will be effective only if they operate with procedural elbow-room to accommodate their new resolve and targets; with the co-operation of colleagues elsewhere in the referral chain; and if there is harmony with co-existing organisational structures and strategies.

Such co-ordination is axiomatic in change management. Policy architects and managers are often driven to eliminate specific inefficiencies within a service. But, whatever the starting point, it is important to:

Think through the system-wide implications of any innovation. Change has no resting point. Change will always be resisted at some point. It is impossible to anticipate all of the consequence of any reform as it travels through the system but this is no reason to ignore the challenge. This starting point will instil the idea that successful interventions rely on distributed leadership and shared responsibilities.

Prompt 3: will reorganisation work?

New leaders who are intoxicated with the prospect of change further fuel perpetual cycles of redisorganization.

Andrew Oxman and colleagues296

Perhaps the most characteristic and most ambitious intervention in the demand management repertoire is the introduction of RMCs. These centralised triaging services or gateways often experienced a troubled introduction, and research shows that the ones that have stayed the course have undergone considerable adaptation from the prototypes.

The core idea behind the introduction of these centralised services was to rationalise the hitherto highly variable GP practices in making referrals. Having a professional service to review referral letters would prompt local GPs to think more carefully about whether or not the referral was necessary, ensure that they had undertaken all the relevant clinical tests prior to referral and consider whether or not the content of their referral letters contained all the necessary information. Without such systematic feedback, it was supposed that GPs would gain little insight into why referrals were redirected or deemed inappropriate and would not change their referral routines.

This menu of aspirations depended crucially on clinical engagement. In order to work successfully, the remit and operation of the RMC had to meet clinical as well as managerial needs. Research shows that this core requirement proved elusive. Several RMCs fell by the wayside. In the ones that prospered, sustainability was fostered by a prolonged and collaborative series of readjustments. In practical terms, clinical engagement was secured by thinking though the following:

• balancing financial ownership of RMCs by PCTs and clinicians, for example by setting up a CIC

• negotiating the precise percentages and type of referrals which should be directed to the RMC

• collaborating in the authorship of referral and triage guidelines and in the redesign of care pathways

• identifying those clinical leads who were able to secure the buy-in of local consultants and GPs

• cultivating and publicising the RMC via face-to-face meetings with local GPs

• negotiating clear agreements about the time scale under which triage would be undertaken

• installing robust tracking systems to ensure that referrals were not lost in the system.

The success of the intervention, in short, rests on a whole series of mini-interventions. Neither is it possible to treat the subsidiary interventions in a tick-box manner. Each one needs to be thought through. Research revealed another set of requirements to deal, for instance, with immediate capacity issues. For example:

• recruiting sufficient, locally respected GPs who were willing to take on the role of RMC triagers

• providing local GPs with direct access to some forms of clinical testing to ensure that all the relevant checks had been undertake prior to referral

• extending local expertise and auxiliary services – such as GPwSIs or Allied Health Professionals – to cover a more sophisticated set of referral routes.

Organisational change has relentless capacity to end as redisorganisation unless a cascade of consequences is thought through. If alternative management systems are provided without any positive attempts engage physicians in the making of the system, and without providing good reasons to change, and without evidence of demonstrable gains, the new layer of service risks creating supply-induced demand and increasing overall referral volumes.

Prompt 4: can intermediaries do the trick?

The go-between wears out a thousand sandals.

Japanese proverb297

Intermediate professional roles have multiplied throughout all health services. The impetus for such change lies in the bright idea of adjusting the workload between senior and junior DMs. Historic work routines often involve a rigid division of labour in which the ‘superior’ carries a number of tasks which do not really require specialist skills and expertise but are not passed on to an unqualified ‘subordinate’. A newly trained and accredited intermediary, the idea goes, can take over a proportion of the decision-making with benefits accruing on workload, value for money and consistency of judgement. One of the major attempts to implement this policy was the introduction of the role of GPwSIs who, in blending the expertise of the GP and the consultant, would improve the management of demand in the system by siphoning off and dealing with an intermediate case mix appropriate to their experience and proficiency.

Research shows that the realisation of this bright idea has met with mixed success, with some GPwSIs wandering awkwardly between pillar and post. Sometimes the new role descended into administrative support if consultants remained in overall control. Sometimes GPs retained referral habits and bypassed or misused the intermediary. Once again, the key to effectiveness is clinical engagement and the development of co-operative working, thus avoiding major territorial claims and power struggles. This, of course, is easier said than done. Research reveals yet again the slow, adaptive, sandal-destroying processes involved as partners have thought through barriers to integrated working. There is no one formula for perfecting the division of labour; case histories reveal many different routes, four of which are summarised here:

• Working and reworking the ‘case mix’. The precise area of expertise that comes with the ‘special interest’ has to be carefully negotiated. One method is for GPwSIs to graduate their responsibilities, starting with tightly defined ailments and symptoms and adding others incrementally.

• General practitioner with special interest posts were funded from quite different NHS posts. Recruitment ‘from within’ had the advantage of building on existing networks and of prior understanding of existing custom and practice (but also of its weaknesses).

• The new role depends on maintaining some equidistance from consultant and GPs, while at the same time being recognised and valued by the patient. Choosing the right location for GPwSI operations has proved significant.

• The GPwSIs’ clinical obligations become clarified via their training and management duties. Having a parallel educative function allows for the coproduction of protocols and guidance on how GPs can improve their referrals.

The introduction of intermediate roles depends for its success on thinking though every detail of the ‘job description’. GPwSIs were introduced enthusiastically on the back of a dozen vague ambitions and often ended up working in isolation, with insufficient clinical back-up. To meet a specific challenge like referral management their roles and responsibilities need to be negotiated, torn up and renegotiated to accommodate changing local contingencies and sensibilities.

Prompt 5: will direct access to tests results reduce excess referrals?

Don’t cut out the middleman – become one.

Entrepreneur Magazine298

The aim of providing GPs with direct access to diagnostic tests that were previously under the remit of consultants is to enable them to distinguish between patients who require management by a secondary care specialist and those who can be managed in primary care. It is a time-honoured idea that has met with mixed success in reducing referrals. Alas, not all tests are created equally and research shows that the precise function and timing of the test is crucial in determining whether or not GP direct access makes a difference.

Researchers have devised many classifications to differentiate the purposes of diagnostic testing. A simple distinction may be made between (1) rule-out tests that screen for serious disorders in the expectation that most patients receive a negative result, (2) differential diagnosis tests to distinguish between several potential conditions in order to initiate the appropriate management regime and (3) clinical indicator tests to measure the progress of known disease in order to refine its treatment or management. Research has shown, as a broad rule of thumb, that direct access to ‘differential diagnostic’ tests had limited impact on demand for specialist care, while direct access to ‘rule-out’ and ‘clinical indicator’ tests had greater capacity to reduce such demand. The immediate practical implication of this finding is that before negotiating direct access to a test it is important to think through carefully its precise function. Although useful, this is too crude a rule of thumb to operate unthinkingly, for there are winners and losers within each type.

Further considerations and expectations need to be managed, such as the following unanticipated consequences. Regardless of the type of tests, research revealed that a small proportion of patients end up with a secondary care consultation even if the results indicate its redundancy. Even in seemingly decisive rule-out tests, some patients pressed for, and some GPs acceded to, requests for further ‘expert’ reassurance. A further unintended outcome is supply-induced demand: widening access to tests may prompt unnecessary, ‘shotgun’ testing.

As with all of the other demand management schemes, a range of supporting practices enhances the value of direct access. Especially when GPs may lack the skills and experience to reach diagnostic conclusions (as in the differential diagnostic tests) and, more especially, when there is disagreement about the value and role of a test at a particular point in the patient pathway, managers should ensure such schemes are also accompanied by:

• clear referral guidelines to advise GPs which patients should be referred for the test

• guidance on how GPs should interpret the test results

• guidance on how patients should be subsequently managed, including patients who may not be reassured by the test results.

Test results, however clinically and statistically precise, do not speak for themselves. It is important to think through how they will be interpreted under direct-access schemes, given the multitude of further pressures on the GP. Supplementary, actionable, bipartisan guidance need to be provided on who to test, what the test does and what follows from the results.

Prompt 6: will guidelines be followed?

We have all a better guide in ourselves, if we would attend to it, than any other person can be.

Jane Austin299

There are clinical guidelines for every medical condition and for every point in the treatment pathway and, thus, no shortage of formal, professionally endorsed advice on when and how to make referrals. Research on the impact of guidelines on practice comes to a crushing and consensual conclusion, namely that the ‘passive’ dissemination of guidelines has failed as a tool for reducing demand. However thoroughly researched and immaculately presented, guidelines will be left on the shelf unless there are further ‘active’ steps to ensure that the advice is followed.

Research has also made useful strides in determining what does and what does not contribute to active utilisation. Guideline uptake is not enhanced if the documentation is enlarged, as it very often is these days, to offer suggestions on overcoming the potential ‘barriers’ to guideline implementation. Paper advice remains paper advice, however thorough and exhaustive. Also ineffective is the much-used process of ‘adapting’ guidelines – by having local expert panels rewrite the advice of national expert panels to local conditions. Paper advice remains paper advice.

Effective guidelines obey Austin’s maxim and depend on practitioners locating useful rules within their own practice. Broadly speaking, guidelines are followed if they are constructed by the same personnel responsible for implementing them. The micro-logistics required to enact a guideline need to be carefully specified and mapped step by step into local organisational structures. Managers need to ask who precisely is responsible for which referral decisions, who is responsible for responding to that decision, and so on. The guidelines then become tailored to the needs and responsibilities of each post holder in the decision-making tree. Even then, the guidance is not issued by decree. Research shows that the real implications of each element of a guideline becomes fully intelligible and actionable only if they are negotiated with each post holder.

Think though the process of getting paper advice into practical decisions. Guidelines, by their very nature, are couched as generalities on the extent of the medical condition, on test sensitivity and so on. Practice responds to the workaday specifics of patient characteristics, available resources and so on. Guidance becomes effective only when translated from the page into an agreed action agenda across each member of the team responsible for a service.

Prompt 7: can productive change be accelerated?

All the real work is done in the rehearsal phase.

Donald Pleasence300

Research has shown, as recounted repeatedly above, that demand management schemes become effective in the longer term only after a period of reconsideration and adjustment. Initial plans often require auxiliary measures to bring them to fruition. A change at one point in a complex system always has implications at dozens of other junctures and these are hard to predict. These findings represent more than a plea for patience. Effective change management is not just a case of waiting for interventions to ‘bed down’. Some programmes will have fatal unintended consequences; others will simply be overwhelmed by adverse conditions elsewhere in the system. Is it possible to distinguish those ideas that have staying power?

There is hope. It resides, inevitably, within the theme of ‘thinking through’. One can think retrospectively (if only?) or concurrently (let’s try?) or prospectively (what if?). The final mode, under a notion sometimes called group model building, may help to identify and build interventions with stamina. The idea, simply said, is to put policy proposals through a series of rehearsals. Any demand management scheme is dependent for its success on the reactions of a whole range of different professional groups. Under group model building, the plan is to bring together representatives to unpick and challenge the underlying ideas behind a new proposal. The group, by turns, questions whether or not the assumed diagnosis of the problem is valid, it evaluates alternative solutions, it maps the roles and responsibilities of the various stakeholders in relation to emerging proposals and it trouble-shoots the surviving ideas to root out potential antagonisms and power plays. All ends, if things go well, in an adapted, bottom-up version of the original plans.

Things may not go well, of course. Group model building, like any other intervention, will only work in the right circumstances and will fail with the inadequate selection of stakeholders, with domineering panellists and if it is managed without foresight.

Think ahead in the planning of any demand management scheme by simulating its system-wide implications. This involves considerably more than putting out policy proposal to ‘consultation’. It involves a systematic series of ‘rehearsals’ – repeated, representative forums attempting to iron out the deep creases in the initial plans. Group model building can never imitate the rough and tumble of real interdisciplinary working but it can initiate and inspire an adaptive process that carries onwards into actual implementation.

Contributions of authors

Ray Pawson, Joanne Greenhalgh and Cathy Brennan conducted the searches, selected and reviewed the papers, conducted the synthesis and wrote the report.

Other contributions from the project team

Judy Wright advised on search strategies. Roberta Longo advised on economic theories underlying demand and capacity problems explored in the review. Laurence Wood chaired the patient group who reviewed and commented on our theories and synthesis. He also advised on stakeholder involvement and dissemination strategies. Lisa Kundi and Hugh Sturgess advised on our theories and reviewed and commented on our synthesis; they also advised on dissemination strategies. Liz Glidewell helped to conduct initial searches and contributed to identification of initial theories. Allan Best reviewed and commented on our report and contributed to discussion on systems thinking throughout the report and on group model building in the conclusions chapter.

Project team

Professor Ray Pawson, Professor of Social Research Methodology, School of Sociology and Social Policy, University of Leeds, Leeds, UK.

Dr Joanne Greenhalgh, Principal Research Fellow in Healthcare Evaluation, School of Sociology and Social Policy, University of Leeds, Leeds, UK.

Dr Cathy Brennan, Lecturer in Public Health, Leeds Institute of Health Sciences, University of Leeds, Leeds, UK.

Dr Roberta Longo, Lecturer in Health Economics, Leeds Institute of Health Sciences, University of Leeds, Leeds, UK.

Judy Wright, Senior Information Specialist, Leeds Institute of Health Sciences, University of Leeds, Leeds, UK.

Dr Elizabeth Glidewell, Lecturer in Primary Care, Leeds Institute of Health Sciences, University of Leeds, Leeds, UK.

Lisa Kundi, Locality Manager, NHS Leeds West Clinical Commissioning Group, Leeds, UK.

Laurence Wood, patient and public representative.

Dr Hugh Sturgess, Director, Pennine MSK Partnership Ltd, Oldham, UK.

Dr Allan Best, Managing Director, InSource, Centre for Clinical Epidemiology and Evaluation, University of British Columbia, Vancouver, BC, Canada.

We would also like to acknowledge contributions from Roshanak Mehdipanah, visiting PhD student from the Department of Experimental and Health Sciences, Universitat Pompeu Fabra, Barcelona, Spain, and Lucy Bailey, undergraduate research scholar, School of Sociology and Social Policy, University of Leeds, Leeds, UK.

Patient group

Mike Edwards.

Elieen Exeter.

Rosemary Hassaman.

Gill Riley.

Advisory group

Professor Jenny Hewison, Professor of the Psychology of Health Care, Leeds Institute of Health Sciences, University of Leeds, Leeds, UK.

Dr Geoff Wong, Clinical Senior Lecturer in Primary Healthcare, Queen Mary University London, London, UK.

Gordon Tollefson, patient and public representative.

Leaf Mobbs, (then) Head of Commissioning, NHS Leeds West Clinical Commissioning Group, Leeds, UK.

Publication

Pawson R, Greenhalgh J, Brennan C, Gildewell E. Do reviews of healthcare interventions teach us how to improve healthcare systems? Soc Sci Med 2014;114:129–37.

Data sharing statement

All of the data is in the original documents on which this report is based.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HS&DR programme or the Department of Health.

Search strategy for Chapter 4

Database: Ovid MEDLINE(R)

Date range searched: 1946 to week 3, November 2013.

Date of search: 25 November 2013.

1. (“referral management” adj3 (centre* or center* or gateway* or clinic or system)).tw. (15)

2. “Referral and Consultation”/ (51,778)

3. (“multi-disciplinary clinic*1” or “multidisciplinary clinic*1”).tw. (347)

4. 2 and 3 (32)

5. *“Referral and Consultation”/ (19,005)

6. exp *Primary Health Care/ or *general practice/ or *family practice/ (87,730)

7. (referral centre* or referral center*).tw. (11,992)

8. 5 and 6 and 7 (17)

9. 1 or 4 or 8 (64)

10. (“referral” adj3 (centre* or center* or gateway* or clinic or system)).tw. (15,221)

11. 5 and 6 and 10 (89)

12. “clinical assessment service*”.tw. (17)

13. “clinical assessment cent*”.tw. (2)

14. 12 or 13 (19)

15. “one stop service*”.tw. (22)

16. “single point of access”.tw. (19)

17. 15 or 16 (41)

18. (triage or referral*).tw. (72,934)

19. 17 and 18 (8)

20. (Triage and referral*).tw. (716)

21. 5 and 20 (120)

22. Triage/ (8072)

23. 5 and 22 (175)

24. 21 or 23 (226)

25. 14 or 19 or 24 (253)

26. 14 or 19 or 24 [final extended referral mgt centre search] (253)

This search was repeated in EMBASE and identified 565 citations.

FIGURE 34.

Flow chart of inclusion of studies into review (RMC).

Search strategy for Chapter 5

EMBASE

Database: EMBASE Classic + EMBASE

Date range searched: 1947 to 7 May 2014.

Date of search: 13 January 2014.

1. GPSI.tw. (16)

2. GP with a Special Interest.tw. (12)

3. Specialist GP.tw. (18)

4. (General Practitioner adj3 special interest).tw. (28)

5. 1 or 2 or 3 or 4 (72)

6. GPSI.tw. (16)

7. GPwSI.tw. (39)

8. 5 or 6 or 7 (101)

9. limit 8 to yr=“1902 - 2013” (99)

FIGURE 35.

Flow chart of inclusion of studies into review (GPwSI).

Search strategy and decision flow chart for direct access to tests, Chapter 6

Database: Ovid MEDLINE(R)

Date range searched: 1946 to week 3, November 2013.

Date of search: 13 January 2014.

1. exp *Primary Health Care/ or *general practice/ or *family practice/ (87,730)

2. “direct access”.tw. (1932)

3. 1 and 2 (86)

4. access.ti. (26,521)

5. 1 and 4 (838)

6. exp “diagnostic techniques and procedures”/ or diagnostic tests, routine/ (5,681,971)

7. 1 and 4 and 6 (93)

8. 3 or 7 (159)

FIGURE 36.

Flow chart of studies included in the review (direct access to tests).

Presentations