Theory and practical guidance for effective de-implementation of practices across health and care services: a realist synthesis

What is being de-implemented?

This attribute is the focus for change, directed either at behaviours (e.g. clinicians or managers) or more broadly at policy level, with both contexts often being targeted simultaneously. 19

Push or driving force

There are several reasons why the de-implementation of a process or procedure becomes prominent at certain times. New evidence can emerge that shows a better way of proceeding, there may be concern about the safety of a procedure or there can be the removal of incentives.

Goal of the action

This attribute describes the goal of the de-implementation strategy. For procedures with safety issues, the objective may be to withdraw the drug or procedure entirely. However, in other contexts, the procedure may be of value to a particular defined group, but not for the majority, and so the de-implementation approach is to reduce the number of people who do not derive benefit. In many cases, the low-value procedure could be replaced by a more effective alternative (i.e. a case of medical reversal that can in certain contexts be an effective stratagem). 77

Degree of intentionality

This attribute highlights the purposefulness of the de-implementation process. The strategy adopted for the abandonment of a procedure may be passive as clinicians increasingly adopt a newer intervention, but this process is not actively initiated or led, leading to the eventual stopping of the procedure. Other strategies may be planned and facilitated, such as the closure of a facility.

Consequences

The results of the de-implementation process may not have been as intended (e.g. the strategy may have had the objective of stopping the procedure, but the strategy achieved only a diminution in practice).

Case studies

Walker and Avant’s74 concept analysis process proceeds through the development of several cases that represent all the attributes of the concept (i.e. a model case, a borderline case that has some attribute and a contrary case that clearly does not contain attributes of the concept). Walker and Avant74 stipulate that the cases can be taken from real-life examples, literature or cases constructed by authors. For all three case examples in our concept analysis, we have drawn on evidence from the literature on de-implementation, rather than construct our own cases. We found that having examples drawn from evidence is a useful starting point for further refinement of the concept.

Model case

A model case is an example that articulates the attributes within the concept. Walker and Avant74 recommend that the model case should be a ‘pure case’. A study by Schondelmeyer et al. 78 serves as a model exemplar.

Borderline case

Borderline cases are examples where most, but not all, attributes are referenced. A study by Kost et al. 79 is outlined below.

Contrary cases

Contrary cases show minimal attributes that signify the concept, and we denote this through a paper by Chamberlain et al. 80 The paper80 tracks National Institute for Health and Care Excellence (NICE) guidelines for the overuse of caesarean sections in women with hepatitis, endometrial biopsies to investigate infertility and varicocele operations to treat infertility in men.

Empirical referents of de-implementation

According to Walker and Avant74, empirical referents are ‘categories of actual phenomena that demonstrates the occurrence of the concept itself’. The focus of most de-implementation strategies is clinician decision-making behaviours, usually to reduce habitual practices. However, other areas for de-implementation would rest at the organisational and policy levels. The driving force of de-implementation is an essential attribute, as it initiates and provides a rationale for de-implementation. The drivers may be on a continuum from strong to weak, as the force of negative evidence or patient safety would push disinvestment quicker than weaker evidence. The action attribute is again on a referent continuum from stopping to a reduction in the procedure. Substitution with a more effective procedure may result in greater de-implementation of the low-value procedure and different strategies maybe required for early and late adopters. As described in the model case, if the de-implementation strategy is planned and facilitated, then there is a greater chance of a reduction in the process or procedure.

Summary

The process of concept analysis enabled the extraction of the defining attributes of de-implementation within health and social care. Five critical attributes were noted as essential to understanding de-implementation. These were (1) the focus of change, (2) the reasons or push to pursue de-implementation, (3) the extent of de-implementation that is achievable, (4) the degree of planning involved and (5) the intended or unintended consequences. The five critical attributes provided a basis for initial theory development. Case studies were helpful to explicate how de-implementation functions in different contexts.

Initial programme theory

The culmination of the work undertaken in Phase 1 was the development of a framework (see Figure 1) that provides an initial explanation of the complexity of de-implementation, and the contextual conditions that underpin the process. This was subsequently used to bound theoretical development. The conceptual components of de-implementation in health care are:

• De-implementation = f (what is being de-implemented; push or driving force; nature of action; degree of intentionality), which results in intentional and unintentional consequences, over time, for different stakeholders.

Our initial perspective was that de-implementation is associated with five main variables that operate in time and context. The first variable is what is being de-implemented. This is the focus of change and can relate to one or more context areas, such as clinician behaviour change or organisational policy. The second variable is the driving force for de-implementation (e.g. lack of evidence of clinical effectiveness or cost-effectiveness, financial constraints or safety issues). The third variable of nature of the action is the continuum of the de-implementation process from complete termination to replacement with another innovation. The fourth variable is intentionality, that is, to what extent the de-implementation process is intentional and planned or passive and opportunistic. Finally, the outcomes of de-implementation can be both intentional and unintended. Through an iterative process of discussion among the project team and referencing back to the initial proposal literature, we developed elements of the initial programme theory. These were in the form of ‘if–then’ statements that linked different components of the initial programme theory. The development and prioritisation of these statements were regularly discussed with key stakeholders throughout Phase 1 via face-to-face meetings and telephone calls. The statements were then used to guide the search and synthesis of evidence from the literature.

Increasing attention to low-value practice

Precommitment is a professional’s statement of intent in which they make visible to others that they will behave in certain ways. Precommitment was generated and communicated in different ways, including the use of invitations to professionals to precommit to de-implementation,87,88 point-of-care reminders for those who did precommit88 and/or general declarations of professional intention (e.g. a poster with a signed photograph of the professional made visible to the patient). 87

A simple public declaration of physicians’ intention to de-implement was tested within outpatient primary care clinics in the USA. It focused on the inappropriate use of antibiotics for acute respiratory infections. 87 These declarations comprised a poster with a signed photograph of the physician made visible to all patients, and committed the physician to the reduction in inappropriate antibiotic prescribing. The intention to de-implement was elicited after a presentation on the study in a standard clinic meeting. If physicians participated in the study, they were informed that this would involve randomisation to display a poster of precommitment within their own patient examination rooms and that, regardless of randomisation, all their antibiotic prescribing data would be monitored. The study intervention ran over a 12-week period, commencing mid-February to accommodate one flu cycle. Baseline data summarising 10 months’ antibiotic prescribing practice were extracted from the electronic health record.

The study was small. A total of 14 physicians completed the study, with data available for 954 patient care episodes. The adjusted baseline rate of inappropriate antibiotic prescribing was similar for the intervention and control practices, at 43.5% and 42.8%, respectively. Nevertheless, at the end of the 12-week trial, the inappropriate prescribing pattern decreased to 33.7% for the intervention clinicians and increased to 52.7% for the controls. The data did not demonstrate any ‘workaround’ by physicians by selecting antibiotic-appropriate diagnosis codes. There was an increase in the reduction of inappropriate prescribing rates for each additional month that the poster was displayed, although this failed to reach statistical significance.

Kullgren et al. 88 incorporated precommitment into a more comprehensive professional behavioural de-implementation intervention that was evaluated in a 12-month stepped-wedge cluster randomised controlled trial (RCT) of 45 primary care physicians in six practices in the USA. The intervention was designed to reduce the incidence of orders of low-value interventions, and was designed to nudge decision-making away from the low-value practice by raising professionals’ awareness of evidence-based recommendations to de-implement very close to the decision point. The precommitment was in the form of an invitation to comply with recommendations for the management of three clinical problems: imaging in uncomplicated low back pain and headaches, and the prescription of antibiotics for acute sinusitis. Precommitment was obtained after a presentation about the de-implementation study. If clinicians did precommit, this was in the form of a signed statement after the initial study presentation. Medical administrative staff were provided with training to identify potential patients who fell into one of the three clinical problems and then applied a Post-it^® note (3M Cynthiana, KY, USA) on the patient’s documentation that was immediately visible to the physician during consultation. Training was in the form of a brief (10-minute) individual session, reinforced with a one-page summary of their role. These staff-applied point-of-care prompts were introduced to physicians, reminding them of their signed precommitment statement. Paper-based prompts were used as these were thought to minimise the risk of computer fatigue and to increase visibility. Prompts were attached to a patient education handout to support physicians’ attempts to reduce low-value interventions. Weekly e-mails with links to resources to improve communication with patients about low-value services were also sent to participating physicians.

Electronic health record data were used to measure the primary effect of the intervention (i.e. any reduction in physicians’ ordering of targeted low-value clinical interventions). Any potential for physicians to order different interventions rather than simply to cease ordering was also evaluated. The collection of electronic ordering data was supplemented by a physician survey that included attitudinal items about intervention components. A sample of physicians was also invited to participate in a telephone interview that explored how their practice was influenced (or not) through their precommitment and associated reminders.

Primary outcome analysis demonstrated no reduction in the incidence of low-value orders other than a small but statistically significant reduction in back pain imaging orders [–1.2%, 95% confidence interval (CI) –2.0% to –0.5%; p = 0.001]. A sensitivity analysis that reviewed clinical encounters with relevant clinical codes, rather than those encounters solely identified by the trained medical administrative staff, showed reductions in inappropriate orders for all three clinical problems during the intervention period. However, these effects were not sustained over a 3-month follow-up period. Reductions in the ordering of imaging for uncomplicated low back pain were mirrored by an increase in the percentage of visits with a potential alternative order of minimal value (1.9%, 95% CI 0.5% to 3.3%; p = 0.007).

Survey data demonstrated that most physicians (63.6%) found patient education handouts helpful, and precommitment itself was thought to be helpful by nearly half (47.7%) of those physicians who completed the survey. Fewer physicians felt that the precommitment reminders (31.8%) and other resources to support communication (20.4%) were helpful. Interview data suggested that the patient education handouts were most helpful in facilitating communication during the clinical encounter, even for ‘the more difficult patients where I would have been tempted not to follow guidelines’. 88

Although the qualitative data indicate that the intervention materials helped in the communication process with patients, the overall results of this study of precommitment as a means of de-implementing low-value interventions were disappointing. The authors conclude that the limited effect of precommitting to reducing the use of low-value treatments without alternative management strategies may be due to the feeling physicians have ‘. . . to do something’. 88

Other strategies identified in the synthesis to increase professionals’ awareness of low-value practices include both the use of suggested alternatives and accountable justification. Both strategies, along with feedback on performance relative to peers, were evaluated in the context of inappropriate antibiotic prescribing in 47 US primary care practices. 90 The ‘suggested alternatives’ intervention comprised an alert on the electronic health record that highlighted to physicians that a prescription of antibiotics would not be generally indicated for the particular diagnosis, and an alternative course of action was suggested. In addition to increasing attention to a low-value practice, the assumption was that ‘prescribers may infer that a suggested alternative ought to be considered, thus reducing the likelihood that an antibiotic would be prescribed’. 90 Although the accountable justification intervention similarly drew attention to the low-value practice, physicians were required to complete an additional activity through the electronic health record to make an inappropriate prescription for antibiotics. This activity was completion of a free-text justification of the treatment decision. Physicians were also informed that this justification would be visible with the patient’s notes as an ‘antibiotic justification note’. A 2 × 2 × 2 factorial design was used to test different combinations of no intervention with these two interventions and a peer comparison intervention across the 47 primary care practices. Findings showed no statistically significant reduction in prescribing rates for the ‘suggested alternatives’ intervention. However, the absolute difference for accountable justification was –18.1% (difference in differences –7.0%, 95% CI –9.1% to –2.9%). Increasing the sense of professional accountability was anticipated to trigger reputational concerns and a desire to act in line with recommendations within clinical guidelines.

Individuals’ concerns about professional reputation

In the Meeker et al. 90 study, independent, statistically significant reductions in inappropriate antibiotic prescribing were seen for both ‘accountable justification’ and ‘peer comparison’ interventions. As above, the requirement to provide a justification for sustaining a low-value, unrecommended intervention may have worked through triggering professionals’ concerns about their reputations in different ways (i.e. reminding clinicians of the low-value nature of the prescribed intervention, forcing an additional level of critical scrutiny of the decision itself and raising concerns about professional reputation through the documentation of an unadvised course of action).

However, ‘peer comparison’ was also shown to be effective. This monthly e-mail-based intervention provided participating physicians with a ranking (i.e. highest to lowest inappropriate prescribing rate). Those in the lowest decile of prescribers were informed that they were ‘top performers’, whereas others were informed that they were ‘not a top performer’. The absolute difference for peer comparison was –16.3% (difference in differences –5.2%, 95% CI –6.9% to –1.6%), with the authors90 pointing to the positive reinforcement provided to those who managed to successfully reduce inappropriate prescribing. Three studies89,91,92 of performance feedback that included an element of social comparison were identified, although in only one study91 the feedback alone was the focus of investigation. Social comparison feedback is a technique that ‘. . . provides participants with information on their own performance compared to their peers to correct misperceptions about norms that may drive overuse of some tests’. 91 Trumbo et al. 89 included comparative performance feedback alongside education and other supportive interventions. However, in a study by Sacarny et al. ,92 feedback to physicians included the statement that performance was being placed under official administrative scrutiny. Although the previous two studies88,90 focused on physicians’ ordering of investigations within inpatient settings, this study focused on ‘off-label’ prescribing of antipsychotic medication for older and disabled adults. All of these studies were conducted in the USA.

The de-implementation intervention investigated by Sacarny et al. 92 attempted to manipulate concerns about professionals’ reputations. The investigation was conducted across a study population of 5055 primary care physicians who were the highest prescribers of an off-label antipsychotic medication enrolled in the Medicare programme. The evaluation design was a placebo-controlled parallel-group investigation of comparative performance in the form of three peer comparison letters to each of 2528 physicians randomised to the experimental intervention. These letters indicated that individuals’ prescribing of antipsychotic drugs was high relative to their peers and was under review by Medicare. The letters were clearly worded to draw attention to inappropriate levels of prescribing and were sent several months apart over the 9-month intervention period. The control group of physicians (n = 2527) received ‘placebo’ letters about medicine regulations. A baseline cohort of patients were identified from those receiving the study medication from any study provider up to 1 year prior to the study, and these patients were classified as being either ‘guideline concordant’ or as having ‘low-value prescribing’ against prespecified clinical criteria.

The intervention demonstrated a decrease of 11.1% (95% CI 9.2% to 13.1%; p < 0.001) in existing and new prescribing practices compared with the control group who received ‘placebo’ letters. The intervention also produced a decrease of 27.1% in the volume of new antipsychotic drug prescriptions over 9 months (95% CI 23.1% to 31.1%; p < 0.001), which persisted over a 2-year period. The authors conclude that letters providing comparative performance data and indication of external surveillance to the ‘highest prescribing physicians in the Medicare programme led to statistically significant, persistent decreases’ in antipsychotic prescribing. 92 The use of both the ‘carrot’ and ‘stick’ approach to providing performance feedback was also attested to in our stakeholder engagement, as in the quotation below from a clinical director of cardiology during a telephone interview:

. . . so I met with the individual [consultants] first and then with their teams, them and their teams and the individual meetings were basically ‘look here’s your data, you’re losing the organisation £1,000,000 a year, you’re keeping people in too long, and more people are dying than they should’. My approach was ‘how can I help you to improve this’ and you know what I also said was ‘obviously if these things can’t improve I’m going to have to look at whether you should be supervising other junior doctors, I’m going to have to look at what you can and can’t do’.

Academic_M5

Adjusting views on normative performance

Ryskina et al. 91 used a single-blinded RCT to evaluate a simple social comparison performance feedback intervention. Social comparison feedback is a technique that ‘. . . provides participants with information on their own performance compared to their peers to correct misperceptions about norms that may drive overuse of some tests’. 91 The clinical focus of the study was in-hospital laboratory routine test ordering, which is long considered to be an overused practice in the USA. In Ryskina et al. ’s91 study, 114 physicians were randomly allocated to one of two groups: a control group that received no feedback or an experimental group that received a personalised e-mail of their routine laboratory ordering (including blood count, metabolic panel, liver function and common coagulation tests) compared with the average of all clinicians from the previous week’s test ordering. These individuals also had access to a continuously updated and personalised dashboard that provided the context to these data, including patient-level details. No statistically significant reduction in the experimental group physicians, in terms of the count of routine laboratory orders placed by each physician per patient-day, was identified, regardless of individual physicians’ pre-intervention ordering rates relative to their peers. However, there was a general reduction in laboratory testing. Ryskina et al. 91 surmised that there may have been ‘cross-contamination’ between the groups, but also that the intervention group did not access their review e-mails and personalised dashboards regularly. This lack of engagement may be because of the clinicians having to physically access the e-mails themselves, presenting an added barrier. However, if the information was automatically accessible without needing to access the data then the difficulty for the physician would be reduced.

In addition to comparative performance feedback, Trumbo et al. 89 provided additional support to enable staff to reduce unnecessary ordering of chest X-ray investigations within two critical care settings [i.e. cardiovascular intensive care units (CVICUs) and medical intensive care units (MICUs)] in one US hospital over a 9-month period. The intervention provided information about Choosing Wisely’s evidence of the consequences of daily routine chest X-rays through a face-to-face session with staff members at each unit, giving feedback on current chest X-ray ordering rates and providing a rationale for reducing routine ordering. ‘Peer champions’ were recruited from each unit following this session. Two advanced nurse practitioners from the CVICU and three physicians (residents) from the MICU volunteered as peer champions. Their role was to co-ordinate data responses and encourage colleagues to adhere to the recommendations, as well as providing weekly feedback on X-ray ordering. Improvement science methods, including Plan, Do, Study, Act cycles, were used to facilitate de-implementation. There were no control settings on the evaluation design. The two critical care units were compared against their pre-intervention X-ray ordering data. The CVICU achieved a small but statistically significant fall in X-ray ordering from a pre-intervention rate of 1.16 (interquartile range 1.06–1.28) chest X-rays per patient per day to a post-intervention rate of 1.07 (interquartile range 0.94–1.21; p < 0.001) chest X-rays per patient per day. The MICU, which had a pre-intervention rate lower than the CVICU, showed no change over a 33-week period. Quantitative data were augmented by observation and semistructured interviews with stakeholders. Reasons for differences between the intensive care units were related to the turnover of staff and reduced quality improvement initiatives in the MICU that showed no significant decrease in X-ray ordering. Trumbo et al. 89 reported ‘messaging about reducing unnecessary tests went well when framed at the unit level but may be counterproductive if used to question individual ordering decisions’. 89 Although it is unclear from the presented data why this conclusion was drawn, two clinical incidents (unrelated to chest X-rays) that occurred during the study time frame may have posed additional de-implementation challenges.

The provision of comparative performance feedback has been linked to both prompt reputational concerns about being a professional outlier and correct views on normative performance. This mechanism is extended in studies that focused on more supportive forms of peer and collegial mentoring. Here, there was an explicit focus on using peer support as a means for de-implementation,93 albeit as part of a complex array of education, training, printed information and ongoing external facilitation. This comprehensive strategy was evaluated in a cluster randomised trial93 covering 765 patients from 72 units in 32 Norwegian nursing homes. The purpose of the trial93 was to limit the use of inappropriate prescribing of antihypertensive medication. The primary outcome was the number of antihypertensive medications prescribed (with description defined as fewer drugs at month 4 than at baseline), together with patients’ systolic blood pressure and pulse measurements.

Findings showed that the medications were significantly reduced in the intervention group (n = 43, 32%) compared with control (n = 11, 10%). Blood pressure increased for the intervention patients from 128 ± 19.5 mmHg to 143 ± 25.5 mmHg at 4 months, but at 9 months this fell back to the pre-intervention average of 134 mmHg, with no effect on pulse. Gulla et al. 93 concluded that in long-term hospital care the introduction of planned medication reviews with collegial mentoring can reduce antihypertensive drug prescribing and potential side effects without increasing patient blood pressure in the longer term. However, following the termination of the study, the difference between the intervention and control units became negligible, with the authors emphasising that continued collegial support and education are essential for achieving sustained de-implementation.

Making it harder to sustain low-value treatment options

Since 2011, primary care physicians in Portugal have had to use an electronic health records system for ordering diagnostic and laboratory tests in primary care. Routine practice required clinicians to order tests either by searching for a specific test or by selecting from a shorter short-cut menu organised around major clinical domains (e.g. basic, cardiovascular, rheumatology, respiratory). Rather than select individual tests from the short-cut menu, physicians were also able to order the full complement of tests by selecting an ‘all’ function for that menu. Concerns that tests were variably over-requested with preventative intentions or to meet patients’ expectations prompted Martins et al. 94 to evaluate the impacts of amending the presentation of choice options within the health record system. In addition to colour coding the strength of recommendations as above, two additional amendments were evaluated:

1. the removal of some diagnostic tests from short-cut menus (including uric acid, serum protein electrophoresis, sedimentation rate, electrocardiogram and chest X-ray)

2. the addition of some diagnostic tests to short-cut menus (including high-density lipoprotein cholesterol).

The electronic health record software modifications were implemented in May 2012, and evaluated in a parallel randomised trial design over 8 months across the equivalent of nine primary health-care centres. We assessed the impact on ordering of (1) tests removed from the electronic health record, (2) tests added to the electronic health record, (3) tests coded with green dots (i.e. those recommended by USPSTF) and (4) tests coded with red dots (i.e. those not recommended by USPSTF). The most significant impact detected in the trial was for the ordering of tests other than electrocardiography, which had been removed from short-cut menus. When ‘low-value’ tests were removed from choice options, although the prescription rates of low-value diagnostic and laboratory tests in the control group were consistent, a reduction from 31.3 prescriptions before software modification to 13.9 prescriptions afterwards was observed. As the authors comment, the ‘results demonstrate the importance of testing and evaluating various aspects of medical informatics programs to improve efficiency and contribute to improved clinical practice and clinical outcomes’. 94

More subtle changes to information displays were evaluated by Soon et al.,95 albeit within a hypothesised practice scenario around the management of low-back pain in primary care. Soon et al. 95 showed that the withdrawal of options from clinicians’ choice menus can positively influence behaviour change and make sustaining low-value practice harder. However, amending the layout of options can, in theory, effect similar changes. The layout changes evaluated in this study included a partition display of treatment options related to the practice scenario. The partition display drew on the assumption that ‘people are less likely to select items listed horizontally than those listed vertically’. 95 In practical terms, all treatment options were open to clinicians, but those options that were considered low value were presented horizontally rather than vertically on a computer screen. These partition display changes were evaluated against a control, where all treatment options were listed horizontally, alongside a default option and a combination of default and partition display. The default option listed only recommended treatment options, together with a selection prompt for clinicians to use to see additional low-value treatment options.

These combinations of displays were evaluated in a 2 × 2 factorial randomised trial among primary care physicians (n = 120) in Australia. Participants were asked to complete four clinical vignettes of patients with low-back pain. Two vignettes linked the scenario with six imaging options, and the other two vignettes linked the scenario with six medication options. Within each scenario, participants were asked to select a minimum of one and maximum of three treatment options. Participants were randomised to the same condition (i.e. control, partition display, default display or partition display plus default) for all four vignettes. The primary outcome was the proportion of scenarios for which study participants opted for a minimum of one low-value option.

Although the partition display group showed no effect in reducing the selection of low-value treatments, the default display group that listed the high-value options reduced the odds of choosing a low-value option by 44% (odds ratio 0.56, 95% CI 0.37 to 0.85; p = 0.006). The partial display group failed to demonstrate a significant impact on low-value option selection independently or on combination with the default display group. This RCT showed the benefits of using electronic systems that make it harder for clinicians to access low-value options and easier to retrieve high-value tests, but in quite transactional ways. Put simply, presenting more effective treatment options and ‘hiding’ low-value options within information displays and treatment options appears to work. Importantly, Soon et al. 95 included an examination of participants’ propensity for critical reflection within decision-making and, on that basis, were able to suggest that ‘these findings suggest that [this approach] may be most effective in situations where thinking “slow” is difficult or discouraged’. 95

Increasing knowledge and skills around the evidence to drive de-implementation

In the Winchester et al. 97 study, an increase in knowledge or skills emerged through the process of persuading physicians and advanced practice providers to reduce inappropriate testing. Increasing knowledge through education was viewed as a low-cost approach, which ‘if applied properly, could help curtail the unnecessary use of medical test and procedures’. 97 Increasing knowledge was considered to change the behaviour of the service providers in this case, as the underlying issue was an identified lack of knowledge as opposed to fixed habits or false beliefs.

In the paper by Menegueti et al. ,99 increasing knowledge and skills via the training was enhanced by the introduction of a daily checklist to support decision-making:

Three days per week, the physician and nurse of the Infection Control Service reviewed the decisions with the clinical staff, and discussed any cases that did not conform to the indication criteria . . . In this phase, the addition of the checklist intended to facilitate the decision-making by the clinical team.

Menegueti et al. 99

In the other evidence included in this CMO, studies focused on de-implementation issues related to medication (i.e. polypharmacy and reduction in inappropriate or harmful medication). In the Gulla et al. 93 study, an increase in knowledge or skills relating to a decrease in the use of antihypertensive drugs in care home patients was assumed through the process of implementing a complex multimodal intervention. However, the authors reported that the effects were not long-lasting, highlighting the importance of sustainability:

. . . we expected that the effect of collegial mentoring, multidisciplinary collaboration between physicians and nursing staff, and education would last beyond the four-month period. However, our results demonstrated that the main effects [of collegial mentoring, collaboration between physicians and nursing staff, and education] were concentrated between baseline and month four, and that significant differences between the intervention and control units diminished over time. Sustained collaboration, education and learning are crucial prerequisites for person-centred care.

Gulla et al. 93

In the Juola et al. 100 study, although nurses’ knowledge about medications was not assessed prior to the intervention, the authors hypothesise that the comprehensiveness of the nurses’ prior education could have been a contributory factor towards intervention success:

Registered and practical nurses in Finland receive a comprehensive education (3.5 and 2.5 years’ duration, respectively). This education includes teaching about medications. This is different to the situation in some other countries where aged care wards are often staffed by aged car workers with lower levels of formal education.

Juola et al. 100

Therefore, attention should be drawn to increased knowledge and the potential of this to improve the capability of staff to de-implement.

Collegiate and peer support

We noted that in some of the evidence presented in CMO 3, the presence of collegiate peer support associated with the educational nature of the intervention. In the Winchester et al. 97 study, for example, materials (i.e. lecture delivery and poster displays distributed throughout the care facility) were designed so as to maximise collaborative peer support without practitioner shaming:

. . . the tone of the lectures and poster was purposefully one of collaboration. None of the materials focused on shaming or punishing providers for ordering rarely appropriate tests. We informed providers that the goal of the initiative was to work together to achieve the best for our patients.

Winchester et al. 97

In the Gulla et al. 93 paper, collegial mentoring forms part of the medication review component of the intervention:

The physician was in charge of all medical decision in the meeting; the nurse provided updated knowledge about the patients’ situation, and the researcher gave provided decision support by initiating discussions, giving support in difficult decisions and supplementing with relevant research on the field (collegial mentoring).

Gulla et al. 93

Being implemented ‘close’ to professionals’ practice

We additionally noted in this CMO that a pattern emerged with interventions designed to include an element of closeness to the practitioners’ own work or workplace. This was often in relation to the professionals’ subject areas, but also included other efforts to fit with the work area. In the Menegueti et al. 99 study, for example, during the training ‘the clinical team was presented with the rates of utilisation of indwelling urinary catheters and the number and rates of CAUTI [catheter-associated urinary tract infection] occurring in the past 6 months’. 99

An interactive educational intervention with actual prescriber patients was described in the Juola et al. 100 paper:

The sessions involved interactive discussion about the list of harmful medication and suitable alternatives. To demonstrate the relevance, nurses were encouraged to present and discuss actual resident cases from their own wards.

Juola et al. 100

Patient-centred discussion formed part of the educational programme in the COSMOS intervention described in the Gulla et al. 93 paper. In the Winchester et al. 97 study, the detail of the printed posters used suggests forethought to fit closely with the professionals’ work:

. . . size was chosen because the primary care clinics have official notice boards which accommodate this poster size. Posters were distributed, two for each clinical site, with the request that they be posted in the providers’ work areas for easy reference.

Winchester et al. 97

The authors conclude that tailoring the intervention to the targeted professionals in this way changed behaviours, as ‘prior data showed that the patterns of rarely appropriate testing differed by specialty. Using education tailored to the audience may have enhanced the effect’. 97

The notion of making education fit closely with the professional’s work in de-implementation activities, as supported by data, also resonated through the telephone interviews undertaken in Phase 3 of the synthesis:

I think a lot of people go back to personal experience and I think what is quite powerful is actually patient stories. I think patient stories when we take it down to the individual can actually be far more powerful than some of the very academic written evidence about some of this stuff.

Nurse and allied health professions clinical director B9

Feedback on individual performance

We noted in this CMO an attempt to test feedback on individual practitioner performance and its potential for enhancing de-implementation efforts. Bregnhøj et al. 98 included feedback on patient medication as one of the combined intervention constituents. GPs in the combined intervention group, following attending the educational meeting, received ‘written recommendation targeting medication problems identified while rating the Medication Appropriateness Index (MAI). Afterwards, the GPs were contacted by telephone by a senior clinical pharmacologist (JS) to discuss any uncertainties concerning the recommendations given’. 98 However, the authors point out that this approach can be more expensive and time-consuming than an educational approach alone.

Validating patients’ health concerns

Diagnostic worries on the part of patients was cited by Fenton et al. 101 and May et al. 104 as one factor associated with overuse of low-value tests in US primary care. Specifically, the use of magnetic resonance imaging for recent-onset uncomplicated low back pain, dual-energy X-ray absorptiometry screening for women aged < 65 years without osteoporosis risk factors and neuroimaging for recent-onset headache. The hypothesis that improving the patient-centredness of professionals’ communication with patients would reduce the need for ordering these investigations was tested by Fenton et al. 101 Although an evaluation failed to demonstrate any overall significant impact of this hypothesis, an observational analysis of this study completed by May et al. 104 provided evidence of communication styles associated with de-implementation.

Fenton et al. 101 designed a brief 20-minute intervention to improve patient-centredness, comprising two visits by standardised patient instructors to US primary care physicians, during which a six-step patient-centred communication strategy was utilised: (1) developing understanding of a patient’s concerns, (2) validating these concerns through empathy and normalisation, (3) providing reassurance when possible, (4) explaining that risks outweigh the benefits of tests, (5) flexibly negotiating alternatives and (6) exploring for residual concerns. Instructors used different techniques to support implementation of this communication strategy, including performing the role of patients requesting two of these low-value tests and feedback on physicians’ performance. The impact of this strategy was evaluated against a control intervention that comprised visits by instructors and the provision by e-mail of guidelines on these clinical topics. No attention appears to have been paid to broader aspects of patient-centredness within the primary care experience.

The patient-centred communication strategy was evaluated in a RCT of 61 physicians. The primary outcome was whether or not physicians ordered any of the three low-value tests in follow-up visits from other standardised patients trained to request the tests. Providing instruction on communication drawing on two of the three low-value tests provided an opportunity to investigate any intervention carry-over to the third test (i.e. neuroimaging for recent-onset headache). Up to three follow-up visits from standardised patients per physician participant over a 3- to 12-month period were planned. Seventy-eight visits for control physicians were completed; of the 77 completed visits for trained physicians, and in addition to a standardised measure of patient-centred communication, the fidelity of the patient-centred communication behaviours was assessed in audio-recordings in all but one of these visits. All participating physicians were asked 2–4 weeks after each visit whether or not they suspected that they had seen a standardised patient. Visits were suspected at similar rates by both groups of physicians. Physicians in the intervention group were also asked to rate the quality and relevance of the training. In summary, and despite a positive qualitative evaluation of the training itself, it was not associated with any difference in the odds of subsequent test ordering from standardised patients (adjusted odds ratio 1.07, 95% CI 0.49 to 2.32).

A post hoc observational analysis of this trial completed by May et al. 104 assessed the association between whether or not low-value tests were ordered during the visits by standardised patients, and five specific communication behaviours related to the strategy outlined above: (1) normalisation, (2) reassurance around specific features of the case, (3) explaining that the risks of testing outweigh the benefits, (4) suggesting that tests will be ordered if symptoms persist, and (5) suggesting evidence-based alternatives for residual concerns. The relationships between these behaviours and whether or not physicians subsequently ordered the requested low-value test were evaluated in a series of regression analyses.

Of these behaviours, the provision of reassurance and alternative evidence-based strategies were used most commonly (95.5% and 97.4%, respectively) and normalisation the least (6.5%), all irrespective of whether or not a low-value test was ordered. Only for risks outweighing benefits (57.1%) and watchful waiting (67.5%) were statistically significant differences observed. Low-value tests were ordered less often when watchful waiting (88% vs. 12%; p < 0.001) and risks outweighing benefits (65% vs. 37%; p = 0.002) were utilised by physicians. Subsequent adjusted analyses demonstrated the overwhelming contribution of watchful waiting communication techniques (marginal effect –38%, 95% CI –44.3% to –31.7%). Although this was a secondary analysis relying on the post hoc coding of patient–physician–consultant communication, the authors104 conclude that watchful waiting, in which patient worries are acknowledged and a future route towards testing is available if problems persist, is a simple and effective communication technique for reducing low-value testing in primary care. The authors104 suggest that this strategy is ‘supported by theories of patient autonomy and control’, and that reassurance alone might be seen by some patients as a passive rejection of their concerns.

However, the use of watchful waiting as a prospective intervention in its own right was a strong theme within the evidence of this synthesis. For example, Keijzers et al. 102 make a compelling case for deliberate clinical inertia, or ‘the art of doing nothing’, as a positive response to situations when there is uncertainty whether or not the benefits of intervention may be outweighed by harm. Keijzers et al. 102 argue that the appropriate context for deliberate clinical inertia is shared decision-making, reinforcing that even doing nothing is an active strategy requiring discussion, knowledge sharing and agreement between patients and professionals. Although patients may have some unrealistic expectations about the value of modern health care, professionals may similarly fear the risk of litigation associated, for example, with missing rare diagnoses. Therefore, knowledge of the strength of evidence, clinical experience and confidence, alongside transparent communication and a fundamental focus on patients’ best interests are all required. 102 Sun et al. 105 highlight that there can be variable requirements of the evidence for watchful waiting. Using the inappropriate use of antibiotics for acute otitis media in children as an exemplar case study, Sun et al. 105 highlighted in an observational analysis of 250 children that, where patients met guideline criteria for prescription of antibiotics for acute otitis media, then there was 100% compliance with a recommendation to prescribe. However, when guidelines indicated that watchful waiting was the recommendation, ‘current practice in this emergency department instituted watchful waiting only 4.9% of the time’. 105

The following framework for the implementation of clinical inertia is presented to reframe ‘doing nothing’ as ‘doing something’:105

• empathy and acknowledgement

• symptom management

• clinical observation

• education about the natural course of the condition

• managing expectations (in the context of patient values and concerns)

• shared decision-making.

This framework for watchful waiting has been developed and evaluated in a number of studies included in the evidence base for this synthesis, largely within the context of reducing low-value or ineffective prescribing practice. For example, Little et al. 106 undertook a randomised trial that involved 315 children aged between 6 months and 10 years who presented with otitis media. The trial was based around general practices in the south-west of England. The watchful waiting intervention was the delay in providing antibiotics until 72 hours post consultation. Importantly, parents could collect the antibiotics if they felt that their child was not improving. The control group received immediate antibiotic medication, and both groups also received written advice sheets appropriate to their group allocation. Results showed that symptoms generally resolved after 3 days, with the children who were prescribed with immediate antibiotics having shorter illness (–1.1 days, 95% CI –0.54 to –1.48 days), fewer nights disturbed (–0.72 days, 95% CI –0.30 to –1.13 days) and slightly less paracetamol consumption (–0.52 spoons/day, 95% CI –0.26 to –0.79 spoons/day). Parents of 36 of the 150 children in the watchful waiting group used the delayed antibiotic prescriptions, with 77% of the parents being very satisfied with the process. Fewer children in the delayed group had diarrhoea, and fewer parents believed in the effectiveness of antibiotics and in the need to see the doctor for future episodes of infection.

The approach to watchful waiting evaluated by Little et al. 106 did contain a safety net for both clinician and parents, in that the antibiotics could be collected at a later date if the child’s condition did not improve. This risk reduction approach to watchful waiting was also alluded to in the stakeholder telephone interviews:

. . . but many years ago I used to work elsewhere in general paediatric practice and I found if you take the 5 minutes to explain to the patient who’s expecting you to give antibiotics for the viral illness, why they don’t need antibiotics, most of them will get it, not all, but most . . . I think also how much you watch and wait will depend on the urgency of the situation but also the competency . . . do you know what I mean? I think, again there were parents that I would see with a child in an emergency room and I would think actually I trust these parents and I’m pretty sure this is a viral and we’ll do whatever we need to do to confirm that and we can give them a safety net and they can go home. You might meet a different set of parents where you think, do you know what I’m not sure they’ll be able to come with this, and maybe we need to help them in a different way. I think it’s a big ask isn’t it, certainly for GP colleagues who are at the front line, I just think and understand that their time is so limited.

Clinical medical director B6

In a US study, MacGeorge et al. 107 recruited, through a participant panel, 134 parents who had children aged ≤ 5 years and who had been advised to undertake a watchful waiting approach to treating their child’s otitis media. An online survey asked the parents to recall the interaction with the health professional, then coded for five aspects of content. The findings showed that up to 27% of the participants did not comply with the watchful waiting advice and immediately obtained the antibiotics for their child. The parents who adhered to the watchful waiting approach had been given greater explanation of the symptoms of otitis media, received instruction on how to monitor and respond to the child’s pain, and provided with a follow-up opportunity. The authors concluded that clinicians can improve patient adherence with a watchful waiting strategy through better education around antibiotic side effects, and reinforce the importance of a ‘safety net’ through advice on monitoring the condition with the option of follow-up contact, which validates patients’ concerns and provides reassurance.

In other clinical contexts, the use of watchful waiting as an effective strategy has been explored using a pragmatic randomised trial design, comparing observation of patients with acute, uncomplicated diverticulitis with the existing practice of administering intravenous antibiotics four times daily for 48 hours. 108 In total, 528 patients were included, with a median time to recovery of 14 days in the observational group and a median time to recovery of 12 days in the standard treatment group (p = 0.15; hazard ratio for functional recovery = 0.91; lower limit of one-sided 95% CI 0.78). Hospital stay was shorter for the watchful waiting group (2 days vs. 3 days; p = 0.006), but findings are limited to confirmed mild forms of diverticulitis. Pradhan et al. ,109 in an overview of systematic reviews and RCTs describing watchful waiting and brief education in the management of low back pain, highlighted the importance of patient communication and education through activity modification and reassurance. Their review, however, showed that for chronic low-back pain, watchful waiting may not be an effective strategy to employ because exacerbation of symptoms increases the likelihood of patient anxiety and depression. To examine the cost savings of a watchful waiting approach to acute otitis media, Sun et al. ’s105 cost–utility analysis of a retrospective sample of 250 children with acute otitis media attending emergency departments in the USA highlighted the potential additional costs of watchful waiting to patients and families. In this case, these costs could include parents’ time away from work, the costs of additional analgesia, and the logistics of future consultations and visits, if required, as part of the watchful waiting strategy.

The importance of health-care provider education about watchful waiting as an active clinical intervention with patients was attested to in a secondary analysis of a scenario-based intervention involving a woman with depression. Meredith et al. 110 examined factors that could increase primary care clinicians’ inclination to favour watchful waiting as a strategy to manage depression. Findings showed that the tendency to use watchful waiting as an approach among the clinicians was low, with only 34 (20%) reporting a high tendency to use watchful waiting. Clinicians were significantly more likely to choose such a strategy if they had greater psychological knowledge or thought that the need to treat the physical or medical condition was more important than treating the depression. Watchful waiting was less likely to be utilised if there was limited access to other mental health professionals. In our telephone interviews, we found that senior clinicians also employed the use of watchful waiting as a technique, but they acknowledged that it was a difficult decision to make:

Oh very very much so, I do see that [watchful waiting], I often and I see colleagues on ward rounds sometimes and the sensible thing to do is nothing . . . doctors, nurses you know all of us in the medical profession, the hardest thing we find is to do nothing.

Clinical medical director M7

The same clinician also commented that a substitution strategy, such as sending to another health-care practitioner, was not useful unless it was seen as effective:

. . . if we’re doing it [substitution] just because we want to give them something, what I mean is if I’m sending to physiotherapy, they don’t need rather than antibiotics they don’t need then I’m just wasting a different resource if you know what I mean. But I think yes there are probably other resources that we can take better advantage of, then you think some of it is probably communication amongst ourselves and patients and families as well.

Clinical medical director M7

Increasing patient involvement in their treatment through shared decision-making and incorporating patient preferences was highlighted by several stakeholders during telephone interviews:

. . . I believe the only way to support patients is to have clearly shared understanding and shared ownership of that risk to a greater extent as you can so that means being fully involved with the patient and shared decision-making, which means making sure that they’ve had the relevant information presented back to them in such a way that they can understand it and weigh it up. And we think we do that very well but very often we don’t for a variety of reasons, time being one of them, but I think if patients understand your process of clinical reasoning and are happy to go along with that thread then it’s much easier to jointly manage that clinical risk and you can reassure them that they don’t need the test or investigation.

Commissioner B10

The context of this CMO configuration recognises the important influence that patients’ expectations can have on the outcome of consultations, particularly in relation to low-value diagnostic testing and drug regimes in primary care. Across the evidence base for watchful waiting, there is consistent recognition of the importance of shared decision-making as the context for de-implementation in this way. However, some de-implementation interventions have focused primarily on a patient perspective. Tannenbaum et al. 111 undertook to reduce the use of benzodiazepines by older adults in Canada by direct patient education. Reflecting evidence of the positive impact of direct-to-consumer marketing by the pharmaceutical industry, their hypothesis was that direct-to-patient education could reduce the inappropriate consumption of benzodiazepines.

The hypothesis was tested in a cluster RCT in which community-dwelling adults in Canada, aged ≥ 65 years and currently receiving benzodiazepine therapy, were allocated to the educational intervention or to usual care. Clustering by the 30 community pharmacies involved was used to evaluate the potential for contamination between service users in the same pharmacy. The intervention itself was underpinned by a programme theory of patient empowerment, in which greater knowledge on the part of service users was expected to drive ‘shared decision-making to deprescribe’ (Tannenbaum et al. ). 111 The educational intervention included a self-assessment about the risks of benzodiazepine use; presentation of the evidence of harm; statements designed to create dissonance about safety; peer champion stories to raise perceptions of self-efficacy; suggestions about alternative strategies for dealing with insomnia/anxiety; and stepwise recommendations for tapering drug use. Complete cessation of benzodiazepine at 6 months post intervention was achieved in 27% of those receiving the educational intervention compared with 5% of those who did not (prevalence difference 23%, 95% CI 14% to 32%). Moreover, nearly two-thirds of all those in receipt of the educational intervention went on to initiate conversations with either their physician or pharmacist, and over half attempted discontinuation in some form.

The programme theory for this intervention was further evaluated in a realist evaluation of the trial data. 112 This provided an opportunity, through a mix of qualitative and quantitative analyses, to disentangle the contribution of potential mechanisms that reduced consumption: (1) increased motivation through greater knowledge and understanding, (2) building capacity to taper through increased self-efficacy or (3) the provision of opportunities to discuss consumption with a health-care provider. The analysis demonstrated that, within the context of reducing benzodiazepine use in older adults, de-implementation mechanisms that ‘target patient motivation and capacity . . . yield successful outcomes where healthcare providers are supportive, and patients do not have competing internal competing desires to remain on drug therapy’. 112

Letting go

Mitchel et al. 115 investigated the reaction of health service staff to the de-implementation of a patient-oriented service, that is the removal of weekend allied health staff, such as dietitians and physiotherapists, on several acute hospital wards. The original de-implementation trial was a two-stepped-wedge cluster RCT conducted between 2014 and 2015 among patients in two hospitals in Australia. The primary outcomes were length of patient stay, unplanned readmission and adverse patient events. Haines et al. 116 concluded that ‘. . . having no weekend allied health service on these wards was no worse than having the newly developed service that was brought back in, and indeed was superior for some outcomes’. 116 The Mitchel et al. 115 study used observation and focus groups with a total of 156 hospital staff before the commencement of trial 1 (i.e. the de-implementation phase). The resultant transcripts were thematically analysed and generated five broad themes: (1) immediate negative reaction with patient and staff consequences, (2) an excuse to save money, (3) incompatibility with health service values, (4) bending rules to circumvent the trial protocol and (5) disinvestment as opportunity. Mitchel et al. 115 used Kubler-Ross’s117 responses to the grief model to explain the reactions of staff to the forthcoming de-implementation, beginning with shock and denial, leading to anger and ending with acceptance. The authors point out that de-implementation directly threatens professional identity that impedes clinicians from questioning their own service provision. This qualitative research showed that the potential negative impact on patients of de-implementing a service, as perceived by health-care providers, needs to be explained to the staff involved. In addition, the research showed that the quality of the service to patients can be maintained and improved through the de-implementation of a service. As Mitchel et al. 115 conclude:

Staff saw the project as a threat to patient safety and quality of care and to meeting patient flow targets, resulting in conflict between the healthcare workers’ professional and organisational identities.

Mitchel et al. 115

Likewise, one consultant physician/dentist commented:

. . . I remember that being quite difficult a scenario to do so because it meant letting go of something which we’ve always done like this and either we haven’t had any major complications, problems or side effects or downfalls to it, but never the less other people did and so there was a difficult process.

Consultant physician/dentist M9

Meaningful engagement

In a narrative review, Day et al. ,2 in examining challenges facing the Choosing Wisely campaign in Australia, note that there is little evidence to show how disinvestment should be conducted. However, the authors2 conclude that if disinvestment is to work then transparent engagement with the clinical community around quality of care and patient safety is essential for success. In relation to the emotional challenges facing clinicians, our interviews show the pressure of maintaining a de-implementation strategy:

. . . if one has had a bad experience [of de-implementation] this is going to scar one quite significantly and the challenges of future cases are going to be potentially if one has had a bad experience.

Consultant physician/dentist M10

Within the context of health technology reassessment, in its broadest meaning, Soril et al. 118 conducted an evidence review of systematic reviews, followed by an expert stakeholder consultation. Their aim was to illustrate concepts that optimise the health technology reassessment process. The review stage synthesised systematic reviews that described or evaluated frameworks or experiences with health technology reassessment. Identified reviews were then appraised to identify primary articles for the analysis. Synthesis of the evidence identified the significant factors involved in health technology reassessment, and a diagram representing these factors was created. This schematic of ‘conceptual groupings’ was then presented to stakeholders, who were made up of clinicians, academics and health managers (sample size not stipulated). Over a 1-month period, stakeholders debated the schematic developed from the literature, through online and face-to-face discussion with the authors, until consensus was reached. The final schematic that emerged outlined the de-implementation process and consisted of six key questions that planners should ask when considering reassessing or de-implementing a health technology.

1. What is the value of the technology?

2. What is the current utilisation gap?

3. What are the available tools and resources?

4. What are the levers for change?

5. What is the desired outcome(s)?

6. Who are the foundational actors?

The central question that underpins health technology reassessment was the identification and engagement of the people involved in the process, and the necessity to have ‘meaningful engagement’, with transparency and authenticity of process delivery with all stakeholders. This was also reflected in our telephone interviews:

. . . if I reflect on the ones we’ve done [de-implementation] in a positive way, it’s always about being honest from the outset and being clear with people and not being ambiguous because once you’re ambiguous and maybe not being truthful from the outset you lose the stakeholders that you need to progress things.

Nurse and allied health professions clinical director B7

Fear

As for engagement with public stakeholders, Daniels et al. 119 looked at how health-care leaders view public involvement in disinvestment decision-making, employing a Q study method and interviews. Fifty-five NHS managers, comprising medical (n = 16), allied health professional (n = 10), nursing leads (n = 8) and administrators (n = 21), agreed to participate in stage 1. Q methodology is a procedure in which statements are rank ordered for significance by individuals and are then sorted through factor analysis. Once this procedure was completed with the 55 NHS managers, 20 participants agreed to telephone interviews based on the Q-sort analysis. The Q-sort produced three broad factors: (1) those health managers who advocated public involvement in de-implementation, (2) managers who were cautious supporters and (3) managers designated the ‘freedom of choice group’, who supported public involvement but believed that the public should be free to participate or not. Through the follow-up interviews, it was apparent that all the groups were more cautious about public involvement in the disinvestment process. Daniels et al. 119 state that ‘. . . in part, respondent fears that public input might take an adversarial form’. The authors highlight the importance of recognising the emotional elements facing clinical managers who make de-implementation decisions when there is a pushback by the public to these decisions:

Participation is required that allows for the expression of discord and emotion alongside consensus-building and rational argument.

Daniels et al. 119

Likewise, one commissioner commented:

And the context of course, so it depends on the level of risk . . . it depends what the stakes are, so if the stakes are low, it’s easier to take the risk isn’t it, if the stakes are high . . . they are unlikely to tolerate that risk for obvious reasons.

Commissioner B10

Information-giving/engagement

In an Australian study, Hodgetts et al. 120 used a co-design approach and deliberative engagement with all stakeholders involved in assisted reproductive technology (ART) treatment. A systematic review showed that ART effectiveness decreases with advancing maternal age, particularly after the age of 42 years, and therefore becomes a low-value treatment. With the findings from the review, stakeholder engagement sessions were instigated, consisting of three groups: (1) a consumer group, comprising patients and partners who had experienced ART; (2) a community forum group who represented the south Australian population more generally; and (3) a group of clinicians with experience in ART. The forums included ‘. . . the provision of accessible information by content experts, facilitation by an independent professional, small and larger group deliberation and the generation of policy guidance’. 120 Each stakeholder group attended two sessions several weeks apart. In the first session, each group considered the results of the review and from these discussions proposed a policy for ART funding in Australia. In the second session, each group received the outputs from the other groups, resulting in further deliberation. Each group’s discussions were transcribed and analysed through a thematic approach. Surprisingly, the consumer groups’ discussions did not show the self-interest of funding of ART that would have been expected. Instead, they recognised the limits on health spending. Similarly, the community group and clinicians also recognised that health provision should meet the needs of all patients, not just one sector, and ART should not be considered any different from other medical conditions. The findings indicate that a strategy of full engagement, including building the communication skills of all stakeholders, can be a way of reducing professionals’ concerns. The importance of engaging the wider stakeholders was also referred to by a health service commissioner whose quote supports the conclusions of Hodgetts et al. :120

They [clinicians] are influenced very much influenced by patient and family experience and patient voice that does help and I think hard facts behind it and showing improvement with regards to things that really speaks home . . .

Consultant physician/dentist M3

Relationships and leadership

In an opinion piece, Gupta and Moriates121 discuss the importance of leadership in creating a culture in which low-value practices are discouraged. The authors121 outline four steps these leaders need to take:

1. engage front-line providers, particularly through patient-centred rationale

2. create infrastructure to support improvements

3. use value-based incentives

4. be visible and show commitment.

In addition, Gupta and Moriates121 state that leaders should also build up an infrastructure that enables staff to communicate and feel confident in making their concerns known through the visibility of shared goals. These leadership qualities were also evident in a three-round Delphi study conducted by Robert et al. 20 Thirty international experts who had extensive experience and responsibility in decommissioning either through research or in practice contexts were recruited. Participants came from Australia (n = 6), Canada (n = 3), the UK (n = 20) and Ireland (n = 1). Their expertise included removal of treatments or prescribed drugs from practice, relocation and reconfiguration of services, and closure of services and departments. The Delphi method is used to ascertain a range of opinions on a particular topic, with the intent of producing a consensus around the topic in question. 122 The method uses a group of participants known as ‘panellists’ who are specially selected for their particular expertise on a topic. There is then a series of two or more sequential questionnaires known as ‘rounds’. Following each round, the researcher collates opinions, and redistributes potential consensus statements to the panellists until there is agreement or consensus around the issue. In the Robert et al. 20 study, three rounds were undertaken. In round 1, participants identified factors that they thought were involved in decommissioning. These responses were analysed through content analysis and formed into an 88-statement Likert scale for round 2. The Likert statements were computed for consensus and those statements receiving low scores were redistributed to participants in round 3, together with the overall results from round 2, with open-ended comments. Opinions from the final round of consensus meetings were grouped into three interlinked categories: (1) change management and implementation, (2) evidence and information and (3) relationships and political dimensions. The consensus from the expert panel showed how ‘messy’ the process of de-implementation is in reality,20 with political and financial decisions eclipsing professional concerns. The unsystematic approach to de-implementation can cause tensions for professionals, and conclusions drawn by Robert et al. 20 from the Delphi study included ‘. . . a strong leadership team, engaging clinical leaders from an early stage, and establishing a clear rationale for change’. 20 The importance of leadership and relationships was also endorsed by the following quotes from clinical directors:

But for bringing around change, you need skilled leaders that’s for sure, I guess that’s what this quality improvement work is like, certainly in our health trust and it’s a national thing but people who are trained to bring about quality improvement through staff engagement and de- . . . whatever it’s called . . . de-implementation is a quality improvement initiative isn’t it.

Clinical medical director B1

. . . you need the basis of good relationships, so before you can even get into being a mature team, because that’s cohesive, you’ve got to spend time on just having those relationships.

Nurse and allied health professions clinical director B7

A necessary and sufficient degree of alignment of multiple stakeholder perspectives

The de-implementation of radical mastectomy was associated with some degree of alignment of different factors, as it occurred ‘at a time when there was convergence in clinicians’ and patients’ awareness of the scientific evidence, patients expressing their preferences for an alternative, and when there were other viable treatments to offer instead of radical mastectomy. 17 Successful de-implementation appears to be dependent on the development of a sufficient degree of alignment between different perspectives, specifically those of relevant stakeholders. The evidence demonstrates a number of ways in which this alignment can be achieved.

The use of a multidisciplinary approach to de-prescribing in primary care settings was investigated in five primary care general practices in England, focusing on shared medicines management for older patients between the GPs and community pharmacists. 123 The Randomised Evaluation of Shared Prescribing for Elderly people in the Community over Time trial implemented an interrupted time series design ‘to estimate the effect of pharmaceutical care on the appropriateness of prescribing; patients’ knowledge, adherence, and quality of life; and the incidence of adverse events. 123 Pharmaceutical care was recommended as an alternative method for reviewing medication prescribed for older people to both improve health and reduce unnecessary hospital admissions, specifically by ‘improving communication between the pharmacist . . . the patient, and the GP’ (e11). Training consisted of theory and practice of pharmaceutical care, and practical exercises in collaboration with the GP. The study failed to demonstrate any statistically significant impact of the complex intervention, which the authors attributed to difficulties the pharmacists had in accessing full patient data and arranging meetings with the GPs:

Pharmacists were also typically unable to gather full and detailed clinical histories, which led to frustration on the part of some GPs when their prescribing was questioned.

e15

Richmond et al. 123 explain the non-significant result to the lack of a relationship between GP and pharmacist, suggesting that without this rapport it was difficult for the pharmacist to influence GP prescribing. However, Spinewine et al. 124 found that adding pharmaceutical care to the usual prescribing management programme in an inpatient hospital setting reduced the overuse, misuse and underuse of medicines for older patients. In a RCT of 203 Belgian patient participants aged ≥ 70 years, pharmaceutical care was provided from admission to discharge and delivered by a specialist clinical pharmacist who had direct contact with the medical team and patients. The appropriateness of prescribing on admission, at discharge and 3 months after discharge was measured, and patients in the intervention group were significantly more likely than control patients to have an improvement in underuse criteria from admission to discharge. This comprehensive intervention resulted in better medication management for the intervention group, as indicated by scores on the MAI, and in the Assessing Care of Vulnerable Elders underuse criteria from admission to discharge (odds ratio 9.1, 95% CI 4.2 to 21.6, and odds ratio 6.1, 95% CI 2.2 to 17.0, respectively).

We also found evidence data to show success of a multiprofessional approach to engagement within long-term care settings. 125 Here, a multidisciplinary approach in the form of the Multidisciplinary Multistep Medication Review (3MR) in nursing home settings was evaluated in a pragmatic clustered randomised trial. The experimental intervention consisted of an assessment of the resident’s perspective, medical history, medications review, and a meeting between the physician and the pharmacist, and implementation of medication changes. This Dutch study had 59 nursing home wards and allocated 16 care of the elderly physicians and 26 wards to the control group, and 16 care of the elderly physicians and 33 wards to the intervention. The 3MR intervention is a four-step process, in which the elderly care physician and pharmacist began by assessing each patient’s situation through a patient-completed questionnaire (with assistance if required) and completing a review of medical charts, including blood results. In step 2, the pharmacist reviewed the prescribed medication using the Screening Tool to Alert to Right Treatment/Screening Tool of Older Persons' Prescriptions (START/STOPP) and Beers Criteria. Step 3 was a multidisciplinary meeting, in which patient data were reviewed and decisions taken to withdrawing inappropriate medications, leading to the final step of implementation. Findings showed that, in comparison with usual care, the multicomponent intervention de-prescribed at least one drug for 39.1% of patients compared with 29.5% in the control group (adjusted relative risk 1.37, 95% CI 1.02 to 1.75). Commenting on their findings, Wouters et al. 125 indicated that the 3MR system needs to be repeated to coincide with changes in the patient’s condition, from prevention to palliative and comforting care. Additionally, the authors125 suggested that further research needs to be done to find better ways of incorporating patient preferences in the de-prescribing process.

The main reasons for the success of this intervention appear to be the structured approach employed and the direct contact between the multidisciplinary team. A structured approach was realised through a therapeutic plan, involving the patient, pharmacist and other professionals. In addition, high levels of pharmacist visibility were reported through the presence of the pharmacist on the unit 4 days per week (e.g. taking part in ward rounds), with access to medical records, and having direct contact with patients and their family carers. The pharmacist conducted a medication assessment on admission and provided a patient record and care plan. The intervention was also noted to contain a degree of opportunity-seeking, whereby the pharmacist discussed the opportunity to improve/optimise prescribing with the prescriber, who could accept or reject the intervention. Furthermore, the pharmacist answered all questions that health-care professionals asked about medications. At discharge, the pharmacist provided written and oral information on treatment changes to the patient or caregiver, as well as written information to the GP.

Other examples of the potential success of structured approaches using multidisciplinary teams across both hospital and community care were reported by Crotty et al. 126 Crotty et al. 126 suggested that multidisciplinary case conferences can increase appropriate prescribing and reduce the use of inappropriate medications in residential care. This clustered randomised trial evaluated the impact of multidisciplinary case conferences on the appropriateness of medications and on patient behaviours in Australian care settings for older people, with 154 participants having medication problems and/or challenging behaviours. The two case conferences were chaired by the GP, held at the residential home and consisted of a geriatrician, a pharmacist, residential care staff and a representative of the Alzheimer’s Association. The case conferences developed a multifaceted approach to the management of the patients, with the patient’s GP augmenting the discussion through the medical records, the Alzheimer’s Association representative contributing with advice on non-pharmaceutical interventions and the care home staff expanding on issues recorded in the patient case notes. Outcomes were measured with the MAI and patient behaviour assessed with the Nursing Home Behavior Problem Scale. Findings showed that medication appropriateness improved in the intervention group (MAI mean change 4.1, 95% CI 2.1 to 6.1) compared with the control group (MAI mean change 0.4, 95% CI –0.4 to 1.2; p < 0.001), with a significant reduction in the MAI score for use of benzodiazepines (mean change control –0.38, 95% CI –1.02 to 0.27, vs. mean change intervention 0.73, 95% CI 0.16 to 1.30; p = 0.017). However, there was no improvement in resident behaviour because of low levels of behavioural problems at baseline. Surprisingly, improvements in medication management were seen only in residents who were discussed in the case conferences; there was no carry-over to other residents in the nursing homes, demonstrating the importance of a co-ordinated approach where all members of the team are aware of the strategies for appropriate medication management.

This section discusses evidence from studies that look at more centralised multiservice health systems, rather than localised interventions. Spinewine et al. 124 suggested that the positive findings of their study were due to the whole-system approach, which involved the multidisciplinary team having high visibility of the pharmacist, who would have built up a relationship with the medical team. The importance of an effective multidisciplinary team for the management of the de-implementation process was reiterated by many participants in our telephone interviews, as this example testifies:

We are quite fortunate in a way that we work in a very close knit team and we all have . . . we come from different backgrounds and we have a very big collaborative in terms of also debating things quite intensely and yes we’re debating it and we’re looking at it but it requires a certain openness and trust of people to actually be able to do it.

Consultant physician/dentist M9

In a cross-sectional survey of eight community services in northern England, Gray et al. 127 assessed the variation in treatment and management of complex wounds. Data from 3179 patients with a total of 5632 complex wounds were evaluated. The data showed a substantial variation in the use of antimicrobial dressings, ranging from 18% to 69%. The authors’ recommend that a systematic approach should be instigated, bringing together clinicians, managers and policy-makers to identify de-implementation priorities. Similarly, Montini and Graham17 conclude their historical analysis of a de-implementation challenge with the view that to ‘. . . facilitate de-implementation . . . intervening in the broader context in which clinicians work – the social, political, and economic realms – rather than focusing on healthcare professionals’ behaviour, may indeed be a fruitful approach to effect change’. 17 Peiro,128 in a narrative review, discusses the complexity of de-implementation across different country contexts. The starting point for this review is the international challenge of disinvestment in low-value practices, with the reluctance of some national contexts to address the controversies associated with that disinvestment. There may be disagreements across different stakeholders as to the evidence for many low-value treatments, together with underlying psychological and sociological barriers that can undermine de-implementation at a much broader level. A plea to build ‘internally sustainable [health] systems’ requires a whole-system approach that is transparent. 128 The paper128 argues for an organisational approach to de-implementation through building trust between stakeholders, and highlights that there may be a small proportion of patients who require treatments that are considered ineffective for the majority.

An exemplar organisational framework to guide de-implementation is described by Parchman et al. 129 Parchman et al. 129 argue for behavioural, organisational and cultural change as requirements to overcoming the challenges of de-implementing low-value care. The framework, which is primarily focused on the challenges of overuse, describes four necessary conditions for de-implementation:

. . . addressing low-value care; build a culture of trust, innovation and improvement; establish shared language and purpose; and commit resources to measurements.

Parchman et al. 129

Pivotal to their framework is the use of ‘sense-making conversations’ where different stakeholders, in this case professionals, teams and groups across health-care services and organisations, can achieve alignment between the potential for harm associated with overuse of low-value interventions and understandings of the scale of the overuse challenge in stakeholders’ daily practice. Alignment in this sense is associated with ‘shared norms of behaviour, shared language, and a shared vision for co-ordinated action around reducing use’. 129 Although essentially a narrative text located at the interface of patient and professional dialogue around decision-making, Parchman et al. 129 appear to reject the concept of ‘value’ in favour of ‘harm’ as the focus for alignment around a common language. The authors129 argue that harm, rather than value, may have more resonance with patients and professionals alike, and that harm can be used to describe de-implementation challenges at a population level. For example, the consequence of overuse of services is reduced access for those patients who may, indeed, benefit from them.

The transferability of economic approaches to the allocation of health resources to de-implementation challenges, such as disinvesting in services, specifically the programme budgeting and marginal analysis method, has been considered. 130 As with other approaches to de-implementation, although there may be a shared goal across stakeholders, such as improving population health, differences in organisational objectives may still minimise the success of specific methods, such as programme budgeting and marginal analysis. Success in these terms related principally from moving to agreement that ‘something needs to be done’, to instrumental changes to service configurations and availability. Mortimer130 argues for greater specification in economic dimensions of de-implementation as a key requirement, including being specific about required budgetary contexts of de-implementation [such as any financial constraints to be delivered, the scope of budgets relevant to the de-implementation initiative and the (lack of) potential for financial recovery], and greater attention to the constituency and ‘democratic’ accountability of stakeholder representation. Budgets should have specific investment proposals linked to de-implementation proposals and an orientation towards ‘shift-lists’, rather than ‘wish-lists’, when the business case is made for de-implementation. Project strategy groups should include stakeholder representatives, but also have advisors who have experience of, and are supportive of, de-implementation. There is a concern that stakeholder engagement in de-implementation is often weighted against disinvestment and towards the status quo, hampered by stakeholders’ specific interests and lack of access to specific knowledge required for success (e.g. of the evidence base).

The use of financial resources in de-implementation, as highlighted by Mortimer,130 was also raised in data collected through telephone interviews, especially linked to patient or service improvements directly relevant to the health-care team:

I would say financial incentives to consultants aren’t necessarily the most viable, I think actually if you say that you can have something back for your services, it’s probably more attractive to a lot of consultants. Some are going to be kind of guided by financial rewards themselves but actually a lot of them will say, ‘if you can have something extra for your service or an extra member of staff or that kind of thing it would actually change practice because actually that would be a win for most of us’.

Consultant physician/dentist B5

The importance of a whole-system approach to directing de-implementation in a health service is also recommended by Hollingworth et al. 131 Hollingworth et al. 131 adopted a multimethod approach that included a rapid systematic review; use of Hospital Episode Statistics; non-participant observation of two primary care trusts; Clinical Commissioning Group meetings, including document analysis; and a case study of de-implementation of carpal tunnel release surgery. The authors131 found a large geographical variation for many procedures, with Clinical Commissioning Group meetings dominated by investment strategies, rather than de-implementation. The authors131 recommend an increased role for central NHS support to encourage commissioners to think about de-implementation by providing relevant evidence to facilitate decision-making. However, Montini and Graham17 argue that historical, economic, professional and social factors can hinder de-implementation efforts. Therefore, if a programme or practice has been in place for a long time and is well integrated into public health systems, it may be highly resistant to de-implementation and may include financial and professional interests that are dependent on the service. It may be difficult to de-implement and replace some low-value programmes without a change in the social milieu surrounding them.