Expectations that publicly funded health research should be productive, in terms of both research publication outputs and contributions to better health outcomes, are becoming increasingly explicit.1,2 This has directed attention to methods for tracking research outputs, where scholarly publication metrics — impact factors and citations — are currently the dominant indices.2,3 Publication of research is expected to disseminate new knowledge and facilitate “real-world” policy and practice impacts.

While the recent emphasis on research productivity spans all types of research,1,4–6 intervention research is particularly relevant, as its findings are likely to be more directly applicable to health policy and practice.7–9 Intervention studies tend to be less prevalent in peer-reviewed journals than descriptive and epidemiological studies, and this has been partly attributed to the practical and scientific challenges of conducting intervention research.9–11

Few studies have empirically investigated the implementation and outputs of health intervention research. As part of a project on the impact of a sample of intervention research funded by the National Health and Medical Research Council (NHMRC), here, we examine the research outputs of these grants. Specifically, we report:

• the descriptive profile of NHMRC-funded intervention research in terms of topics, settings, funding terms, and stages of development of the interventions; and
• whether and how statistically significant intervention effects on primary outcome variables influenced research productivity.

Methods

Data were collected between 23 July 2012 and 10 December 2013 on studies funded by the NHMRC between 1 January 2003 and 31 December 2007. Studies were eligible if they fitted our definition of health intervention research, which was: “any form of trial or evaluation of a service, program or strategy aimed at disease, injury or mental illness prevention, health promotion or psychological intervention, conducted with general or special populations, or in clinical or institutional settings”. Clinical trials of potentially prescribable drugs, vaccines and diagnostic tests were excluded.

Eligibility was assessed by two coders who reviewed titles, application abstracts, end-of-project reports to the NHMRC and publications arising from the grant. The 5-year period was selected to allow enough time for completion of the research and publication of the findings, balanced against limiting recall bias about studies completed too long ago.

Descriptive profile

Basic information on sample grants was collated, including the duration of funding and the topic of the intervention. The studies were classified according to “stage of intervention development” based on definitions from a previously published guide, distinguishing controlled interventions (efficacy), those carried out in real-life conditions (effectiveness), those that were replication or adaptation studies in different settings, or dissemination studies.12 Additional information was gathered from online surveys of chief investigators. A full description of the data collection process and response rates is provided elsewhere.13

Bibliometric analysis

To collect consistent information, we reviewed all publications submitted by chief investigators, and conducted literature searches (in Web of Science and Google Scholar databases) for the years following the commencement of each completed grant. Publications were reviewed to check if they were related to the grant in question. Key search terms included chief investigators’ names, grant numbers, project titles, intervention descriptions and relevant health issues. In the case of grants for which publications of study results could not be found, we attempted to contact chief investigators, including previous non-responders.

Assessing published results of intervention research

Two assessors reviewed publications that reported results of interventions to identify whether there were any statistically significant changes to the primary outcomes proposed in the research application summary. Where there was any uncertainty, decisions about what constituted primary outcomes were checked by other authors in a panel process. We classified interventions as: (i) those that showed statistically significant effects on primary outcomes; (ii) those with “mixed” results (eg, significant changes for some but not all primary outcomes), or if unintended or secondary outcomes were emphasised; and (iii) those that found no statistically significant effects.

Ethics approval

This project had approval from the University of Sydney Human Research Ethics Committee (15003). All chief investigators were assured that projects would not be identified because of anticipated sensitivities about publication output, ineffective interventions or lack of real-world impact.

Results

Completion

Sixty-six (80%) of the 83 intervention studies we identified were completed at the time of data collection, and 13 were ongoing. The status of four was unknown, with no responses from chief investigators. Of the 13 that were ongoing, reasons stated for incompleteness included problems recruiting study participants, being part of larger international trials or being longitudinal studies with longer follow-up. The proportion that were incomplete or ongoing was highest for the eight studies that commenced in 2007, the most recently sampled year, and included three grants scheduled for completion in 2011 or 2012.

Description of funded intervention research projects

The mean duration of funding of the 66 completed projects was 3 years (range, 2–5 years). Interventions included treatment and management (30 studies), screening and early intervention (12 studies), and primary prevention (24 studies), implemented in clinical or community settings, with many dealing with aspects of chronic disease. Topics reflected a variety of health disciplines, including medicine, psychiatry, psychology, dietetics, dentistry, physiotherapy and nursing. In terms of stage of intervention development, most focused on intervention efficacy (28 studies) or effectiveness (27 studies); 10 were replications or adaptations of an intervention in a new setting or population group; and one tested dissemination of the intervention.

Intervention effects

We could not locate published results on primary outcomes for 12 of the completed studies. There were equal numbers of studies that produced statistically significant effects (including “mixed” results; 27 studies) and those that did not show significant effects (27 studies). An example of mixed results was a school intervention that prevented (or delayed) age-related increases in students’ alcohol consumption, but did not reduce the prevalence of students’ depressive symptoms, which had been nominated as the primary objective.

Publication outputs

Publications related to each completed grant were categorised according to whether they reported on intervention effects or on “other” descriptive topics, such as measurement, intervention feasibility, epidemiological questions, or commentaries. The mean number of published articles per grant was 3.3 (range, 0–13), with 2.0 reporting results. Many investigators reported that their publication process was ongoing; eight had not yet published any articles, and twelve had not published articles on intervention effects. Among grants with published results, those with and without significant intervention effects had similar numbers of “other” publications (mean, 1.3 per grant), although the latter had smaller numbers of publications reporting intervention results and of total publications (Box).

Discussion

Our study describes the publication outputs for intervention studies funded by the NHMRC from 2003 to 2007, inclusive, and provides a benchmark to inform expectations about the publication yield of such research. We found that publications covered many aspects of intervention development8,12,14 and were not restricted to intervention effects, although studies reporting no statistically significant intervention effects produced slightly fewer results-based publications.15

While the number of publications is not an indicator of relevance to health policy,6 publication volume remains a basic metric of academic productivity.2,16 Analysis of Australian health promotion intervention research has previously identified between one and seven publications per study,17 while another Australian study of primary care research reported a mean of 2.3 publications per grant (range, 0–7 publications).18 However, the contexts and funding sources for these two studies and our study vary, and there is no endorsed benchmark for assessing numbers of publications across different areas of research.

In relation to our estimates, we acknowledge that later assessment may be required to capture complete publication outputs, and that the impact of non-responders on estimates (whether they would be lower or higher) is unknown. Further, the output estimates in our study cannot be extrapolated to non-intervention research.

Our findings on the stage of intervention development are consistent with those of other reviews of intervention research.9,10 As research type is not routinely documented by the NHMRC, the proportion of available funding that is invested in intervention research is currently unknown. However, our methods indicated that intervention research accounted for a small proportion of NHMRC grants in this period, although the interventions studied related to national health priorities and major causes of mortality and morbidity across Australia.

While intervention research typically tests effectiveness, the statistical significance of study results is not an indicator of study value.2 Some studies reporting non-significant results generated findings with important implications for policy and practice — for example, that an intervention should be discontinued or modified. It was beyond the scope of our study to critically appraise the methods of each funded study, and thus assess whether studies had sufficient statistical power to detect the changes they hypothesised.

While it is often claimed that researchers are discouraged by the difficulty of publishing statistically non-significant findings, we found no evidence for this. However, the length of time to intervention study completion and the relatively small number of intervention study publications may constitute disincentives for researchers to embark on these kinds of studies, particularly as there are no established methods to demonstrate other forms of impact, such as measures of policy change and influence on practice. Reviews of research funding have called for an increase in intervention research and for strategies to help remove the barriers faced by intervention researchers (such as dedicated funding for intervention research, longer funding periods, support for pilot studies and separate review panels).19 Meanwhile, policy agencies have suggested similar remedies to redress their concerns about a lack of definitive evidence on effective interventions in many areas.20

This is the first independent study to document the publication outputs of a set of intervention studies funded through a major national funding body. Tracking research publication outputs is important as a mechanism to ensure accountability in expenditure of public funds and, potentially, as a basis for quality improvement of research funding systems. Ongoing investigations of this kind are needed to provide information on whether current research investment patterns match the need for evidence about health care interventions.

Identifying when the wrong treatments are chosen, and putting a stop to it, is an important way to improve the quality of health care. There are many ways to do this. Strategies include using clinical guidelines or decision support systems, clinician education, clinical engagement, peer review, and adjusting the pricing, funding and availability of individual treatments. All of these methods can be useful. But there are currently few concerted efforts to evaluate and benchmark treatment choices at the hospital level, and to use this information to drive improvement.

Inappropriate care is a longstanding concern in health policy. For over 40 years, small-area analyses have shown significant geographic variation in the rates at which different subpopulations are given common surgical procedures. 1,2 The Organisation for Economic Co-operation and Development has recently published this kind of analysis, including a chapter on Australia showing the rates of nine types of hospital admission in different Medicare Local areas, adjusted for age and sex.3 A more detailed atlas of variation and supporting studies has also been published by the Australian Commission on Safety and Quality in Health Care.4

Wide variation in practice patterns has been attributed to clinicians interpreting evidence and guidelines inconsistently, with patients in different areas being under- and overtreated as a result.

A newer but burgeoning body of literature has looked at inappropriate care in a different way, focusing on the interventions themselves rather than their usage. This focus has led to the development of lists of treatments for disinvestment, either because they have not been proven effective, or because testing has shown that they are ineffective or inferior to a substitute treatment.

Both of these approaches to inappropriate care have shortcomings. Small-area variation analysis has uncovered disturbing variation in patterns of care, but has not produced meaningful policy or practice change.5 This is probably owing to the fact that variation analyses rarely take into account legitimate drivers of clinical variation such as patient morbidity and patient preferences, and so are unable to convincingly differentiate warranted from unwarranted variation.6,7

Although disinvestment work has produced much more compelling evidence of inappropriate care, it has struggled to achieve meaningful policy change. As this study shows, clinicians can continue to use a treatment long after it has been declared inappropriate.

There are few interventions that are ineffective for all patients and indications. Typically, the value of a treatment varies for different types of patients. This complicates measurement of inappropriate care and the development of policies to reduce it. Across-the-board funding cuts, or even funding cuts among specific subgroups of patients, may ignore clinical heterogeneity and deny funding for valuable, as well as ineffective, care.8 A patient may have characteristics that typically rule out a treatment, but have other characteristics that mean they are not well represented in clinical trial samples, or are not eligible for treatments that are recommended as more effective. Approaches are emerging that may help health care organisations and clinicians to distinguish, at the site of care, between patients for whom a certain treatment is warranted and those for whom it is not.8 But in many cases the coded data do not capture all of the relevant clinical variation. This presents a serious challenge to measuring and comparing clinical choices in different regions or among providers.

We sought to develop a practical way to identify which hospitals are most likely to be choosing inappropriate treatments. Our methods draw on the analytical strengths of variation and disinvestment analysis. From disinvestment analysis, we took a selection of treatments that evidence clearly shows should not be done routinely, or at all. From variation analysis, we focused on outlier hospitals — those providing a “do not do” or “do not do routinely” treatment at rates that are far in excess of the national average.

This approach overcomes several key deficiencies of variation and disinvestment research. First, focusing on procedures listed for disinvestment means that high usage rates can be convincingly linked to inappropriate practice. This is not the case in variation analysis, where the procedures analysed are generally considered effective, and where high relative usage in one area can be due to underprovision in other areas.

Second, using variation to focus on providers with high rates of potentially inappropriate care is a practical way to pursue disinvestment. Rather than advocating the removal of clinician discretion with respect to the procedures analysed, this approach advocates monitoring, with priority given to the providers where care is clearly out of step with both clinical evidence and standard practices, and therefore very likely to be inappropriate.

Third, focusing on hospitals rather than geographic areas allowed us to correct for several major deficiencies in other variation analyses. Focusing on hospitals (and specific specialties within those hospitals) allows for analysis of microcultures of care, which are likely to be obscured when practice patterns are aggregated to a regional level. Further, geographically aggregated analyses generally only make crude adjustments for patient morbidity differences, while our use of rich patient-level data allowed us to correct for variation in morbidity to a much greater extent. Finally, there are few viable policy options for dealing with variation at the geographic level. Developing strategies aimed at hospital-level accountability for practice patterns may be more productive in addressing inappropriate care.

Methods

Data

De-identified patient-level data about all public and private hospital separations (discharge, deaths and transfers) for the financial year 2010–11 were obtained from the Australian Institute of Health and Welfare after approval by each state and territory. The dataset included 8 720 771 records from 709 separate public hospital sites in all states except the Australian Capital Territory (private hospitals in each state were all grouped with a single code). Data were released as one record per admission, so it was not possible to link records to derive data on a per-person basis. Names of public hospitals were suppressed as part of the approval process.

Approval from an ethics committee was not required, but data confidentiality requirements were imposed as part of data release.

Selecting treatments for assessment

Potentially ineffective treatments were drawn from published lists of, or recommendations about, inappropriate care. These include a list of procedures identified as potential disinvestment candidates,9 a list of procedures where there had been a “reversal of evidence” — that is, where subsequent evidence had shown that early treatment recommendations were no longer appropriate10 — and examination of decisions of two national health technology assessment bodies: the Medical Services Advisory Committee in Australia and the National Institute for Health and Care Excellence in England. Only guidance published before our data period (2010–11) was used.

To select do-not-do treatments for analysis, these lists were whittled down by excluding recommendations that were vaguely expressed and treatments that were unable to be reliably coded using the International Classification of Diseases Australian Modification diagnosis and procedure codes or did not take place in hospitals. A further filter excluded treatments for which the evidence originally cited was weak or was contradicted by subsequent evidence, or was not supported by clinical experts we consulted. Finally, since we are investigating variation, we excluded cases with five or fewer occurrences and combined overlapping advice (Box 1). At each of these steps we took a conservative approach to reduce the chance that our analysis — a proof of concept — would be rejected based on the examples we used. We also looked at examples of treatments with recommendations against being performed “routinely”, but these were selected opportunistically (as with do-not-do advice, the evidence was evaluated and clinical experts were consulted).

Do-not-do and do-not-do-routinely advice was expressed as “do not (routinely) do procedure x for diagnosis y”. Clinical experts reviewed those cases which had multiple diagnoses listed to ensure that none of the additional diagnoses or procedures might provide a justification for the do-not-do (or do-not-do-routinely) procedure. Coding assignment of diagnoses and procedures was reviewed by an independent health information manager.

The filtering process excluded practically all of the original candidates of potentially inappropriate care. From the original lists, only five procedures were clearly potentially inappropriate and could be analysed in the data (Box 2). From the many do-not-do-routinely procedures, three were selected as exemplars.

Measuring use of the selected treatments

To look at variation in a fair and meaningful way, we measured the proportion of patients in a hospital who have the relevant diagnosis (such as osteoarthritis of the knee) who also received the do-not-do procedure (such as arthroscopic lavage or debridement). This partly addresses a criticism of the geographic research that it does not standardise adequately for differences in underlying rates of disease.

With this metric, we only compared hospitals that are able to provide the do-not-do treatment. These hospitals both perform the relevant procedure and treat patients (> 5) with the relevant morbidity.

Results

Incidence of each of the five do-not-do procedures is shown in Box 3.

The incidence of the five identified do-not-do treatments was quite low, with a total of 5888 procedures identified in the dataset. While this was a very small proportion of all patients, it is 4.5 per cent of all patients receiving the relevant procedures (or in one case, combination of procedures). It should be noted that this is a lower bound as we did not measure compliance with all do-not-do guidance for these procedures.

Among hospitals that perform the relevant procedure and also treat patients (> 5) with the relevant morbidity, we found that the incidence of the do-not-do procedures was highly variable (Box 4).

For all the do-not-do treatments, the outliers with the highest rates were a long way from the average. There were 25 hospital departments that provided a do-not-do treatment more than three times as often as the average hospital (within hospital comparator groups). Eight hospital departments provided do-not-do treatments at over five times the average rate, while three departments did so at over 10 times the national rate.

Again, we saw substantial variability in hospital procedure rates for the procedures that should not be done routinely (Box 5), with some hospitals clear outliers. The hospitals with the highest rates for the three do-not-do-routinely treatments offered them at more than nine times, six times and double the average rate.

Discussion

This study is the first attempt to quantify the extent of inappropriate care in Australian hospitals for a range of conditions using routine data. We have shown that it is possible to use routine hospital data to identify the incidence of potentially inappropriate care.

Importantly, the procedures used here as examples have either been shown in academic studies to be inappropriate or are recommended against in guidelines, or both. What we have shown is that, despite this advice, and even defunding in the Medicare Benefits Schedule, the procedures are still being performed. Guidelines and funding policies are clearly not sufficient to solve this problem.

Limitations

This study has a number of limitations. First, it cannot be a basis for generalising about the overall incidence of inappropriate care in Australian hospitals. We used a small, non-representative sample of hospital procedures and analysed their incidence in a single year. Because inappropriate care appears to be relatively infrequent, there may be instability in the incidence of our indicator conditions.

Second, the inappropriate care identified in this study can only be considered potentially (rather than definitively) inappropriate. Some of the identified inappropriate treatments may be coding errors or may be justified on the basis of a rare combination of patient characteristics. Routine data can only identify what is occurring, not why. We therefore suggest that the indicator be labelled “potentially inappropriate care”, although the procedures themselves would remain indicators of, prima facie, inappropriate care.

Third, we were not able to analyse rates of inappropriate care in individual private hospitals.

Fourth, we were unable to link data across settings or time. A person who had multiple treatments, one of which was a do-not-do treatment, would thus be counted once in the numerator and multiple times in the denominator. This makes our prevalence estimates conservative.

What should be done to ensure safe and effective care?

Identifying potentially inappropriate practice is irrelevant unless action occurs. The steps outlined below will allow governments to monitor more and ultimately reduce inappropriate care in Australian hospitals.

First, forms of inappropriate care should be identified and communicated in a more consistent and accessible way. This should be a role for the Australian Commission on Safety and Quality in Health Care or the health productivity and performance commission foreshadowed in the 2014 federal Budget to take over functions from the Commission together with a number of other performance-related national authorities. Such a Commission should, with clinical involvement, identify additional procedures that represent potentially inappropriate care. It might draw on parallel efforts such as NPS MedicineWise’s Choosing Wisely program and the Royal Australasian College of Physician’s EVOLVE initiative. The Commission should maintain a centralised, easily accessible and continuously updated list of questionable procedures for clinicians to use.

Second, existing data should be used to measure and benchmark a wider range of inappropriate care. In this study, data access agreements meant we could only analyse a handful of do-not-do treatments. Many more can be analysed using additional sources of clinical guidance and data that the Australian government already possesses, and that the Commission should have access to. Studies of geographic variation have linked patient data across datasets and over time.23 Linkage of this kind could allow the Commission to do more; for example, analysing inappropriate use of treatments that should not be first-line interventions. It could also link patient records across different parts of the health system to cover hospitals, the Pharmaceutical Benefits Scheme and the Medicare Benefits Scheme.

Third, the Commission should advise states and hospitals about their rates of questionable care. States should give outlier hospitals a chance to improve, but if high rates persist there should be an external clinical review. The reviews should investigate all aspects of clinical decision making in the relevant department or specialist area. This would include confirming that the data accurately recorded the treatments that were chosen and assessing whether these choices were clinically valid.

States should prioritise investigation of hospitals with high rates of potentially inappropriate care, both for practical reasons and because low rates may reflect cases where most procedures were actually legitimate. States should also consider the volume of patients with the relevant morbidity. Our cut-off (> 5) could result in investigations that affect the care of very few people. Choosing thresholds that trigger investigations is ultimately a normative decision. It involves a trade-off of unnecessary investigatory burden and waste of resources against potential clinical risks to patients associated with inappropriate care.

Ultimately, patients should have a reasonable expectation of receiving appropriate care. Our study has shown that procedures that are contrary to contemporary clinical evidence are being done in Australia, and that some hospitals seem to provide a very high rate of these procedures. These hospitals should be alerted to the fact of their aberrant practice and be subject to clinical review if that practice continues, helping to ensure that hospital care in Australia is evidence-based, effective and safe.

1 
Selection process for do-not-do treatments

3 
Incidence of do-not-do treatments, 2010–11

4 
Distribution of proportion of relevant patients receiving a do-not-do treatment in public hospitals with capacity to perform the do-not-do treatment, 2010–11

---

*Average rates among comparator hospitals: hyperbaric oxygen therapy (2.9%); arthroscopic lavage or debridement of the knee (3.3%); vertebroplasty (5.4%); oophorectomy (1.4%); uterine nerve ablation (2.1%).

5 
Distribution of proportion of patients receiving a do-not-do-routinely treatment in public hospitals with capacity to provide the treatment, 2010–11

---

*The 10% of hospitals with the highest rates are identified as outliers.

The costs and harms associated with systemic overdiagnosis and overtreatment are receiving international attention. Clinicians and their professional organisations are obliged to assume resource stewardship as an ethical responsibility as embodied in the 2002 millennium professional charter.

Low-value care is use of an intervention where evidence suggests it confers no or very little benefit on patients, or risk of harm exceeds likely benefit, or, more broadly, the added costs of the intervention do not provide proportional added benefits. The lattermost concept involves considering benefit and cost relative to those of alternative care options, as embodied in cost-effectiveness analysis.1 Examples of high-value and low-value care occupy a matrix which emphasises this relativity of benefit over cost (Appendix 1). Care may be high (or low) value if, despite higher (or lower) costs, it confers proportionately greater (or lower) benefit. Choosing low-value care consumes resources that could have been expended on alternative forms of care conferring greater levels of benefit, either to the patient in question or to other patients.

Efforts to reduce low-value care run counter to the dominant financial incentives in our fee-for-service (private sector) and activity-based funding (public sector) systems that reward volume over value, challenge the cultural assumption that more is better, and raise concerns about stinting on necessary care. Historically, most research and effort in quality improvement has focused on underuse of high-value care2 — a legitimate concern — rather than overuse of low-value care.

How much low-value care exists?

It has been claimed that at least 20% of health care expenditure in the United States is wasted on activities that add no value.3 Studies using US Medicare claims data suggest that almost half of beneficiaries receive some form of low-value care.4 While comparable statistics for Australian health care are not available, reviews of Medicare Benefits Schedule (MBS) items have suggested that at least 150 commonly used tests and procedures are associated with little high-quality evidence of benefit, and that for some there is evidence of harm for their assigned indications.5 To date, fewer than 5% of MBS items have been closely scrutinised for their evidence-based worth. Operations such as arthroscopic debridement for uncomplicated knee osteoarthritis are frequently performed despite randomised trials showing no benefit.6 Investigation requests — such as those for vitamin B₁₂, folate7 and vitamin D8 assays, and for computed tomography scans for back pain and chest diseases9 — have surged in recent years despite considerable doubt as to their usefulness to patient care. Screening and diagnostic tests and procedures predominate over therapeutic agents in most studies of overuse.4 Some are of high value in high-risk populations (such as screening colonoscopy in patients younger than 60 years of age with premalignant colon conditions or family history of bowel cancer) but assume much lower value when extended to low-risk populations (patients older than 75 years of age with no risk factors).5 Overuse may also partly explain the marked geographical variation in age- and sex-standardised rates of cardiac catheterisation (7.4-fold variation) and hysterectomies (4.0-fold variation).10 In 2006, the Productivity Commission estimated that the efficiency of Australian health care could be improved by up to 20% by aligning performance with best practice across a range of service areas.11

The science of measuring overuse is in its infancy — a recent review noted that only 37 fully specified measures exist for overuse, compared with hundreds for underuse.12 Measures may take several forms (see Box) but all are constrained by the lack of systematic collection of granular clinical data at the level of individual patient care that captures the indications for the intervention (why was it given?) and the views and preferences of patients (in cases of marginal benefit, was there a strong patient preference to receive it?). Such nuanced data are necessary in deciding when the same service is high value in one patient but low value in another. Electronic health records which mandate insertion of indications, linked to utilisation databases (pathology, radiology, MBS, Pharmaceutical Benefits Scheme), may enable development of data repositories that are capable of interrogation at the individual level using unique identifiers. Although routine data can suggest that a low-value treatment has been provided, they cannot reveal why it occurred.

Attention must also be given to potential unintended adverse consequences of overuse measurement such as underuse of indicated care, biased patient selection, harm to doctor–patient relationships, and shifts to alternative tests and treatments in clinical settings not subject to measurement.13

What strategies can be deployed to reduce low-value care?

Currently, there is considerable government interest in Australia in using policy levers to minimise overservicing and contain health care costs. Strategies focus on decision making involving consumers (demand side) or providers (supply side). It should be noted at the outset that the impact of these strategies appear unaffected by whatever system of health care financing (fee-for-service, capitation, or managed care) predominates in particular clinical settings.14

Demand-side levers targeting patients principally include financial incentives and education. Increasing patient cost-sharing by way of copayments for services is a blunt instrument and can lead to reduced use of both low-value and high-value care, as most patients cannot differentiate between the two. Indeed, patients overestimate benefit or underestimate harm in about two-thirds of clinical decisions amenable to benefit–risk analysis.15 Furthermore, most research on consumer education campaigns, including public reporting on provider performance, suggest that such campaigns are weak instruments for changing patient behaviour.16 Patient information and decision-aid approaches in the context of shared decision making show promise,17 but are currently confined to relatively few clinical scenarios.

Supply-side levers, aimed at health care providers, depend heavily on financial incentives which may be service specific (eg, pay-for-performance and prior authorisation) or population based (eg, risk-sharing, in which providers accept financial responsibility for total costs of care under bundled or blended care payment schemes). However, evidence of effectiveness of such incentives remains limited,18 and Australian trials of prepaid coordinated care (a blended payment scheme) for managing patients with diabetes have not been finalised. Removing or lowering Medicare rebates or diagnosis-related group payments for low-value interventions is another option. However, past efforts to substantially revise the MBS have failed and very few items have been removed from the schedule in the past decade. However, a major review process is currently underway. One innovative option is indication-specific pricing whereby remuneration for an intervention depends on the amount of benefit it confers for a specific indication, with the base price anchored to the indication for which it provides the most value, or linked to a preset value.19

As for information-based supply-side interventions, clinical practice guidelines, clinical audits and feedback, academic detailing, decision support and other professional educational interventions exert limited impact in curtailing inappropriate care. Embedding of decision support, quality measurement and instant feedback into electronic health records and handheld-device applications may help to reduce low-value care at the point of care. Examples include lung cancer guidelines in the US20 and various guidelines from Cancer Council Australia.

While waiting for digital solutions, analysing claims-based data and identifying providers (as groups or as individuals) with inordinately higher rates of use compared with peer averages may be useful in terms of stimulating practice reviews for evidence of overuse. While such datasets are inexpensive, widely available, and population-based, they lack detailed clinical information. Population-based outcome monitoring linked to professional incentives is another option; this approach is finding favour in the US in accountable care contracts which encourage decreased use of low-value care through partial capitation or shared savings.

What is the role of clinicians in recognising and minimising low-value care?

What constitutes low-value care lies in the eye of the beholder. Most clinicians regard very few interventions as conferring no benefit in all clinical circumstances. They are reticent in labelling a practice low value if, in their personal experience at least, the practice has stood the test of time and conferred benefit on some patients with no safety concerns.

Altering such perceptions is not easy. First, evidence of treatments that are of little or no benefit is viewed sceptically if it: is perceived as lacking objectivity, consistency or clinical plausibility; is not equally applicable to all individuals (different magnitude of effects or different patient valuations of benefit among different patient groups); or challenges strongly held professional beliefs based on personal experience and peer opinion. Information about what constitutes ineffective treatment is diffuse, sometimes low quality and hard to use.

Second, various cognitive factors distort perceptions of low-value care,21 reflecting clinician desire to avoid potential injustice to individuals, and their own sense of regret or medicolegal liability, from withholding interventions that may possibly bestow some benefit. In many such cases, the absence of any other viable treatment option pushes the clinician to offer the only available, but still low-value, treatment.

Third, external factors such as quality-of-care metrics, competitiveness in the medical marketplace, and organisational aspirations to be “centres of excellence” all encourage “thoroughness”, pro-intervention bias and desire to meet expectations of patients, referring clinicians and corporate stakeholders at the risk of overservicing.

However, professional organisations in the US have begun the task of defining instances where the use of a particular intervention is likely to constitute low-value care. The American Board of Internal Medicine Foundation’s Choosing Wisely campaign involves more than 60 specialty societies, each identifying at least five commonly encountered scenarios in their specialty where specific interventions should be avoided.22 Examples of how these can be applied in clinical practice are described in Appendix 2.

Critics worry that US specialty groups have found it easier to recommend reduction in use of services by other specialties compared with those within their own specialty, or focus solely on low-cost diagnostic tests, rather than address high-cost interventions specific to their own craft group.23 Nevertheless, it represents a beginning and the Royal Australasian College of Physicians24 and the National Prescribing Service (http://www.choosingwisely.org.au) are mounting similar campaigns. The US Preventive Services Task Force, the National Quality Forum and the American College of Physicians’ High Value Care platform all seek to educate front-line clinicians on methods for identifying unnecessary and wasteful services.1

In the United Kingdom, the Academy of Medical Royal Colleges has released a report outlining the rationale and principles behind reducing low-value care, illustrated by numerous case studies.25 Specialty societies and researchers are employing expert panels equipped with the best available research syntheses to rank the level of appropriateness of different interventions using consensus methods and Delphi rounds.26 These activities recognise that most interventions are neither low value nor high value in all cases, and seek instead to identify their nuanced low-value and high-value applications according to clinical context.

Should clinicians be allowed to be the final arbiters of low-value care? In making complex and highly individualised clinical decisions, nuanced clinical judgements of experienced, well informed clinicians will arguably outperform any service-level measurement and incentive program aimed at recognising and reducing low-value care. Attempts to distil and apply evidence-informed “do not do” rules at the population level are bedevilled by the rule of legitimate exceptions when care is directed at specific individuals.

Moreover, in a recent national US survey, 92% of physicians said they felt responsible for ensuring patients avoid unnecessary tests and procedures, and 58% believed that physicians were best positioned to do so.27 Arguably, deciding how and when to use limited health care resources are clinical questions that are best addressed by those with sufficient training and experience. Clinicians are more likely to respond to evaluations of quality by their peers than to adverse comments from non-clinical reviewers or managers, or to threats of financial sanctions and penalties for seemingly low-value care based solely on macro level analysis.

At the micro level, some physicians are establishing programs in their areas of practice that identify and remove low-value care using evidence of effectiveness, cost awareness, and regular audits and feedback.28 Training and education programs can enhance and assess the ability of aspiring clinicians to recognise and practice high-value care.29,30 Such bottom-up approaches are a good place to start and public policy interventions should support clinician-led efforts to seek professional consensus on what constitutes low-value care and the best means for reducing it.