Our preferred alternative: a professional community standard

For these reasons, we prefer a professional community standard for distinguishing correct from incorrect diagnosis.19 By this we mean the following: ‘correct’ diagnoses are those that are accepted as correct within a relevant professional community. This recognises that medical and scientific expertise is important when adjudicating between correctness and incorrectness, and that health-related judgements and actions that occur outside the clinic can reasonably be judged against the knowledge of relevant professional communities. However, it also acknowledges that standards for determining correct action are actively produced by members of a professional community, meaning that evidence is necessary but never sufficient to the task.

Sometimes, the active creation of ‘correct’ diagnoses is obvious: for example, in the negotiation of the relatively new diagnosis of pre-cancers, such as ductal carcinoma in situ (DCIS) or high- and low-grade squamous intraepithelial lesion.20 ,21 However, this also occurs more subtly. For example, the shift from morphological criteria to immunophenotyping as the basis for classifying and diagnosing leukaemia and lymphomas involved active creation and coordination of equivalences between seemingly incommensurable clinical understandings, pathological descriptions, qualitative and quantitative diagnostic criteria, and professional interests in an otherwise clearly defined clinical domain.22 ,23

Professions are communities of practice; members of these communities produce and instantiate the best available standard for practice at any given time. Explicit standard production is preferable because this requires greater justification and accountability from participants. But standards can also develop implicitly, first instantiated through practice routines and then tacitly shared. A new test or drug, for example, can enter everyday practice through individual practitioners and their networks, without any formal endorsement from expert bodies.

These observations again revise the definition of overdiagnosis to the following. Overdiagnosis occurs when a health-related problem is identified and labelled (an intervention may also be offered), this identification and labelling would be accepted as correct in a relevant professional community, but it does not benefit.

A professional community standard: implications

There are three implications of recognising that the standard for ‘correct’ judgements and actions is negotiated intersubjectively in professional communities.

First, overdiagnosis cannot be identified until a standard exists. There are several consequences of this. One is that the line between overdiagnosis and error, particularly false positives, may be less clear than is usually assumed. To date, the idea that overdiagnoses are correct and false positives incorrect has been central to the conception of overdiagnosis. But a tacit or explicit standard must exist, and be recognisable, for correct judgements and actions to be distinguished from incorrect ones. This means that, particularly in the earliest days of a practice or technology before standards are established, overdiagnosis and false positives may be indistinguishable.

This is further complicated by conflict between professionals. Often, subspecialties or different disciplines hold competing understandings of conditions. Take peripheral vascular disease (atherosclerosis of the lower limb), which pathologists diagnose by narrow calcified vessel walls, cardiologists by low blood pressure in the legs, physiotherapists by the presence of claudication and radiologists by narrowed vessels in a contrast study.24 ,25 Each discipline focuses on a different object—vessel walls, blood pressure, pain and so on—demonstrating that standards are acts of coordination that somehow fold these different objects into one disease. If we acknowledge that the line separating ‘correct’ from ‘incorrect’ is negotiated within professional communities, and that different communities may work from quite different conceptions of conditions, it becomes apparent that disagreements about the prevalence of overdiagnosis may sometimes be disagreements within or between professional communities about what the reference standard for identifying a condition should be.

Second, acknowledging that standards are produced in and by professional communities forces attention to the process by which they are developed, and so provides better analytic tools for dealing with disagreement over overdiagnosis. Recall the example of peripheral vascular disease. Advocates of a correspondence standard may insist that science should be able to determine which definition is truly ‘correct’, and thus dismiss the other partial or ‘incorrect’ accounts. But this is of limited use as all disciplines will have scientific cases to make. The professional community standard encourages us to ask how competing standards are arrived at and by whom, and to understand disagreements about overdiagnosis as the result of differing processes of production, rather than as a result of one group being ‘right’ and the others ‘wrong’.

The third implication follows: the professional community conception is the best basis for considering the vested interests, political pressures, and social norms and conventions that have been key concerns for those studying the drivers of overdiagnosis.26

Benefit and harm in individuals and populations

The definitions quoted at the beginning of this paper all contain a particular implicit conception of non-benefit: the condition diagnosed would not have caused symptoms in the individual in their lifetime, such that there is no benefit to be had from the diagnosis. These definitions implicitly work at the level of the individual. They assume a (perhaps hypothetical) patient and doctor, and attempt to point to situations in which that doctor overdiagnoses that patient. We propose, in contrast, that because overdiagnosis is almost exclusively observable at the level of populations, and non-observable at the level of the individual, it should be defined at the level of populations. This move is also made implicitly in Hofmann's nomenclature, although it is not explicitly defended in those terms.3

This issue can be usefully considered via Widdows and Cordell's distinction between aggregative and corporate communal goods.28 ‘Aggregative communal goods’ are constituted of goods that attach to individuals: these individual goods aggregate to a communal good. In contrast, corporate goods attach to whole communities, emerging only as a property of that community. An increase in average lifespan or a decrease in cancer incidence are aggregative communal goods; a universal healthcare system is a corporate communal good.

Overdiagnosis occurs in individuals, and these instances aggregate in populations. Critically, overdiagnosis can only be observed statistically, estimated via studies of whole populations. If a cancer screening programme increases the incidence of early-stage disease without reducing the incidence of late-stage disease, for example, overdiagnosis will be statistically inferred. This may be occurring because there is no treatment available, because treatment is not beneficial or (as in Marcus' tumour A and tumour B types)2 because screening-identified conditions would have remained unsymptomatic, so there was no benefit of detection to outweigh the resulting harmful labelling and intervention. The screening programme is harming, or in the best case failing to benefit, individuals. These individual cases aggregate to the observable population-level rate of overdiagnosis, but the individual cases can rarely be identified.3

Imagine two asymptomatic women diagnosed with early-stage invasive breast cancer following mammographic screening. Both have a mastectomy. For each, the harm of removing a breast is readily observable. However, the benefit for each woman cannot be known because it is unknown how each woman's life would have proceeded without the diagnosis and surgery. Perhaps one of those women was going to develop life-threatening invasive breast cancer and the other was not. Perhaps, if left unscreened, one of these women would have died of a heart attack within 5 years (before being diagnosed with breast cancer) but the other would have developed life-threatening invasive cancer in 10 years. In either of these scenarios, one woman has likely been overdiagnosed, and the other not. However, this cannot be known for the individual women. This is what Hofmann describes as the problem of counterfactuality: that it is impossible to know what would have happened if a screened person had not been screened.3 Hofmann is correct about counterfactuality at an individual level; we would add to his observation, however, that each woman represents a case of early-stage breast cancer in a population, and their cases can be aggregated with others to estimate overdiagnosis in that population.

This pattern can be seen across a range of conditions. For example, when ventilation-perfusion scanning was replaced by more sensitive CT pulmonary angiography, rates of diagnosis of pulmonary embolism increased; population-level data were required to reveal that this had increased rates of complications of anticoagulation but not reduced mortality rates.29 The diagnostic criteria for mild cognitive impairment and ‘pre-dementia’ are broadening, supported by new cognitive screening tests, cerebrospinal fluid biomarkers and neuroimaging. Population data suggest that such diagnoses often spontaneously resolve, are not consistently associated with progression to dementia and produce a range of harms.30 Changes in testing technology and threshold changes have increased the prevalence of ‘pre-diabetes’ twofold to threefold, with massive prevalence in some populations (eg, 50% in Chinese adults). However, population-level data suggest this is not beneficial and may be producing new harms and costs.31 The counterfactuality problem makes it impossible for an individual diagnosed with pulmonary embolism, pre-dementia or pre-diabetes to know what would have occurred had they not received the diagnosis and intervention. However, their outcomes can be aggregated, and this—at a population level—can allow overdiagnosis to be inferred.

So, although overdiagnosis does occur in individuals, it can rarely be observed in individuals.ⁱⁱⁱ It follows that, to be useful, the definition of overdiagnosis must be expressed at a population level rather than an individual level. Even if we were able to generate a satisfying individual-level definition, it could rarely be used. A useful definition of overdiagnosis will operate at a population level and can be used as a guide to policy and practice. The goal would be to alter diagnostic practices in such a way that cases of overdiagnosis—the (largely unidentifiable) cases that would have aggregated to create the population-level rate of overdiagnosis—do not occur.

Expressing the benefit and/or harm component of overdiagnosis in population terms requires thinking statistically or probabilistically. In addition, because treatment may cause harm in order to benefit, these two must be held in relation to one another in the definition. Cancer treatment, for example, is often harmful (eg, causing nausea or bodily disfigurement) but can prevent early death, providing recipients with the benefit of decades more of life. Because of this, we judge it to be beneficial on balance despite the harms. Following this, the component of overdiagnosis relating to consequences should be expressed in terms of an unfavourable balance between benefits and harms.

The unfavourable balance between benefits and harms relates to the aggregative good or bad of overdiagnoses. However, overdiagnosis also produces corporate harms. Overdiagnosis mostly arises from system-level shifts: changing the definition of conditions, introducing new technologies and aggressively promoting testing. Because the balance between benefits and harms caused by the diagnosis is unfavourable, the economic cost of these activities to the health system is not justified by their outcome. They effectively divert resources from more to less effective care.32 This has an aggregative component (the harms experienced by individuals whose care is compromised). But it is also a corporate harm: it produces a less responsive, less trustworthy, less effective healthcare system.^iv Taking the resource allocation implications of overdiagnosis seriously entails identifying services most likely to do aggregative or corporate harm or wrong. This leads to the next question: how to define which benefits, harms or wrongs matter.

Who should judge benefit and harm?

There are two possible answers to the question of who should determine which benefits, harms or wrongs matter, each supported by reasonable arguments. Although these answers may seem incommensurable, they are likely to be combined in practice.

The first answer is: individuals receiving interventions should be free to judge which benefits, harms and costs matter to them. This is a proceduralist view: it suggests that what matters is due process. Individuals, in this view, should not be coerced into interventions; some would suggest they should also be supported to choose in accordance with their values.35 This view loosely reflects classical liberalism, founded in the intrinsic political and moral significance of (1) having one's autonomy respected, and (2) having the liberty or freedom to pursue what one values. Such freedoms are unquestionably important, but this model is not uncomplicated.

Aronowitz has made the important observation that once diagnosed people may act in accordance with the worst-case scenario rather than their own situation. For example, rates of prophylactic mastectomy have been increasing. Women diagnosed as ‘high risk’ for breast cancer (eg, through genetic testing) have been electing to have mastectomies. Women diagnosed with stage I, II and III breast cancer in one breast have been electing to have both breasts removed.11 Other work has also shown an increasing US trend for American women diagnosed with DCIS, a pre-cancer or cancer risk factor, to opt for bilateral mastectomy.36 Importantly, although the likelihood of benefit of prophylactic mastectomy will likely differ between these groups, uptake has been steadily increasing across all groups.11 Women may, of course, have excellent and carefully considered reasons for such choices. However, this analysis suggests it is also possible that risk perception can lead us to accept interventions that are disproportionate to the problem, particularly if healthcare systems actively make them available to us.

Other challenges face the proceduralist view. It may be difficult for clinicians to balance the preferences of individuals against resource implications for the system. Shared decision-making is resource-intense, and dedicated efforts are required to ensure equitable outcomes.37 ,38 There is also a disjunct between this method of judging the balance of benefits to harms and our contention that overdiagnosis can only be defined at population level. While there may be intrinsic moral goods in allowing or supporting each individual to judge what matters to her, this provides little guidance for determining whether or not a pattern of care is causing overdiagnosis in a population.

An alternative could be characterised as a substantive view: that is, a view that experts or authorities should define which benefits, harms and costs are important in each case, and determine how they should be measured. Overdiagnosis cannot be estimated in populations, and medicine cannot function, without some substantive view of which benefits and harms matter. Those who emphasise a substantive view may prioritise minimisation of harm (even at the expense of individual choice) or hold that decision-making should rely on expert, not lay, knowledge. A substantive position also has weaknesses: experts are likely to disagree on which harms or benefits matter, and may seek to coercively impose the interventions they prefer onto individuals. Somehow the significant heterogeneity of risk, disease and illness experience needs to be adequately reflected in the substantive position taken. And even if experts attempt to take ‘the patient perspective’ into account, they can never completely capture and reflect what matters to any given individual. To quote Greenhalgh and colleagues:… [Patients] live in the messy, idiosyncratic, and unpredictable world of a particular person in a particular family context (or, for some, in a context of social isolation and/or abandonment by family) … patients' priorities ‘on average’ … can never fully capture the situated, fluctuating granularity of what matters most to a particular patient and carer at a particular point in the illness journey.39

Greenhalgh and colleagues go on to argue that evidence-based medicine (which is, at its heart, an attempt to integrate best evidence of benefits and harms into clinical practice) is biased in systematic ways that de-privilege human experience and exclude what matters to people.^v Thus, while a birds-eye, population-level, average view of benefits and harms is necessary even to determine whether overdiagnosis is occurring, it needs to be nuanced and responsive to heterogeneity in experience and values, and it can never be sufficient, particularly when intervening with individuals.

Although they may seem incommensurable, proceduralist and substantive views are likely to be combined in practice. Substantive assessments of benefit and harm may be strengthened if they are procedurally just (eg, by including citizens and being transparent). Conversely, individuals cannot make good choices about interventions without some substantive content to consider. The balance between substantive and procedural concerns will likely shift depending on the case. If experts agree that a test or procedure is extremely harmful to many people, for example, few would argue that all patients should be offered the test and invited to fully consider its benefits and harms. More frequently, however, the benefits and harms of an intervention will be equivocal. Diagnosis of and intervention for cardiovascular disease risk in most people, for example, is of equivocal benefit.40 In such situations, the determination of the rate of overdiagnosis will need to be contextualised in careful qualification about how benefits and harms were selected and valued, it may be appropriate to run processes such as community juries that seek citizen guidance for policymaking, and shared decision-making may become especially important in the clinic.

Three types of overdiagnosing action with different normative implications

The degree of harm, avoidability of harm and actors' goals interact to produce a typology of overdiagnosing actions. We propose these as ideal types in a sociological sense: they are referents against which messy, real cases can be compared. They do not seem morally equivalent. We have named the three types predatory, tragic and misdirected overdiagnosis. Recall that, in all three types, the balance between benefits and harms must be unfavourable (as specified in the definition of overdiagnosis), and that the greater the harm, the more morally problematic. This is assumed in the following sections.

Procedural justice in identifying benefits, harms and professional standards

Throughout, we have emphasised the potential conflicts in determining reasonable actions, and relevant benefits or harms. Many epidemiologists and economists are currently working to quantify the benefits, harms and costs of potentially overdiagnosing interventions. To do this, they need to make assumptions about the standard for reasonable actions as these separate ‘correct’ from ‘incorrect’ care, and so separate overdiagnosis from error. They also need to define the benefits, harms and costs that they will measure and model. But these assumptions are often implicit. The overdiagnosis movement may benefit from a more procedurally just approach to addressing these fundamental problems. For example, community juries or the principles of accountability for reasonableness could be used to determine what benefits, harms and costs should matter and why.48 ,49 In addition to being goods in themselves, such processes would provide fairer means for negotiating differences between professional standards and communities.

Who is morally responsible for overdiagnosis, including the opportunity costs?

The literature is not clear on who should reasonably be held responsible for fixing the problem of overdiagnosis. It is variously implied that policymakers, clinicians, lawmakers, manufacturers and/or citizens should act. There are many open questions in this area. Is it reasonable, for example, to expect citizens to engage with detailed information and make their own decisions if experts cannot reach agreement on what is reasonable? Or if predatory overdiagnosis is identified, who should respond to contain it? We have shown that overdiagnosis is multifactorial, and it seems likely that different actors may be responsible for different components. For example, regulators or the courts may be responsible for sanctioning the worst cases of predatory overdiagnosis, while individual clinicians may only be responsible for avoiding misdirected overdiagnosis by examining their clinical goals and adhering to a justifiable professional standard. Our analysis emphasises that overdiagnosis is an artefact of systems of healthcare. This suggests we should be wary of overindividualising moral responsibility for the problem. In fact, some of the core elements of our system—such as defining and maintaining the integrity of professional standards—require collective effort for their achievement and are corporate goods in themselves.

Individuals will carry some moral responsibility for some forms of overdiagnosis, perhaps most strongly predatory overdiagnosis. However, collective actors such as states or professional bodies may reasonably be held responsible for providing the structures needed to minimise the chance of future overdiagnosis. So, for example, if professional standards are in conflict, and some suggest harms are large and avoidable, there may be a shared obligation to reach fair agreement on which standard is correct. If the conclusion is that large avoidable harms are occurring, there may be a collective obligation to strongly guide individual clinicians (who rely on the standard) away from testing, rather than maintaining a laissez-faire practice environment with substantial potential for individual clinical discretion. If predatory actors continue to promote alternative pro-testing standards, and funding arrangements make testing easy, collective actors such as colleges, expert bodies and the research community may be obliged to actively promulgate their agreed standard, criticise problematic standards, remove funding for testing and/or sanction predatory actors.

Recognising overdiagnosis as a matter for distributive justice

Overdiagnosis is a matter for distributive justice in two competing ways. First, a pressing moral problem with overdiagnosing interventions is their structural effects on the healthcare system. Overdiagnosis often arises from the application of expensive new testing or treatment technologies. This diverts resources away from basic health services and public health interventions, and may threaten the sustainability of universal healthcare where it exists.32 This is likely to effect the least-well-off, who are least able to pay for services, most strongly. This is a matter for justice and a loss of a corporate good, and can only be addressed at a systems level.

Conversely—and paradoxically—limiting diagnostic categories may redefine individuals as non-diseased, or even malingering, which may remove the health and social care that follows official medical labels. A diagnosis can provide access not just to medical care but also to social, welfare and other services. For example, being diagnosed with a mental illness may be a pathway not only to drug treatment, but also to psychological treatment, social supports or a disability pension. A diagnosis of attention-deficit hyperactivity disorder may facilitate not only drug treatment, but also extra help in the classroom, or parenting support. While medicalising ‘normal’ life experiences can be very reasonably critiqued and may do harm in itself,5 ,26 there can also be beneficial social effects of receiving a medical label.

Thus, efforts to reduce overdiagnosis may have negative social consequences for the people who rely on the social power of that diagnosis. This is not morally neutral, particularly when public services are under threat. It may also especially disadvantage the least-well-off who may lose access to services without the power of a diagnostic label. This is the paradox: overdiagnosis may misallocate needed funds away from universal healthcare, which will disadvantage the least-well-off; but curbing overdiagnosis may also remove services from the least-well-off.

There is no easy answer, and as previously noted, different groups may be differently responsible for this problem. Researchers, for example, should take care to include all potential benefits and harms, and their distribution, in estimating the effects of interventions to limit overdiagnosis. Policymakers, colleges and other groups with responsibility for practice change should be especially attentive to effects in already-disadvantaged communities. This includes not just the effects of change (eg, the possibility that social services may be lost if a medical diagnosis is removed) but also the effects of remaining with the status quo (eg, the harm that may be done if people continue to be overdiagnosed). It should be noted that the possibility of unintended harms should not prevent action: rather, action should be taken in such a way as to monitor and minimise new harms. Returning to our earlier discussion of identifying benefits and harms, it is likely that potential harms or loss of benefits will only be fully understood if affected communities are included in decision-making.

Responding to the experience of those who may have been overdiagnosed

The final unresolved moral challenge of overdiagnosis is in relation to the harms that may be done by the way overdiagnosis is communicated. Intervention to limit overdiagnosis is generally motivated by a desire to limit harms done to populations by overdiagnosing interventions. However, this has implications with regard to the counterfactual problem. Recall the two women, both diagnosed and treated with breast cancer, one of whom has been overdiagnosed. With low awareness of overdiagnosis, both women are likely to think themselves saved by screening and the onerous treatment they subsequently undertook. As awareness of overdiagnosis is raised, both women are likely to question whether or not they have been overdiagnosed. They may, as a result, experience distress or existential doubts and lose trust in the institution of medicine. This is not a reason to avoid communicating about overdiagnosis, but it does suggest a need for care and the provision of appropriate supports to those affected.