Statistics from Altmetric.com
Overdiagnosis is a significant problem for healthcare, harming individuals and incurring costs. One potential barrier to addressing overdiagnosis concerns the difficulty of pinning down exactly what we mean by the term. To this end, Carter et al make a welcome contribution to the growing debate.1 In particular, the authors raise important points about the social harms of overdiagnosis and the need for structural interventions; their typology is useful for discussing ethical issues in specific cases of overdiagnosis and may ground legal and institutional actions; and they contribute to investigation of the psychological and sociological processes that lead to overdiagnosis. Our main reservation hinges on their definition of overdiagnosis, which we feel may fail to map current usage.
On their account, overdiagnosis of condition A in a population occurs when that condition is correctly identified and labelled (diagnosed) but the diagnosis and/or intervention is, on balance, more harmful than beneficial. The correctness of the diagnosis relies upon acceptance by a relevant professional community. Overdiagnosis must be defined at a population level, given the difficulties of knowing which specific individuals diagnosed with condition A suffer more harm than benefit.
The authors' definition seeks to explicate key features attributed to overdiagnosis, namely that the diagnosis is correct but that there is no benefit. They start by characterising the “rough meaning-in-use for overdiagnosis [as] ‘a correct diagnosis that does not benefit’” (ref. 1, p. 2). This claim follows numerous quotations from the literature summarising dominant views about the nature of overdiagnosis. These views are remarkably consistent in their central claim, namely that overdiagnosis involves a condition or disease that would: ‘not have harmed’;2 ‘not present symptomatically’;3 ‘never cause symptoms or death’;4 ‘not have been identified clinically’5 and so forth. In turn, many of these views are based upon epidemiological evidence of increasing rates of diagnosis with little or no concomitant decreases in mortality, indicating the identification of a large reservoir of subclinical disease.6 ,7 Thus from the existing literature, the key feature of overdiagnosis is the identification of non-harmful disease. In characterising this as ‘a correct diagnosis that does not benefit’, the authors sidestep this central, and we would argue pathognomonic and essential feature of overdiagnosis as the identification of harmless disease. Instead of connecting the lack of benefit with the innocuous nature of the disease identified, the authors take a much wider frame to include the overall balance of benefits and harms of both diagnosis and any subsequent intervention. In doing so, they widen the definition to include any conditions in which there is a negative balance of benefits to harms, irrespective of the nature of the disease in question.
This overinclusiveness leads to a number of counterintuitive conclusions. First, it seems to include conditions for which the subsequent intervention is more harmful than beneficial, yet which do not usually count as instances of overdiagnosis. For example, when coronary artery disease (CAD) was correctly identified according to relevant professional standards and treated with ligation of the internal mammary artery, the ‘label and/or intervention carrie[d] an unfavourable balance between benefits and harms’ (ref. 1, p. 4), because the intervention was in itself harmful (involving surgery) and there was no therapeutic benefit. This example seems to fall within the authors' definition of overdiagnosis. However, we believe that most commentators would not classify CAD as overdiagnosis on the basis of harmful therapy. Similarly on their account we could claim that type 1 diabetes is overdiagnosed when harmfully treated, whether with starvation or intensive lowering of blood glucose, but again, this does not seem to capture the central concern about overdiagnosis.
Carter et al's definition may include diseases just because they currently lack effective treatment. For many patients, the diagnosis of Huntington's disease may cause significant psychological harm that outweighs any benefit. If empirical research with affected people found this to be the case, Huntington's disease would meet the authors' definition of overdiagnosis. Yet it is unlikely that leaders in the field would accept this as overdiagnosis because Huntington's disease is not a harmless condition. Any diseases with high burdens of morbidity and mortality but which currently lack effective interventions may meet the authors' definition yet fail to align with prevailing accounts of overdiagnosis. And there are further situations where a correct diagnosis might cause more overall harm than benefit, such as cancer diagnoses in resource-poor contexts where there is no access to treatment.
Overdiagnosis should be defined in a way that separates it from other examples of harm following a correct diagnosis. In our account, we explicitly recognise that the correct diagnosis of a condition may be on balance more harmful than beneficial for reasons other than overdiagnosis.8 These include side effects, complications or overtreatment. We argue that overdiagnosis can be distinguished from these other situations by linking overdiagnosis to the failure to track harmful disease, consistent with the current meaning-in-use in the literature. Carter et al may claim that our examples of correct diagnoses leading to harm would be excluded on their account because the overall balance of benefits and harms would be favourable at a population level, yet this seems to be the wrong justification, because it allows any condition with a treatment that has more harms than benefits to fall within the definition of overdiagnosis. In contrast, overdiagnosis should be linked to loss of prognostic value: in conditions that are overdiagnosed, the diagnosis fails to distinguish harmful from non-harmful instances. Our definition focuses solely on diagnosis: ‘overdiagnosis has no additional power over simple description of the relevant physical findings or symptoms/experiences’ (ref. 8, p. 585). We do not rely upon including the consequences of related interventions in a balance of benefits and harm.
A second potential problem with the authors' reliance on a negative balance of benefits and harms, rather than the notion of non-harmful disease, arises if the expected harms following diagnosis and intervention change. Current assessments of overdiagnosis may apparently dissolve with changes in therapy. If the treatment for screening-detected breast cancer were a 5 min warm shower, then the calculation of benefits and harm would alter considerably. In such a situation, the diagnosis of screening-detected breast cancer might be deemed overall to be beneficial, and therefore, no longer a case of overdiagnosis on the authors' account. Yet we contend that even if no harm ensues, breast cancer overdiagnosis exists, because at least some of the treated breast cancers are instances of harmless disease. Likewise it might be possible that benign effective treatments emerge for prostate cancer, attention deficit hyperactivity disorder (ADHD), gestational diabetes, chronic kidney disease (CKD) and so forth, but this does not mean that these conditions would no longer be overdiagnosed. Rather, it would mean that there is less harm from the overdiagnosis than before. But the advent of harmless interventions would not alter the population-level feature suggestive of overdiagnosis which is rapidly rising incidence with little or no changes in mortality, due to the identification of cases of non-harmful disease.7
How we define disease is central to concerns about overdiagnosis. Carter et al eschew an objectivist approach to disease in favour of a professional community standard. While we recognise problems with objectivist accounts of disease, we question the apparent assumption that professional communities reject some kind of realism in developing disease definitions. More importantly, the authors' definition hinges upon the notion that a diagnosis is correct according to a relevant professional standard. This seems problematic, as many claims about overdiagnosis9–11 are actually a protest against a relevant professional community standard. These protests are underpinned by a view that the relevant professional community is wrong; that their definitions (of gestational diabetes, of CKD, of ADHD and so forth) do not accurately track the condition in question. Thus contra the authors, it seems that many in the field do hold on to some kind of disease realism. The realists may be wrong of course, but it seems counterintuitive to allow for diagnostic correctness to be determined in the absence of any underlying conception of disease when this runs contrary to claims about overdiagnosis.
The lack of clarity as to who or what counts as a relevant professional community is potentially problematic. This opens the way to multiple and competing claims about index conditions from different professional communities. Indeed, the authors refer to the problem of conflict between professionals (p. 3) but do not offer any method for adjudicating between competing claims about correct diagnosis, which thereby undermines their definition. We agree that many disease boundaries are vague (Rogers W, Walker MJ. The line-drawing problem. J Med Phil 2016 (forthcoming), making it inevitable that diagnostic criteria must be specified, and that at the moment, professional groups perform this role with little scrutiny. However, as well as clarifying disease definition processes, we also need clarity as to who has the authority to determine diagnostic boundaries.
Our final concern with the author's definition lies in their appeal to population-level benefits and harms. We agree that it is impossible to identify which particular individuals are overdiagnosed in a population. But it is possible to estimate the scale of the problem by the discrepancy between rising rates of diagnosis and unchanging mortality and/or instances of advanced disease. Prima facie this is evidence that some or many of the diagnoses are of harmless disease. It is not clear what a population-level calculation of harms and benefits adds to this, and as we have argued above, including the harms and benefits of associated interventions further complicates the picture. The authors appeal to utility, claiming that a population-level definition will be more useful than an individual-level definition (ref. 1, p. 4). However, requiring a population-level calculation of harms and benefits seems well nigh impossible in practice, as individuals vary considerably in their valuing of different aspects of healthcare.
Turning to the ethical typology, we agree that degree of harm and the extent to which it can be avoided point to important ethical considerations, as do actors' motives. Nonetheless, we disagree about some of the particulars. Predatory overdiagnosis is explained using the promotion of a new therapeutic indication created for an existing drug. In this example, a harmful treatment for a concocted condition is taken up by a self-interested group of practitioners. This is clearly ethically wrong, but we would hesitate to claim that drug company employees and avaricious practitioners constitute a relevant professional community with the authority to mandate a ‘correct diagnosis’. The description here seems to be of disease-mongering rather than overdiagnosis, in which case the diagnosis may well not be accepted as correct by a relevant professional community. In our discussion of this distinction, we argue that disease-mongering may overlap with some forms of overdiagnosis, such as that created by the expansion of disease categories (‘misclassification overdiagnosis’ on our typology), but the example in the authors' paper is of disease-mongering and medicalisation rather than overdiagnosis (ref. 8, p. 583).
Tragic overdiagnosis involves unavoidable harms despite beneficent motivations. In the neuroblastoma example, the underlying cause of the overdiagnosis is lack of knowledge; the harm became apparent only in the light of research evidence. This raises questions about the boundary between tragic and misdirected overdiagnosis. At what point is there sufficient research evidence to claim that any further overdiagnosis is misdirected or predatory rather than tragic? The authors note that these categories blur, but given the current state of knowledge about the situations in which overdiagnosis occurs, this may not be a useful category. Responsible and well-informed practitioners should now be sufficiently aware of the potential for overdiagnosis such that they cannot claim tragic ignorance.
We provide another way of thinking about the potential blameworthiness of particular cases of overdiagnosis related to intentions and knowledge, but without making categorical ethical judgements.8 Our typology makes a distinction between maldetection overdiagnosis that occurs as the result of an epistemic gap (the inability to differentiate aggressive from indolent screening-detected disease), and misclassification overdiagnosis that arises through the intentional widening of disease boundaries. Maldetection overdiagnosis shares some features with tragic overdiagnosis, in that those seeking to diagnose may have beneficent motives; the harm ensues because there is no way of sorting the harmful from the harmless disease. But as overdiagnosis can be predicted, we prefer to avoid a term such as ‘tragic’. Our discussion of misclassification overdiagnosis identifies the role that motives may play when professional communities or expert committees determine disease boundaries; but we prefer our descriptive category given the impossibility of accurately assessing the motives of a group of individuals.
In the authors' typology, clinicians, corporations and states are the actors whose motives regarding overdiagnosis are subject to assessment. At one level, this makes sense as without these actors, there can be no overdiagnosis. But this approach ignores the motives of patients, who may be equal actors in at least some cases of overdiagnosis. It is now current practice in Australia not to promote prostate cancer screening and only to offer men the opportunity to discuss this should they so wish. Thus it seems possible that future prostate cancer overdiagnosis might be motivated by patients, reflecting their values and preferences, rather than the motives of clinicians or corporations. The authors may counter that if, at a population level, men prefer prostate cancer screening, then this alters the balance of benefits and burdens and prostate cancer overdiagnosis no longer exists. But this does not alter the fact that many cases of prostate cancer are harmless, making their diagnosis overdiagnosis, regardless of who triggers the diagnostic process.
In conclusion, Carter et al advance the debate about defining overdiagnosis in interesting and provocative ways. In our view, their definition does not accurately map the central concern about harmless disease identified by other scholars in the area, but we look forward to further discussion of this and other points raised here.
Contributors WAR and YM both developed the critical analysis. WAR led on the drafting. YM reviewed drafts for revisions. Both authors signed off on the final version.
Funding WAR is supported by an Australian Research Council Future Fellowship (FT130100346).
Competing interests None declared.
Provenance and peer review Commissioned; internally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.