Overdiagnosis in primary care: framing the problem and finding solutionsBMJ 2018; 362 doi: https://doi.org/10.1136/bmj.k2820 (Published 14 August 2018) Cite this as: BMJ 2018;362:k2820
- Minal S Kale, assistant professor1,
- Deborah Korenstein, chief of the general internal medicine service2
- 1Division of General Internal Medicine, Icahn School of Medicine at Mount Sinai, New York, USA
- 2Department of Medicine and Center for Health Policy and Outcomes, Memorial Sloan Kettering Cancer Center, New York, NY 10017, USA
- Correspondence to: D Korenstein
Overdiagnosis, is defined as the diagnosis of a condition that, if unrecognized, would not cause symptoms or harm a patient during his or her lifetime, and it is increasingly acknowledged as a consequence of screening for cancer and other conditions. Because preventive care is a crucial component of primary care, which is delivered to the broad population, overdiagnosis in primary care is an important problem from a public health perspective and has far reaching implications. The scope of overdiagnosis as a result of services delivered in primary care is unclear, though overdiagnosis of indolent breast, prostate, thyroid, and lung cancers is well described and overdiagnosis of chronic kidney disease, depression, and attention-deficit/hyperactivity disorder is also recognized. However, overdiagnosis is a known consequence of all screening and can be assumed to occur in many more clinical contexts. Overdiagnosis can harm patients by leading to overtreatment (with associated potential toxicities), diagnosis related anxiety or depression, and labeling, or through financial burden. Many entrenched factors facilitate overdiagnosis, including the growing use of advanced diagnostic technology, financial incentives, a medical culture that encourages greater use of tests and treatments, limitations in the evidence that obscure the understanding of diagnostic utility, use of non-beneficial screening tests, and the broadening of disease definitions. Efforts to reduce overdiagnosis are hindered by physicians’ and patients’ lack of awareness of the problem and by confusion about terminology, with overdiagnosis often conflated with related concepts. Clarity of terminology would facilitate physicians’ understanding of the problem and the growth in evidence regarding its prevalence and downstream consequences in primary care. It is hoped that international coordination regarding diagnostic standards for disease definitions will also help minimize overdiagnosis in the future.
RG is a 48 year old woman who sees a new primary care physician. She has a body mass index of 30 and reports that her mother has diabetes. The physician is concerned about RG’s obesity and, given her age and family history, orders a hemoglobin A1C (HbA1C) test to screen for diabetes. Her HbA1C is 6.0% and the patient is told she has pre-diabetes. She is encouraged to modify her diet, increase her exercise, and consider a drug to prevent diabetes and its complications. She makes a few changes to her diet and starts taking metformin. She experiences side effects from the metformin but is happy when her next HbA1C test result is 5.8%. Has this patient experienced an early diagnosis or overdiagnosis?
In our healthcare systems, people interact with their primary care providers to engage in patient centered, comprehensive, and continuing care that includes disease prevention/health promotion, education and counseling, and diagnosis and treatment of new or ongoing problems.1 The specter of overdiagnosis, largely unrecognized and unmentioned, lurks behind much diagnostic testing performed in primary care.
Overdiagnosis is defined as the diagnosis of a condition (often subsequently treated), that would otherwise not cause symptoms or harm to a patient during his or her lifetime.2 It has long been recognized as a consequence of cancer screening but in recent years has increasingly been acknowledged as an important consequence of any diagnostic testing in the absence of symptoms.345 However, overdiagnosis remains the “elephant in the examination room” for a variety of patient, provider, and sociocultural reasons, including confusion surrounding the meaning of the term, lack of awareness outside the realm of cancer screening, and expectations about healthcare. Because primary care is delivered, at least ideally, to the entire population, the problem of primary care related overdiagnosis is of particular importance in the realm of public health.
In this review, we present an evidence based summary of the scope of overdiagnosis related to screening of asymptomatic patients in primary care and provide clinical examples of how overdiagnosis is manifest in the primary care setting. We also clarify definitions of overdiagnosis and related terms, discuss drivers and consequences, and clarify the ways in which it can be documented and studied. Lastly, we suggest future steps to guide research and the implementation of discussions about overdiagnosis in primary care.
Sources and selection criteria
We began with a review of the literature to understand the scope of the published evidence related to overdiagnosis in primary care, defining overdiagnosis as identification of a condition that would not cause clinical harm during the patient’s lifetime.6 We searched PubMed, Embase, and Web of Science on 29 August 2017 using combinations of the following terms: “over diagnosis”, overmedicaliz(s)ation, “medical overuse” [MeSH], “general practice”, “primary health care” [MeSH], and “general practice” [MeSH], with no language restriction. In addition, we hand searched references from relevant identified articles. We excluded letters and articles published only in abstract form. We excluded articles in which the definition of the term “overdiagnosis” differed from our operational definition. After reviewing 582 titles and abstracts, we identified 71 publications related to overdiagnosis in primary care, including primary studies (n=19), systematic reviews and meta-analyses (n=5), narrative reviews (n=26), editorials (n=14), guidelines (n=4), and other (n=3).
To understand the prevalence of overdiagnosis more fully, we performed a second search of the same three databases on 8 November 2017 to identify studies quantifying overdiagnosis but not limited to primary care. The second search identified 1073 papers, of which 46 quantified overdiagnosis. After reference tracking we added eight papers for a total of 54 papers.
In papers from our first search (overdiagnosis in relation to primary care), cancer was most commonly mentioned, specifically prostate (n=16) and breast (n=12), with fewer discussions of lung (n=5), colon (n=4), and thyroid (n=2) cancers (appendix table A). We also found articles about chronic kidney disease (n=5), depression (n=4), neuroblastoma (n=3), and attention-deficit/hyperactivity-disorder (n=5), with additional diseases mentioned in a single paper (appendix table B). From our search of overdiagnosis quantification, which was not limited to primary care, we found that overdiagnosis of breast cancer was most commonly measured (24 papers), followed by prostate cancer (n=10), lung cancer (n=8), and thyroid cancer (n=4). Other cancers and non-cancer problems were covered in eight studies (appendix table C). The detailed content of the identified papers informed our analysis and discussion.
Context of overdiagnosis
In the case scenario, RG’s diagnosis of pre-diabetes may represent overdiagnosis. Although it is impossible to identify overdiagnosis at the point of care with an individual patient, the potential for overdiagnosis can be understood on the basis of the evidence of benefits and harms of screening. In the case of screening for diabetes, only recently has limited direct evidence shown that screening improves clinical outcomes; in one European study of 1 900 000 patients, screening for diabetes in high risk people was associated with statistically significant reductions in all cause mortality and cardiovascular disease events at 10 years (hazard ratio 0.84 (confidence interval 0.80 to 0.89) for cardiovascular events).7 Randomized trials have shown that pre-diabetes interventions (lifestyle and drugs) can prevent or delay the progression to type 2 diabetes.8 However, a meta-analysis of prospective observational trials of progression rates from pre-diabetes to diabetes found that more than half of people with pre-diabetes do not have diabetes after 10 years, suggesting that this condition often does not progress to clinically important disease, so that treatment in this situation would fulfill the definition of overdiagnosis.9
Given the high prevalence of diabetes, the many complications of untreated disease, and the benefit of screening at a population level, screening for diabetes is recommended in high risk groups in many countries including the United Kingdom,10 Canada,11 and the United States,8 despite the potential for overdiagnosis. In RG’s case, screening was consistent with guidelines from these countries. The case of pre-diabetes illustrates the potential for overdiagnosis that exists in any screening scenario. Overdiagnosis was first described in relation to cancer screening.12 However, the potential for overdiagnosis accompanies screening for non-cancer conditions including hypertension, hyperlipidemia, and even a “screening” physical examination in an asymptomatic person. Theoretically, the potential for overdiagnosis would be eliminated if important disease that would threaten health and require treatment could be accurately identified. However, perfect tests do not exist and all diseases occur along a clinical spectrum, so overdiagnosis remains.
Overdiagnosis, then, is inherent to the modern practice of healthcare, which seeks to diagnose and mitigate disease before it is clinically evident. Figure 1 illustrates patterns of disease progression after diagnostic testing. The test in question could be an imaging test, a laboratory test, a component of the physical examination, or even an interview question. Tests are intended to identify clinically meaningful disease (green line), which if untreated will go on to threaten health or even become fatal. Treatment of these diseases will improve health. A subset of disease, however, will progress more slowly and will not threaten health before the patient’s death from other causes (purple line). Finally, some disease will never progress at all (pink line). The identification of patients with slowly progressive or non-progressive disease represents overdiagnosis because treatment will not improve these patients’ health will but still expose them to potential harms.
Some level of overdiagnosis is unavoidable, owing in part to the unacknowledged trade-off between minimizing underdiagnosis and tolerating overdiagnosis. To optimize health, we seek to reduce underdiagnosis, or the failure to identify a disease that ultimately threatens a person’s health.13 Ideally, screening limits the underdiagnosis of early disease that is destined to progress. When determining the effectiveness of screening, we evaluate the balance between clinical benefit (such as improved mortality) and clinical harm (such as complications from diagnostic tests and treatment) in a screened population. Going back to our case of pre-diabetes, a few studies show that screening for diabetes leads to lower mortality and mixed evidence suggests that tighter control of type 2 diabetes leads to a reduction in some macrovascular complications.14 Although overdiagnosis of both diabetes and pre-diabetes is a consequence of screening, guideline panels believe that harms related to overdiagnosis are offset by the population level benefits of early diagnosis (and subsequent treatment) of true disease and therefore recommend risk assessment and targeted screening. Limiting screening to people at the highest risk can minimize (though not eliminate) overdiagnosis while maximizing benefit. However, to reduce overdiagnosis and its resultant harms further we must also understand the factors that drive it.
Drivers of overdiagnosis
Several factors work independently and together to encourage overdiagnosis (table 1).
Broadening disease definitions
Many diseases exist on a severity spectrum. Disease cut-off points must be selected (for example, the blood pressure at which hypertension is considered) and are often chosen to minimize underdiagnosis, sometimes at the expense of high rates of overdiagnosis. Figure 2 illustrates the impact of different diagnostic thresholds on numbers of patients diagnosed, harmed from disease, and harmed from treatment. In the absence of screening, many patients have delayed diagnosis, all patients have the potential to be harmed by their disease, and few are harmed from treatment of the disease. When screening is performed with a high diagnostic threshold for disease (for example, screening for diabetes with a high glucose cut-off point), more patients are diagnosed compared with no screening, fewer have clinical harm from disease, somewhat more are potentially harmed by treatment (because more are treated), and a small amount of overdiagnosis occurs. Finally, screening with a low diagnostic threshold (for example, screening for diabetes with a low glucose cut-off point) leads to more patients being diagnosed with the disease, many patients being overdiagnosed, more being harmed from treatment, and the fewest being harmed by untreated disease.
Going back to our case, RG has been diagnosed with pre-diabetes, a relatively recently defined “disease.” Traditionally, glucose cut-off values defining diabetes were based on the risk of retinopathy.29 However, a desire to prevent other complications of diabetes led to the establishment of various diabetes precursor conditions such as impaired fasting glucose, impaired glucose tolerance, and borderline HbA1C values, with cut-off values defined on the basis of the risk of progression to frank diabetes.30 The creation of these “at risk for diabetes” categories led to the creation of the umbrella term pre-diabetes, which in effect broadened the population of patients diagnosed with diabetes in some form. However, the diagnostic tests used to identify pre-diabetic conditions have variable accuracy for predicting diabetes and different performance characteristics in different populations.31 Reliance on the fiction of a “one size fits all” test with a low diagnostic threshold to diagnose pre-diabetes (such as HbA1C of 5.7%) leads to more diagnoses of pre-diabetes with fewer of these people developing diabetes. Moreover, expanding the spectrum of diabetes to include pre-diabetes promotes the false idea that untreated pre-diabetes will universally lead to diabetes. The notion of a “pre” disease condition is not limited to diabetes; it is also discussed for conditions such as osteopenia32 and for cardiovascular risk factors such as hypertension and hyperlipidemia.33 In the case of hypertension, the expansion of guideline definitions of hypertension to include lower systolic blood pressures has identified patients with a lower risk of poor cardiovascular outcomes who have “hypertension,” thereby subjecting them to treatment that is unlikely to benefit them.
The increasing availability and use of advanced technology also contributes to overdiagnosis. For example, in the eight years after high resolution computed tomography pulmonary angiography became available to diagnose pulmonary embolism, the number of cases doubled compared with the previous five years.34 In the absence of overdiagnosis, an increase in the number of cases should have led to fewer deaths from untreated pulmonary embolism. However, mortality did not change over that time period; this implies that the additional cases of pulmonary embolism were clinically insignificant and overdiagnosed.34
Incidentalomas provide another example of the impact of the widespread use of advanced imaging. About 5-15% of all abdominal imaging tests performed in asymptomatic people contain incidental findings.35 Despite recognition that most of these findings are of little clinical significance, their presence often triggers a cascade of unnecessary sequential diagnostic tests. On the rare occasion that an incidentally found abnormality appears to be clinically important, clinicians or patients may erroneously conclude that more imaging saves lives.
Worry from incidental findings is heightened when the incidental finding represents a potentially fatal tumor, such as thyroid cancer. Since the introduction of neck ultrasound in the 1980s, thyroid imaging has become common and the incidence of thyroid cancer has increased globally, mainly as a result of a rise in small papillary carcinomas, with little change in mortality from thyroid cancer. In this context, nearly half of thyroid cancers in men and more than 80% in women in high resource settings are estimated to represent overdiagnosis, amounting to more than 500 000 cases.36 The problem of incidental findings of unclear importance may be compounded by genetic testing, both in clinical and direct-to-consumer settings, the explosive growth of which may result in widespread overdiagnosis owing to the identification of gene carriers who may never develop disease.37
Public health screening programs
The use of screening programs as a disease prevention and control strategy is an important driver of overdiagnosis. By design, screening programs presume that a reservoir of undiagnosed disease exists and that screening will lead to a lower clinical impact of, or mortality from, that disease. However, a screening program will identify all disease along the spectrum of clinical severity and may tend to detect more indolent disease, with little to distinguish between severe and indolent disease.38 Screening therefore always results in some degree of overdiagnosis. Screening for breast cancer provides an important example. Estimated rates of overdiagnosis of breast cancer with screening vary across countries and populations, but the presence of overdiagnosis is universally accepted.39 A UK panel estimated that 11% of screen detected breast cancers represent overdiagnosis,40 and rates of overdiagnosis of breast cancer in Europe are estimated to range between 1% and 10%.22 Within the US system there is more overdiagnosis in higher income populations, which may reflect better access to care, particularly screening.41
Culture around medicine and health
In many countries there is great public enthusiasm for cancer screening and a relative lack of concern about the potential for overdiagnosis. In a study of the general US population, most adults believed that routine cancer screening is almost always a good idea, with 75% also believing that finding cancer early saves lives most or all of the time.42 A more recent survey assessed people’s tolerance of overdiagnosis in the context of effective screening. Although responses varied widely and there is no “correct” answer, participants tolerated a median of 113-150 cases of overdiagnosis per 1000 people screened to save one life.43 Notably, most survey respondents reported never having previously heard of overdiagnosis despite more than half having been screened for cancer.
Reimbursement structures may also drive overdiagnosis. In fee for service healthcare systems direct financial incentives may encourage testing regardless of clinical appropriateness. In the US, the ownership of imaging equipment by physicians is associated with more testing with similar clinical outcomes, implying that financial incentives motivate clinical behavior towards excessive testing.26 Similarly, a lower share of public health expenditure (that is, more reliance on direct payments from individuals and private health insurance) has been found to be associated with a higher incidence of thyroid cancer across different healthcare systems, with no difference in mortality from thyroid cancer,44 which again suggests profit driven testing.
Financial incentives can also operate through more complex mechanisms involving the drug and medical device industries, which may contribute to overdiagnosis in several ways. Firstly, industry may seek to expand drug markets by working to broaden disease definitions to create more patients who are eligible for certain drugs through influencing guidelines (for example, guidelines for lipid lowering), diagnostic criteria (such as in the Diagnostic and Statistical Manual of Mental Disorders (DSM)), and the content of medical education.45464748 In countries that allow direct-to-consumer advertising, drug advertising may obscure the line between disease and normal variants, driving patient requests for prescriptions. Furthermore, disease awareness campaigns may serve as proxies for efforts to increase patient requests for diagnostic testing for the disease in question and ultimately for drugs to treat it.49
Limitations in evidence application
Evidence based medicine, the application of the best medical research to the clinical care of individual patients, is understood to optimize medical decision making and improve patient outcomes.50 However, problems with physicians’ application of evidence and limitations in the evidence itself can contribute to overdiagnosis. Physicians have a poor understanding of quantitative information and test and treatment performance, which probably contributes to unnecessary testing and overdiagnosis.5152 Several studies have examined the role of cognitive biases and heuristics on medical decision making. Doctors’ susceptibility to decision making biases such as insensitivity to known probabilities, availability, and confirmation affects their ability to interpret information, including physical examination findings and diagnostic test results.5354 Some cognitive errors may contribute to overdiagnosis. One example is the representativeness bias, in which an individual is assumed to belong to or to be representative of a category (such as a disease) on the basis of similarity to other characteristics in the category. In practice, representativeness bias may lead a physician to overestimate the benefit of an intervention in a patient with a lower risk of disease, expecting that patient to experience similar benefit to one with more severe disease who is at higher risk of complications. Similarly, availability bias, the tendency for recent experiences to affect decision making, might lead a physician to pursue diagnostic testing aggressively because of a recent negative experience with a different patient in whom a diagnosis was missed. Many of these cognitive errors can lead physicians to overestimate the benefit of diagnosing and subsequently treating mild disease, thus pushing them to practice in a way that fosters overdiagnosis.
Important limitations with the evidence related to diagnostic test accuracy and the effectiveness of treatments also enable overdiagnosis. With regard to diagnostic testing, studies evaluating diagnostic accuracy often ignore the problem of disease spectrum, and high sensitivity tests may largely detect clinically insignificant disease,55 as seen with computed tomography pulmonary angiography, described above.28 Similar problems occur in studies evaluating the effects of treatments. Clinical trials generally report the average effect across a population; however, the average treatment effect applies poorly to individuals, with the benefit in a particular patient being related to disease severity.56 Valuing and applying the same intervention to people at lower risk of benefit to those at high risk may lead to treatment of overdiagnosed disease and unnecessary exposure of patients to potential treatment toxicities. By contrast, targeting high risk patients for intervention can maximize net benefit in the population. For example, re-analysis of data from the National Lung Screening Trial showed that using a risk of death from lung cancer based approach to screening would prevent the greatest number of deaths among those at highest risk, with very few deaths among low risk patients who would not be screened.57
Magnitude of the problem
The magnitude of overdiagnosis is unclear. The limitation in our understanding is in part attributable to the challenges of studying overdiagnosis and in part due to relative lack of attention to the problem in many clinical areas, particularly outside of cancer. Our literature review yielded few reliable estimates of rates of overdiagnosis in non-cancer clinical areas.
Overdiagnosis has been quantified for many cancers (table 2). The best evidence exists for prostate and breast cancers.2 In these two diseases, estimates of the proportion of disease that represents overdiagnosis vary widely across studies, reflecting both the challenges of quantifying overdiagnosis and the widely variable rates of overdiagnosis based on patient characteristics. Between 2.9% and 88.1% of prostate cancers have been estimated to represent overdiagnosis.
The scope of overdiagnosis in diseases other than cancer has not been well defined.61 Although screening for chronic non-cancer conditions such as hypertension, diabetes, and depression comprises much of the routine work of primary care, there are few estimates of the magnitude of overdiagnosis of these disorders. In our literature review, we found only a few studies that quantified overdiagnosis of non-cancer conditions, namely chronic kidney disease, and abdominal aortic aneurysm (AAA). In kidney disease, authors simply reported rising rates of diagnosis and suggested that the rise probably reflected overdiagnosis without quantifying the proportion of actual overdiagnosis.62 The paper on AAA used evidence on the risk of rupture to estimate rates of overdiagnosis for screen detected AAAs of various sizes; the number of AAAs representing overdiagnosis ranged from 11.5% for >54 mm aneurysms to 87% for 26-29 mm aneurysms.63
Consequences of overdiagnosis
Overdiagnosis has many consequences, some theoretical and others more quantifiable. The effects are multi-pronged, affecting the individual, healthcare system, and society at large.
Overtreatment refers to the unnecessary treatment of a condition. It occurs whenever overdiagnosed disease is treated and can affect the individual patient as well as the wider healthcare system. Overdiagnosed disease provides no opportunity for treatment benefit so the individual incurs only harms. These potential harms include direct negative consequences of the unnecessary treatment itself (such as a wound infection after thyroidectomy to treat an overdiagnosed thyroid cancer) and indirect harms related to the consequences of resultant downstream services (such as palpitations resulting from an incorrect dose of replacement levothyroxine after thyroidectomy). The individual patient is also affected by the opportunity costs related to treatment—for example, time away from usual activities while recovering from surgery.
The effects of overtreatment on the healthcare system and society are less obvious and challenging to measure. Given limited total capacity for healthcare delivery, overtreatment of one patient may limit healthcare access for another person who may truly need care, causing harm to people other than the overtreated patient, which on a broad scale amounts to societal harm.64 In addition, health system investment in overtreatment deflects resources from other pressing medical needs and represents lost opportunity to improve the health of the public.6566
It is difficult to estimate psychological harm from overdiagnosed disease because the patient often does not know the “disease” represents overdiagnosis. Patients with recognized overdiagnosed thyroid cancer who opted against aggressive care have reported anxiety and feelings of isolation and secret keeping,67 though evidence in other clinical settings is limited. However, the psychological ramifications of disease diagnoses probably apply to overdiagnosis as well as to legitimate diagnoses.68 New diagnosis of a chronic disease may require patients to adjust their life expectations and employment, owing to prognostic and functional considerations and matters related to disease treatment.69 Some may experience depression and anxiety after a new diagnosis.7071 For example, an international study of breast cancer survivors found an increased risk of suicide up to 25 years after diagnosis,70 and patients with ductal carcinoma in situ, which may represent overdiagnosed breast cancer, often experience persistent anxiety related to fear of recurrence and death.7273 In the context of screening, psychological harms related to unexpected diagnoses of prostate cancer and AAA have been described, though evidence for most diseases is lacking.7475
Patients can be affected by being “labeled” with a disease diagnosis in many adverse ways. In children, the diagnosis of a benign heart murmur can lead to unnecessary restrictions on activity.7677 In adults the impact of labeling is more difficult to ascertain. Interestingly, patients who undergo early imaging for mild acute low back pain, which is likely to reveal insignificant anatomic abnormalities, are more likely than non-imaged patients to be out of work with disability one year later.78 This suggests that the knowledge of an identifiable abnormality may affect people’s perceptions about their own health and engagement in society.
The potential financial harms of overdiagnosis are enormous and contribute to waste in healthcare systems.79 In the US, the cost of breast cancer overdiagnosis alone (both ductal carcinoma in situ and invasive breast cancer) in women aged 40-59 years has been estimated at $1.2bn (£0.91bn; €1.03bn) a year.80 As health systems around the world struggle to reduce unsustainably high costs,81 the elimination of overdiagnosis is an attractive way to save money without compromising, and in fact improving, public health.
Clarity of terminology is crucial to efforts to enhance the understanding of overdiagnosis and minimize its impact. Table 3 describes terms that are related to overdiagnosis and that may be confused with true overdiagnosis.
In our literature review, we noted that several of these terms are often conflated with overdiagnosis. Many articles using the term overdiagnosis were really describing misdiagnosis, defined as the diagnosis of the wrong disease.85 For example, in one cross sectional study of overdiagnosis of heart failure diagnosed by primary care practitioners, researchers found that one in six people probably did not have heart failure, but had another disorder instead.88 Although it is tempting to use the term overdiagnosis to describe the mislabeling of these patients (because the term conjures the concept of an excess of a diagnosis), this example reflects an inaccurate diagnosis and not true overdiagnosis. For patients incorrectly labeled as having heart failure, their symptoms (presumably breathlessness) were not caused by heart failure so the diagnosis was in fact a medical error.
Overdiagnosis and overmedicalization are also often conflated. The confusion regarding these terms is natural for semantic reasons, and they have overlapping themes, concepts, and drivers.4 In overmedicalization, a phenomenon that is part of the normal human experience is recontexualized as disease and (often) treated as such. Overmedicalization shares many of the same consequences of overdiagnosis (such as labeling, overtreatment, and excessive cost). For example, advances in medical technology and changes in culture have led to overmedicalization of the process of dying. This has increased the use of more intensive procedures among those who are dying, with no benefit to individuals, and has resulted in a rise in deaths in hospital.89
The term “disease mongering” is related to overmedicalization but it is used to describe situations in which outside forces, mainly the drugs industry, encourage overmedicalization for the purpose of creating new drug markets.9091 For example, premenstrual dysphoric disorder, first mentioned in the DSM fourth edition in 1994,9293 represents a severe form of premenstrual symptoms. A drug targeting premenstrual dysphoric disorder, Sarafem, was developed, approved by the Food and Drugs Administration, and heavily marketed to physicians and (in the US) consumers. However, Sarafem, which was approved by the FDA in 2001, was simply rebranded fluoxetine, which, in the form of Prozac, went off patent during that same year.94 Although many women experience severe menstrual symptoms,95 the relabeling of the experience as a disease is likely to have led to more widespread, and possibly unnecessary, drug treatment. Advertising is a powerful tool to facilitate disease mongering and foster overdiagnosis, as is shown in the case of Sarafem and others. In the US, direct-to-consumer advertising of testosterone replacement products was associated with testosterone testing and initiation of treatment with testosterone replacement therapy96; many of these cases probably represented overdiagnosis.
It is also important to distinguish misuse, or preventable complications of care, from overdiagnosis.82 Misuse may arise while treating an overdiagnosed disease, and some may argue that any treatment of such a disease is a case of misuse. However, given the difficulty in discerning overdiagnosis in a clinical scenario in real time, this is more of a theoretical concern. The growing emphasis on avoiding low value care is also relevant to the discussion of overdiagnosis. Low value care is defined as care that results in little benefit relative to its cost.8687 Although care related to overdiagnosis is inherently low value, not all low value care is related to overdiagnosis.
Finally, medical overuse is a term used by some authors interchangeably with overdiagnosis. Overuse is the use of unnecessary health services, either tests or treatments, for which potential harms outweigh potential benefits.82 The association between overuse and overdiagnosis is complicated (see fig 3). Overuse is broader than overdiagnosis and can be both a cause and a driver of overdiagnosis.
Figure 3 describes testing and screening scenarios that may lead to overdiagnosis, the consequences of overdiagnosis, and the association between overdiagnosis and the overuse of medical services (unnecessary testing and treatment). Both appropriate and inappropriate testing, as well as broadening of diagnostic thresholds, can lead to overdiagnosis, which in turn can lead to labeling and unnecessary testing and treatment.
There are several approaches to quantifying overdiagnosis; each has its own set of assumptions that lead to bias and limitations. These biases in part contribute to the wide ranges of rates of overdiagnosis among studied disease, as seen in table 2.
Table 4 describes approaches to quantifying overdiagnosis. Several methods are used and, confusingly, the incidence of overdiagnosis can be expressed as a proportion of screen detected cases or as a proportion of all cases of disease.9798 The excess incidence approach leverages long term follow-up of groups that had been exposed or unexposed to a diagnostic test, generally a screening test for cancer. Patients’ exposure to testing can occur within a randomized trial or in real world settings. After a lengthy follow-up time, the number of cancers in the screened group is compared with the number in the group that did not undergo screening; cancers in the control group would have been detected clinically or incidentally during other testing. After enough time, the number of cancers in the control group will catch up with that in the screened group—the two groups should have the same number of clinically important cancers over time but early detection as a result of screening should be associated with better prognosis in the screened group. Extra cancers in the screened group, the excess incidence, represent overdiagnosed cancers that were never destined to become clinically important. For example, a Canadian study compared rates of breast cancer over time before and after the introduction of population based screening.99 By assessing the increase in the total number of cancers after screening was introduced, the authors estimated that 5.4% of invasive breast cancers were overdiagnosed. The excess incidence approach is considered the most reliable method for estimating overdiagnosis, but it is resource intensive and requires very long follow-up.
A second approach to estimating rates of overdiagnosis, the lead time approach, uses modeling based on estimates of expected rates of disease. Known treatment patterns, response rates, rates of disease progression, and likelihood of mortality from competing causes are used to estimate predicted disease survival times and overall survival times with screening. In these models, the proportion of patients in the screened group predicted to die of their disease beyond their overall predicted life expectancy represents overdiagnosis. Investigators have used a lead time approach to estimate rates of overdiagnosis associated with lung cancer screening. Simulating rates of lung cancer development, progression, detection, follow-up, treatment, and survival, investigators estimated that a mean of 11.9% (range 5.5-23.2%) of screen detected cancers were overdiagnosed.100
The third and least accurate method for estimating rates of overdiagnosis relies on disease characteristics. This approach uses pathologic features of screen detected disease to predict future clinical behavior; disease that is predicted to never become clinically important (or to become important only after the patient is expected to die from other causes) defines overdiagnosis. For example, a study of prostate cancer reported an increase in small and very early stage tumors over time as screening rates rose; the proportion of these low risk tumors reflected the rate of overdiagnosis with screening.101 Although this method is informed by an understanding of the association between disease characteristics and prognosis, it involves many assumptions and offers only a rough estimate of rates of overdiagnosis, though future growth in biomarkers to distinguish indolent from aggressive disease may improve its accuracy.
Minimizing and managing overdiagnosis
Because overdiagnosis is an expected consequence of screening in asymptomatic people, some degree of overdiagnosis will persist. However, the ethical obligation to avoid patient harm compels physicians to minimize the prevalence of overdiagnosis,102 which can be done by improving the understanding of overdiagnosis, optimizing disease definitions, and considering overdiagnosis when making clinical decisions. We discuss these three goals and strategies to achieve them.
Firstly, enhancing the evidence base related to overdiagnosis would improve understanding. We need better estimates of rates of overdiagnosis of non-cancer conditions for which there is widespread screening in primary care, including hypertension, mental health disorders, and hyperlipidemia. Although there are methodological concerns, even imprecise estimates of rates of overdiagnosis of these common conditions would be helpful. In part the evidence could be advanced through required reporting of overdiagnosis in studies involving changes in diagnostic thresholds and in studies of new diagnostic tests or screening approaches.
Secondly, the medical community could minimize overdiagnosis by optimizing disease definitions. Because broadening of definitions of disease can lead to overdiagnosis, any changes should use a systematic, transparent approach where benefits and harms are explicit, especially when they lead to an increase in the prevalence of disease, and broadening of definitions should require evidence of clinical benefit. International standards for defining disease and altering disease definitions already incorporate these concepts; such standards could be institutionalized by professional societies and guideline developing organizations.103 Primary care providers could minimize overdiagnosis by avoiding unnecessary screening and testing; a more consistent approach to defining disease would facilitate best clinical practice and benefit patients overall.
Thirdly, overdiagnosis must be better managed. Guidelines related to screening examinations and changes to disease definitions must acknowledge the potential for overdiagnosis and attempt to quantify it. Discussion of possible overdiagnosis is already commonly recommended in cancer screening guidelines104105106 and has been alluded to outside of cancer; for example, in the National Institute for Health and Care Excellence hypertension guideline.107 Broader incorporation of concerns about overdiagnosis into guidelines would be facilitated by the inclusion of such concerns into guideline development standards from organizations such as the Institute of Medicine and the Guidelines International Network.60108
Primary care clinicians are also responsible for improving the management of overdiagnosis and can use several strategies. Firstly, clinicians must inform patients about overdiagnosis and incorporate it into clinical decision making. Currently, few patients who undergo cancer screening report discussing overdiagnosis with their physician, and it is likely that even fewer discuss overdiagnosis when they undergo screening for non-cancer conditions such as diabetes.109110 There are challenges to discussing overdiagnosis with patients; the concept may be difficult to understand and some may not recognize overdiagnosis as a real problem.111112 Patients who do appreciate the potential for overdiagnosis may be reluctant to raise concerns with their doctors. Patient decision aids can facilitate these discussions and enhance patient knowledge and informed screening choices.113 They are particularly suited to decisions related to population based cancer screening, where benefits and harms are similar across the population.65Figure 4 illustrates overdiagnosis in the context of benefits and harms of prostate cancer screening.
Primary care clinicians can also minimize overdiagnosis through thoughtful management approaches and referral practices. Conservative management of indolent disease that probably represents overdiagnosis can minimize harm from resultant overtreatment. Such strategies are recommended for early stage prostate and thyroid cancers,114115 and they are increasingly discussed in the context of breast cancer.116 A recommendation for conservative management from the clinician has a big influence on the patient’s decision117 and is likely to be particularly potent in the context of an ongoing primary care relationship, empowering primary care clinicians to optimize patient care. Similarly, clinicians can selectively refer patients to specialists who are similarly committed to minimizing overdiagnosis, and these consultants also have substantial influence.118
Finally, greater clarity about the term overdiagnosis, with a broadly shared definition, would contribute to awareness of the problem and uniformity of approaches to curtail it. Lack of agreement regarding the term overdiagnosis and its conflation with related phenomena such as overuse, overtreatment, and misdiagnosis reduces practitioners’ understanding of overdiagnosis and its implications in everyday clinical practice. The dissemination of this definition will be key; it will require the cooperation of journal editors and would be enhanced by outreach in the lay press.
It is impossible to know whether the diagnosis of pre-diabetes in our case scenario will lead to clinical benefit or if it represents overdiagnosis. However, it is likely that neither the patient nor her doctor considered overdiagnosis when the initial HbA1C test was sent. Better understanding of overdiagnosis by physicians and how best to manage it, along with an appreciation of the phenomenon by patients, will be important as we try to minimize both its prevalence and its harms in primary care.
How patients were involved in the creation of this article
To gain insight into overdiagnosis from the patient’s perspective, we solicited critical feedback from two members of the Memorial Sloan Kettering Cancer Center’s Patient and Family Advisory Council for Quality. They noted the importance of mentioning that patients may be reluctant to challenge their doctors about diagnostic testing, and this concept was incorporated into the manuscript in the Minimizing and managing overdiagnosis section.
Future research questions
How can new technologies be harnessed to identify biomarkers of cancers and conditions that are likely to represent overdiagnosis?
At what step in the clinical decision making process should overdiagnosis be discussed and what are best communication practises?
How should overdiagnosis best be explained to clinicians and patients to optimize understanding and minimize bias?
We acknowledge the contributions from Brooke Barrow and Antonio DeRosa.
Series explanation: State of the Art Reviews are commissioned on the basis of their relevance to academics and specialists in the US and internationally. For this reason they are written predominantly by US authors
Contributors: MSK and DK substantially contributed to the conception, analysis, data interpretation, manuscript drafting, and critical revision of this article. Both authors gave final approval of this version of the manuscript for publication. Both authors agree to be accountable for all aspects of the work to ensure that questions related to the accuracy and integrity of any part of the work are appropriately investigated and resolved.
Funding: This study was not funded. MSK is funded by a career development award (K07CA187071) from the National Cancer Institute of the National Institutes of Health. DK’s work on this project was supported in part by a Cancer Center Support Grant to Memorial Sloan Kettering Cancer Center (P30 CA0087848). MSK’s work was supported in part by an NIH grant (NCI K07CA187071).
Competing interests: The authors have read and understood BMJ policy on declaration of interests and declare that we have no interests.
Provenance and peer review: Commissioned; externally peer reviewed.