The problem of over-diagnosis

(cross posted at freeforall)

Aaron Carroll has a good post on over-diagnosis. The paper in the BMJ about which Aaron is writing defines over diagnosis this way:

Narrowly defined, overdiagnosis occurs when people without symptoms are diagnosed with a disease that ultimately will not cause them to experience symptoms or early death. More broadly defined, overdiagnosis refers to the related problems of overmedicalisation and subsequent overtreatment, diagnosis creep, shifting thresholds, and disease mongering, all processes helping to reclassify healthy people with mild problems or at low risk as sick.

The problems of over-diagnosis include health impacts of treatments, stigma from labeling, worry and cost. The essence of the problem is that these costs (monetary and intangible) are increasing while the benefits of any treatment available are not (mortality reduction, morbidity decline). The figures from the paper provide examples in which death rates from certain Cancers are stable over time, while cases are increasing, typically due to improved ability to detect disease and/or more attempts to do so via broader screening. Below I reproduce just one of the 5 graphs demonstrating this phenomenon with Thyroid Cancer:

Increased diagnosis suggests an explosion of cases, while death from this cause are flat. This epidemiology can produce different survival rates across nations, for example, as Aaron clearly describes in his post if different nations diagnose disease earlier; but early diagnosis only helps if it provides the route to an effective treatment. Thus, in many cases, differences in survival do not really provide a means of capturing any net benefits of treatment. However, our cultural default is to assume that more is better, and to wonder how any information from a test hurt? And even asking the question of whether some efforts to detect and/or treat disease are worth it will bring howls of protest and accusations of your wanting to kill someones grandmother in the U.S. political context.

There is no technical way to address this problem. What is needed most is cultural change, that first and foremost leads each of us, and us collectively, to ask the following questions about any test or treatment:

  • Does it extend life?
  • Does it improve quality of life?
  • How much does it cost? (with a broad notion of costs, not only financial)

Only by answering these 3 questions can you have a hope of determining whether something is “worth it.” It is often hard to answer these questions well, or clearly, and I don’t mean to minimize this. However, the first step is to learn to even ask them.


Author: Don Taylor

Don Taylor is an Associate Professor of Public Policy at Duke University, where his teaching and research focuses on health policy, with a focus on Medicare generally, and on hospice and palliative care, specifically. He increasingly works at the intersection of health policy and the federal budget. Past research topics have included health workforce and the economics of smoking. He began blogging in June 2009 and wrote columns on health reform for the Raleigh, (N.C.) News and Observer. He blogged at The Incidental Economist from March 2011 to March 2012. He is the author of a book, Balancing the Budget is a Progressive Priority that will be published by Springer in May 2012.

13 thoughts on “The problem of over-diagnosis”

  1. What are you intending to argue here? Because the data you present is not persuasive to a case of over diagnosis. Rather it could be used to argue that improved early detection has drastically improved the outcome of the diagnosed patient. I don’t see how you can tell the difference between 1) early diagnosis identifies people who can therefore get more effective treatment, and 2) early diagnosis identifies many people who would never have gotten sick and treats them. The first is a boon, the second a problem. I think there are epidemiological approaches that can help, but this doesn’t.

    1. The BMJ article from which the graph came cites evidence from randomized trials which compare screened and unscreened populations, which ought to show that screened populations have lower death rates from screened cancers, but fail to show this difference. The thyroid cancer graph was given as an example of a different but related phenomenon, namely, the trends in diagnosis and in mortality from 1975 to 2005, showing that even though rates of diagnosis had risen due to more diagnostic interventions, the death rates had not declined in the same time frame.

      Prostate cancer screening has been back in the news recently. There is one other issue with PSA testing which deserves special attention. has a summary of a case in which a resident followed prostate cancer screening guidelines but was later sued because evidence-based guidelines do not stand up well in court; the community standard of care in Virginia was to order PSA on everyone over 50.

      Tests like PSA are subject to two kinds of bias which may make them appear to save more lives than they actually do. One is that they detect disease earlier, which means that the time from diagnosis to death is increased (greater survival times) even if the early diagnosis and intervention do not change the natural history of the disease. The second kind of bias results from the fact that slow-growing (and therefore less aggressive) tumors are over-represented in periodic screening tests. Rapidly growing tumors, the ones that actually kill you, are easily missed in periodic screening because they are undetectable in one testing period and have already progressed significantly when they are picked up at the next screening period.

      One of the factors that threatens to bankrupt the country is the overdoing of so many available interventions. The BMJ article focuses on just one, namely overdiagnosis.

    2. @Paul Orwin
      the figures are based on a meta type analysis of RCTs of available therapies, so the argument is that there is true over diagnosis (identifying cases whose course couldn’t be positively altered by therapy. Here is source paper that the new BMJ paper cited

      There are several good questions down this thread, esp about difficulty in assigning cause of death and dying ‘with’ something, and the paper noted above addresses all these issues. To summarize, my argument: (1) there is overdiagnosis; (2) this is a very hard concept for Americans (especially) to get our heads around; and (3) there is no technical solution….meaning we will have to learn to discuss limits openly to have a hope of muddling through a little better in this area.

  2. In an aging population, a constant rate of thyroid cancer deaths per capita is actually fairly impressive. I don’t suggest doing the work for a blog post (and there is no way I would do it) but an even more interesting graph would show age adjusted death rates. This is very important since earlier diagnosis causing a 10% drop in death rates 8masked by a 10% increase due to population aging) would be very desireable.

  3. The overall point of this post is an important and right one.

    But I also question the usefulness of the graph presented. In addition to the points already made here, I’d add that attributing cause of death to a particular disease is sometimes tricky, especially with many cancers. If an undiagnosed tumor in organ A metasticizes and new tumors of organ B appear, then it’s easy to see how the subsequent death of the patient would be classified as being caused by “Cancer of the B.” If the patient had been diagnosed as having “Cancer of the A” before tumors appeared in B, then its more likely that the cause of death would be classified as A. If we were diagnosing more Cancers of the A, but not doing anything to improve survival rates from Cancer A, then you expect “classified deaths” to go up, ceteris paribus. The fact that they are flat may indicate that diagnosis is leading to effective treatments, but that this positive is canceled out in the death rate number by the fact that physicians can now link more deaths to the original cancer.

    1. While second cancers can certainly occur in patients already diagnosed with one cancer, the main problem with reading the graphs arises from using the same scale on the vertical axis (rates per 100,000 people) for both the cancer diagnoses and cancer deaths. Since deaths are much less common than diagnoses, the graph does not provide much room to show any decrease in death rates; those rates are crowded into a space near the floor of the graph, potentially making differences in death rates more difficult to see graphically. There are ways to work around this rather common problem in graphic displays of data, but the BMJ did not employ any of them for this particular graph.

  4. Cultural change? Really?

    We have an institutional structure that rewards and promotes predatory medicine. Every technical advance is dressed up, culturally, in righteous rationalizations, but used to extract more money from a system, which is already consuming twice the fraction of GDP that it should, even in gold-plated form, and delivering lousy outcomes, from high rates of medical bankruptcy to stagnant or declining life expectancy.

    The quibbling about statistical adjustments and graphical presentation is not unimportant, but like “culture” in general, it is subsidiary. Yes, our best boffins should be thinking in terms of extending life and life quality, and curing or preventing diseases that compromise the life quality and expectancy of the young ought to get much more weight that the cancers and other diseases that accompany a normal old age. And, that more sophisticated thinking ought to percolate down into, and inform the general consensus and conventional wisdom. It ain’t going to happen, though, if it isn’t being financed by, and used to govern, our institutions of medical research and care.

  5. Dr. Nortin Hadler’s outstanding book “The Last Well Person: how to stay well well despite the health care system” was my intorduction to this concept he and it has had quite a profound effect on my thinking. His three questions to ask before accepting any screening, alone, could slash billions from US sickness creation and improve outcomes greatly.

  6. Oh, and this:

    All of medicine and medical economics is whistling past an open grave that is going to widen into an abyss as the limits to growth and the end of business as usual kick in. When you have so many complex systems emerge (no design) in ways that are totally dependent on cheap energy to permit interest based financing, the end of cheap energy and the start of wrenching contractions is fateful.

  7. There is a question that comes even before these: “Is the information from this test going to make a difference to this person’s treatment?” Especially for indolent diseases, that answer is way too often “no.”

Comments are closed.