Overestimating Overdiagnosis


In a recent article in the New England Journal of Medicine[1], H. Gilbert Welch and associates draw from the Surveillance, Epidemiology, and End Results (SEER) program, spanning the years from 1975 to 2012, to estimate the extent of overdiagnosis in mammographic screening. Using yet another variation of indirect deduction, this time focusing on tumor size, it was determined that “only 30 of the 162 additional small tumors per 100,000 women that were diagnosed were expected to progress to become large,” implying that 81.5% of women were overdiagnosed by screening. This conclusion sets a record high for the calculated extent of overdiagnosis. Given that there are no patient-specific data about mammography use and compliance in these massive SEER reviews, one has to wonder how far these indirect methodologies can take us. Is the next stop 100% overdiagnosis?

While criticisms of the indirect methodology will be levied, then defended ad nauseum, one question always looms in this debate when discussing the overdiagnosis of invasive breast cancer (as distinct from DCIS) – why don’t we see evidence of these overdiagnosed tumors clinically? Of course, the epidemiologists have beaten that argument down to their own satisfaction by reminding us that these overdiagnosed tumors are all removed, making it impossible to directly measure overdiagnosis.

Still, what is happening at the level of the individual patient? At the level of histology and tumor biology? For such an allegedly pervasive phenomenon, there must be a clinical correlate. In fact, the only true overdiagnosis that can be directly documented would be more accurately labeled as misdiagnosis, e.g., when a complex sclerosing lesion is confused with invasive carcinoma, something that can trip up even the experts[2]. But this scenario doesn’t even begin to explain the staggering numbers for overdiagnosis being generated indirectly and likewise spilling into proposed informed consents for screening.

Using the strict definition of overdiagnosis – tumors that never progress – there are only two scenarios that could be happening at the histologic level, that is, tumor regression or, alternatively, tumor quiescence where the cancer simply reaches a certain size and stops. Either way, the tumor does not become clinically evident during the life of the patient.

Of these two possible explanations, the focus recently has been more on complete tumor regression, a shifty target since the evidence has already disappeared. Dr. Welch and associates will sometimes reference themselves in claiming that evidence supports regression[3], yet these are not studies where direct observation confirms that cancers melt away. Instead, the authors simply layer more indirect evidence on top of the indirect premise, creating a circle of self-validation.

And this is where a different perspective emerges through direct patient care. While there is some evidence histologically for “burned out” DCIS, the microscopic evidence from which to theorize the same for invasive regression is scant, at best. In contrast to the claim that we can only use indirect methods to study overdiagnosis, we clinicians have the occasional opportunity to witness tumor regression in the clinic by virtue of those women who refuse biopsy of screen-detected abnormalities. Or, after biopsy, patients refuse further treatment. Curiously, some will return for follow-up, often still declining further intervention. For us, this sequence happens once or twice a year, mounting to a fair number of instances over time. The collective experience of thousands of radiologists over decades seems to be similar to our facility where we have never seen a single case of tumor regression.

The more plausible explanation would be the option of tumor quiescence – that is, indolent tumors that reach a certain size and stop growing, remaining silent throughout the life of the patient. The easily accessible repository in which to document this phenomenon – directly – would be autopsy studies, in which these cancers should be evident in numbers out of proportion to expected disease prevalence. Oddly, there is so much momentum by the overdiagnosis bandwagon that it is not uncommon to hear it stated that autopsy studies of occult breast cancer show “pretty much the same” as what one sees in prostate cancer where overdiagnosis is readily evident. But this is not the case.

In the 1990s, there was enormous enthusiasm for the detection and treatment of DCIS, as it was felt that we could nip invasion in the bud every time. To counter this trend, an epidemiologist set about to show how common DCIS could be found in autopsy studies. Indeed, the review[4] indicated modest potential for overdiagnosis in DCIS (range – 0 to 14.7% in seven autopsy studies, drawn largely from the pre-mammographic era). As one would expect, the higher rates were seen when more slides were examined, the mean number of slides per breast in the studies ranging from 9 to 275.

Seemingly forgotten today, this comprehensive autopsy review secondarily included coverage of invasive disease as well. Given that disease prevalence in the U.S. for invasive breast cancer among the living is around 1% in an unscreened population, one would expect the autopsy data to generate a number well in excess of 1% if widespread tumor quiescence were at work. In fact, the discovery of occult invasive breast cancer in these seven autopsy studies ranged from 0 to 1.8%, with a median of 1.3%, a strong indictment against the alleged phenomenon of tumor quiescence. Of special interest, the lead author of this 1997 autopsy review was H. Gilbert Welch, cutting his teeth on DCIS, who many years later would move on to invasive disease. In 2012, Bleyer and Welch would claim that, over the past 30 years, 1.3 million women in the U.S. have been overtreated for pseudocancers of the breast, and that we continue at the rate of 70,000 per year[5].

If neither tumor regression nor tumor quiescence is at work, then most likely, the magnified illusion of overdiagnosis is occurring due to old-fashioned length bias with very long natural histories for many breast cancers. By using indirect methodology, these tumors appear as “excess cancers” when adjustments are not made accordingly.

Overdiagnosis is intertwined with length bias – the former being tumors that never progress, the latter relating to tumors that slowly progress. Yet, the two cousin concepts can be indistinguishable in large-scale population studies, depending on how the data is handled. And given the choice of picking overdiagnosis as one’s operative word versus length bias, which is the more powerful iconoclastic bombshell? “Overdiagnosis” sends shivers up the public spine, whereas “Length Bias” only draws a collective yawn.



[1] Welch HG, Prorok PC, O’Malley AJ, Kramer BS. Breast-cancer tumor size, overdiagnosis,, and mammography screening effectiveness. N Engl J Med 2016; 375:1438-1447.


[2] Elmore JG, Longton GM, Carney PA, et al. Diagnostic concordance among pathologists interpreting breast biopsy specimens. JAMA 2015; 313:1122-1132.


[3] Zahl PH, Maehlen J, Welch HG. The natural history of invasive breast cancers detected by screening mammography. Arch Intern Med 2008; 168:2311-2316.


[4] Welch HG, Black WC. Using autopsy series to estimate the disease “reservoir” for ductal carcinoma in situ of the breast: how much more breast cancer can we find? Ann Intern Med 1997; 127:1023-1028.


[5] Bleyer A, Welch HG. Effect of three decades of screening mammography on breast-cancer incidence. N Engl J Med 2012; 367:1998-2005