In the ongoing controversy about overdiagnosis as one of the harms of breast cancer screening, one of the most important observations is the disease reservoir at autopsy. With the decline in autopsies as a standard practice, our data is historical. But in this case, the historicity works to our advantage, in that autopsy disease reservoirs were established largely in the pre-mammographic era, such that we get to see how often subclinical disease lingered silently in the natural state.
The numbers generated from these old studies have been hanging around a long time, only to resurface in the current era, wildly distorted through a double misstep. First: When these studies are referenced, DCIS and invasion are often lumped together and called “breast cancer” (see Part One July editorial on “Polysemy” – a word or phrase with more than one related meaning). If the two entities were present at autopsy in their usual ratio seen through screening, I would probably not be writing this article. But as we will see shortly, the autopsy series reveal DCIS almost entirely. Second: The next misstep is quoting only the very highest numbers (again, applicable only for DCIS), so it’s not uncommon to hear that the disease reservoir for “breast cancer” in autopsy series is 30-40% (“comparable to prostate or thyroid cancer”). Few seem to remember that there is a wide range for DCIS found at autopsy, and the low end of the published range happens to be zero.
Why is accuracy suddenly so important when discussing the disease reservoir? In recent years, critics of screening are generating sky-high numbers for overdiagnosis, once applied only to DCIS, but now with invasive carcinoma in the crosshairs. All methodologies used to generate these whopping numbers are through indirect observations, accompanied by the claim that “it’s all we’ve got” since there is “no way to directly observe overdiagnosis.” But that’s not entirely true.
Overdiagnosis, as distinguished from length bias, is based on the concept that tumors either regress or become quiescent with no further growth. In a nutshell, with true overdiagnosis, tumors never progress during the life of the patient. Note: this is a shade different than “slow-growing tumors” that are unlikely to kill the host, wherein screen-detection does not improve outcomes. It is these slow-growers that are responsible for length bias in screening studies (with the bias overpowered by mortality reductions). While length bias is a cousin concept to overdiagnosis, it doesn’t have near the firepower when drawing media attention. Who stays awake at night worrying about length bias?
If overdiagnosis of invasive breast cancer is real, then most agree that we can’t recognize it in an individual patient newly diagnosed. However, we are able to make some generalizations about the existence of overdiagnosis through direct observations regarding the natural history of invasive breast cancer. If some breast cancers never progress and therefore never kill the host, then there are only two possible scenarios, and they should be observable – regression or quiescence. “Slow growing” doesn’t cut it – that’s the biology that generates length bias. In contrast, overdiagnosis translates to “pseudocancers.”
How could we directly observe tumor regression or quiescence? Regression is the more challenging scenario because the evidence “disappears.” Some epidemiologists claim, by the way, that they have proven the existence of regression such that it is now a known “fact.” But upon reviewing these studies, you’ll find layer upon layer of indirect evidence sandwiched between two slices of precarious assumptions. In contrast to this deductive process, we do have a way to directly observe the natural history of untreated invasive breast cancer to see if it can regress.
There are some patients who go untreated after a positive biopsy of a screen-detected cancer, yet for whatever reason, still opt to return for their next mammogram. Until recently, there was no systemized approach to track the natural history of screen-detected breast cancer after these long delays. We relied on radiology leaders and colleagues who told us they’d “never seen a single case of cancer regression.” And this was a data deficit that I described in my book – Mammography and Early Breast Cancer Detection (devoting 2 chapters to overdiagnosis). Essentially, we were relying on hearsay to support the notion that tumor regression of invasive cancer was pure bunk.
Well, it’s no longer hearsay. In the July 2017 issue of the Journal of the American College of Radiology, Ed Sickles led a group of breast radiologists in a massive effort to identify tumor regression rates (Arleo EK, Monticciolo DL, et al. J Am Coll Radiol 2017; 14:863-867). While certainly not a prospective randomized trial for Level One evidence, this still goes down in history as the first organized attempt to document the validity of breast cancer regression in screen-detected tumors.
From 42 actively practicing Society of Breast Imaging fellows, their entire practice results were tallied for a 10-year period, looking for untreated, screen-detected cancers. Of the 6,865,324 screening mammograms performed, there were 25,281 screen-detected invasive cancers and 9,360 screen-detected in situ cancers. Of these, there were 240 cases of untreated invasive breast cancer and 239 cases of untreated DCIS who had imaging follow-up. The number of cases that either decreased in size or regressed completely on the next mammogram was ZERO.
Could some of these still have been “overdiagnosed?” Yes, but recall that we’re trying to explain critics like H. Gilbert Welch et al who have announced repeatedly and publicly that we currently diagnose “pseudocancers” with mammographic screening to the tune of 70,000 cases of invasive cancer every year. While the authors of this new survey of breast radiologists shy away from a direct assault against the pseudocancer claim and overdiagnosis, they point out the inescapable conclusion: While it is conceivable that some of these cancers might be overdiagnosed, the mere fact that NONE of them regressed on imaging after their next mammogram means that extending the interval for screening from one year to every 2 years, will have ZERO impact on reducing the rate of overdiagnosis. The cancers will still be there next year, waiting.
If invasive cancers don’t regress (note that DCIS didn’t regress in the Sickles study either) then how could there still be overdiagnosis in the group of 479 cancers noted above? Why didn’t the authors go ahead and kick the overdiagnosis enthusiasts while they were down? Because there still could be tumor quiescence. That is, the tumors didn’t regress, but they didn’t grow either – a tumor-host stand-off. And this is where the autopsy data are so critical. The lack of regression does not rule out tumor quiescence. It takes a ONE-TWO punch to knock out overdiagnosis – no regression (as strongly indicated above) and no quiescence (as supported by the invasive disease reservoir, or lack thereof, in autopsy studies).
If tumor quiescence is so incredibly common to account for 70,000 (invasive) pseudocancers every year (over 1,000,000 women walking around today having been treated for fake cancer), then we should see high numbers of invasive cancer in the pre-mammographic autopsy series where disease reservoir was measured. Welch et al can generate incredibly high numbers by capitalizing on slow-growing tumors and length bias, re-christening these tumors as “pseudocancers” through skillful statistical shenanigans. But if we’re going to look for 70,000 pseudocancers a year, it is unavoidable – there must be either: 1) frequent tumor regression, or 2) tumor quiescence as reflected in a high disease reservoir at autopsy.
Yes, it is probably true that one-third of screen-detected invasive cancers would have been cured without screen-detection, but this is a different concept than true overdiagnosis where tumors never emerge clinically. When length bias is the culprit, the favorable tumors will be cured with or without screen-detection. If the patient lives long enough, these tumors will eventually emerge without screening (therein lies the rub – life expectancy). This is why it is so important to correct for lead time at the end of a screening study.
Without knowing the autopsy data, it seems much more probable that tumor quiescence would be the mechanism behind overdiagnosis, rather than tumor regression, the latter now known to be fiction through a direct observation rate of ZERO. If tumors are stagnating at a certain point in their biologic history, then they will accumulate over time and appear in autopsy series in disproportionate numbers compared to what is seen clinically. And if overdiagnosis of invasive breast cancer is as common as being touted (70,000 per year is a “conservative” calculation, according to Welch et al), then back-of-the-envelope calculations indicate that autopsy findings ought to be similar to what is seen in prostate cancer, or even worse. As it turns out, they’re not even close. Importantly, the autopsy series need to come from a population that has not undergone mammographic screening during life, given that removal of tumors would falsely lower autopsy incidence.
In sharp contrast to invasive breast cancer, the disease reservoir for prostate cancer at autopsy is roughly proportional to a man’s age. As was seen in a review of the world literature addressing autopsy studies (Haas GP, et al. Canadian Journal of Urology 2008; 15:3866-3871), by age 30, one finds 31% of men with foci of prostate cancer, and by age 80, over 80% will have prostate cancer. And what about disesase reservoir in the living? – In a prostate cancer prevention trial of finasteride (New Engl J Med 2003; 349:215-224), the participants with normal PSA and normal digital exam were offered a random biopsy – a rather startling 15% of those participating had prostate cancer present in the very limited tissue sampling.
Speakers at the podium, still today, will routinely cast breast cancer in the same light as prostate and thyroid cancer, claiming comparable autopsy findings. So, what are the numbers for invasive breast cancer at autopsy? Take a guess. You’ve seen the slides at meetings. You’ve seen the reference made in published articles. You’ve heard the CME tapes. “Autopsy studies show up to 35-40% of patients have occult breast cancer when they die.” The correct number? The incidence of invasive breast cancer at autopsy is 1%. Disengaging DCIS from invasive breast cancer is an eye-opener!
Nearly all the “breast cancers” found in autopsy studies are DCIS. Even DCIS is overstated when one looks at the photomicrographs from those historic studies and sees that many of lesions would today be called Atypical Ductal Hyperplasia. Furthermore, a small area of low grade DCIS without calcium is not going to be a clinical problem, so yes, disease might be indolent, but it’s not going to be identified on mammography – thus, it’s not an issue. You can’t overdiagnose if you can’t see the lesion on breast imaging. Still, these small non-calcific areas of low grade DCIS (and borderline lesions) greatly inflate the autopsy numbers. In contrast to mammographically occult low grade lesions, elevated PSAs in men prompt biopsies, which then reveal the occult indolent cancers, making overdiagnosis/overtreatment a significant problem.
In the photo above, the lesion is virtually identical to Duct Lesion #10 in Rosai’s 1991 landmark article (Am J Surg Path 1991; 15:209-221) where interobserver variability in borderline epithelial lesions was initially described. For lesion #10, three experienced pathologists called it hyperplasia, one called it ADH and one called it DCIS. Among 5 pathology experts, there was not complete agreement on a single lesion in the study of both ductal and lobular lesions. The prevalence of DCIS in autopsy series are inflated to an unknown degree by these difficult, borderline lesions. (Note: Dr. Page later countered the Rosai article to a degree by showing the interobserver variability is greatly reduced if there is standardization of criteria (complete agreement among 6 pathologists in 58% of the lesions); however, while the variability was remarkably tight from the viewpoint of pathology, in contrast, from the clinical standpoint where decisions are made, 42% of the cases still had some disagreement among experts.)
I’m not going to deny that overdiagnosis occurs in DCIS. Most are in agreement that some (an unknown number) of these lesions do not progress, or would not progress during the lifetime of the patient, especially lower grade DCIS. Hopefully. the much anticipated “Comparison of Operative to Monitoring and Endocrine Therapy Trial” (COMET) will sort this out. Good luck to Dr. Hwang and all those participating.
My intent is to draw attention to the remarkably low disease reservoir of invasive breast cancer, no different than disease prevalence in the living. If you perform the initial mammographic screen (prevalence screen) on 1,000 women at age 50, you will find 1% of them have breast cancer, the same number as found at autopsy (albeit some of these are DCIS). But now add the mammographically occult cancers we can find using screening MRI wherein Dr. Christiane Kuhl’s series (in women with normal mammograms, normal exams, and most with normal ultrasound), identified an additional 2.2% with cancer on the prevalence screen, again some with DCIS. However, if you remove the DCIS from both mammographic screening and MRI screening, you’ll still have 2% of women with invasive breast cancer as detectable disease prevalence, higher than what is found at autopsy. These numbers do not support overdiagnosis of invasive cancer. They do not support tumor quiescence.
Stated more emphatically, the autopsy data is the strongest evidence we have against widespread overdiagnosis when it comes to invasive cancer, especially in light of the new information that excludes tumor regression by direct observation. So why do we shoot ourselves in our collective feet by playing into the hands of the epidemiologists who are using every trick in the book to malign breast cancer screening? Why, in the non-radiologist world, have many adopted the 30% overdiagnosis rate, not based on the epidemiologic claims or reasoning, but by believing and promoting incorrect autopsy data? Say what you want about DCIS, but when it comes to invasive cancer, the 30% overdiagnosis rate (by strict criteria where tumors are considered “pseudocancers”) is simply not true.
Now – to my source for the 1% disease reservoir for invasive breast cancer. In the 1990s, when many of us (incorrectly) believed that the key to eradicating breast cancer was to find all cases of DCIS, a young internist (with a specific interest in the hazards of screening healthy populations for a variety of diseases) decided to stake his claim to fame on exposing overdiagnosis in DCIS. He decided to perform a combined review of all available autopsy studies at the time, which consisted of 852 patients total, mostly from the pre-mammography era. The title of his article was: “Using Autopsy Series to Estimate the Disease ‘Reservoir’ for Ductal Carcinoma In Situ of the Breast: How Much More Breast Cancer Can We Find?” The article was published in a journal that has always been a haven to anyone willing to shed light on the harms of screening – Ann Intern Med 1997; 127:1023-1028.
In the 7 autopsy series in this comprehensive combined analysis, the overdiagnosis rate, that is, the disease reservoir rate, was 0 to 14.7%, with the higher rates seen when more slides were reviewed (the range here was 9 slides per breast to 275 slides per breast). Granted, one of the 7 studies revealed a 39% incidence of DCIS (borderline photomicrographs notwithstanding) in 109 autopsies when considering only those patients who were of screening age (one must keep in mind that these are often forensic autopsies that include younger patients). However, the median prevalence of DCIS when all patients were considered in the seven studies was only 8.9%.
Seemingly forgotten today, this autopsy review secondarily included invasive cancers as well. The range was very tight, including the “275 slides per breast” study. For invasive disease, the range was from 0 to 1.8% with a median of 1.3%, a conclusive indictment against the alleged phenomenon of tumor quiescence. As for the outlier study above, where 39% had DCIS with extensive sampling, 275 slides per breast were only able to identify invasive cancer in 0.9%.
So, if we don’t see invasive cancers regress clinically, and we don’t see excessive invasive cancers that are quiescent until death of the host, there is no observable support for overdiagnosis of invasive breast cancer when one examines untreated natural history. And while one can argue that this is still indirect evidence, it is far closer to direct observation of the “black hole of overdiagnosis” than are the large scale epidemiologic reviews that don’t include specific mammography data as to who did and who didn’t comply. And, the lack of disease reservoir is much closer to the truth than the abuse of the historical screening trials for mammography where lead time is not corrected at the end of the study, allowing an “excess” of cancers in the screened group.
Oh, I nearly forgot. The lead author of the DCIS exposé was Dr. H. Gilbert Welch, prominent screening expert from Dartmouth who, subsequent to his little-noted 1997 autopsy article, has gone on to fame by exposing the extraordinarily high rates of overdiagnosis with regard to screen-detected invasive cancer. His indirect evidence of the “black hole” proved beyond a shadow of doubt to many, including what seems like every science journalist in America (hyperbole notwithstanding), that mammographic screening did more harm than good, by diagnosing “pseudocancers” in 70,000 women every year, totaling to 1.3 million women over the past 30 years (Bleyer and Welch, New Engl J Med 2012; 367:1998-2005). “All those unfortunate women undergoing mutilating surgery, unnecessary radiation and chemotherapy for benign disease,” wrote one journalist.
But Dr. Welch didn’t stop with a measly 70,000 overdiagnosed cancers per year, or one-third of invasive cancers. Abandoning the concept of tumor quiescence (perhaps due to his own 1997 findings), he has embraced tumor regression as the likely mechanism at work, and in 2016, he upped the ante to an indirect estimate of overdiagnosis that exceeded 80% (N Engl J Med 2016; 375:1438-1447). And this is why the Society of Breast Imaging study quoted above is so critical, and so wonderfully timed – ZERO regression. And then, from Dr. Welch’s own work – ZERO tumor quiescence of invasive disease reflected at autopsy.
While we can’t determine if a specific cancer is a “pseudocancer” or not, there are direct observations that can be made, as we’ve seen above, that can confirm or deny overdiagnosis. Indeed, the very first trick played upon an audience listening to exaggerated claims of overdiagnosis is the mistaken belief that one can only calculate overdiagnosis indirectly. That’s where the magic begins – you see with your own eyes that the elephant on the stage has disappeared, but you also know it’s a trick.
“But it’s all moot,” you might say. “What difference does it make what you call it, if a screen-detected cancer is not going to be a threat to a patient’s life? Who cares whether it’s length bias or overdiagnosis? You’re splitting hairs. The problem is overtreatment.”
And you would be justified in your criticism…at first glance. In September, I’ll conclude with Part 3 – The Alchemy of Overdiagnosis.
For Dr. Hollingsworth’s book on the nuances of screening (including two chapters on Overdiagnosis), Click on: http://www.mcfarlandpub.com/book-2.php?id=978-1-4766-6610-5