Uncertainty versus certainty in the mammography wars

As I write this, I am winging my way home from the 2014 meeting of the American Association for Cancer Research (AACR, Twitter hashtag #AACR14) in San Diego. (OK, I’m revising this to fit the format and, of course, the Insolence of this particular blog. Shockingly, I didn’t have as much time to blog in San Diego as I had thought I would. Go figure.) Basically, the AACR meeting is one of the largest meetings of basic and translational cancer researchers in the world. I try to go every year, and pretty much have succeeded since around 1998 or 1999. As an “old-timer” who’s attended at least a dozen AACR meetings and presented many abstracts, I can see various trends and observe the attitudes of researchers involved in basic research, contrasting them to that of clinicians. One difference is, as you might expect, that basic and translational researchers tend to embrace new findings and ideas much more rapidly than clinicians do. This is not unexpected because the reason scientists and clinical researchers actually do research is because they want to discover something new. Physicians who are not also researchers become physicians because they want to take care of patients. Because they represent the direct interface between (hopefully) science-based medicine and actual patients, they have a tendency to be more conservative about embracing new findings or rejecting current treatments found not to be effective.

While basic scientists are as human anyone else and therefore just as prone to be suspicious and dismissive of findings that threaten their scientific world view, they can (usually) eventually be convinced by experimental observations and evidence. As I’ve said many times before, the process is messy and frequently combative, but eventually science wins out. Not infrequently, it takes far longer than in retrospect we think it should have, an observations frequently exploited by advocates of pseudoscience and quackery to claim that their pseudoscience or quackery must be taken seriously because “science was wrong before.” To this, I like to paraphrase Dara O’Briain’s famous adage that just because science doesn’t know everything doesn’t mean you can fill in the gaps with whatever fairy tale that you want. But I digress (although only a little).

In accepting the validity of science that indicates either that a medical intervention that was commonly used either doesn’t help, doesn’t help as much as we thought it did, or can even be harmful, they have to contend with the normal human reluctance to admit to oneself that what one was doing before might not have been of value (or might have been of less value than previously believed) or even that, worst of all, it might have caused harm. Or, to put it differently, physicians understandably become acutely uncomfortable when faced with evidence that the benefit-risk profile of common treatment or test might not be as favorable as previously believed. Add to that the investment that various specialties have in such treatments, which lead to financial conflicts of interest (COI) and desires to protect turf (and therefore income), and negative evidence can have a hard go among clinicians.

Examples abound in the history of medicine. In my own specialty, some surgeons resisted the transition over to doing needle biopsies of breast cancers because that meant more radiologists and fewer surgeons would be doing breast biopsies. There was also resistance to the concept that doing only partial mastectomies (“lumpectomy”) was safe and effective. Urologists weren’t too thrilled with the development of drugs that could shrink benign prostatic hypertrophy (BPH), the widespread adoption of which rapidly resulted in a dramatic decline in the number of transurethral resections of the prostate (TURPs) performed, a “bread and butter” procedure done by urologists back when I was in medical school and residency that made up large percentages of urologists’ practices. The proliferation of angiographic stenting of coronary blockages led to turf battles over which lesions needed surgical bypass and which ones could be safely stented by cardiologists, battles that continue today. Tonsillectomy is another good example of a procedure that was widely done (I had one when I was around 6) but now seldom done because in the vast majority of cases before it was unnecessary.

All of which brings us to screening mammography, in my usual logorrheic meandering (and, above all, self-deprecating) way, because there were two new studies from JAMA last week that add to the drip, drip, drip of mammography studies that all seem to be converging on similar conclusions. Once again, I must emphasize that I am not talking about diagnostic mammography; i.e., mammography ordered to evaluate a lump or other symptoms in the breast. Nothing—I repeat—nothing has changed with respect to that. If a woman has a lump in her breast or concerning symptoms, mammography is indicated. What I am referring to is screening mammography performed on women without symptoms in order to detect breast cancer before it becomes symptomatic. I cannot emphasize that enough.

Yet another mammography study…times two

Last week, as has been happening more frequently, there was a study published that cast more doubt on current mammography guidelines. It is not like the Canadian National Breast Screening Study whose results were published a couple of months ago in that it does not find that screening mammography does not result in a benefit. However, it does find that the benefit-risk ratio of routine screening mammography is not as favorable as believed in the past. Rather, it is more like studies by Gilbert Welch, a Norwegian study, and studies highlighting the problem of overdiagnosis, in which significant percentages of screening-detected cancers are overtreated and some even spontaneously regress. Before I get to the study itself, I often like to give you a flavor of how the press reported the study. An example of pretty good reporting is the news story from NPR covering the study. Other news reports tended to make it look as though mammography were useless.

Here’s the study by Lydia E. Pace and Nancy L. Keating, entitled A Systematic Assessment of Benefits and Risks to Guide Breast Cancer Screening Decisions. It’s not behind a paywall, indicating that JAMA must have thought it important enough that it should be widely available. It’s also pretty well written and (mostly) understandable, even if you don’t necessarily have a high level of expertise in mammography and epidemiology. What the authors did was to search MEDLINE for relevant randomized clinical trials (RCTs), meta-analyses, systematic reviews, and observational studies from 1960 to January 19, 2014, as well as the references of key articles, reviews, meta-analyses, and practice recommendations, which were searched manually. They also included meta-analyses of randomized clinical trials (RCTs) of mammography screening examining breast cancer mortality. From 525 articles identified, 20 meta-analyses met these criteria, and the authors, quite appropriately, focused most of their attention on the 5 meta-analyses published after 2006, which include the last RCT of mammography.

Another thing that the authors did was to emphasize their estimates of number needed to invite (NNI) to screening to save one life, as well as absolute risk ratios. Frequently, results of clinical trials (not just RCTs of mammography) are presented as relative risk reductions (e.g., “mammography decreases the risk of dying of breast cancer by 19%), which can be misleading if the absolute risks and risk reduction are low. I’ve described this phenomenon before in the context of RCTs of adjuvant chemotherapy for breast cancer, where a 30% relative risk reduction might actually mean only a 2-3% absolute decrease in the risk of dying of early stage breast cancer. In adjuvant chemotherapy for breast cancer, the absolute magnitude of the survival benefit increases along with the absolute risk of dying. In other words, a given adjuvant chemotherapy regimen might produce a similar, say, 30% decrease in the risk of dying for all breast cancers of a certain subtype, but that 30% means more on an absolute basis for nastier tumors. Put it this way. If you have a 50% chance of dying of your disease, a 30% relative risk reduction of that risk translates to a 15% lower chance of dying on an absolute basis. If, on the other hand, you only have a 10% chance of dying of your disease, that same 30% relative risk reduction translates to only a 3% absolute risk reduction.

A similar principle exists for screening tests, and this review explains that quite well, boiling it down into a couple of concise tables. Since the article is free, I don’t feel obligated to reproduce Table I and Table II so I’ll just discuss them. Table I shows pooled mortality reductions due to mammography for different age ranges. I could discuss how the authors derived their estimates ad nauseam. Certainly, if anyone wants to get into that level of detail, he’d have to have a detailed understanding of the studies that went into coming up with these numbers. Given that the numbers are roughly consistent with estimates that Laura Esserman derived in her review and a fair number of other estimates, I’ll spare you. Basically, Table I was derived from Nelson et al., the meta-analysis conducted for the USPSTF. The key parts of the estimates is this: The younger the patient group, the higher the NNI is, ranging from 377 for women aged 60-69 to save one life, resulting in an absolute risk reduction of dying from breast cancer in the next ten years of approximately 0.3%, to 1,904 for women aged 40-49, translating to an absolute risk reduction of 0.05%.

Table II shows the estimated benefits and harms of screening mammography for women in the various age ranges studied for 10,000 women undergoing screening using Surveillance, Epidemiology, and End Results (SEER) estimates from a recent review by Welch and Passow. Not unsurprisingly, the error bars are large. The lower number reflects a minimal breast cancer mortality of 5% based on RCTs reporting no benefit, while the upper number reflects a reduction of 36% based on the clinical trial reporting the highest benefit, the always-cited Swedish trial by Tabár et al.

Also included are estimates of overdiagnosis, false positives screening studies requiring more imaging and/or biopsy, and the number of patients who underwent one or more unnecessary biopsies during the ten year period. Again, the numbers for overdiagnosis are similar, actually slightly lower, than the numbers I’ve discussed before, which have generally converged around an estimate of 20% or slightly more of screen-diagnosed breast cancers turning out to be overdiagnosed. I note that in all age ranges the upper estimates of overdiagnosis approach 40% but that in women aged 40-49 the upper estimate is over half. For women starting screening at age 40 or 50, the chance of at least one false-positive result is 61%, while it’s 49.7% for women undergoing screening between the ages of 66 and 74.

These results lead the authors to conclude:

Although some of the challenges of mammography can be resolved with further research to guide individualized decisions and thoughtful development and dissemination of decision aids, better breast cancer screening tests are needed. More sophisticated tools, for example, could distinguish aggressive vs indolent tumors, reducing the burden of overtreatment.⁸⁶ Mammography screening appears to be associated with reduced breast cancer mortality, but for some patients, the harms may outweigh the benefits. Until better screening methods are available, improved understanding of these harms, enhanced strategies to identify the highest-risk patients, and tools to help patients and clinicians incorporate these in their decisions should be research priorities.

Or, as one article reporting on the study characterized it, Stop One-Size-Fits-All Mammography. This is a reasonable take-home message from the study. The problem, of course, is that our risk assessment tools and models are still, even a decade into the genomic revolution, grossly inadequate to the task.

Adding to the confusion is a review article that appeared in the same issue of JAMA by Louise C. Walter and Mara A. Schonberg, entitled Screening Mammography in Older Women: A Review. It begins with a patient story:

Mrs M is 91 years old and has had annual screening mammograms since age 50. She lives alone in her apartment and independently performs all activities of daily living. Her chronic medical conditions include hypertension and osteopenia, and she was diagnosed with intermittent claudication in 2010. Her medications include valsartan, furosemide, and isosorbide dinitrate. Mrs M had her only child at age 16, experienced menopause at age 50, and never used postmenopausal estrogen or progesterone. She had a negative breast biopsy in 1984. Her daughter died of breast cancer at age 37.

In 2008 at age 87, Mrs M had an abnormal screening mammogram with microcalcifications in the left medial inferior breast, interpreted as category 3 (Breast Imaging-Reporting Data System [BI-RADS]; probably benign). She subsequently underwent scheduled 6-month follow-up diagnostic mammograms over the next 18 months. The third mammogram, performed in 2010, showed interval increase in the number of heterogeneous microcalcifications and was classified as a BI-RADS category 5 (highly suggestive of malignancy). The lesion was not amenable to biopsy under stereotactic guidance. Therefore, she underwent excisional biopsy of the left breast lesion using needle localization. The biopsy identified ductal carcinoma in situ (DCIS), intermediate grade without comedo-type necrosis, on 2 of 9 slides. Estrogen receptor staining was strongly positive. One area of DCIS was less than 1 mm from the anterior margin so she underwent reexcision and no residual DCIS was identified. Mrs M met with a radiation oncologist who did not recommend radiation therapy, and she declined tamoxifen due to concerns about adverse effects. She continues to undergo annual mammograms, which have been negative, and she is seen by a breast oncologist every 6 months.

Perspectives
Mrs M: I get mammograms every year. I know you don’t get them all your life.

Dr P: I think people might say: “What are you doing getting mammograms in a 91-year-old?” but you have to meet this lady. She is a lot more likely to live to be 100 than I am.

It might be surprising to you, but there are no RCTs examining screening mammography that included women over 74. All the major studies that showed a decrease in death due to breast cancer in women screened by mammography excluded older women. Most guideline panels and organizations have therefore recommended that screening mammography in women 75 and older be decided based on an individualized approach that weighs potential benefits and harms of screening in the context of a woman’s overall health, life expectancy, and preferences. It’s a shocking hole in our evidence base with regards to screening mammography, but there it is. Clinicians, as is so often the case, have to apply an incomplete data set lacking rigorous studies to individual patients.

To examine the issue of mammography in women 75 and older, the investigators attempted to look at the balance of risk and benefit in such women. It’s not just a woman’s risk of breast cancer, as in the first study, that drives the risk-benefit consideration. It’s also how long a woman is likely to live, which drives the risk that a women will (1) develop breast cancer in her lifetime and (2) that that breast cancer will kill her before something else does, such as old age. The authors therefore searched PubMed for English-language studies in peer-reviewed journals published from January 1, 1990, to February 1, 2014, to identify risk factors for late-life breast cancer in women aged 65 years and older and to quantify the benefits and harms of screening mammography for women aged 75 years and older. Their first search identified 864 articles, of which 24 provided data on risk factors for breast cancer in women aged 65 years and older. The second search resulted in 2,121 articles of which 13 provided data on survival benefits and/or false-positive results and biopsies following screening mammography specifically for women aged 75 years and older.

Their conclusions/recommendations from their review of the literature included:

• Advancing age is the greatest risk factor for developing breast cancer. Risk factors that reflect hormonal exposures in the distant past (eg, age at first birth, age at menarche) may be less predictive of late-life breast cancer than factors indicating recent hormonal exposures (eg, high bone mass or obesity).
• Randomized trials of screening mammography did not include women aged 75 years and older. Observational studies suggest benefit of continued screening mammography in older women who have a life expectancy of more than 10 years.
• Potential harms of screening mammography include overdiagnosis (ie, finding a breast cancer that would not have clinically surfaced in the absence of screening) as well as pain and anxiety associated with false-positive results and biopsies.
• Breast cancer treatments are effective among older women in good health; however, the harms of breast cancer treatment increase as women age, particularly among older women with limited life expectancy.
• Providing information about life expectancy along with potential benefits and harms of screening mammography may help older women’s decision making about screening mammography.

Regarding #4 above, the authors also note that screening women over 74 will only be effective if effective treatments exist for early stage aysmptomatic breast cancer in such women. Unfortunately, few breast cancer treatment clinical trials included women aged 75 years and older or women with multiple comorbidities or frailty. In other words, not only are women older than 74 not represented in RCTs of mammography, but they are underrepresented in RCTs testing breast cancer treatments. Indeed, it’s worse than that because it is women 65 and older who are grossly underrepresented in breast cancer clinical trials. Confusing things even more is that breast cancers diagnosed in older women tend to more indolent tumors, exactly the kind that are more frequently overdiagnosed, thanks to length bias. Thus, after a certain age (what that age is isn’t clear), overdiagnosis tends to increase with age, both because older women tend to develop more indolent tumors and, to put it bluntly but very simply, there’s less time for diagnosed cancers to kill them due to fewer years of life expectancy as one ages. This led to an estimate by the authors that roughly 32% of breast cancers diagnosed by screening mammography in older women will be overdiagnosed.

Overall, the authors’ recommendations are reasonable, but, as is the case in the first article, maddeningly vague, leaving the primary doctor to do pretty much what he or she thinks is best, with little hard guidance as to what is best. However, adding these two studies to the pile of studies that have been published over the last six years or so, I am definitely coming to agree with Pace and Keating that the era of one-size-fits-all mammography is almost certainly coming to an end. Indeed, even the Komen Foundation seems to recognize this in its statement about the first study. If you read it, you’ll notice a subtle backing-away from a concrete recommendation that screening begin at 40, a bit of a disclaimer, with a shift in emphasis towards the need for better tools to evaluate a woman’s risk of breast cancer to guide screening decisions. To that I’d add that we desperately need better tests for diagnosed cancers and DCIS, in order to determine who does and doesn’t require aggressive treatment.

In the meantime, however, clinicians are left with uncertainty, and clinicians, understandably, don’t like uncertainty.

Putting it into context

Being a translational researcher in addition to a surgeon, I’m a bit “schizophrenic” in that I live in both worlds, the world of the clinician and the world of the basic/translational scientist. Because until relatively recently I didn’t do much clinical research, I tended to think much more like a basic scientist in some areas but, when it came to routine practice, I thought much more like a clinician. So it was that I didn’t think much about the guidelines for mammography until around six years ago or so. The guidelines were the guidelines, and I followed them, not really showing much interest in the science behind them. Then in 2008, I rediscovered the concept of overdiagnosis, something that doctors tend not to be taught about as much as they should be and therefore something that they either know little about or tend to dismiss. The result was my first post about how difficult it is to show that early detection of cancer results in improved outcomes, thanks to the issues of lead time bias and length bias. Then in 2009, I noted a review/commentary article by Laura Esserman regarding the need to rethink how we screen for breast and prostate cancer, highlighting the high numbers of people who need to be screened to prevent one death from the cancer screened for and the not inconsequential harm of overtreatment that can occur.

It seemed like karmic justice that two weeks later the United States Preventive Services Task Force (USPSTF) published its suggested guidelines with regard to screening mammography, in which it recommended that screening begin at age 50, not 40, and that it be conducted every other year rather than every year. I commented on the guidelines, of course, noting that they were not the final word, a prediction about as safe as predicting that the sun will rise in the east or that this year’s election will be contentious—or that radiologists will tend to react…badly to such studies.

It was also my first acutely-uncomfortable experience with new findings regarding mammographic screening, as I found myself in the distinct minority among my colleagues in not being alarmed by the results or feeling the need to explain them away. Worse, this was the first time that I had ever been in any sort of position where I was asked to help craft the official public response by my cancer center, a task that had to be completed within a day or so because of the number of inquiries we were getting. It was my first real experience trying to communicate science in a way that had real consequences. Blogging is just blogging, and I can spout off more or less what I’m thinking without worrying overmuch about it. In contrast, this time, I was trying to help craft a statement as part of my real world job, a statement that could have real effects on patients.

The reason it was so hard is because over the years the guidelines that stated that screening mammography should start at age 40 and be done yearly had become very entrenched, and persuading women that this was a good thing to do required a simple, concrete message. The USPSTF guidelines laid bare in a highly publicized story how messy the scientific process and, in particular, the process of translating basic and clinical science into guidelines is. Predictably, cancer quacks seized on these guidelines that “mammography doesn’t work,” while advocate groups were dismayed and had to defend their guidelines, while some radiologists, led by Daniel Kopans, a man for whom no hyperbole is too ridiculous when defending mammography—used what can only be described as despicable claims to protect their turf (and still do). For example:

This is simply malicious nonsense,” said Dr. Daniel Kopans, a senior breast imager at Massachusetts General Hospital in Boston. “It is time to stop blaming mammography screening for over-diagnosis and over-treatment in an effort to deny women access to screening.”

No, Dr. Kopans, the New York Times did not kill your patient. (I know the linked article is about prostate cancer, but the same arguments apply.)

Dr. Kopans’ statement reveals what to me as a cancer surgeon and scientist is an incredible ignorance—most likely willful—of what, exactly, overdiagnosis actually is, and an utter cluelessness about the actual biology of breast cancer, in particular its variability in biological aggressiveness, that are utterly staggering in their breadth and depth. His statement was the real malicious nonsense, and unfortunately he is a font of these sorts of belligerent statements. No one was (or is) trying to “deny women access to screening. What these two reviews are trying to do is to assess the true benefits and harms of mammography, just as the NEJM study by Bleyer and Welch tried to do. (That was the study that provoked Dr. Kopans’ outburst in 2012.) Seriously. Such studies are, quite simply, bringing claims for mammographic screening “down to earth” and demonstrating that, in the real world, outside of the carefully regimented and controlled confines of RCTs, mammography probably doesn’t do as well as its boosters claim. Also, like all medical interventions, it has a risk-benefit profile. Contrary to what its boosters claim, those risks include overdiagnosis and overtreatment. In his denial of the existence of overdiagnosis due to mammography, Dr. Kopans reminds me of antivaccinationists and anthropogenic global warming denialists in his ability to deny science. Based on his statements, I conclude that he clearly prefers certainty and can’t handle doubt and uncertainty.

It’s instructive to look at what the American Cancer Society has recommended over the years, which is conveniently found in a table form on the ACS website. It wasn’t until 1997 that the ACS started recommending yearly mammography for women between 40 and 49. Before that, the recommendation was for a baseline mammogram at or around age 40 and then mammography every 1 to 2 years. Before that, the recommendation was for a woman to consult her physician, and before that the recommendation was: “May have mammogram if they or their mother or sisters had breast cancer.” This occurred not long after two other occurrences. First, there was an NIH consensus conference on mammography in women aged 40-49 held earlier that year.

The press release about that conference and the controversies discussed sounds eerily familiar to much of what I just discussed above. Second, the National Cancer Advisory Board recommended “that the National Cancer Institute (NCI) advise women 40 to 49 to get screening mammograms every one to two years if they are at average risk for breast cancer.” Interestingly, no mention was made of overdiagnosis and overtreatment; the only harms listed were false positives, which, it was noted, are more common in younger women. The NCAB also noted that its decision reversed a previous decision in 1993 to withdraw its recommendation for routine screening mammography between ages 40 and 49. Not coincidentally, the NCI reversed itself less than a week after the ACS and American College of Radiology (ACR). Although the NCI denied that its decision was influenced by it, there was intense political pressure brought to bear, with the Senate previously having unanimously passed a resolution urging mammograms for women in their 40s. In addition, Sen. Arlen Specter, R-Pa., had threatened the NCI director’s job over the controversy, as his Senate subcommittee that oversees federal health spending practically ordered the NCI to do the same, and several legislators wrote or called the NCAB with the same request.

No, I don’t believe for a minute that politics didn’t have anything to do with that decision, but I do now understand how everything old is new again. Most people aren’t aware of the long strange trip mammography took to become what it is now. Politics has been and continues to be a factor in determining mammography guidelines.

Of course, the American College of Radiology, as has been the case every time a study has been published that found that either the risks of mammography were higher than thought before or the benefits not as great as previously believed, is at it again, this time with the Society of Breast Imaging. The two professional societies issued a joint press release last week, and it’s full of the usual spin:

The American College of Radiology (ACR) and Society of Breast Imaging (SBI) agree with statements made by Pace and Keating, in their recent article published in the Journal of the American Medical Association (JAMA), that women should discuss mammography with their doctor and breast cancer diagnosis and treatment may one day be more individualized. However, at present, breast cancer screening based primarily on risk — as discussed in the JAMA article — would miss the overwhelming majority of breast cancers present in women and result in thousands of unnecessary deaths each year.

Utter piffle, so much so that another post might well be required to explain why. Suffice to say that the press release trots out the same tired (and discredited) arguments that somehow the CNBSS has been “discredited” (it hasn’t) and a paper purporting to show that overdiagnosis is not nearly as big a problem as recent studies have indicated, which might require an entire new post of its own. In the meantime, suffice also to say that there’s less to the attacks on, for example, Bleyer and Welch’s study from 2012 than certain radiologists would like you to believe.

What to do?

Contrary to the hysterical denunciations of the recent studies that have led me to this conclusion, women won’t be dying in droves if screening recommendations are made less intense. Most developed countries don’t start screening until age 50 for most women, and several recommend screening only every two years. Yet, these countries have breast cancer mortality rates comparable to what is observed in the US. That means one of two things: Either Europeans have better treatments for breast cancer than we do, sufficient to compensate for later diagnosis due to less intensive mammographic screening, or starting screening at age 50 doesn’t detectably affect mortality from breast cancer compared to starting at age 40. I doubt that Europeans have treatments for breast cancer that are that much better than those in the US; the second explanation is more likely the correct one.

As a surgeon and a scientist, I’ve found the resurrection of the debate over the efficacy of screening mammography to prevent breast cancer deaths to be simultaneously frustrating, educational, and confusing. I also realize that, although I specialize in the surgical treatment of breast cancer, in a way I’m not “on the front lines” in that I don’t see a lot of healthy women and advise them about whether to get their yearly mammogram beginning at age 40 or not. By the time they see me, they either have breast cancer, a suspicious lump (which means they will get more imaging if they haven’t already had it) or a mammographic abnormality that can’t be needle-biopsied and needs a surgical biopsy. That being said, as hard as it is for me sometimes to believe, I have served as leader of our breast cancer program and still work closely with existing leaders. I cannot avoid “taking a stand.”

Until recently, my stand was that we should stick with the existing guidelines until our professional organizations can examine the science and come up with new guidelines that reflect more recent studies. However, more and more I am of the opinion that we do over-screen. I still believe that mammography is of benefit in women of average risk beginning at age 50, but I’m becoming much less convinced of its utility in women under 50, at least those without identifiable risk factors for breast cancer. Such women should indeed have an individualized approach to screening. I say that, even though I recognize that our currently existing tools are grossly inadequate to give any more than a rough estimate. However, I also recognize that part of the equation has to be an accurate understanding of the absolute numbers and what the individual woman being screened values.

What that will mean remains to be worked out. Again. And the process will be messy. Same as it ever was. That will leave huge challenges to science and medical communicators. I hope we’re up to them.