I’ve written more times than I can remember about the phenomenon of overdiagnosis and the phenomenon that is linked at the hip with it, overtreatment. Overdiagnosis is a problem that arises when large populations of asymptomatic, apparently healthy people are screened for a disease or a condition, the idea being that catching the disease at an earlier stage in its progression will allow for more successful treatment. Two prominent examples include—of course—screening for breast cancer with mammography and screening for prostate cancer with prostate-specific antigen (PSA) testing, and I’ve written about the problem of overdiagnosis with each of them on many occasion. Basically, overdiagnosis occurs when the screening test picks up what we call “preclinical” disease (i.e., disease that hasn’t become symptomatic) that, if left untreated, would never become symptomatic or endanger the health or life of the patient). Although intuitively, it seems to the lay public (and, truth be told, most doctors) that detecting cancer earlier must be inherently better, it turns out that it’s way more complicated than you think. There is a price to be paid for early diagnosis in the form of overtreatment of disease that doesn’t need treatment and for disease that is destined to threaten the life of the patient earlier treatment doesn’t always result in better outcomes. Also, whenever you screen for a condition in asymptomatic people, you will always—always—find much more of it, and the significance of those added diagnoses is not always clear, as a new study in JAMA Oncology shows.
Before I get to the meat of the study, from my perspective, nowhere is the problem of overdiagnosis and overtreatment in cancer screening as pronounced than in the condition known as ductal carcinoma in situ (DCIS). DCIS is commonly referred to as “stage 0” breast cancer and is characterized by milk duct cells that appear malignant but remain confined to the milk ducts. In other words, they haven’t invaded the tissue surrounding the ducts. In general, DCIS is treated similarly to breast cancer, with surgical excision, either by mastectomy or breast conserving surgery, followed by radiation therapy if breast conserving surgery is used. Then, depending on its hormone receptor status, adjuvant treatment consists of blocking estrogen for five years. The rationale for this treatment is the view of DCIS as being a precursor to fully invasive breast cancer and that treating the DCIS will prevent the development of breast cancer. Over the last couple of decades, however, it has become clear that not all DCIS is created equal. Much of it will never progress to breast cancer in the lifetime of the woman (particularly if the woman is older, which means less time for fully malignant transformation to occur). Evidence suggesting this includes studies showing an increase in DCIS incidence by 16-fold since the 1970s, when mammography started to be introduced on a large scale, with little change in the incidence of invasive cancer. Today, 20-25% of mammography-detected breast cancer diagnoses are DCIS; forty years ago, DCIS was an uncommon diagnosis, except associated with an invasive cancer.
Screening for breast cancer tends to be based on the concept that progression from normal duct cells to invasive cancer follows more or less a linear progression, and then invasive cancer grows in the primary organ and eventually spreads, either through lymph vessels or the blood, to distant sites. By this paradigm, effective screening should result in finding disease earlier and thus preventing patients destined to develop metastatic disease from progressing to that point because the disease is removed before it can metastasize. This should result in what is referred to as a “stage shift,” in which the incidence of metastatic disease at diagnosis and of locally advanced disease declines, “shifting” the stage lower, to smaller, more localized disease. Unfortunately, although there is evidence of a small stage shift for breast cancer, the decline in incidence of advanced disease is far lower than the amount of DCIS diagnosed, meaning that most DCIS is almost certainly overdiagnosed. Indeed, in the accompanying editorial to this study, Laura Esserman and Christina Yau note that long-term epidemiology studies have demonstrated that the removal of 50,000 to 60,000 DCIS lesions annually has not been accompanied by a reduction in the incidence of invasive breast cancers. The problem, of course, is that we don’t know which cases of DCIS are destined to progress and which are not; so we have to treat them all the same.
This new study introduces a new wrinkle into the situation. You’ve probably seen the stories about it, such as one in the New York Times entitled Doubt Is Raised Over Value of Surgery for Breast Lesion at Earliest Stage. The reason is that the study questions whether treatment of DCIS actually decreases a woman’s risk of dying of breast cancer, based on an analysis of the Surveillance, Epidemiology, and End Results (SEER) database, a massive database maintained by the National Cancer Institute of cancer cases and outcomes. Specifically, the investigators (Narod et al) from the Women’s College Research Institute, Women’s College Hospital, and the Dalla Lana School of Public Health at the University of Toronto looked at the SEER18 database, which covers approximately 28% of the U.S. population. (The SEER database is maintained by a number of local registries; so it is basically a database of multiple local cancer registries.) From the SEER database, Narod et al identified 108,196 patients with DCIS diagnosed from 1988 to 2011 who formed the cohort studied. Patients with microinvasion (a tiny area of invasive cells) were excluded because that is generally considered invasive cancer. They then calculated a standardized mortality ratio (SMR), which is the ratio of the risk of dying in the study cohort compared to the risk of death in all women in the US population.
The first finding of the study is that the breast cancer–specific mortality over 20 years was 3.3% (95% CI, 3.0%-3.6%), which is similar to the lifetime risk of all women of dying of breast cancer, according to the American Cancer Society (2.7%). Next, the risk of death from breast cancer among all women with DCIS was 1.8 times greater than that of the US population (SMR, 1.8 [95% CI, 1.7-1.9]). Not surprisingly, the SMR decreased with increasing age at diagnosis, from 17.0 for women with DCIS before age 35 years to 1.4 for women older than 65 years. This is not surprising, given that younger women have a much longer time to live during which breast cancer can develop than older women. It makes intuitive sense. Indeed, as Esserman and Yau observe, DCIS in women under 40 is likely biologically different, because it would not in most cases be detected by mammographic screening and would thus be first noticed as a lump or other symptom. For these women, continuing current treatment protocols is probably appropriate.
There were other risk factors besides age at diagnosis of DCIS that predicted breast cancer mortality. For example, black women had a higher risk of death from DCIS than white, non-Hispanic women (HR 2.42 [95% CI, 2.05-2.87]; p < 0.001). Other factors were related to the tumor itself and included tumor size, grade, and the presence of a characteristic of DCIS called comedonecrosis, which has long been known to be a marker of more aggressive DCIS. The results of this analysis were robust and did not change substantially when patients with microinvasion or patients with a breast cancer death or recurrence within 6 months of the DCIS diagnosis were included.
Another result of the study is that aggressive treatment of nearly all DCIS does not appear to lead to a reduction in breast cancer mortality, which is consistent with an analysis of the NSABP B-17 and B-24 trials published four years ago. Basically, adding radiation therapy after lumpectomy also appears to have no effect on the risk of breast cancer mortality and might even be slightly higher. What this result suggests is that radiation therapy might not be necessary for the majority of DCIS cases.
The last finding is one that will be harder for doctors and patients to wrap their brains around, which is that DCIS might be more of a risk factor for ultimately developing breast cancer than a precursor to breast cancer. The reason is how similar ipsilateral (same side) and contralateral (opposite side) recurrence rates were, 5.9% and 6.2%, respectively, at 20 years. Recurrence of DCIS was not associated with increased risk of death for breast cancer, but an invasive recurrence was, hazard ratio 18.1 for ipsilateral recurrences, 13.8 for contralateral recurrences. As the authors put it:
The finding of greatest clinical importance was that prevention of ipsilateral invasive recurrence did not prevent death from breast cancer. Women with DCIS who developed an ipsilateral invasive in-breast recurrence were 18.1 times more likely to die of breast cancer than women who did not. For patients who had a lumpectomy, the use of radiotherapy reduced the risk of developing an ipsilateral invasive recurrence from 4.9% to 2.5% but did not reduce breast cancer–specific mortality at 10 years (0.9% vs 0.8%). Similarly, patients who underwent unilateral mastectomy had a lower risk of ipsilateral invasive recurrence at 10 years than patients who underwent lumpectomy (1.3% vs 3.3%) but had a higher breast cancer–specific mortality (1.3% vs 0.8%). Patients who had a mastectomy had cancers with a larger mean size and higher grade than patients who had a lumpectomy (eTable 5 in the Supplement). After adjustment for tumor size, grade, and other factors, the difference in survival for mastectomy vs lumpectomy was not significant (HR, 1.20 [95% CI, 0.96-1.50]; P = .11).
In other words, the type of treatment for DCIS didn’t matter. The risk of death from breast cancer was the same regardless, and very low. Moreover, these data challenge the existing paradigm of DCIS as a precursor of breast cancer:
Only a fraction of treated DCIS lesions progress to invasive breast cancer,13 but in the absence of treatment, the risk of invasive cancer is much higher.14 Also, mortality from breast cancer in women with DCIS increases substantially following the development of an invasive local recurrence.4,5 However, if DCIS were truly a (noninvasive) precursor of breast cancer, then a woman with DCIS should not die of breast cancer without first experiencing an invasive breast cancer (ipsilateral or contralateral), and the prevention of an invasive recurrence should prevent her death from breast cancer. Surprisingly, the majority of women with DCIS in the cohort who died of breast cancer did not experience an invasive in-breast recurrence (ipsilateral or contralateral) prior to death (54.1%). Furthermore, preventing the invasive in-breast recurrence (with mastectomy or radiotherapy) does not reduce mortality from breast cancer. This is in keeping with the findings of other studies.
This study is, of course, not without weaknesses. Any study that relies on the SEER database will have the weaknesses inherent in how the SEER database is put together. The database has been in existence many years and undergone multiple revisions. It’s also a bit slow to add data elements considered important to breast cancer treatment, such as HER2 status, as I discovered in 2009 when I wanted to do a project involving the SEER database. For instance, it’s possible that some cases of DCIS actually had microinvasion. Also, the authors didn’t have data on which DCIS cases were detected by mammography and which were symptomatic, although it’s likely that a small minority were symptomatic given that the vast majority of DCIS is detected by mammography. There were also no data on margin status. A positive margin (a margin in which tumor cells actually reach the surgical margin) is associated with a higher risk of recurrence. There are also vagaries of how multiple recurrences are coded.
So where to go from here? In an very rare occurrence, I cannot agree with Monica Morrow about this:
Dr. Monica Morrow, chief breast cancer surgeon at Memorial Sloan Kettering Cancer Center, said it made more sense to view D.C.I.S. as a cancer precursor that should be treated the way it is now, with a lumpectomy or mastectomy. She questioned whether those women who were treated and ended up dying of breast cancer anyway had been misdiagnosed.
In some cases, pathologists look at only a small amount of tumor, Dr. Morrow said, and could have missed areas of invasive cancer. Even the best mastectomy leaves cells behind, she added, which could explain why a small number of women with D.C.I.S. who had mastectomies, even double mastectomies, died of breast cancer.
Given the large numbers, I find it to be bordering on wishful thinking to speculate that so many of these women had been misdiagnosed that it significantly affected the results of this study. Yes, pathologists only look at a relatively small sampling of tumor, because to look at the whole thing for every patient would be so labor-intensive and expensive as to be impractical. The same could be said of any operations we do, and we’ve accepted this limitation since pathologists started looking at tissues under the microscope. It’s highly unlikely that rates of misdiagnosis are so high as to have made a difference.
What I think this study really means is that there is an urgent need for better molecular markers to determine which cases of DCIS are dangerous and which are not, something someone well known to readers of this blog has joined with others in arguing before in the medical literature. Until we have such prognostic markers, we will have this dilemma. Esserman and Yau suggest:
Narod and colleagues4 have assembled an impressive analysis on the basis of SEER data. There are limitations of SEER, but the large numbers and long-term follow-up provide a compelling case that it is time for change. The community of radiologists and surgeons needs to be part of the call for change. Given the low breast cancer mortality risk, we should stop telling women that DCIS is an emergency and that they should schedule definitive surgery within 2 weeks of diagnosis. The sum total of the data on DCIS to date now suggest that:
- Much of DCIS should be considered a “risk factor” for invasive breast cancer and an opportunity for targeted prevention.
- Radiation therapy should not be routinely offered after lumpectomy for DCIS lesions that are not high risk because it does not affect mortality.
- Low- and intermediate-grade DCIS does not need to be a target for screening or early detection.
- We should continue to better understand the biological characteristics of the highest-risk DCIS (large, high grade, hormone receptor negative, HER2 positive, especially in very young and African American women) and test targeted approaches to reduce death from breast cancer.
Of these, I think #4 is the most important, but unfortunately this approach will take years to bear fruit. In the meantime physicians and patients are left to struggle through as best we can with incomplete data. Nothing should yet change in the near term in our approach to DCIS in young women and with unfavorable characteristics, but it’s becoming increasingly clear that something needs to change is in our approach to lower risk forms of DCIS.
