Breast cancer is now recognized to be a group of genetically distinct diseases with different behaviors. Outcomes are maximized when therapy is tailored to patient and disease characteristics. Ductal carcinoma in situ (DCIS) can be eliminated with mastectomy, but this approach is considered unnecessarily radical in most cases for an entity that is non–life threatening, usually asymptomatic, and may not progress to invasive cancer. Breast-conserving surgery (BCS) consisting of removal of the DCIS to clear margins (tumor not touching the inked surface) is the minimal standard of care. The need for negative margins is already incorporated in guidelines for DCIS management based on strong evidence from randomized clinical trials ( 1–3 ), as well as a meta-analysis of the individual patient data from the randomized trials of BCS vs BCS and radiotherapy (RT) ( 4 ). The main topic of debate now is whether more widely clear margins of resection can eliminate the need for RT.
In this issue of the Journal, Wang et al. ( 5 ) use a network meta-analysis of the impact of different margin widths on local recurrence (LR) in DCIS to reach the somewhat surprising conclusion that wider margins should be a “priority” for all DCIS patients, regardless of the receipt of RT. This recommendation is based upon an analysis of 24 cohorts from 21 studies (three randomized trials) involving 7564 patients treated over 25 years. A margin width of 10 mm or greater was observed to decrease the risk of LR compared with a margin of 2 mm or greater (odds ratio = 0.46; 95% confidence interval = 0.29 to 0.69), leading to a recommendation by the authors for wide excision. This would represent a major change in the current standard surgical approach to DCIS, as illustrated even by the selected dataset used in the meta-analysis, where margins greater than 1 cm were obtained in only 4% of patients treated with RT and in 9% of those treated with excision alone. Before embarking on such a change in practice, clinicians need to ask whether the observed association between larger margins and lower LR is valid, whether there is evidence of a causal relationship, and whether this change would improve the health of women with DCIS.
In any observational study, there is concern that the results could be influenced by important confounding factors. This is a particularly challenging problem in this meta-analysis incorporating the results of many observational studies where patients were not randomly assigned to the different margin widths. In fact, the authors acknowledge the possibility that bias caused by unobserved selection effects across studies cannot be excluded, particularly when only five of 21 studies included in the meta-analysis reported margin widths of 1 cm or more. This subset of patients may have differed from the much larger cohort used to determine recurrence for patients with smaller margins in ways that could result in the lower rate of LR. For example, they may have been older age ( 4 ), had lower tumor grade ( 4 ), or greater use of boost dose of RT ( 6 ), or tamoxifen ( 3 )—all factors that can have an important effect on risk of LR. Furthermore, the treatment cohorts may have differed with regard to margin processing and evaluation, which was not standardized between institutions or even within institutions over time. The patients in the five studies evaluating 1 cm margins were generally treated in a more recent time period than those in the other 16 studies, and rates of LR have declined over time because of improvements in imaging and pathology independent of margin width ( 7 ). The inability to control for this complex, but potentially powerful, residual confounding may limit the validity of the findings.
Another challenge to this study is the uncertainty in the inference of causality ( 5 ). It is just as clinically plausible to hypothesize that tumors with more favorable prognosis were more likely to result in more widely clear margins. Results from a prospective study of a large group of patients with DCIS conducted by the Eastern Cooperative Oncology Group ( 8 ) do not support a causal relationship between margin width and LR. In that 671-patient study, a minimal margin width of 3 mm was required, complete embedding and sequential sectioning of the specimen were specified, and post-excision mammograms were mandated, removing many of the sources of variation present in retrospective studies. In this context, the actuarial risk of LR at 5 years did not differ statistically significantly between patients with margins of 1 cm or greater (n = 329) compared with those with smaller margins (n = 332), even after stratification by grade.
The final question concerns the potential consequences of adopting a margin threshold of 1 cm as a priority for women with DCIS. Re-excision is already a common procedure; in a recent population-based study ( 9 ), 31% of women with DCIS successfully treated with breast conservation underwent re-excision ( 9 ). This proportion would undoubtedly increase if a policy of more widely clear margins is adopted because approximately half of the re-excisions in a recent report were performed to achieve an arbitrary margin width greater than tumor cells not touching ink ( 10 ). More importantly, this recommendation may result in more women being deemed ineligible for BCS by their surgeons initially and more women choosing to undergo mastectomy after an initial BCS attempt fails to achieve the “best” margin width because fear of LR is strongly associated with mastectomy use ( 11 ). Prospective trials have not identified a subset of women not benefiting from RT, and more accurate estimates of RT use in DCIS than the 50% cited by Wang et al. ( 5 ) suggest rates above 70% ( 11 , 12 ). This suggests that aggressive surgical treatment will do more harm than good because only a minority of patients treated without RT might benefit from a wider margin, whereas a much larger group would be subject to overtreatment.
Ultimately, this article illustrates the challenges to making clinical policy recommendations based on observational data. Meta-analyses, regardless of the sophistication of the analytic technique used, are only as good as the data they include. The Early Breast Cancer Trialists Collaborative overview analysis has been such a powerful force in invasive breast cancer management because it includes patient-level analysis of data from prospective randomized trials. This approach resulted in the identification of the survival benefit associated with local control that was not seen in any of the individual trials of BCS with and without RT ( 13 ) and clarified estimates of the effect of many of the systemic therapies used today. Although trials are generally superior to retrospective approaches, we must ask in the end whether a study of margin width is worthy of the large investment such a trial would necessitate. This is not an ethical issue, as suggested by Wang et al. ( 5 ), because the “right” margin width remains uncertain. But there is an issue of feasibility. Although margin width remains a clinical dilemma, in the National Surgical Adjuvant Breast and Bowel Project B17 and B24 trials that required margins of tumor not touching ink, only 72 (2.8%) of 2612 participants treated with BCS with and without RT died of breast cancer after 15 years of follow-up ( 3 ). Thus, any net benefit of more widely clear margins on the overall health of women with DCIS would be extremely small or negligible. The patient and financial resources needed for a study to quantify such a small difference are likely to do more good if devoted to identifying those factors that result in the progression of DCIS to invasive cancer—knowledge that is critical to our ability to tailor the extent of treatment and optimize net benefit in individual patients. Observational research on the margin question would be greatly facilitated by the adoption of a standard method of specimen processing and examination. At present, there is no compelling evidence that bigger is better for margins in DCIS, and physicians should be cautious about the routine use of large resections or re-excision to achieve an arbitrary negative margin width.
The authors declare no conflicts of interest.