NSABP B-35 margin-width analysis
ForPostmenopausal HR+ DCIS after lumpectomy, WBI and 5yr endocrine therapy
5.6% vs 4.0%
1mm cutoff, absolute difference 1.6%; 2mm cutoff 5.3% vs 3.8%
TL;DR10yr IBTR 5.6% vs 4.0% at 1mm cutoff (abs diff 1.6%), 5.3% vs 3.8% at 2mm (abs diff 1.5%).
Reported via The ASCO Post →
The RT-relevant read is that every patient here got whole-breast irradiation plus 5 years of endocrine therapy, so the 1.5% to 1.6% margin penalty is the residual after full adjuvant treatment. That gates transfer: it says nothing about a close margin when RT is omitted or refused, which is where the margin question actually bites.
In a postmenopausal woman with HR+ DCIS whose lumpectomy margin is under 2mm and who will complete whole-breast RT plus 5yr endocrine therapy, this argues re-excision buys little; it does not extend to premenopausal, HR-negative, or RT-omitted pts.
Every patient here received whole-breast irradiation with an optional boost plus 5yr endocrine therapy, so 1.5% to 1.6% is the residual margin effect after full adjuvant treatment. Boost use by margin group is not reported in source, which matters: it determines whether a close margin was tolerated or quietly compensated.
A margin under 2mm carried a 10yr IBTR of 5.3% vs 3.8% for ≥2mm, an absolute 1.5% penalty, in the largest randomised-trial population yet examined. That directly gates the return-to-theatre decision, though margin width was not randomised and the narrow group is enriched for disease that could not be cleared.
9 details 1 trial watching
Secondary margin-width analysis of NRG Oncology/NSABP B-35, a double-blind randomised trial of tamoxifen vs anastrozole that enrolled 3,104 postmenopausal women 2003-2006. Because local recurrence did not differ between the two endocrine arms, the arms were pooled and margin width analysed across the whole population. Margin data were collected prospectively by participating pathologists.
Postmenopausal women with hormone receptor-positive DCIS treated with lumpectomy. Two overlapping analysis cohorts: n=2,707 with margins classifiable as <1mm (close/indefinite) vs ≥1mm, and n=2,546 with the closest margin measured, permitting a <2mm vs ≥2mm cutoff.
All patients received whole-breast irradiation with an optional boost. Dose, fractionation and boost uptake are not reported in the source, so the RT exposure behind these recurrence rates cannot be characterised beyond "whole breast, boost optional".
Primary endpoint of interest: cumulative incidence of ipsilateral breast tumor recurrence at 10 years, analysed at both the 1mm and 2mm cutoffs. A secondary analysis of all breast cancer events, including contralateral disease, was also performed.
The prevailing 2mm threshold rests largely on the SSO/ASTRO/ASCO DCIS consensus, whose meta-analytic base drew heavily on series with variable and often absent adjuvant therapy. This analysis puts the same question to a uniformly irradiated, uniformly endocrine-treated randomised trial population, which is why the residual margin effect looks so much smaller.
Margin width was not randomised, and patients re-excised on trial carry their final margin, so the narrow-margin group is enriched for disease that could not be cleared. Source reports no hazard ratios, confidence intervals or event counts, and no multivariable adjustment for grade, size, age or boost use.
The claim is that a 1.5% to 1.6% absolute 10-year difference does not justify routine reoperation, which is a value judgment about the trade against anxiety, cosmesis and cost rather than a statistical one. The difference was statistically significant at the 1mm cutoff, so the argument turns on clinical meaningfulness, and a patient who weighs local recurrence heavily could reasonably read the same number differently.
| Cutoff | Narrow margin | Wider margin | Absolute difference |
|---|---|---|---|
| 1 mm | 5.6% (n=502) | 4.0% (n=2,205) | 1.6% |
| 2 mm | 5.3% (n=879) | 3.8% (n=1,667) | 1.5% |
Prospectively collected margin data from a large RCT population directly contests the 2mm re-excision threshold, but the margin comparison itself is non-randomised and abstract-only.
- Does the finding hold when whole-breast RT is omitted? not yet Assessment of Biosignature Classification of DCIS for RadioTherapy Benefit Post Lumpectomy (ABCD RT) Phase 3n=5270 · primary completion 2039-07 · randomises RT omission in biosignature-low DCIS
- Boost use and dose by margin group
- Applicability to premenopausal or HR-negative DCIS
📚 Sources · 📄 1 paper
Abstract
The longer read
The interesting thing about this analysis is not the effect size but the population it was measured in. The 2mm margin threshold for DCIS entered guidelines through a consensus process built on pooled series in which adjuvant treatment was heterogeneous: some patients irradiated, many not, endocrine therapy inconsistently given and inconsistently recorded. In that setting margin status is doing a lot of work, because it is one of the few levers acting on residual disease. B-35 is the opposite situation. Every woman received whole-breast irradiation, every woman was assigned 5 years of an endocrine agent, and the endocrine question was itself randomised. What the margin analysis therefore measures is the residual contribution of surgical clearance after two other effective local and systemic treatments have already been applied. A 1.6% absolute difference at 10 years is what is left over, and the honest reading is that it is small because the other two modalities absorbed most of the risk, not because margins are irrelevant.
That framing also sets the boundary of the result. A close margin in a woman who declines radiotherapy, or who is not a candidate for endocrine therapy, is a different question, and this dataset cannot speak to it. The commentary accompanying the presentation makes exactly this point, and it deserves more weight than it usually gets: in practice a meaningful fraction of DCIS lumpectomy patients do not complete both adjuvant treatments, and for them the pre-existing threshold logic still applies. The generalisation risk here runs in one direction, toward under-treating the patient who was never going to receive the full trilogy of therapy that makes the narrow margin safe.
Methodologically the comparison is observational inside a randomised trial, which is a real limitation rather than a formality. Margin width was not assigned; it emerged from tumour biology, extent of disease and surgical judgment. Patients re-excised during the trial are counted at their final margin, so the group that remained under 1mm is enriched for disease that could not be cleared, which is plausibly a marker of extent rather than a simple measurement artefact. That biases toward showing a worse outcome in the narrow group, and the observed difference is still only 1.6%. The direction of the bias therefore strengthens rather than weakens the authors' conclusion. What is missing is the other half of the picture: no hazard ratios, no confidence intervals, no event counts, and no adjustment for grade, lesion size, age or whether a boost was delivered. Boost use in particular is worth knowing, because a surgeon confronting a close margin and a radiation oncologist adding a boost are two responses to the same problem, and if boost was preferentially given to the narrow-margin group then part of what is being measured is compensation rather than tolerance.
The difference at the 1mm cutoff was statistically significant, so the argument is explicitly about clinical meaningfulness rather than about whether an effect exists. That is a legitimate argument, and one that trades a real 1.5% to 1.6% against reoperation, poorer cosmesis, delay and cost. Different patients will price that trade differently, and this analysis is better understood as licensing a genuine conversation than as retiring the 2mm rule. Peer-reviewed publication with adjusted estimates, event counts and boost data would settle how much of the residual difference survives.