Introduction

Neoadjuvant chemotherapy (NAC) is currently accepted as a preferred option for treating breast cancer, and its usage has increased over time [1]. NAC has several advantages over upfront surgery, including early observation of response to systemic treatment and modification of adjuvant treatment and breast-conserving surgery (BCS) for patients with clinically large tumors who initially require total mastectomy [2, 3]. Furthermore, several trials have shown that NAC has an equivalent effect on survival outcomes as adjuvant chemotherapy [4].

The major concern regarding BCS is the resection margin (RM). Current guidelines strongly recommend achieving “no tumor on ink” for invasive breast cancer [5,6,7]. The importance of no ink on the stained margin after BCS is associated with the risk of local recurrence (LR) [8, 9]. However, the application of the RM definition after NAC is unclear. Guidelines lack clear evidence of appropriate width for RM after NAC, and few studies have reported the effect of margin status on oncological outcomes [10]. In addition, the significance of clear RM for surgical specimens could be weakened after NAC because some tumors shrink with scattered or multifocal patterns [11], and minimally remaining lesions might be effectively eradicated in the era of a newly effective regimen of cytotoxic drugs and radiation treatment (RT) [2, 12].

Moreover, previous studies recommended a margin width of ≥ 2 mm for specimens after surgery for ductal carcinoma in situ (DCIS) [13]. The significance of RM status remains unclear when patients show pathologic complete response (pCR) with residual DCIS in the breast. Thus, surgeons are required to consider additional resection for involved or close RM, against the preference to preserve the breasts as much as possible after NAC.

This study aimed to investigate the effect of RM on LR by comparing patients with involved RM and close (≤ 2 mm) or clear RM. Previous studies only included a small number of patients because only few patients refuse further re-excision for RM after BCS. To the best of our knowledge, this is the largest study of three major institutions in South Korea to analyze the effect of RM after BCS following NAC.

Patients and methods

Study design

The study protocol was reviewed and approved by the review board (IRB) of the following three institutions in Korea: Asan Medical Center (AMC), Samsung Medical Center (SNH), and Seoul National University Hospital (SNUH) (IRB No.: AMC, 2017–1341; SMC, 2021–03-096–003; SNUH, 2014–015-1210). The protocol was reviewed and approved by our institution, and the study followed the Declaration of Helsinki and good clinical practice guidelines. The requirement for informed consent was waived.

We obtained baseline clinicopathologic data and reviewed detailed information of female patients with breast cancer who underwent curative BCS for invasive cancer between January 2008 and December 2016. All patients received NAC followed by surgery and adjuvant whole-breast radiation therapy (WBRT) (Supplementary table S1). We excluded patients with stage IV breast cancer, recurrent breast cancer, bilateral breast cancer, or synchronous or metachronous cancer in other organs. In case of close or involved RM, surgeons further resected the breast in the direction of reported margin based on clinical experience of each physicians. Patients who underwent further resection via total mastectomy were also excluded, whereas those who completed the surgical treatment with partial resection were included. Initial breast cancer was clinically staged according to the 8th American Joint Committee on Cancer staging criteria. All patients were diagnosed with invasive breast cancer using core needle biopsy, fine-needle aspiration or vacuum-assisted breast biopsy of abnormal findings on breast sonography or mammography at each institution. Hormone receptor (HR) status, including estrogen and/or progesterone receptors, was reviewed by pathologists from each institution based on immunohistochemistry findings, with positivity defined as > 1% or Allred scores of 3–8 [14]. Human epidermal growth factor receptor type 2 (HER2) status was assessed using anti-HER2 antibodies and/or fluorescence in situ hybridization (FISH) or silver in situ hybridization (SISH). When the result of HER2/chromosome enumeration probe 17 (CEP17) ratio was > 2.0 on FISH or SISH, tumor was regarded as HER2 positive according to American Society of Clinical Oncology/College of American Pathologists (ASCO/CAP) guideline. Data on the Ki-67 labeling index were not collected because of different cut-off values used in institutions. As we focus on the effect of residual tumors on local recurrence of breast, pCR was defined as the absence of residual invasive cells in the breast. Regarding the preoperative radiographic diagnosis before surgery, all of institutions had conducted breast MRI to assess the response to neoadjuvant chemotherapy before, in the middle of, and after administration of neoadjuvant chemotherapy. Additionally, preoperative sonography was once more conducted to precisely check the tumor size and range before surgery.

Assessment of resection margin

Involved RM was defined as the presence of ink on the radial margin of the final surgical specimen, regardless of the intraoperative frozen section results. Margin widths were reviewed based on data from histology reports from each institution, and RM was classified into close and clear according to widths of ≤ 2 mm and > 2 mm, respectively. Data of superficial and deep RM were not collected because they were previously reported to not be significant factors affecting LR [15]. Furthermore, detailed pathologic reports for close margin were not also collected as one of institutions in our study did not report the type of tumors for close RM.

Recurrence and recurrence-free survival

LR, the primary endpoint of this study, was defined as the first recurrence in any quadrant of the ipsilateral breast. Recurrence at the breast skin was excluded from the LR. LR-free survival (LRFS) was defined as the interval between the dates of surgery and pathologic confirmation of LR. Assuming that neither regional recurrence nor distant metastasis (DM) is associated with the effect of RM on LR, the events without concurrent LR were not regarded as censoring events when analyzing LRFS. DM-free survival (DMFS) was defined as the time interval between the date of surgery and the time of radiologic or pathologic confirmation of distant metastasis. Recurrence-free survival (RFS) was defined as the interval between the date of surgical treatment and the date of diagnosis of any recurrence including LR, regional recurrence and DM.

Statistical analyses

All analyses were performed using SPSS (version 25.0; IBM Corp., Armonk, NY, USA). Demographic and clinicopathologic variables were compared using Student’s t-test for continuous variables and Pearson’s χ2 test for categorical variables. Survival analyses were performed using the log-rank test to analyze the difference in survival outcomes between groups, and the curves were derived using the Kaplan–Meier method. Cox proportional hazard regression was used to adjust for variables affecting pathologic response and survival outcomes and it was also used to estimate the hazard ratio. Propensity score matching procedure was also performed to reduce the effects of several clinicopathological variables using “MatchIt” R package (version 3.6.3; R Core Team, Vienna, Austria). Statistical significance was set at p values < 0.05. All curves were drawn using GraphPad Prism™ (version 9.0; GraphPad Software, San Diego, USA).

Results

Patient demographics and characteristics

We identified 2,803 patients who underwent NAC followed by BCS and met the inclusion criteria. Nineteen patients who underwent total mastectomy for involved RMs were excluded. The median follow-up period was 62.3 months (range, 0.4–157.2 months). Clinicopathological characteristics of all patients are listed in Supplementary table S2.

Patients were classified into subgroups according to their RM status and pathologic response (Table 1). We divided them into three broad groups: patients with pCR (RpCR, subgroups 5–8, n = 786), patients with non-pCR and clear or close RM (R0, subgroup 1–2, n = 1,949), and patients with non-pCR and involved RM (R1, subgroups 3–4, n = 68), with median follow-up periods of 70.4, 71.7, and 71.6 months, respectively (Table 2). Patients in the RpCR group were significantly older at the time of operation and had lower clinical T stage, lower HR positivity, higher HER2 positivity, and higher histologic grade than those in the other groups. Among the 786 patients who had pCR in the breast, 500 (subgroup 8, 63.6%) had no tumor and 286 (subgroups 5–7, 36.4%) had residual in situ lesion.

Table 1 Subgroup classification according to RM status and pathologic response
Full size table
Table 2 Clinical characteristics of patients according to the resection margin and pCR status
Full size table

Survival outcomes

In total, 23, 5, and 121 ipsilateral breast tumor recurrence (IBTR) events were noted in the RpCR, R1, and R0 groups, respectively. The five-year LRFS rates were 97.4%, 91.5% and 94.0% for the RpCR, R1, and R0 groups, respectively. The Kaplan–Meier curves revealed that patients in the RpCR group had higher LRFS than those in the R1 (hazard ratio [HR], 2.55; 95% confidence interval [CI], 0.97–6.72; log-rank p = 0.049) and R0 (HR, 2.15; 95% CI, 1.37–3.35; p = 0.001) groups (Fig. 1a). In contrast, the LRFS in the R0 group was not significantly different from that in the R1 group (HR, 1.20; 95% CI, 0.49–2.93; p = 0.692). DMFS and RFS of patients in the RpCR group were also significantly higher than those in the other groups (p < 0.001, Fig. 1b-c). There was no significant difference in DMFS (p = 0.598) and RFS (p = 0.338) between patients in the R1 and R0 groups (p < 0.001).

Fig. 1
figure 1

The Kaplan–Meier curves showing the survival outcomes according to the resection margin status and pathologic response for all patients (a-c) and after propensity score matching (d-f). The hazard ratio was calculated using a univariate Cox regression analysis. Abbreviations: R1: involved resection margin group; R0: clear or close resection margin group; RpCR: pathologic complete response group; CI: confidence interval; LRFS: local recurrence-free survival; DMFS: distant metastasis-free survival; RFS: recurrence-free survival

Full size image

Importantly, to minimize the effect of confounding factors between the R0 and R1 groups (supplementary table S3), we performed 1:3 propensity score matching by incorporating clinicopathologic variables, yielding 66 and 198 patients in the R0 and R1, respectively. The variables were not significantly different between the two groups after propensity score matching, and no significant difference in survival outcomes was observed between the two groups (Fig. 1d-f).

Among patients with non-pCR, there was no significant difference in LRFS between those with involved (subgroup 3,4), close (subgroup 2), and clear RMs (subgroup 1) (p = 0.492) (Fig. 2a). Additionally, for the 68 patients with involved RM in the R1 group (subgroups 3,4), no difference in LRFS was observed with respect to the pathology of tumors with RM (DCIS vs. invasive cancer, HR, 0.54; 95% CI 0.09–3.26; p = 0.497) (Fig. 2b).

Fig. 2
figure 2

Kaplan–Meier curves of the patients without pCR according to margin widths (a) and involved tumor types (b). The hazard ratio was calculated using a univariate Cox regression analysis. Abbreviations: LRFS: local recurrence-free survival; R1: involved resection margin group; R0: clear or close resection margin group; IDCa: invasive ductal carcinoma; DCIS: ductal carcinoma in situ; CI: confidence interval; pCR: pathologic complete response

Full size image

Focusing on the surgeon’s point of view, we analyzed the LRFS of non-pCR patients, including those with pCR and residual DCIS (subgroup 6–7). Similarly to the aforementioned results, the log-rank test showed RM status was not a risk factor for LRFS (subgroup 1, 5 vs. subgroup 2, 6 vs. subgroup 3, 4, 7, p = 0.317).

Multivariate analysis for LRFS between R0 and R 1

Clinical T and N stage, HR status, and histologic grade were significantly associated with LR according to the log-rank test. Cox regression analysis that was adjusted for other prognostic variables revealed no significant difference in LRFS between the R0 and R1 groups (HR, 2.05; 95% CI, 0.64–6.58, p = 0.227) (Table 3, Supplementary fig. S1). Moreover, all the abovementioned factors were not significant variables predicting LR.

Table 3 Univariate and multivariate analyses for local recurrence-free survival among patients without pCR
Full size table

Subgroup analysis according to subtypes

Depending on pre-chemotherapy pathology reports, 1,960 patients could be distinguished according to different tumor subtypes: 908 (46.3%) with HR + /HER2-, 303 (15.5%) with HR + /HER2 + , 228 (11.6%) with HR-/HER2 + , and 521 (26.6%) with HR-/HER2-. The beneficial effect of clear RM on LRFS was not observed for all subtypes of non-pCR tumors (Fig. 3a–d). Furthermore, the log-rank test showed no difference in DMFS and RFS between the R0 and R1 groups for all subtypes (Fig. 3e, f, Supplementary fig. S2). Especially, patients with HR-/HER2- subtype in the R1 group had a tendency of having poorer DMFS and RFS rates than those in the R0 group.

Fig. 3
figure 3

The Kaplan–Meier curves of survival outcomes stratified according to HR and HER2 status. The Kaplan–Meier curves of LRFS between the R0 and R1 according to tumor subtypes are shown (a–d). Additionally, the survival graphs of DMFS and RFS of HR-/HER2- subtype are shown (e–f). The hazard ratio was calculated using a univariate Cox regression analysis. Abbreviations: LRFS: local recurrence-free survival; HR: hormone receptor; HER2: human epidermal growth factor receptor 2; R1: involved resection margin group; R0: clear or close resection margin group; CI: confidence interval

Full size image

Survival analysis among patients with residual DCIS

We further investigated the effect of residual DCIS on RM in the RpCR group (subgroups 5–8). We identified 286 patients with pCR with DCIS alone after surgery: 12 with involved (subgroup 7), 10 with close (subgroup 6), and 264 with clear (subgroup 5) RM. During the follow-up period, there were 1 LR events and 11 LR events in patients with involved RM (subgroups 7) and clear or close RM (subgroup 5–6), respectively. The five-year LRFS rates were 96.1% and 90.0% for the subgroup 5–6 and subgroup 7, respectively. The log-rank test revealed no significant difference in LRFS between the two groups (HR, 2.49; 95% CI, 0.32–19.37; p = 0.366) (Fig. 4).

Fig. 4
figure 4

The Kaplan–Meier LRFS curves according the RM status among 286 patients showing pCR with residual in situ lesion in the RpCR group. The hazard ratio was calculated using a univariate Cox regression analysis. Abbreviations: RM: resection margin; RpCR: pathologic complete response group; LRFS: local recurrence-free survival; CI: confidence interval; pCR: pathologic complete response

Full size image

Discussion

In the current study, we could not determine whether clear RM status after NAC and BCS followed by WBRT was associated with improved benefit for LRFS in all subtypes. However, we observed that achieving clear RM in cases of tumors showing pCR with residual DCIS would not result in a survival benefit. Our results suggest that struggling to gain a clear RM to reduce LR risk, as opposed to the expectation for cosmetic benefit after NAC, does not always lead to better prognosis.

A few retrospective studies have investigated the effect of clear RM on LR. Wimmer et al. [16] retrospectively analyzed 416 patients who underwent BCS after NAC and observed no significant difference in LRFS, DFS, and overall survival with respect to margin widths (RM > 1 mm vs. 0 < RM ≤ 1 mm; 5-year LRFS, 91% vs. 94%; p = 0.940). Similarly, Lin et al. [10] analyzed 161 patients who underwent BCS after excluding those with involved RM and reported similar results that specimens with RM ≥ 1 mm had no benefit for LRFS compared with those with RM < 1 mm (HR, 0.44; 95% CI 0.14–1.38; p = 0.161). All the abovementioned studies suggested that the definition of RM as “no tumor on ink” would be safe for application in the NAC setting. However, these studies analyzed for a small number of patients, including patients with pCR into the R0 group and excluded patients with involved RM. In another study investigating a large number of patients, Tyler et al. [17] conducted a population-based analysis of 10,863 patients and reported similar LRFS between patients with involved, close, and clean margins (p = 0.084). Additionally, a lower BCSS rate was observed in patients with positive RM than in those with clear RM (Cox regression analysis, p = 0.024). They concluded that omitting further re-excision would be acceptable for carefully selected patients with positive margin status. However, systemic chemotherapy was administered to only 36.4% of patients in this study, and there was no mention of whether it was administered in the adjuvant or neoadjuvant setting.

The risk of RM involvement after NAC was three times higher (2.5%–7.8%) than that after upfront surgery in a meta-analysis study [18]. Of all patients in our study, 2.9% of patients had involved RM but they did not have worse LRFS compared to patients with clear or close RM. We assumed several explanations for this result. First, although the ink on the specimen is present, there is a possibility of no residual tumor in the remnant cavity. Tang et al. [19] compared pathologic reports of lumpectomy margins with those of shaved cavitary margins and concluded that the lumpectomy margin is not reliable for predicting cavity status with an overall accuracy of 64.9%. Second, improvement in high-resolution magnetic resonance imaging and multiparametric evaluation enables precise measurement of residual tumors after NAC [20, 21]. This would allow surgeons to include all residual lesions in the resection volume, although some tumors shrink in a multifocal pattern. Third, a retrospective study reported a negative correlation between clinical tumor size and BCS conversion rate [22]. Patients with large tumors at the time of pre-treatment might have undergone mastectomy, which could lead to selection bias, making it impossible for clinically large tumors to be analyzed in this study. Lastly, a high dose of RT boost would have affected the better prognosis for patients with positive RM [23].

The response to NAC differs across various breast cancer subtypes. Among the subtypes, TNBC shows the highest sensitivity to NAC and pCR rate [24, 25]. Additionally, the shrinkage pattern of tumor regression after NAC is different among the subtypes, and the HR + /HER2- subtype usually shrinks in a scattered pattern, while HER + or TNBC tumors show concentric patterns [26, 27]. It can be inferred that the effect of RM status on recurrence may differ across subtypes; however, we did not find a difference in LR risk according to RM for all subtypes. Especially, patients with TNBC in the R1 group had a tendency of poorer DMFS and RFS than those in the R0 group. TNBC has no benefit owing to hormone therapy or HER2-targeted therapy, implying that entire tumor resection without residual lesion might be important for survival outcomes. Moreover, the number of TNBC tumors which were down-staged in tumor size after NAC was significantly higher in the R0 group than in the R1 group in our study (73.9% vs. 37.5%, p = 0.035). As TNBC with residual lesion or refractory to NAC has a higher probability of systemic recurrence than other subtypes, the difference in response rates to NAC between the two groups may have resulted in the tendency of poorer survival of patients in the R1 group [25, 28]. Although there was no statistically significant difference, it may be due to the small number of patients with TNBC subtype.

While the margin width for DCIS is well known to be enough with < 2 mm at upfront surgery, there is no criteria of allowed width for residual DCIS after NAC [13, 29]. Few studies have investigated how residual DCIS, which is regarded as pCR, would affect survival outcomes. We demonstrated that the prognosis of patients with residual DCIS was not significantly affected by RM status despite the small number of subgroups. Several reports have shown that patients with residual DCIS after NAC do not have different survival outcomes compared with patients with no tumors in the breast [30, 31]. The results suggest that residual DCIS after NAC is clinically or pathologically different from usual DCIS; thus, there may be no need to worry about residual DCIS on RM.

To the best of our knowledge, this is the largest study to analyze the effect of RM after BCS following NAC; however, our study had several limitations. The study was conducted in multiple institutions; thus, there was hidden bias owing to the heterogeneity of data, such as surgeon factors, methods for pathologic review, and surveillance strategies. Nevertheless, we could include a large number of cases from three institutions and combine the patient population more closely with real-world data. In addition, due to the nature of retrospective study, we could not review several clinical, radiologic and pathologic variables such as patients’ preference or feasibility, microcalcification and multiplicity that would have affected the pathologic response and survival outcomes. Further study would be needed for more reliable results incorporating abovementioned features in the analysis. Additionally, we classified patients who had no more breast tissue for resection in the direction of the involved or close margin into the R1 group, raising the possibility of selection bias. But the number of those patients was so small according to medical records that it could be ignored. Also, we defined the pCR as the absence of invasive breast tumor in the breast instead of both breast and axilla that are widely accepted. This can lead to selection bias, but the results would not be different from ours as our study focused on the effect of RM on local recurrence in the breast. Finally, patients who received NAC generally had a high probability for systemic recurrence, and almost half of the patients (53.5%) showed DM without LR. Focusing on the primary endpoint, we did not censor patients at the time of distant metastasis so the LR rate could have been overestimated considering the competing risk. To overcome this issue, we further analyzed LRFS after censoring patients at the time of all recurrences, but still no significant difference in LRFS was observed according to RM status (p > 0.05, data not shown).

In conclusion, our study showed no difference in LRFS rates for involved or close RM compared with clear RM after BCS following NAC. We suggest that the relative risk of LR according to margin status after NAC might differ from that of upfront surgery, and the definition of “no tumor on ink” should be revisited. Additionally, for patients who achieved pCR with DCIS alone, DCIS on RM did not increase the risk of LR. Clinicians should not overlook the importance of clear surgical margins; however, re-excision for close or involved margins after enough volume excision to reduce the LR risk can be omitted for selected patients. Further studies comparing patients with involved or close RM after NAC and those after upfront surgery would validate our results.