Predicting non-sentinel node metastasis during surgery in breast cancer patients with one to three positive sentinel node(s) on a frozen biopsy result after neoadjuvant chemotherapy

Purpose Our goal was to develop a tool that could accurately predict the possibility of non-sentinel lymph node metastasis (NSLNM) during surgery, allowing a surgeon to decide the extent of further axillary lymph node dissection (ALND) intraoperatively for patients with one to three positive sentinel lymph node(s) (SLN) after neoadjuvant chemotherapy (NAC). Methods In a retrospective analysis of the Asan Medical Center (AMC) database, we included 558 patients’ records who were treated between 2005 and 2019. Using chi-square and logistic regression with a bootstrapped, backward elimination method, 13 factors were assessed for their utility in predicting NSLNM. Based on the results of the univariate analysis for statistical signicance, the number of positive SLN(s), number of frozen nodes, progesterone receptor (PR) positivity and clinical N stage were selected for the multivariate analysis and used to generate a nomogram for predicting residual nodal disease. The resulting nomogram was validated using a more recent, different time window patient group at the AMC. the univariate analysis: age at diagnosis, tumour grade, hormone receptor score, HER2 status, classication into four subtypes (HR+/HER2−, HR+/HER2+, HR-HER2+, HR−/HER2−), Ki-67, clinical T stage and N stage before NAC and its degree of response to therapy, number of metastatic sentinel nodes, total number of submitted sentinel nodes for frozen section biopsy and the greatest tumour invasion depth of the sentinel nodes.


Introduction
Although previous trials of the survival bene t for patients who underwent neoadjuvant chemotherapy (NAC) failed to demonstrate a relative superiority over patients treated with adjuvant chemotherapy, the addition of NAC for treatment of eligible breast cancer patients has been widely accepted. NAC can help reduce the need for total mastectomy, full axillary lymph node dissection (ALND) and the associated morbidity without increasing locoregional recurrence [1]. Sentinel lymph node biopsy (SLNB) after NAC in patients with clinically positive axilla resulted in acceptable accuracy, establishing it as a viable axillary management strategy [1][2][3]. Our institutional practice pattern has also included SLNB as the initial approach for the axilla after NAC unless the patient had a signi cant disease burden remained or progressive disease.
According to the recent National Comprehensive Cancer Network (NCCN) guidelines, complete ALND has been the standard surgical management for patients who presented with node-positive breast cancer after NAC [4]. Non-sentinel lymph nodes (NSLNs) were found to be tumour free in a signi cant proportion of breast cancer patients with positive SLNs who had received ALND. Among 160 patients with macrometastasis to SLNs, Dingemans et al. reported that 59% of primary breast cancer patients lacked NSLNM. These patients might received unnecessary ALND, which could have resulted in signi cant complications and provided no therapeutic bene ts [5]. Jeruss et al. conducted a study that included 104 patients who received NAC, had a positive SLN andunderwent ALND between 1997 and2005 [6]. Of their research cohort, 44% did not have positive non-SLNs. They looked at factors such as lymphovascular invasion (LVI), method of SLN metastasis detection, multicentricity, ALN status at presentation and pathological tumour size to see if they could predict additional NSLNM. Based on this nding, they derived the MD Anderson nomogram, which has a signi cant AUC. In addition, 132 patients were followed prospectively between 2001 and 2007 in a study by Gimbergues et al. All patients were given NAC and had SLN biopsy with ALND levels I and II [7]. They reported that 47.1% of their patient population did not have NSLNM, and they tested the accuracy of previous nomograms from the Memorial Sloan-Kettering Cancer Center, the MD Anderson Cancer Center and the Tenon Hospital in Paris, with AUC values ranging from 0.7 to 0.8.
However, these nomograms are mostly based on factors from a nal pathology report after surgery, such as LVI, pathologic tumour size and the size of SLN metastasis, among others. Thus, when indicated, patients are expected to undergo further axillary dissection on a separate schedule. In this study, we retrospectively analysed patient data from the Asan Medical Center (AMC) to develop a nomogram that could help predict the possibility of NSLNM based on the clinical information available before a planned surgery.

Patients
We reviewed data from patients who had breast surgery with ALND after NAC between 2005 and 2019.
Patients with one to three metastasis-positive sentinel node(s) treated with standard axillary procedure were included in the study. According to the oncologists at the AMC, all included patients received a full course of standard neoadjuvant therapy. As a result of ALND, we could identify patients with or without residual nodal disease based on the nal pathology reports. We excluded patients who had bilateral breast cancer, in ammatory breast cancer or had distant metastasis at the time of presentation, as well as those who had more than four sentinel nodes positive for metastasis and went straight to ALND without SNB or SNB only. We also excluded patients whose NAC was incomplete due to intolerance or refusal. Finally, 558 patients were chosen for further analysis. The records of 384 patients treated between 2005 and 2016 were utilized to develop the prediction model, and the data of 174 patients treated from 2017 to 2019 were used to validate the generated prediction model. The data of the patients were reviewed for the total number of metastatic nodes on the nal pathology report, intraoperative frozen section biopsy result of sentinel node(s), presence or absence of additional metastatic non-sentinel nodes and number, tumour invasion depth, tumour biology, initial clinical stage before chemotherapy and radiology report of ultrasonogram or MRI of breast before and after chemotherapy.
This study was approved by the Institutional Review Board of the Asan Medical Center, Seoul, South Korea (20171341). Because the study was based on retrospective clinical data, informed consent was not required.
Preoperative chemotherapy and SNLB mapping method NAC was administered to a patient every 3 weeks, and a regimen was selected from among standard proposed regimens based on the clinical stage or tumour biology of a patient. Although standard regimens are constantly evolving, the oncologists at our institution generally followed the most recent NCCN guideline of the time. The surgery was performed 3-4 weeks after the completion of preoperative chemotherapy. We assessed a patient's response to NAC using either an ultrasonogram or an MRI of the breast before and after treatment. According to the Revised Response Evaluation Criteria in Solid Tumors (RECIST guideline, version 1.1), we de ned partial remission as a decrease of more than 30% in the sum of the longest diameters of the target lesions compared with baseline. In addition, we used the term complete remission when we discovered the disappearance of all target lesions. All tumours that did not meet the above criteria were classi ed as stable disease.
We used 99m Tc-sulphur colloid diluted in normal saline as a radiopharmaceutical agent with gamma probe detection (NeoProbe2000, US surgical, Norwalk, CT) for SLN identi cation. We injected the mapping agent periareolarly and massaged the breast for 5 minutes. Along with the most radioactive nodes, clinically enlarged, rm or palpable axillary lymph nodes without an active gamma signal were excised and counted as part of the total number of SLNs.

Statistical analysis
To identify signi cant factors that predict the possibility of residual disease in non-sentinel nodes, we rst included the following parameters in the univariate analysis: age at diagnosis, tumour grade, hormone receptor score, HER2 status, classi cation into four subtypes (HR+/HER2−, HR+/HER2+, HR-HER2+, HR−/HER2−), Ki-67, clinical T stage and N stage before NAC and its degree of response to therapy, number of metastatic sentinel nodes, total number of submitted sentinel nodes for frozen section biopsy and the greatest tumour invasion depth of the sentinel nodes.
We used the data from 384 patients treated between 2005 and 2016 for the development of the prediction model and the data from 174 patients treated from 2017 to 2019 for validation. In the development set, univariate assessment of these factors was performed using a logistic regression model. A multivariable logistic regression model was used to further analyse and generate a prediction model for the possibility of residual nodal disease after NAC when one to three sentinel nodes were positive intraoperatively. The predictors for the multivariable model were selected using backward elimination in more than half of the 1000 bootstrap resamples. The nal model was estimated using penalised maximum likelihood and was presented as a nomogram. The discrimination ability of the nomogram was assessed by using area under the receiver operating characteristic curve (AUC). The calibration ability was assessed using the calibration plot and the Hosmer-Lemeshow test. We performed an internal validation with bootstrapping with 1000 iterations, calculating optimism-corrected AUC (C statistics). In validation set, the discrimination and calibration abilities were also evaluated. All tests were two-sided, and p-value of less than 0.05 was considered statistically signi cant. Statistical analysis was conducted with SPSS statistics version 23.0 (IBM Corp., Armonk, USA) and R (version 3.6.1; R Foundation for Statistical Computing, https://www.Rproject.org). Table 1 shows the demographics of the 558 patients whose clinical data was used to develop the nomogram. In total, the majority of patients were under the age of 50 (63.6%), had a single SLN metastasis at the time of surgery (50.9%), had three to ve frozen biopsy sent for pathologic con rmation of SLN status (66.1%), were in clinical T stage 2 (63.4%). Furthermore, a substantial portion of our patients had N1 stage disease both before (74.4%) and after (75.1%) NAC. The majority of tumours were low grade (81%), oestrogen receptor positive (80.6%), progesterone receptor positive (66.5%), HER2 negative (78.3%), biological subtype of hormone receptor positive and HER2 negative (78.3%). As shown in the Table 1, the baseline characteristics of patients with or without residual nodal disease differed signi cantly. The residual nodal disease group had more oestrogen receptor-positive patients (85.7%, p = 0.007) and more progesterone receptor-positive patients (72.2%, p = 0.011) than the no nodal residual disease group. In addition, patients with nodal residual disease had a higher initial N stage (p < 0.001), a higher number of positive SLN (p < 0.001), a higher pathologic T stage (p = 0.001) and a higher N stage (p < 0.001) than those without nodal residual disease.  Table 1 shows the comparison of detailed clinicopathologic factors between the baseline and validation groups based on residual nodal disease.

Factors for predicting NSLNM and the development of a nomogram
Our goal was to develop a tool that could link routinely measured clinical factors to the actual probability of NSLNM during surgery. Table 3 displays the detailed analysis results. In the univariate analysis, the number of positive SLN(s), number of frozen nodes, tumour grade, ER and PR positivity, clinical N stage and biological subtype were found to be signi cantly associated with the possibility of residual nodal disease. The odds ratio increased in parallel with the number of positive SLN(s) and the initial N stage. In a multivariate stepwise logistic regression analysis, the number of metastatic SLN(s), number of frozen SLN(s), PR positivity and preoperative clinical N stage were found to be the independent predictors of NSLNM. These four variables were included to develop the nomogram (Fig. 1.)

Discussion
Despite the fact that traditional axillary surgery remains one of the standard management options, optimal treatment of the axilla has been an evolving area aimed at reducing its related morbidity. According to the most recent NCCN guideline for invasive breast cancer, we should perform standard ALND or SLNB in selected cases when nodes are clinically negative after NAC and FNA or core biopsy is positive prior to preoperative chemotherapy [4]. ALND, as a means for achieving local disease control, carries an indisputable and often unacceptable risk of complications such as seroma, infection and lymphedema [8].
However, previous studies that analysed patients with primary breast cancer or who underwent NAC found that 40-60% of those who underwent ALND had no residual axillary disease [5][6][7]. In our study population, 56% of patients had no residual nodal metastasis after completion of axillary dissection, which was performed due to a positive sentinel nodal biopsy result at the time of surgery. As a result, a substantial portion of patients may have been subjected to the signi cant morbidity of extensive axillary surgery without receiving any clinical bene t. Thus, the goal of our study was to nd a tool that allows a surgeon to be more selective in choosing a subgroup of patients who may be spared from the possible morbidity of ALND. Our nomogram is composed of four variables: number of metastatic SLN(s), number of frozen nodes, PR positivity and preoperative clinical N stage. These parameters were available before proceeding to a full ALND.
Several models have been proposed to predict the presence of NSLNM for breast cancer patients with or without NAC [6, [9][10][11][12][13]. One of the most widely used nomograms was developed by Van Zee et al., which included eight statistically signi cant variables of pathological size, LVI, method of detection, number of positive SLNs, multifocality and number of negative SLNs [13]. ER status and nuclear grade were included in the model but failed to show signi cant association with the likelihood of NSLNM. The overall discriminative ability of this nomogram, as measured by the ROC curve, was 0.76 for the retrospective population. The AUC value for the corresponding prospective population was 0.77. However, this model is only applicable to patients without NAC. Patients who were treated with NAC may need another version of the nomogram to accurately predict NSLNM. Moreover, the variables, such as pathological tumour size and LVI, may not always be available in a routine frozen section pathology during surgery. patients' data from different time windows. Aside from the difference in the number of patients included, the data involved 57 patients who were SLN negative but underwent ALND, whereas our study included only patients with one to three positive SLN(s) for analysis.
The aforementioned studies and this study share the goal of accurately predicting NSLNM, which may help a surgeon be more selective in nding full-extent ALND candidates. However, de-escalation of axillary surgery to eligible patients might raise concerns about possible residual metastatic nodal disease and the associated risk of tumour recurrence in the future. Nguyen et al. discovered a signi cant shift in the axillary surgery trend for clinical N1 patients treated with NAC, with SLN surgery becoming more common while ALND becoming less common [15]. Although de-escalation of axillary surgery after NAC has been an increasing trend, signi cant prospective data regarding disease recurrence and related survival are lacking [16].
A retrospective study compared the survival result between SLNB alone and full-extent ALND in patients with one to three positive sentinel nodes on intraoperative frozen biopsy after preoperative chemotherapy [17]. They were cautious about reducing the extent of axillary surgery, but they also demonstrated that limiting axillary surgery may be feasible in some selected patients with favourable tumour biology. Until further conclusive clinical data is published, we believe that a more cautious approach to ypN1 breast cancer patients is still appropriate, but it is worthwhile to try to be more selective in choosing eligible patients for reduced range of axillary management.
There are several limitations to this study. Given the single institutional, retrospective nature of this study, we acknowledge the potential existence of selection bias among eligible patients. There was observed heterogeneity in baseline patient characteristics between the test group and validation group. The baseline group had a higher proportion of clinical N0 patients, whereas the validation cohort had a higher proportion of clinical N2 patients. However, throughout the entire patient data analysis, we found signi cant differences in the baseline clinicopathologic factors between the cohort with non-sentinel node metastasis and the cohort without residual nodal disease. In addition, it is signi cant to show that this is the data that most accurately re ects real-world practice and demonstrates the recent trend of changes in surgical methods used by operators. Despite its different period of treatment time window, the resulting nomogram was validated only in a patient cohort from a single institution. External validation with su cient number of patients and patients with different background demographic data needs to be done to further validate the correlation of this proposed nomogram.

Conclusions
The ultimate goal would be to tailor appropriate axillary surgery based on each patients' disease status, so that only patients who are expected to bene t from ALND are subjected to the possible morbidity. Moreover, we hope to spare patients who may not gain signi cant bene t from the extensive procedure.
For patients with one to three positive SLN(s) after preoperative chemotherapy, this nomogram could provide clinically useful information to a surgeon about whether to proceed with further axillary dissection.
As a result, we may have an additional guiding tool to decide ALND intraoperatively, allowing a patient to avoid a separate surgery session of ALND.

Declarations
Funding This study was supported by the Asan Institute for Life Sciences, Asan Medical Center, Seoul, Korea (Grant Number 2020-0037). Figure 1