The contribution of axillary lymph node volume to recurrence-free survival status in breast cancer patients with sub-stratication by molecular subtypes and pathological complete response

Purpose This study sought to examine the contribution of axillary lymph node (LN) volume to recurrence-free survival (RFS) in breast cancer patients with sub-stratication by molecular subtypes, and full or nodal PCR. Methods The largest LN volumes per patient at pre-neoadjuvant chemotherapy on standard clinical breast 1.5-Tesla MRI, 3 molecular subtypes, full, breast, and nodal PCR, and 10-year RFS were tabulated (N = 110 patients from MRIs of I-SPY-1 TRIAL). A volume threshold of two standard deviations was used to categorize large versus small LNs for sub stratication. In addition, “normal” node volumes were determined from a different cohort of 218 axillary LNs. Results LN volume (4.07 ± 5.45 cm 3 ) were signicantly larger than normal axillary LN volumes (0.646 ± 0.657 cm 3 , P = 10 − 16 ). Full and nodal pathologic complete response (PCR) was not dependent on pre-neoadjuvant chemotherapy nodal volume (P > .05). The HR+/HER2– group had smaller axillary LN volumes than the HER2 + and triple-negative groups (P < .05). Survival was not dependent on pre-treatment axillary LN volumes alone (P = .29). However, when substratied by PCR, the large LN group with full (P = .011) or nodal PCR (P = .0026) both showed better recurrence-free survival than the small LN group. There was signicant difference in RFS when the small node group was separated by the 3 molecular subtypes (P = .036) but not the large node group (P = .97). Conclusions This study found an interaction of axillary lymph node volume, pathological complete responses, and molecular subtypes that inform recurrence-free survival status. Improved characterization of the axillary lymph nodes has the potential to improve the management of breast cancer patients.


Introduction
Recurrence-free survival (RFS) is a common measure of treatment outcome in breast cancer patients.
Pathologic complete response (PCR) is pathologic determination of breast tumor and axillary lymph nodes (LNs) response to neoadjuvant chemotherapy (NAC) and is possibly a surrogate marker for RFS.
However, the ability of PCR to predict RFS in breast cancer is controversial. [1][2][3][4][5][6] A few studies have shown that breast MRI tumor volume is a moderate predictor of PCR with an area under the receiver-operative curve (AUC) of 0.7-0.8 and RFS AUC of 0.6-0.8. 6,7 Moreover, patients with different molecular subtypes also respond differently to NAC, affecting PCR pro les and RFS. 8 For example, patients with hormonal receptor positive/human epidermal growth factor receptor 2 negative (HR+/HER2-) molecular subtypes are known to be less responsive to NAC compared to patients with a triple-negative (TN) or HER2 + molecular subtype. 8 PCR, tumor volume, and molecular subtype seem to have a complex relationship with RFS with no single parameter able to accurately predict RFS.
Locally advanced breast cancer, de ned as breast cancer that has spread beyond the breast to the skin, chest wall, or axillary LNs but not to other organs, is now considered a manageable disease. However, breast cancer is more concerning when metastasizes. 9 Most of the lymphatic uid in the breasts passes through the axillary LNs which are the major conduits by which breast cancer cells metastasize. Axillary LN status may thus have prognostic value in determining RFS. Indeed, axillary LN involvement is known to increase the risk of recurrence in breast cancer. 10,11 Compared to a LN biopsy which is limited to a small region of a node and/or a few nodes, MRI has the potential to visualize most axillary LNs noninvasively in 3-dimensions and in situ. With current standards, MRI of axillary LNs is challenging because axillary LNs are small and often not in the eld of view. When visualized, most have low sensitivity as current detectors and standard breast MRI protocols are not currently optimized for axillary LN imaging.
Although LN involvement by pathology has been associated with poor RFS in breast and other cancers, the contribution of axillary LN volume alone as well as its association with PCR, and receptor status in determining RFS in breast cancer has not been adequately investigated. 12,13 This study sought to examine the contribution of axillary LN volume on RFS in breast cancer patients with sub-strati cation by molecular subtypes, and full or nodal PCR. We tested the hypothesis that large pre-treatment axillary LN volume is associated with poor RFS in breast cancer patients. We also sought to investigate the interaction amongst axillary LN volume, PCR, and molecular subtypes that inform RFS.

Data Sources and Study De nitions
The data used in this study was obtained from the American College of Radiology Imaging Network (ACRIN) 6657 study with Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and molecular Analysis (I-SPY 1 TRIAL). 6 anthracycline-cyclophosphamide regimen with or without taxane. I-SPY 1 TRIAL MR data was acquired across multiple sites on 1.5-T MRI scanners.
In this study, large axillary LNs were de ned as any axillary LNs that were two standard deviations larger than the average volume of normal nodes. For volume quantitation, the single largest axillary LN was contoured from each of the 110 patients from the I-SPY 1 TRIAL, as the largest axillary LN was the most suspicious of metastasis in all patients. Refer to Fig. 1 for a owchart of the study design. Normal node volume was obtained using MRIs of breast cancer patients at Stony Brook University Hospital on a 1.5-T GE scanner from 01/01/2010 to 07/30/2018 (I-SPY 1 TRIAL data had no contralateral breast or axillary LN images). The breast and axillary LNs contralateral to the diseased breast were not diagnosed with cancer and assumed to be normal. 218 normal contralateral axillary LNs were collected from 71 breast cancer patients. 192 of the 218 normal contralateral axillary LNs have been previously reported. 16 The prior article studied whether NAC-induced change of axillary LN size could be used as a predictor of PCR Page 4/17 using AUC analyses while this study considered if pre-NAC axillary LN sizes can differentiate patient RFS with Kaplan-Meier estimates.

Study Design
The following 3 parameters were obtained from the I-SPY 1 TRIAL clinical and outcome dataset: a) 3 different molecular subtype categories, b) full, breast, and nodal PCR status, and c) RFS at 10 years. Due to some missing nodal volume, molecular subtype, PCR, and RFS data, analysis of each variable was done with varying sample sizes as shown in Tables 1 and 2. The study was performed in accordance with the research guidelines approved by the Stony Brook University Institutional Review Board.   1 Data are mean ± standard deviation. 2 Data in parenthesis are range.
Hormone receptor positivity and HER2 expression were determined from pre-NAC biopsy of the primary breast tumor by immunohistochemistry, uorescent in situ hybridization, and Allred score. 3 following molecular subtypes were used for analysis: HR+/HER2-, HER2+, and triple negative (TN). HR+/HER2are patients that are negative for HER2 and positive for either estrogen and/or progesterone receptor. HER2 + are positive for HER2 regardless of estrogen and/or progesterone positivity status. TN patients are negative for HER2, estrogen and progesterone receptors.
Breast PCR (ypT0/is) was de ned by surgical pathology as the absence of invasive cancer in breast, irrespective of remaining in situ cancer in the primary tumor. Nodal PCR (ypN0) was de ned by surgical pathology as the absence of invasive cancer in the axillary LNs. PCR or full PCR (ypT0/is, ypN0) is a combination of breast and nodal PCR again de ned by surgical pathology as no residual invasive disease in either breast or axillary LNs after NAC. Patients who achieved PCR will be referred to as complete responders and patients who did not achieve PCR will be referred to as non-complete or partial responders.
RFS time was documented in months from the date of chemotherapy initiation and listed as having a recurrence or remaining recurrence-free (or censored) at 10 years.
For axillary LN volumes, all visible axillary LNs on the pre-NAC MRI rst post-contrast image (around 2 minutes post-contrast) were manually segmented on ITK-SNAP 3.8.0 (http://itksnap.org) by RC, JK, and VT under the guidance of expert breast radiologists CB, RP, and PF each with more than 20 years of experience. The entire lymph node was contoured to determine the volume, including the fatty hilum. For consistency, only the largest axillary LN prior to NAC in each patient were used for volume analysis.
Retrospective review under guidance of breast radiologists con rmed that essentially all (95%) of the largest axillary LNs were the most suspicious of malignancy due to characteristics such as large size, loss of fatty hilum, and clustering.  18 There was no signi cant difference in survival for patients who achieved PCR and patients who achieved nodal PCR (P > .05). There was also no signi cant difference in survival for patients who did not achieve PCR and patients who did not achieve nodal PCR (P > 0.05). Survival was similar across all 3 molecular subtypes (P = .21, Fig. 1B).
Patients were divided into two groups using a threshold of two standard deviation from mean of normal nodes (1.96 cm 3 ). With this threshold, 52.7% of patients were grouped into the large axillary LN group and the rest were grouped into the small axillary LN group. There was no signi cant difference in RFS between large and small axillary LNs pre-NAC (P = .29, Fig. 1C). The characteristics of the small axillary LN and the large axillary LN group are seen in Table 1. Between the two subgroups, signi cant differences were seen in the racial composition and molecular subtypes.

Sub-strati cation by PCR and Node Size
Subgroup analyses were done with full PCR status, nodal PCR status, and molecular subtypes. In the small axillary LN group, there was no difference whether the patients achieved full PCR (P = .46; Fig. 2A) or nodal PCR (P = .51; Fig. 2B). However, in the large axillary LN group, patients who achieved PCR had an 88.2% RFS rate compared to a 48.7% RFS rate of patients who did not (P = .011; Fig. 3C) and patients who achieved nodal PCR had an 81.5% survival rate compared to 41.4% survival rate of patients who did not have a PCR (p = .0026; Fig. 2D).
Within the nodal non-responder and responder groups, there was no difference in RFS whether there were small or large nodes with P values of .1 and .56, respectively ( Fig. 3A and B).
Kaplan-Meier curves were analyzed for small and large axillary LNs with strati cation into molecular subtypes. For small axillary LNs, log-rank test found that there is signi cant difference in between the three molecular subtypes (P = .036; Fig. 4A), where HR+/HER2-had a signi cantly better survival outcome than the HER2 + group (P = .017) but not the TN group (P = .70). There was no difference between the HER2 + and TN groups (P = .11). For large axillary LNs, similar log-rank test did not nd any difference in RFS for the three molecular subtypes (P = .97; Fig. 4B).
Within each molecular subtype, pre-treatment axillary LN size did not have any effect on 10-year RFS with P values of .068, .25, and .34 for the HR+/HER2-, HER2+, and TN subgroups, respectively (Fig. 5A, B, and C).

Discussion
Axillary LN volumes (average = 4.07 cm 3 ) of the I-SPY 1 TRIAL breast cancer patients were signi cantly larger than our normal axillary LN volumes (average = 0.646 cm 3 ). There are no similar studies with which to quantitatively compare to our study but our ndings are in agreement with the notion that axillary LNs < 1.5 cm diameter are generally considered normal-appearing. 19  There are several limitations in this study. A limitation is the small sample sizes after sub-strati cation into axillary LN volumes, PCR status, and molecular subtypes. These ndings need to be replicated on a large sample sizes to be generalizable. Another limitation was that there was no pathological con rmation of the diseased nodes. It is challenging to identify the same nodes on MRI that was biopsied. Nonetheless, the largest axillary LNs in this study were con rmed to be the most suspicious of malignancy due to morphological features, such as loss of fatty hilum, increased size, and/or thickening of cortex by our expert radiologists. Additionally, although radiologists tend to measure long and short diameter, we did not make this measurement because volumetric measurements were more reliable especially for small size objects. Future studies to improving MRI spatial resolution and contrast of axillary LNs and applying texture analysis and arti cial intelligence to the axillary LNs, among others, could yield additional clinically useful information. Patients with different molecular biomarker subtypes usually received different adjuvant treatments, e.g. hormonal therapy in HR + patients, and additional HER2 targeting therapy for maintenance in HER2 + patients. These could affect the RFS. However, the sample size was insu cient to further divide based on post NAC treatment.
Conclusions LN = lymph node, AUC = area under the receiver-operator curve, CI = con dence interval, HER2 = human epidermal growth factor receptor 2, HR = hormone receptor, NAC = neoadjuvant chemotherapy, PCR = pathologic complete response, RFS = recurrence-free survival, TN = triple negative Declarations Ethical Approval and Consent to participate Ethical approval is not necessary as this is a publicly available, deidenti ed dataset. Figure 1 Flowchart of study design and lymph node grouping. Numeric values are shown as mean ± standard deviation. FOV: eld-of-view; RFS: recurrence-free survival.

Figure 2
Kaplan-Meier recurrence-free survival estimates for PCR & nodal PCR (A), molecular subtype (B), and volume (C) subgroups. The log-rank test P value is shown for B and C. Shaded areas signify 95% con dence interval thus non-overlapping curves signify P < .05. PCR: pathologic complete response; TN: triple negative. . The log-rank test P value is shown. A volume threshold of 1.96 cm3 was used. Shaded areas signify 95% con dence interval thus non-overlapping curves signify P < .05. PCR: pathologic complete response.

Figure 4
Kaplan-Meier recurrence-free survival estimates for nodal non-PCR (A) and nodal PCR (B) subgroups.
The log-rank test P value is shown. Shaded areas signify 95% con dence interval thus non-overlapping curves signify P < .05. PCR: pathologic complete response.